Hi all,
Here's v2 series to improve resctrl selftests with generalized test framework and rewritten CAT test. In contrast to v1, this version does not include L2 CAT test because it needs further work. In general, I feel that v2 is in much better shape than v1 because I ended up addressing a few small things beyond what came up during v1 review.
The series contains following improvements:
- Excludes shareable bits from CAT test allocation to avoid interference - Replaces file "sink" with a volatile variable - Alters read pattern to defeat HW prefetcher optimizations - Rewrites CAT test to make the CAT test reliable and truly measure if CAT is working or not - Introduces generalized test framework making easier to add new tests - Lots of other cleanups & refactoring
This series have been tested across a large number of systems from different generations.
v2: - Postpone adding L2 CAT test as more investigations are necessary - Add patch to remove ctrlc_handler() from wrong place - Improvements to changelogs - Function comments improvements & comment cleanups - Move some parts of the changes into more logical patch - If checks: buf == NULL -> !buf - Variable naming: - p -> buf - cbm_mask_path -> cbm_path - Function naming: - get_cbm_mask() -> get_full_cbm() - cache_size() -> cache_portion_size() - Use PATH_MAX - Improved cache_portion_size() parameter names - int count -> unsigned int - Pass filename to measurement taking functions instead of resctrl_val_param - !lines ? : reversal - Removed bogus static from function local variable - Open perf fd only once, reset & enable in the innermost test loop - Add perf fd ioctl() error handling - Add patch to change compiler optimization prevention "sink" from file to volatile variable - Remove cpu_no and resource (the latter was added in v1) members from resctrl_val_param (pass uparams and test where those are needed) - Removed ARRAY_SIZE() macro - Add patch to rename "resource_id" to "domain_id"
Ilpo Järvinen (26): selftests/resctrl: Don't use ctrlc_handler() outside signal handling selftests/resctrl: Split fill_buf to allow tests finer-grained control selftests/resctrl: Refactor fill_buf functions selftests/resctrl: Refactor get_cbm_mask() and rename to get_full_cbm() selftests/resctrl: Mark get_cache_size() cache_type const selftests/resctrl: Create cache_portion_size() helper selftests/resctrl: Exclude shareable bits from schemata in CAT test selftests/resctrl: Split measure_cache_vals() selftests/resctrl: Split show_cache_info() to test specific and generic parts selftests/resctrl: Remove unnecessary __u64 -> unsigned long conversion selftests/resctrl: Remove nested calls in perf event handling selftests/resctrl: Consolidate naming of perf event related things selftests/resctrl: Improve perf init selftests/resctrl: Convert perf related globals to locals selftests/resctrl: Move cat_val() to cat_test.c and rename to cat_test() selftests/resctrl: Open perf fd before start & add error handling selftests/resctrl: Replace file write with volatile variable selftests/resctrl: Read in less obvious order to defeat prefetch optimizations selftests/resctrl: Rewrite Cache Allocation Technology (CAT) test selftests/resctrl: Create struct for input parameters selftests/resctrl: Introduce generalized test framework selftests/resctrl: Pass write_schemata() resource instead of test name selftests/resctrl: Add helper to convert L2/3 to integer selftests/resctrl: Rename resource ID to domain ID selftests/resctrl: Get domain id from cache id selftests/resctrl: Add test groups and name L3 CAT test L3_CAT
tools/testing/selftests/resctrl/cache.c | 274 +++++---------- tools/testing/selftests/resctrl/cat_test.c | 328 +++++++++++------- tools/testing/selftests/resctrl/cmt_test.c | 76 +++- tools/testing/selftests/resctrl/fill_buf.c | 132 ++++--- tools/testing/selftests/resctrl/mba_test.c | 26 +- tools/testing/selftests/resctrl/mbm_test.c | 28 +- tools/testing/selftests/resctrl/resctrl.h | 117 +++++-- .../testing/selftests/resctrl/resctrl_tests.c | 205 +++++------ tools/testing/selftests/resctrl/resctrl_val.c | 56 +-- tools/testing/selftests/resctrl/resctrlfs.c | 246 ++++++++----- 10 files changed, 821 insertions(+), 667 deletions(-)
perf_event_open_llc_miss() calls ctrlc_handler() to cleanup if perf_event_open() returns an error. Those cleanups, however, are not the responsability of perf_event_open_llc_miss() and it thus interferes unnecessarily with the usual cleanup pattern. Worse yet, ctrlc_handler() calls exit() in the end preventing the ordinary cleanup done in the calling function from executing.
ctrlc_handler() should only be used as a signal handler, not during normal error handling.
Remove call to ctrlc_handler() from perf_event_open_llc_miss(). As unmounting resctrlfs and test cleanup are already handled properly by error rollbacks in the calling functions, no other changes are necessary.
Suggested-by: Reinette Chatre reinette.chatre@intel.com Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/cache.c | 1 - 1 file changed, 1 deletion(-)
diff --git a/tools/testing/selftests/resctrl/cache.c b/tools/testing/selftests/resctrl/cache.c index bcbca356d56a..8aa6d67db978 100644 --- a/tools/testing/selftests/resctrl/cache.c +++ b/tools/testing/selftests/resctrl/cache.c @@ -41,7 +41,6 @@ static int perf_event_open_llc_miss(pid_t pid, int cpu_no) PERF_FLAG_FD_CLOEXEC); if (fd_lm == -1) { perror("Error opening leader"); - ctrlc_handler(0, NULL, NULL); return -1; }
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
perf_event_open_llc_miss() calls ctrlc_handler() to cleanup if perf_event_open() returns an error. Those cleanups, however, are not the responsability of perf_event_open_llc_miss() and it thus interferes
responsibility
unnecessarily with the usual cleanup pattern. Worse yet, ctrlc_handler() calls exit() in the end preventing the ordinary cleanup done in the calling function from executing.
ctrlc_handler() should only be used as a signal handler, not during normal error handling.
Remove call to ctrlc_handler() from perf_event_open_llc_miss(). As unmounting resctrlfs and test cleanup are already handled properly by error rollbacks in the calling functions, no other changes are necessary.
Suggested-by: Reinette Chatre reinette.chatre@intel.com Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
Reviewed-by: Reinette Chatre reinette.chatre@intel.com
Reinette
MBM, MBA and CMT test cases call run_fill_buf() that in turn calls fill_cache() to alloc and loop indefinitely around the buffer. This binds buffer allocation and running the benchmark into a single bundle so that a selftest cannot allocate a buffer once and reuse it. CAT test doesn't want to loop around the buffer continuously and after rewrite it needs the ability to allocate the buffer separately.
Split buffer allocation out of fill_cache() into alloc_buffer(). This change is part of preparation for the new CAT test that allocates a buffer and does multiple passes over the same buffer (but not in an infinite loop).
Co-developed-by: Fenghua Yu fenghua.yu@intel.com Signed-off-by: Fenghua Yu fenghua.yu@intel.com Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/fill_buf.c | 26 +++++++++++++--------- 1 file changed, 15 insertions(+), 11 deletions(-)
diff --git a/tools/testing/selftests/resctrl/fill_buf.c b/tools/testing/selftests/resctrl/fill_buf.c index 0d425f26583a..6f32f44128e1 100644 --- a/tools/testing/selftests/resctrl/fill_buf.c +++ b/tools/testing/selftests/resctrl/fill_buf.c @@ -135,33 +135,37 @@ static int fill_cache_write(unsigned char *buf, size_t buf_size, bool once) return 0; }
-static int fill_cache(size_t buf_size, int memflush, int op, bool once) +static unsigned char *alloc_buffer(size_t buf_size, int memflush) { unsigned char *buf; - int ret;
buf = malloc_and_init_memory(buf_size); if (!buf) - return -1; + return NULL;
/* Flush the memory before using to avoid "cache hot pages" effect */ if (memflush) mem_flush(buf, buf_size);
+ return buf; +} + +static int fill_cache(size_t buf_size, int memflush, int op, bool once) +{ + unsigned char *buf; + int ret; + + buf = alloc_buffer(buf_size, memflush); + if (!buf) + return -1; + if (op == 0) ret = fill_cache_read(buf, buf_size, once); else ret = fill_cache_write(buf, buf_size, once); - free(buf);
- if (ret) { - printf("\n Error in fill cache read/write...\n"); - return -1; - } - - - return 0; + return ret; }
int run_fill_buf(size_t span, int memflush, int op, bool once)
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
MBM, MBA and CMT test cases call run_fill_buf() that in turn calls fill_cache() to alloc and loop indefinitely around the buffer. This binds buffer allocation and running the benchmark into a single bundle so that a selftest cannot allocate a buffer once and reuse it. CAT test doesn't want to loop around the buffer continuously and after rewrite it needs the ability to allocate the buffer separately.
Split buffer allocation out of fill_cache() into alloc_buffer(). This change is part of preparation for the new CAT test that allocates a buffer and does multiple passes over the same buffer (but not in an infinite loop).
Co-developed-by: Fenghua Yu fenghua.yu@intel.com Signed-off-by: Fenghua Yu fenghua.yu@intel.com Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
Could you please list the changes to a patch in this area instead of lumping all in the cover letter? Without this it is not clear what, if anything, changed in the new version.
tools/testing/selftests/resctrl/fill_buf.c | 26 +++++++++++++--------- 1 file changed, 15 insertions(+), 11 deletions(-)
diff --git a/tools/testing/selftests/resctrl/fill_buf.c b/tools/testing/selftests/resctrl/fill_buf.c index 0d425f26583a..6f32f44128e1 100644 --- a/tools/testing/selftests/resctrl/fill_buf.c +++ b/tools/testing/selftests/resctrl/fill_buf.c @@ -135,33 +135,37 @@ static int fill_cache_write(unsigned char *buf, size_t buf_size, bool once) return 0; } -static int fill_cache(size_t buf_size, int memflush, int op, bool once) +static unsigned char *alloc_buffer(size_t buf_size, int memflush) { unsigned char *buf;
- int ret;
buf = malloc_and_init_memory(buf_size); if (!buf)
return -1;
return NULL;/* Flush the memory before using to avoid "cache hot pages" effect */ if (memflush) mem_flush(buf, buf_size);
- return buf;
+}
+static int fill_cache(size_t buf_size, int memflush, int op, bool once) +{
- unsigned char *buf;
- int ret;
- buf = alloc_buffer(buf_size, memflush);
- if (!buf)
return -1;- if (op == 0) ret = fill_cache_read(buf, buf_size, once); else ret = fill_cache_write(buf, buf_size, once);
- free(buf);
- if (ret) {
printf("\n Error in fill cache read/write...\n");return -1;- }
The changelog does not motivate the removal of this error message. It seems ok at this point since the only failing functions already print their own error message. Without a motivation of this change it is not clear if it is actually intended.
In any case, this looks good.
Reviewed-by: Reinette Chatre reinette.chatre@intel.com
Reinette
On Tue, 28 Nov 2023, Reinette Chatre wrote:
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
MBM, MBA and CMT test cases call run_fill_buf() that in turn calls fill_cache() to alloc and loop indefinitely around the buffer. This binds buffer allocation and running the benchmark into a single bundle so that a selftest cannot allocate a buffer once and reuse it. CAT test doesn't want to loop around the buffer continuously and after rewrite it needs the ability to allocate the buffer separately.
Split buffer allocation out of fill_cache() into alloc_buffer(). This change is part of preparation for the new CAT test that allocates a buffer and does multiple passes over the same buffer (but not in an infinite loop).
Co-developed-by: Fenghua Yu fenghua.yu@intel.com Signed-off-by: Fenghua Yu fenghua.yu@intel.com Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
Could you please list the changes to a patch in this area instead of lumping all in the cover letter? Without this it is not clear what, if anything, changed in the new version.
Okay, I'll try to keep them per patch.
+static int fill_cache(size_t buf_size, int memflush, int op, bool once) +{
- unsigned char *buf;
- int ret;
- buf = alloc_buffer(buf_size, memflush);
- if (!buf)
return -1;- if (op == 0) ret = fill_cache_read(buf, buf_size, once); else ret = fill_cache_write(buf, buf_size, once);
- free(buf);
- if (ret) {
printf("\n Error in fill cache read/write...\n");return -1;- }
The changelog does not motivate the removal of this error message. It seems ok at this point since the only failing functions already print their own error message. Without a motivation of this change it is not clear if it is actually intended.
Hi,
I don't have recollection of how it ended up into this patch as it's so long time ago already... But in any case, it naturally seemed to fit into the next patch that removes the extra call level so I moved the removal of the duplicated error printout to that patch instead.
In any case, this looks good.
Reviewed-by: Reinette Chatre reinette.chatre@intel.com
There are unnecessary nested calls in fill_buf.c: - run_fill_buf() calls fill_cache() - alloc_buffer() calls malloc_and_init_memory()
Simplify the code flow and remove those unnecessary call levels by moving the called code inside the calling function.
Resolve the difference in run_fill_buf() and fill_cache() parameter name into 'buf_size' which is more descriptive than 'span'. Also, while moving the allocation related code, rename 'p' into 'buf' to be consistent in naming the variables.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/fill_buf.c | 54 +++++++--------------- tools/testing/selftests/resctrl/resctrl.h | 2 +- 2 files changed, 18 insertions(+), 38 deletions(-)
diff --git a/tools/testing/selftests/resctrl/fill_buf.c b/tools/testing/selftests/resctrl/fill_buf.c index 6f32f44128e1..b9303a9d819b 100644 --- a/tools/testing/selftests/resctrl/fill_buf.c +++ b/tools/testing/selftests/resctrl/fill_buf.c @@ -51,29 +51,6 @@ static void mem_flush(unsigned char *buf, size_t buf_size) sb(); }
-static void *malloc_and_init_memory(size_t buf_size) -{ - void *p = NULL; - uint64_t *p64; - size_t s64; - int ret; - - ret = posix_memalign(&p, PAGE_SIZE, buf_size); - if (ret < 0) - return NULL; - - p64 = (uint64_t *)p; - s64 = buf_size / sizeof(uint64_t); - - while (s64 > 0) { - *p64 = (uint64_t)rand(); - p64 += (CL_SIZE / sizeof(uint64_t)); - s64 -= (CL_SIZE / sizeof(uint64_t)); - } - - return p; -} - static int fill_one_span_read(unsigned char *buf, size_t buf_size) { unsigned char *end_ptr = buf + buf_size; @@ -137,12 +114,25 @@ static int fill_cache_write(unsigned char *buf, size_t buf_size, bool once)
static unsigned char *alloc_buffer(size_t buf_size, int memflush) { - unsigned char *buf; + void *buf = NULL; + uint64_t *p64; + size_t s64; + int ret;
- buf = malloc_and_init_memory(buf_size); - if (!buf) + ret = posix_memalign(&buf, PAGE_SIZE, buf_size); + if (ret < 0) return NULL;
+ /* Initialize the buffer */ + p64 = buf; + s64 = buf_size / sizeof(uint64_t); + + while (s64 > 0) { + *p64 = (uint64_t)rand(); + p64 += (CL_SIZE / sizeof(uint64_t)); + s64 -= (CL_SIZE / sizeof(uint64_t)); + } + /* Flush the memory before using to avoid "cache hot pages" effect */ if (memflush) mem_flush(buf, buf_size); @@ -150,7 +140,7 @@ static unsigned char *alloc_buffer(size_t buf_size, int memflush) return buf; }
-static int fill_cache(size_t buf_size, int memflush, int op, bool once) +int run_fill_buf(size_t buf_size, int memflush, int op, bool once) { unsigned char *buf; int ret; @@ -164,16 +154,6 @@ static int fill_cache(size_t buf_size, int memflush, int op, bool once) else ret = fill_cache_write(buf, buf_size, once); free(buf); - - return ret; -} - -int run_fill_buf(size_t span, int memflush, int op, bool once) -{ - size_t cache_size = span; - int ret; - - ret = fill_cache(cache_size, memflush, op, once); if (ret) { printf("\n Error in fill cache\n"); return -1; diff --git a/tools/testing/selftests/resctrl/resctrl.h b/tools/testing/selftests/resctrl/resctrl.h index a33f414f6019..08b95b5a4949 100644 --- a/tools/testing/selftests/resctrl/resctrl.h +++ b/tools/testing/selftests/resctrl/resctrl.h @@ -92,7 +92,7 @@ int write_bm_pid_to_resctrl(pid_t bm_pid, char *ctrlgrp, char *mongrp, char *resctrl_val); int perf_event_open(struct perf_event_attr *hw_event, pid_t pid, int cpu, int group_fd, unsigned long flags); -int run_fill_buf(size_t span, int memflush, int op, bool once); +int run_fill_buf(size_t buf_size, int memflush, int op, bool once); int resctrl_val(const char * const *benchmark_cmd, struct resctrl_val_param *param); int mbm_bw_change(int cpu_no, const char * const *benchmark_cmd); void tests_cleanup(void);
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
There are unnecessary nested calls in fill_buf.c:
- run_fill_buf() calls fill_cache()
- alloc_buffer() calls malloc_and_init_memory()
Simplify the code flow and remove those unnecessary call levels by moving the called code inside the calling function.
Resolve the difference in run_fill_buf() and fill_cache() parameter name into 'buf_size' which is more descriptive than 'span'. Also, while moving the allocation related code, rename 'p' into 'buf' to be consistent in naming the variables.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
Reviewed-by: Reinette Chatre reinette.chatre@intel.com
Reinette
Callers of get_cbm_mask() are required to pass a string into which the capacity bitmask (CBM) is read. Neither CAT nor CMT tests need the bitmask as string but just convert it into an unsigned long value.
Another limitation is that the bit mask reader can only read .../cbm_mask files.
Generalize the bit mask reading function into get_bit_mask() such that it can be used to handle other files besides the .../cbm_mask and handles the unsigned long conversion within get_bit_mask() using fscanf(). Change get_cbm_mask() to use get_bit_mask() and rename it to get_full_cbm() to better indicates what the function does.
Also mark cache_type const while at it and remove useless comments that are related to processing of CBM bits.
Co-developed-by: Fenghua Yu fenghua.yu@intel.com Signed-off-by: Fenghua Yu fenghua.yu@intel.com Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/cat_test.c | 7 +-- tools/testing/selftests/resctrl/cmt_test.c | 5 +- tools/testing/selftests/resctrl/resctrl.h | 2 +- tools/testing/selftests/resctrl/resctrlfs.c | 51 +++++++++++++++------ 4 files changed, 41 insertions(+), 24 deletions(-)
diff --git a/tools/testing/selftests/resctrl/cat_test.c b/tools/testing/selftests/resctrl/cat_test.c index 224ba8544d8a..92f0f3ce90e5 100644 --- a/tools/testing/selftests/resctrl/cat_test.c +++ b/tools/testing/selftests/resctrl/cat_test.c @@ -93,25 +93,20 @@ int cat_perf_miss_val(int cpu_no, int n, char *cache_type) int ret, pipefd[2], sibling_cpu_no; unsigned long cache_size = 0; unsigned long long_mask; - char cbm_mask[256]; int count_of_bits; char pipe_message; size_t span;
- /* Get default cbm mask for L3/L2 cache */ - ret = get_cbm_mask(cache_type, cbm_mask); + ret = get_full_cbm(cache_type, &long_mask); if (ret) return ret;
- long_mask = strtoul(cbm_mask, NULL, 16); - /* Get L3/L2 cache size */ ret = get_cache_size(cpu_no, cache_type, &cache_size); if (ret) return ret; ksft_print_msg("Cache size :%lu\n", cache_size);
- /* Get max number of bits from default-cabm mask */ count_of_bits = count_bits(long_mask);
if (!n) diff --git a/tools/testing/selftests/resctrl/cmt_test.c b/tools/testing/selftests/resctrl/cmt_test.c index 50bdbce9fba9..d6ddacfb4416 100644 --- a/tools/testing/selftests/resctrl/cmt_test.c +++ b/tools/testing/selftests/resctrl/cmt_test.c @@ -75,17 +75,14 @@ int cmt_resctrl_val(int cpu_no, int n, const char * const *benchmark_cmd) unsigned long cache_size = 0; unsigned long long_mask; char *span_str = NULL; - char cbm_mask[256]; int count_of_bits; size_t span; int ret, i;
- ret = get_cbm_mask("L3", cbm_mask); + ret = get_full_cbm("L3", &long_mask); if (ret) return ret;
- long_mask = strtoul(cbm_mask, NULL, 16); - ret = get_cache_size(cpu_no, "L3", &cache_size); if (ret) return ret; diff --git a/tools/testing/selftests/resctrl/resctrl.h b/tools/testing/selftests/resctrl/resctrl.h index 08b95b5a4949..7a2b4da684ed 100644 --- a/tools/testing/selftests/resctrl/resctrl.h +++ b/tools/testing/selftests/resctrl/resctrl.h @@ -99,7 +99,7 @@ void tests_cleanup(void); void mbm_test_cleanup(void); int mba_schemata_change(int cpu_no, const char * const *benchmark_cmd); void mba_test_cleanup(void); -int get_cbm_mask(char *cache_type, char *cbm_mask); +int get_full_cbm(const char *cache_type, unsigned long *mask); int get_cache_size(int cpu_no, char *cache_type, unsigned long *cache_size); void ctrlc_handler(int signum, siginfo_t *info, void *ptr); int signal_handler_register(void); diff --git a/tools/testing/selftests/resctrl/resctrlfs.c b/tools/testing/selftests/resctrl/resctrlfs.c index 5ebd43683876..211fc7647659 100644 --- a/tools/testing/selftests/resctrl/resctrlfs.c +++ b/tools/testing/selftests/resctrl/resctrlfs.c @@ -196,30 +196,29 @@ int get_cache_size(int cpu_no, char *cache_type, unsigned long *cache_size) #define CORE_SIBLINGS_PATH "/sys/bus/cpu/devices/cpu"
/* - * get_cbm_mask - Get cbm mask for given cache - * @cache_type: Cache level L2/L3 - * @cbm_mask: cbm_mask returned as a string + * get_bit_mask - Get bit mask from given file + * @filename: File containing the mask + * @mask: The bit mask returned as unsigned long * * Return: = 0 on success, < 0 on failure. */ -int get_cbm_mask(char *cache_type, char *cbm_mask) +static int get_bit_mask(const char *filename, unsigned long *mask) { - char cbm_mask_path[1024]; FILE *fp;
- if (!cbm_mask) + if (!filename || !mask) return -1;
- sprintf(cbm_mask_path, "%s/%s/cbm_mask", INFO_PATH, cache_type); - - fp = fopen(cbm_mask_path, "r"); + fp = fopen(filename, "r"); if (!fp) { - perror("Failed to open cache level"); - + fprintf(stderr, "Failed to open bit mask file '%s': %s\n", + filename, strerror(errno)); return -1; } - if (fscanf(fp, "%s", cbm_mask) <= 0) { - perror("Could not get max cbm_mask"); + + if (fscanf(fp, "%lx", mask) <= 0) { + fprintf(stderr, "Could not read bit mask file '%s': %s\n", + filename, strerror(errno)); fclose(fp);
return -1; @@ -229,6 +228,32 @@ int get_cbm_mask(char *cache_type, char *cbm_mask) return 0; }
+/* + * get_full_cbm - Get full Cache Bit Mask (CBM) + * @cache_type: Cache type as "L2" or "L3" + * @mask: Full cache bit mask representing the maximal portion of cache + * available for allocation, returned as unsigned long. + * + * Return: = 0 on success, < 0 on failure. + */ +int get_full_cbm(const char *cache_type, unsigned long *mask) +{ + char cbm_path[PATH_MAX]; + int ret; + + if (!cache_type) + return -1; + + snprintf(cbm_path, sizeof(cbm_path), "%s/%s/cbm_mask", + INFO_PATH, cache_type); + + ret = get_bit_mask(cbm_path, mask); + if (ret) + return -1; + + return 0; +} + /* * get_core_sibling - Get sibling core id from the same socket for given CPU * @cpu_no: CPU number
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
Callers of get_cbm_mask() are required to pass a string into which the capacity bitmask (CBM) is read. Neither CAT nor CMT tests need the bitmask as string but just convert it into an unsigned long value.
Another limitation is that the bit mask reader can only read .../cbm_mask files.
Generalize the bit mask reading function into get_bit_mask() such that it can be used to handle other files besides the .../cbm_mask and handles the unsigned long conversion within get_bit_mask() using fscanf(). Change get_cbm_mask() to use get_bit_mask() and rename it to get_full_cbm() to better indicates what the function does.
"to better indicates" -> "to better indicate"
Also mark cache_type const while at it and remove useless comments that are related to processing of CBM bits.
Co-developed-by: Fenghua Yu fenghua.yu@intel.com Signed-off-by: Fenghua Yu fenghua.yu@intel.com Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
...
@@ -229,6 +228,32 @@ int get_cbm_mask(char *cache_type, char *cbm_mask) return 0; } +/*
- get_full_cbm - Get full Cache Bit Mask (CBM)
- @cache_type: Cache type as "L2" or "L3"
- @mask: Full cache bit mask representing the maximal portion of cache
available for allocation, returned as unsigned long.
- Return: = 0 on success, < 0 on failure.
- */
+int get_full_cbm(const char *cache_type, unsigned long *mask) +{
- char cbm_path[PATH_MAX];
- int ret;
- if (!cache_type)
return -1;- snprintf(cbm_path, sizeof(cbm_path), "%s/%s/cbm_mask",
INFO_PATH, cache_type);- ret = get_bit_mask(cbm_path, mask);
- if (ret)
return -1;- return 0;
Can this just be "return get_bit_mask()" ? But actually, I would like to propose that this also returns a failure if the returned mask is zero. This would make the code more robust, especially looking ahead at utility like cache_portion_size() that divides by this full mask.
Reinette
get_cache_size() does not modify cache_type so it could be const.
Mark cache_type const so that const char * can be passed to it. This prevents warnings once many of the test parameters are marked const.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/resctrl.h | 2 +- tools/testing/selftests/resctrl/resctrlfs.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/tools/testing/selftests/resctrl/resctrl.h b/tools/testing/selftests/resctrl/resctrl.h index 7a2b4da684ed..62bad95ea7d0 100644 --- a/tools/testing/selftests/resctrl/resctrl.h +++ b/tools/testing/selftests/resctrl/resctrl.h @@ -100,7 +100,7 @@ void mbm_test_cleanup(void); int mba_schemata_change(int cpu_no, const char * const *benchmark_cmd); void mba_test_cleanup(void); int get_full_cbm(const char *cache_type, unsigned long *mask); -int get_cache_size(int cpu_no, char *cache_type, unsigned long *cache_size); +int get_cache_size(int cpu_no, const char *cache_type, unsigned long *cache_size); void ctrlc_handler(int signum, siginfo_t *info, void *ptr); int signal_handler_register(void); void signal_handler_unregister(void); diff --git a/tools/testing/selftests/resctrl/resctrlfs.c b/tools/testing/selftests/resctrl/resctrlfs.c index 211fc7647659..1ea610fa65b1 100644 --- a/tools/testing/selftests/resctrl/resctrlfs.c +++ b/tools/testing/selftests/resctrl/resctrlfs.c @@ -138,7 +138,7 @@ int get_resource_id(int cpu_no, int *resource_id) * * Return: = 0 on success, < 0 on failure. */ -int get_cache_size(int cpu_no, char *cache_type, unsigned long *cache_size) +int get_cache_size(int cpu_no, const char *cache_type, unsigned long *cache_size) { char cache_path[1024], cache_str[64]; int length, i, cache_num;
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
get_cache_size() does not modify cache_type so it could be const.
Mark cache_type const so that const char * can be passed to it. This prevents warnings once many of the test parameters are marked const.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
Reviewed-by: Reinette Chatre reinette.chatre@intel.com
Reinette
CAT and CMT tests calculate size of the cache portion for the n-bits cache allocation on their own.
Add cache_portion_size() helper that calculates size of the cache portion for the given number of bits and use it to replace the existing span calculations. This also prepares for the new CAT test that will need to determine the size of the cache portion also during results processing.
Rename also 'cache_size' local variables to 'cache_total_size' to prevent misinterpretations.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/cat_test.c | 10 +++++----- tools/testing/selftests/resctrl/cmt_test.c | 8 ++++---- tools/testing/selftests/resctrl/resctrl.h | 15 +++++++++++++++ 3 files changed, 24 insertions(+), 9 deletions(-)
diff --git a/tools/testing/selftests/resctrl/cat_test.c b/tools/testing/selftests/resctrl/cat_test.c index 92f0f3ce90e5..cfcdb7bd600f 100644 --- a/tools/testing/selftests/resctrl/cat_test.c +++ b/tools/testing/selftests/resctrl/cat_test.c @@ -91,7 +91,7 @@ int cat_perf_miss_val(int cpu_no, int n, char *cache_type) { unsigned long l_mask, l_mask_1; int ret, pipefd[2], sibling_cpu_no; - unsigned long cache_size = 0; + unsigned long cache_total_size = 0; unsigned long long_mask; int count_of_bits; char pipe_message; @@ -102,10 +102,10 @@ int cat_perf_miss_val(int cpu_no, int n, char *cache_type) return ret;
/* Get L3/L2 cache size */ - ret = get_cache_size(cpu_no, cache_type, &cache_size); + ret = get_cache_size(cpu_no, cache_type, &cache_total_size); if (ret) return ret; - ksft_print_msg("Cache size :%lu\n", cache_size); + ksft_print_msg("Cache size :%lu\n", cache_total_size);
count_of_bits = count_bits(long_mask);
@@ -136,7 +136,7 @@ int cat_perf_miss_val(int cpu_no, int n, char *cache_type) /* Set param values for parent thread which will be allocated bitmask * with (max_bits - n) bits */ - span = cache_size * (count_of_bits - n) / count_of_bits; + span = cache_portion_size(cache_total_size, l_mask, long_mask); strcpy(param.ctrlgrp, "c2"); strcpy(param.mongrp, "m2"); strcpy(param.filename, RESULT_FILE_NAME2); @@ -158,7 +158,7 @@ int cat_perf_miss_val(int cpu_no, int n, char *cache_type) param.mask = l_mask_1; strcpy(param.ctrlgrp, "c1"); strcpy(param.mongrp, "m1"); - span = cache_size * n / count_of_bits; + span = cache_portion_size(cache_total_size, l_mask_1, long_mask); strcpy(param.filename, RESULT_FILE_NAME1); param.num_of_runs = 0; param.cpu_no = sibling_cpu_no; diff --git a/tools/testing/selftests/resctrl/cmt_test.c b/tools/testing/selftests/resctrl/cmt_test.c index d6ddacfb4416..51cbfc392f66 100644 --- a/tools/testing/selftests/resctrl/cmt_test.c +++ b/tools/testing/selftests/resctrl/cmt_test.c @@ -72,7 +72,7 @@ int cmt_resctrl_val(int cpu_no, int n, const char * const *benchmark_cmd) { const char * const *cmd = benchmark_cmd; const char *new_cmd[BENCHMARK_ARGS]; - unsigned long cache_size = 0; + unsigned long cache_total_size = 0; unsigned long long_mask; char *span_str = NULL; int count_of_bits; @@ -83,10 +83,10 @@ int cmt_resctrl_val(int cpu_no, int n, const char * const *benchmark_cmd) if (ret) return ret;
- ret = get_cache_size(cpu_no, "L3", &cache_size); + ret = get_cache_size(cpu_no, "L3", &cache_total_size); if (ret) return ret; - ksft_print_msg("Cache size :%lu\n", cache_size); + ksft_print_msg("Cache size :%lu\n", cache_total_size);
count_of_bits = count_bits(long_mask);
@@ -107,7 +107,7 @@ int cmt_resctrl_val(int cpu_no, int n, const char * const *benchmark_cmd) .setup = cmt_setup, };
- span = cache_size * n / count_of_bits; + span = cache_portion_size(cache_total_size, param.mask, long_mask);
if (strcmp(cmd[0], "fill_buf") == 0) { /* Duplicate the command to be able to replace span in it */ diff --git a/tools/testing/selftests/resctrl/resctrl.h b/tools/testing/selftests/resctrl/resctrl.h index 62bad95ea7d0..57fe42f4eda2 100644 --- a/tools/testing/selftests/resctrl/resctrl.h +++ b/tools/testing/selftests/resctrl/resctrl.h @@ -117,4 +117,19 @@ int show_cache_info(unsigned long sum_llc_val, int no_of_bits, unsigned long max_diff_percent, unsigned long num_of_runs, bool platform, bool cmt);
+/* + * cache_portion_size - Calculate the size of a cache portion + * @cache_size: Total cache size in bytes + * @portion_mask: Cache portion mask + * @full_cache_mask: Full Cache Bit Mask (CBM) for the cache + * + * Return: The size of the cache portion in bytes. + */ +static inline int cache_portion_size(unsigned long cache_size, + unsigned long portion_mask, + unsigned long full_cache_mask) +{ + return cache_size * count_bits(portion_mask) / count_bits(full_cache_mask); +} + #endif /* RESCTRL_H */
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
CAT and CMT tests calculate size of the cache portion for the n-bits cache allocation on their own.
Add cache_portion_size() helper that calculates size of the cache portion for the given number of bits and use it to replace the existing span calculations. This also prepares for the new CAT test that will need to determine the size of the cache portion also during results processing.
Rename also 'cache_size' local variables to 'cache_total_size' to prevent misinterpretations.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
Reviewed-by: Reinette Chatre reinette.chatre@intel.com
Reinette
CAT test doesn't take shareable bits into account, i.e., the test might be sharing cache with some devices (e.g., graphics).
Introduce get_mask_no_shareable() and use it to provision an environment for CAT test where the allocated LLC is isolated better. Excluding shareable_bits may create hole(s) into the cbm_mask, thus add a new helper count_contiguous_bits() to find the longest contiguous set of CBM bits.
create_bit_mask() is needed by an upcoming CAT test rewrite so make it available in resctrl.h right away.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/cat_test.c | 12 ++- tools/testing/selftests/resctrl/resctrl.h | 3 + tools/testing/selftests/resctrl/resctrlfs.c | 89 +++++++++++++++++++++ 3 files changed, 100 insertions(+), 4 deletions(-)
diff --git a/tools/testing/selftests/resctrl/cat_test.c b/tools/testing/selftests/resctrl/cat_test.c index cfcdb7bd600f..2c8b37c0a179 100644 --- a/tools/testing/selftests/resctrl/cat_test.c +++ b/tools/testing/selftests/resctrl/cat_test.c @@ -89,15 +89,19 @@ void cat_test_cleanup(void)
int cat_perf_miss_val(int cpu_no, int n, char *cache_type) { + unsigned long full_cache_mask, long_mask; unsigned long l_mask, l_mask_1; int ret, pipefd[2], sibling_cpu_no; unsigned long cache_total_size = 0; - unsigned long long_mask; int count_of_bits; char pipe_message; size_t span;
- ret = get_full_cbm(cache_type, &long_mask); + ret = get_full_cbm(cache_type, &full_cache_mask); + if (ret) + return ret; + /* Get the exclusive portion of the cache */ + ret = get_mask_no_shareable(cache_type, &long_mask); if (ret) return ret;
@@ -136,7 +140,7 @@ int cat_perf_miss_val(int cpu_no, int n, char *cache_type) /* Set param values for parent thread which will be allocated bitmask * with (max_bits - n) bits */ - span = cache_portion_size(cache_total_size, l_mask, long_mask); + span = cache_portion_size(cache_total_size, l_mask, full_cache_mask); strcpy(param.ctrlgrp, "c2"); strcpy(param.mongrp, "m2"); strcpy(param.filename, RESULT_FILE_NAME2); @@ -158,7 +162,7 @@ int cat_perf_miss_val(int cpu_no, int n, char *cache_type) param.mask = l_mask_1; strcpy(param.ctrlgrp, "c1"); strcpy(param.mongrp, "m1"); - span = cache_portion_size(cache_total_size, l_mask_1, long_mask); + span = cache_portion_size(cache_total_size, l_mask_1, full_cache_mask); strcpy(param.filename, RESULT_FILE_NAME1); param.num_of_runs = 0; param.cpu_no = sibling_cpu_no; diff --git a/tools/testing/selftests/resctrl/resctrl.h b/tools/testing/selftests/resctrl/resctrl.h index 57fe42f4eda2..a911b08fa595 100644 --- a/tools/testing/selftests/resctrl/resctrl.h +++ b/tools/testing/selftests/resctrl/resctrl.h @@ -99,7 +99,10 @@ void tests_cleanup(void); void mbm_test_cleanup(void); int mba_schemata_change(int cpu_no, const char * const *benchmark_cmd); void mba_test_cleanup(void); +unsigned long create_bit_mask(unsigned int start, unsigned int len); int get_full_cbm(const char *cache_type, unsigned long *mask); +int get_shareable_mask(const char *cache_type, unsigned long *shareable_mask); +int get_mask_no_shareable(const char *cache_type, unsigned long *mask); int get_cache_size(int cpu_no, const char *cache_type, unsigned long *cache_size); void ctrlc_handler(int signum, siginfo_t *info, void *ptr); int signal_handler_register(void); diff --git a/tools/testing/selftests/resctrl/resctrlfs.c b/tools/testing/selftests/resctrl/resctrlfs.c index 1ea610fa65b1..3a9ed918d657 100644 --- a/tools/testing/selftests/resctrl/resctrlfs.c +++ b/tools/testing/selftests/resctrl/resctrlfs.c @@ -228,6 +228,44 @@ static int get_bit_mask(const char *filename, unsigned long *mask) return 0; }
+/* + * create_bit_mask- Create bit mask from start,len pair + * @start: LSB of the mask + * @len Number of bits in the mask + */ +unsigned long create_bit_mask(unsigned int start, unsigned int len) +{ + return ((1UL << len) - 1UL) << start; +} + +/* + * count_contiguous_bits - Returns the longest train of bits in a bit mask + * @val A bit mask + * @start The location of the least-significant bit of the longest train + * + * Return: The length of the contiguous bits in the longest train of bits + */ +static unsigned int count_contiguous_bits(unsigned long val, unsigned int *start) +{ + unsigned long last_val; + unsigned int count = 0; + + while (val) { + last_val = val; + val &= (val >> 1); + count++; + } + + if (start) { + if (count) + *start = ffsl(last_val) - 1; + else + *start = 0; + } + + return count; +} + /* * get_full_cbm - Get full Cache Bit Mask (CBM) * @cache_type: Cache type as "L2" or "L3" @@ -254,6 +292,57 @@ int get_full_cbm(const char *cache_type, unsigned long *mask) return 0; }
+/* + * get_shareable_mask - Get shareable mask from shareable_bits + * @cache_type: Cache type as "L2" or "L3" + * @shareable_mask: Shareable mask returned as unsigned long + * + * Return: = 0 on success, < 0 on failure. + */ +int get_shareable_mask(const char *cache_type, unsigned long *shareable_mask) +{ + char mask_path[PATH_MAX]; + + if (!cache_type) + return -1; + + snprintf(mask_path, sizeof(mask_path), "%s/%s/shareable_bits", + INFO_PATH, cache_type); + + return get_bit_mask(mask_path, shareable_mask); +} + +/* + * get_mask_no_shareable - Get Cache Bit Mask (CBM) without shareable bits + * @cache_type: Cache type as "L2" or "L3" + * @mask: The largest exclusive portion of the cache out of the + * full CBM, returned as unsigned long + * + * Parts of a cache may be shared with other devices such as GPU. This function + * calculates the largest exclusive portion of the cache where no other devices + * besides CPU have access to the cache portion. + * + * Return: = 0 on success, < 0 on failure. + */ +int get_mask_no_shareable(const char *cache_type, unsigned long *mask) +{ + unsigned long full_mask, shareable_mask; + unsigned int start, len; + + if (get_full_cbm(cache_type, &full_mask) < 0) + return -1; + if (get_shareable_mask(cache_type, &shareable_mask) < 0) + return -1; + + len = count_contiguous_bits(full_mask & ~shareable_mask, &start); + if (!len) + return -1; + + *mask = create_bit_mask(start, len); + + return 0; +} + /* * get_core_sibling - Get sibling core id from the same socket for given CPU * @cpu_no: CPU number
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
CAT test doesn't take shareable bits into account, i.e., the test might be sharing cache with some devices (e.g., graphics).
Introduce get_mask_no_shareable() and use it to provision an environment for CAT test where the allocated LLC is isolated better. Excluding shareable_bits may create hole(s) into the cbm_mask, thus add a new helper count_contiguous_bits() to find the longest contiguous set of CBM bits.
create_bit_mask() is needed by an upcoming CAT test rewrite so make it available in resctrl.h right away.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
tools/testing/selftests/resctrl/cat_test.c | 12 ++- tools/testing/selftests/resctrl/resctrl.h | 3 + tools/testing/selftests/resctrl/resctrlfs.c | 89 +++++++++++++++++++++ 3 files changed, 100 insertions(+), 4 deletions(-)
diff --git a/tools/testing/selftests/resctrl/cat_test.c b/tools/testing/selftests/resctrl/cat_test.c index cfcdb7bd600f..2c8b37c0a179 100644 --- a/tools/testing/selftests/resctrl/cat_test.c +++ b/tools/testing/selftests/resctrl/cat_test.c @@ -89,15 +89,19 @@ void cat_test_cleanup(void) int cat_perf_miss_val(int cpu_no, int n, char *cache_type) {
- unsigned long full_cache_mask, long_mask; unsigned long l_mask, l_mask_1; int ret, pipefd[2], sibling_cpu_no; unsigned long cache_total_size = 0;
- unsigned long long_mask; int count_of_bits; char pipe_message; size_t span;
- ret = get_full_cbm(cache_type, &long_mask);
- ret = get_full_cbm(cache_type, &full_cache_mask);
- if (ret)
return ret;- /* Get the exclusive portion of the cache */
How about: "Get largest contiguous exclusive portion of the cache" ?
- ret = get_mask_no_shareable(cache_type, &long_mask); if (ret) return ret;
@@ -136,7 +140,7 @@ int cat_perf_miss_val(int cpu_no, int n, char *cache_type) /* Set param values for parent thread which will be allocated bitmask * with (max_bits - n) bits */
- span = cache_portion_size(cache_total_size, l_mask, long_mask);
- span = cache_portion_size(cache_total_size, l_mask, full_cache_mask); strcpy(param.ctrlgrp, "c2"); strcpy(param.mongrp, "m2"); strcpy(param.filename, RESULT_FILE_NAME2);
@@ -158,7 +162,7 @@ int cat_perf_miss_val(int cpu_no, int n, char *cache_type) param.mask = l_mask_1; strcpy(param.ctrlgrp, "c1"); strcpy(param.mongrp, "m1");
span = cache_portion_size(cache_total_size, l_mask_1, long_mask);
span = cache_portion_size(cache_total_size, l_mask_1, full_cache_mask);diff --git a/tools/testing/selftests/resctrl/resctrl.h b/tools/testing/selftests/resctrl/resctrl.h index 57fe42f4eda2..a911b08fa595 100644 --- a/tools/testing/selftests/resctrl/resctrl.h +++ b/tools/testing/selftests/resctrl/resctrl.h @@ -99,7 +99,10 @@ void tests_cleanup(void); void mbm_test_cleanup(void); int mba_schemata_change(int cpu_no, const char * const *benchmark_cmd); void mba_test_cleanup(void); +unsigned long create_bit_mask(unsigned int start, unsigned int len); int get_full_cbm(const char *cache_type, unsigned long *mask); +int get_shareable_mask(const char *cache_type, unsigned long *shareable_mask);
Does this need to be exported?
+int get_mask_no_shareable(const char *cache_type, unsigned long *mask); int get_cache_size(int cpu_no, const char *cache_type, unsigned long *cache_size); void ctrlc_handler(int signum, siginfo_t *info, void *ptr); int signal_handler_register(void); diff --git a/tools/testing/selftests/resctrl/resctrlfs.c b/tools/testing/selftests/resctrl/resctrlfs.c index 1ea610fa65b1..3a9ed918d657 100644 --- a/tools/testing/selftests/resctrl/resctrlfs.c +++ b/tools/testing/selftests/resctrl/resctrlfs.c @@ -228,6 +228,44 @@ static int get_bit_mask(const char *filename, unsigned long *mask) return 0; } +/*
- create_bit_mask- Create bit mask from start,len pair
Please add a space after the comma.
- @start: LSB of the mask
- @len Number of bits in the mask
- */
+unsigned long create_bit_mask(unsigned int start, unsigned int len) +{
- return ((1UL << len) - 1UL) << start;
+}
+/*
- count_contiguous_bits - Returns the longest train of bits in a bit mask
- @val A bit mask
- @start The location of the least-significant bit of the longest train
- Return: The length of the contiguous bits in the longest train of bits
- */
+static unsigned int count_contiguous_bits(unsigned long val, unsigned int *start) +{
- unsigned long last_val;
- unsigned int count = 0;
- while (val) {
last_val = val;val &= (val >> 1);count++;- }
- if (start) {
if (count)*start = ffsl(last_val) - 1;else*start = 0;- }
- return count;
+}
/*
- get_full_cbm - Get full Cache Bit Mask (CBM)
- @cache_type: Cache type as "L2" or "L3"
@@ -254,6 +292,57 @@ int get_full_cbm(const char *cache_type, unsigned long *mask) return 0; } +/*
- get_shareable_mask - Get shareable mask from shareable_bits
- @cache_type: Cache type as "L2" or "L3"
- @shareable_mask: Shareable mask returned as unsigned long
- Return: = 0 on success, < 0 on failure.
- */
+int get_shareable_mask(const char *cache_type, unsigned long *shareable_mask)
It looks like this can be static.
+{
- char mask_path[PATH_MAX];
- if (!cache_type)
return -1;- snprintf(mask_path, sizeof(mask_path), "%s/%s/shareable_bits",
INFO_PATH, cache_type);- return get_bit_mask(mask_path, shareable_mask);
+}
+/*
- get_mask_no_shareable - Get Cache Bit Mask (CBM) without shareable bits
- @cache_type: Cache type as "L2" or "L3"
- @mask: The largest exclusive portion of the cache out of the
full CBM, returned as unsigned long
- Parts of a cache may be shared with other devices such as GPU. This function
- calculates the largest exclusive portion of the cache where no other devices
- besides CPU have access to the cache portion.
- Return: = 0 on success, < 0 on failure.
- */
+int get_mask_no_shareable(const char *cache_type, unsigned long *mask) +{
- unsigned long full_mask, shareable_mask;
- unsigned int start, len;
- if (get_full_cbm(cache_type, &full_mask) < 0)
return -1;- if (get_shareable_mask(cache_type, &shareable_mask) < 0)
return -1;- len = count_contiguous_bits(full_mask & ~shareable_mask, &start);
- if (!len)
return -1;- *mask = create_bit_mask(start, len);
- return 0;
+}
/*
- get_core_sibling - Get sibling core id from the same socket for given CPU
- @cpu_no: CPU number
Reinette
measure_cache_vals() does a different thing depending on the test case that called it: - For CAT, it measures LLC misses through perf. - For CMT, it measures LLC occupancy through resctrl.
Split these two functionalities into own functions the CAT and CMT tests can call directly. Replace passing the struct resctrl_val_param parameter with the filename because it's more generic and all those functions need out of resctrl_val.
Co-developed-by: Fenghua Yu fenghua.yu@intel.com Signed-off-by: Fenghua Yu fenghua.yu@intel.com Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/cache.c | 66 ++++++++++++------- tools/testing/selftests/resctrl/resctrl.h | 2 +- tools/testing/selftests/resctrl/resctrl_val.c | 2 +- 3 files changed, 43 insertions(+), 27 deletions(-)
diff --git a/tools/testing/selftests/resctrl/cache.c b/tools/testing/selftests/resctrl/cache.c index 8aa6d67db978..129d1c293518 100644 --- a/tools/testing/selftests/resctrl/cache.c +++ b/tools/testing/selftests/resctrl/cache.c @@ -147,7 +147,7 @@ static int get_llc_occu_resctrl(unsigned long *llc_occupancy) * * Return: 0 on success. non-zero on failure. */ -static int print_results_cache(char *filename, int bm_pid, +static int print_results_cache(const char *filename, int bm_pid, unsigned long llc_value) { FILE *fp; @@ -169,35 +169,51 @@ static int print_results_cache(char *filename, int bm_pid, return 0; }
-int measure_cache_vals(struct resctrl_val_param *param, int bm_pid) +/* + * perf_event_measure - Measure perf events + * @filename: Filename for writing the results + * @bm_pid: PID that runs the benchmark + * + * Measures perf events (e.g., cache misses) and writes the results into + * @filename. @bm_pid is written to the results file along with the measured + * value. + * + * Return: =0 on success. <0 on failure. + */ +static int perf_event_measure(const char *filename, int bm_pid) { - unsigned long llc_perf_miss = 0, llc_occu_resc = 0, llc_value = 0; + unsigned long llc_perf_miss = 0; int ret;
- /* - * Measure cache miss from perf. - */ - if (!strncmp(param->resctrl_val, CAT_STR, sizeof(CAT_STR))) { - ret = get_llc_perf(&llc_perf_miss); - if (ret < 0) - return ret; - llc_value = llc_perf_miss; - } + ret = get_llc_perf(&llc_perf_miss); + if (ret < 0) + return ret;
- /* - * Measure llc occupancy from resctrl. - */ - if (!strncmp(param->resctrl_val, CMT_STR, sizeof(CMT_STR))) { - ret = get_llc_occu_resctrl(&llc_occu_resc); - if (ret < 0) - return ret; - llc_value = llc_occu_resc; - } - ret = print_results_cache(param->filename, bm_pid, llc_value); - if (ret) + ret = print_results_cache(filename, bm_pid, llc_perf_miss); + return ret; +} + +/* + * measure_llc_resctrl - Measure resctrl llc value from resctrl + * @filename: Filename for writing the results + * @bm_pid: PID that runs the benchmark + * + * Measures llc occupancy from resctrl and writes the results into @filename. + * @bm_pid is written to the results file along with the measured value. + * + * Return: =0 on success. <0 on failure. + */ +int measure_llc_resctrl(const char *filename, int bm_pid) +{ + unsigned long llc_occu_resc = 0; + int ret; + + ret = get_llc_occu_resctrl(&llc_occu_resc); + if (ret < 0) return ret;
- return 0; + ret = print_results_cache(filename, bm_pid, llc_occu_resc); + return ret; }
/* @@ -252,7 +268,7 @@ int cat_val(struct resctrl_val_param *param, size_t span) }
sleep(1); - ret = measure_cache_vals(param, bm_pid); + ret = perf_event_measure(param->filename, bm_pid); if (ret) goto pe_close; } diff --git a/tools/testing/selftests/resctrl/resctrl.h b/tools/testing/selftests/resctrl/resctrl.h index a911b08fa595..d35e3ba4dfa2 100644 --- a/tools/testing/selftests/resctrl/resctrl.h +++ b/tools/testing/selftests/resctrl/resctrl.h @@ -114,7 +114,7 @@ int cmt_resctrl_val(int cpu_no, int n, const char * const *benchmark_cmd); unsigned int count_bits(unsigned long n); void cmt_test_cleanup(void); int get_core_sibling(int cpu_no); -int measure_cache_vals(struct resctrl_val_param *param, int bm_pid); +int measure_llc_resctrl(const char *filename, int bm_pid); int show_cache_info(unsigned long sum_llc_val, int no_of_bits, size_t cache_span, unsigned long max_diff, unsigned long max_diff_percent, unsigned long num_of_runs, diff --git a/tools/testing/selftests/resctrl/resctrl_val.c b/tools/testing/selftests/resctrl/resctrl_val.c index 88789678917b..a07ba336db48 100644 --- a/tools/testing/selftests/resctrl/resctrl_val.c +++ b/tools/testing/selftests/resctrl/resctrl_val.c @@ -830,7 +830,7 @@ int resctrl_val(const char * const *benchmark_cmd, struct resctrl_val_param *par break; } else if (!strncmp(resctrl_val, CMT_STR, sizeof(CMT_STR))) { sleep(1); - ret = measure_cache_vals(param, bm_pid); + ret = measure_llc_resctrl(param->filename, bm_pid); if (ret) break; }
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
measure_cache_vals() does a different thing depending on the test case that called it:
- For CAT, it measures LLC misses through perf.
- For CMT, it measures LLC occupancy through resctrl.
Split these two functionalities into own functions the CAT and CMT tests can call directly. Replace passing the struct resctrl_val_param parameter with the filename because it's more generic and all those functions need out of resctrl_val.
Co-developed-by: Fenghua Yu fenghua.yu@intel.com Signed-off-by: Fenghua Yu fenghua.yu@intel.com Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
tools/testing/selftests/resctrl/cache.c | 66 ++++++++++++------- tools/testing/selftests/resctrl/resctrl.h | 2 +- tools/testing/selftests/resctrl/resctrl_val.c | 2 +- 3 files changed, 43 insertions(+), 27 deletions(-)
diff --git a/tools/testing/selftests/resctrl/cache.c b/tools/testing/selftests/resctrl/cache.c index 8aa6d67db978..129d1c293518 100644 --- a/tools/testing/selftests/resctrl/cache.c +++ b/tools/testing/selftests/resctrl/cache.c @@ -147,7 +147,7 @@ static int get_llc_occu_resctrl(unsigned long *llc_occupancy)
- Return: 0 on success. non-zero on failure.
*/ -static int print_results_cache(char *filename, int bm_pid, +static int print_results_cache(const char *filename, int bm_pid, unsigned long llc_value) { FILE *fp; @@ -169,35 +169,51 @@ static int print_results_cache(char *filename, int bm_pid, return 0; } -int measure_cache_vals(struct resctrl_val_param *param, int bm_pid) +/*
- perf_event_measure - Measure perf events
- @filename: Filename for writing the results
- @bm_pid: PID that runs the benchmark
- Measures perf events (e.g., cache misses) and writes the results into
- @filename. @bm_pid is written to the results file along with the measured
- value.
- Return: =0 on success. <0 on failure.
I do not think this is accurate. It looks like this function returns the return value of print_results_cache() which returns errno on failure. If this is the case then I think this proves that returning a positive integer on failure should be avoided since it just creates traps.
- */
+static int perf_event_measure(const char *filename, int bm_pid) {
- unsigned long llc_perf_miss = 0, llc_occu_resc = 0, llc_value = 0;
- unsigned long llc_perf_miss = 0; int ret;
- /*
* Measure cache miss from perf.*/- if (!strncmp(param->resctrl_val, CAT_STR, sizeof(CAT_STR))) {
ret = get_llc_perf(&llc_perf_miss);if (ret < 0)return ret;llc_value = llc_perf_miss;- }
- ret = get_llc_perf(&llc_perf_miss);
- if (ret < 0)
return ret;
- /*
* Measure llc occupancy from resctrl.*/- if (!strncmp(param->resctrl_val, CMT_STR, sizeof(CMT_STR))) {
ret = get_llc_occu_resctrl(&llc_occu_resc);if (ret < 0)return ret;llc_value = llc_occu_resc;- }
- ret = print_results_cache(param->filename, bm_pid, llc_value);
- if (ret)
- ret = print_results_cache(filename, bm_pid, llc_perf_miss);
- return ret;
+}
Perhaps print_results_cache() can be made to return negative error and this just be "return print_results_cache(...)" and the function comment be accurate?
+/*
- measure_llc_resctrl - Measure resctrl llc value from resctrl
llc -> LLC
- @filename: Filename for writing the results
- @bm_pid: PID that runs the benchmark
- Measures llc occupancy from resctrl and writes the results into @filename.
llc -> LLC
- @bm_pid is written to the results file along with the measured value.
- Return: =0 on success. <0 on failure.
same issue ?
Reinette
On Tue, 28 Nov 2023, Reinette Chatre wrote:
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
measure_cache_vals() does a different thing depending on the test case that called it:
- For CAT, it measures LLC misses through perf.
- For CMT, it measures LLC occupancy through resctrl.
Split these two functionalities into own functions the CAT and CMT tests can call directly. Replace passing the struct resctrl_val_param parameter with the filename because it's more generic and all those functions need out of resctrl_val.
Co-developed-by: Fenghua Yu fenghua.yu@intel.com Signed-off-by: Fenghua Yu fenghua.yu@intel.com Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
tools/testing/selftests/resctrl/cache.c | 66 ++++++++++++------- tools/testing/selftests/resctrl/resctrl.h | 2 +- tools/testing/selftests/resctrl/resctrl_val.c | 2 +- 3 files changed, 43 insertions(+), 27 deletions(-)
diff --git a/tools/testing/selftests/resctrl/cache.c b/tools/testing/selftests/resctrl/cache.c index 8aa6d67db978..129d1c293518 100644 --- a/tools/testing/selftests/resctrl/cache.c +++ b/tools/testing/selftests/resctrl/cache.c @@ -147,7 +147,7 @@ static int get_llc_occu_resctrl(unsigned long *llc_occupancy)
- Return: 0 on success. non-zero on failure.
*/ -static int print_results_cache(char *filename, int bm_pid, +static int print_results_cache(const char *filename, int bm_pid, unsigned long llc_value) { FILE *fp; @@ -169,35 +169,51 @@ static int print_results_cache(char *filename, int bm_pid, return 0; } -int measure_cache_vals(struct resctrl_val_param *param, int bm_pid) +/*
- perf_event_measure - Measure perf events
- @filename: Filename for writing the results
- @bm_pid: PID that runs the benchmark
- Measures perf events (e.g., cache misses) and writes the results into
- @filename. @bm_pid is written to the results file along with the measured
- value.
- Return: =0 on success. <0 on failure.
I do not think this is accurate. It looks like this function returns the return value of print_results_cache() which returns errno on failure. If this is the case then I think this proves that returning a positive integer on failure should be avoided since it just creates traps.
- */
+static int perf_event_measure(const char *filename, int bm_pid) {
- unsigned long llc_perf_miss = 0, llc_occu_resc = 0, llc_value = 0;
- unsigned long llc_perf_miss = 0; int ret;
- /*
* Measure cache miss from perf.*/- if (!strncmp(param->resctrl_val, CAT_STR, sizeof(CAT_STR))) {
ret = get_llc_perf(&llc_perf_miss);if (ret < 0)return ret;llc_value = llc_perf_miss;- }
- ret = get_llc_perf(&llc_perf_miss);
- if (ret < 0)
return ret;
- /*
* Measure llc occupancy from resctrl.*/- if (!strncmp(param->resctrl_val, CMT_STR, sizeof(CMT_STR))) {
ret = get_llc_occu_resctrl(&llc_occu_resc);if (ret < 0)return ret;llc_value = llc_occu_resc;- }
- ret = print_results_cache(param->filename, bm_pid, llc_value);
- if (ret)
- ret = print_results_cache(filename, bm_pid, llc_perf_miss);
- return ret;
+}
Perhaps print_results_cache() can be made to return negative error and this just be "return print_results_cache(...)" and the function comment be accurate?
I think, I'll just change all "return errno;" to "return -1" before this, however, one open question which impacts whether this is actually Fixes class issue:
It seems that perror()'s manpage doesn't answer one important question, whether it ifself can alter errno or not. The resctrl selftest code assumes it doesn't but some evidence I came across says otherwise so doing return errno; after calling perror() might not even be valid at all.
So I'm tempted to create an additional Fixes patch about the return change into the front of the series.
Hi Ilpo,
On 12/7/2023 6:32 AM, Ilpo Järvinen wrote:
On Tue, 28 Nov 2023, Reinette Chatre wrote:
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
...
- /*
* Measure llc occupancy from resctrl.*/- if (!strncmp(param->resctrl_val, CMT_STR, sizeof(CMT_STR))) {
ret = get_llc_occu_resctrl(&llc_occu_resc);if (ret < 0)return ret;llc_value = llc_occu_resc;- }
- ret = print_results_cache(param->filename, bm_pid, llc_value);
- if (ret)
- ret = print_results_cache(filename, bm_pid, llc_perf_miss);
- return ret;
+}
Perhaps print_results_cache() can be made to return negative error and this just be "return print_results_cache(...)" and the function comment be accurate?
I think, I'll just change all "return errno;" to "return -1" before this, however, one open question which impacts whether this is actually Fixes class issue:
It seems that perror()'s manpage doesn't answer one important question, whether it ifself can alter errno or not. The resctrl selftest code assumes it doesn't but some evidence I came across says otherwise so doing return errno; after calling perror() might not even be valid at all.
So I'm tempted to create an additional Fixes patch about the return change into the front of the series.
I would not trust errno to contain code of earlier calls after a call to perror(). If errno is needed I think it should be saved before calling perror(). Looking at perror() at [1] I do not see an effort to restore errno before it returns, and looking at the implementation of perror() there appears to be many opportunities for errno to change.
Reinette
[1] https://sourceware.org/git/?p=glibc.git%3Ba=blob%3Bf=stdio-common/perror.c%3...
On Thu, 7 Dec 2023, Reinette Chatre wrote:
On 12/7/2023 6:32 AM, Ilpo Järvinen wrote:
On Tue, 28 Nov 2023, Reinette Chatre wrote:
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
...
- /*
* Measure llc occupancy from resctrl.*/- if (!strncmp(param->resctrl_val, CMT_STR, sizeof(CMT_STR))) {
ret = get_llc_occu_resctrl(&llc_occu_resc);if (ret < 0)return ret;llc_value = llc_occu_resc;- }
- ret = print_results_cache(param->filename, bm_pid, llc_value);
- if (ret)
- ret = print_results_cache(filename, bm_pid, llc_perf_miss);
- return ret;
+}
Perhaps print_results_cache() can be made to return negative error and this just be "return print_results_cache(...)" and the function comment be accurate?
I think, I'll just change all "return errno;" to "return -1" before this, however, one open question which impacts whether this is actually Fixes class issue:
It seems that perror()'s manpage doesn't answer one important question, whether it ifself can alter errno or not. The resctrl selftest code assumes it doesn't but some evidence I came across says otherwise so doing return errno; after calling perror() might not even be valid at all.
So I'm tempted to create an additional Fixes patch about the return change into the front of the series.
I would not trust errno to contain code of earlier calls after a call to perror(). If errno is needed I think it should be saved before calling perror(). Looking at perror() at [1] I do not see an effort to restore errno before it returns, and looking at the implementation of perror() there appears to be many opportunities for errno to change.
Reinette
[1] https://sourceware.org/git/?p=glibc.git%3Ba=blob%3Bf=stdio-common/perror.c%3...
I already spent some moments in converting all return error -> return -1, since all such places do perror() calls anyway (which I also converted to ksft_perror() or ksft_print_msg() where perror() didn't make any sense) there's not much added value in returning the errno which was not correctly done in the existing code anyway.
On 12/7/2023 10:33 AM, Ilpo Järvinen wrote:
I already spent some moments in converting all return error -> return -1, since all such places do perror() calls anyway (which I also converted to ksft_perror() or ksft_print_msg() where perror() didn't make any sense) there's not much added value in returning the errno which was not correctly done in the existing code anyway.
Thank you very much Ilpo.
Reinette
show_cache_info() calculates results and provides generic cache information. This makes it hard to alter pass/fail conditions.
Separate the test specific checks into CAT and CMT test files and leave only the generic information part into show_cache_info().
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/cache.c | 42 +++++----------------- tools/testing/selftests/resctrl/cat_test.c | 30 ++++++++++++++-- tools/testing/selftests/resctrl/cmt_test.c | 32 +++++++++++++++-- tools/testing/selftests/resctrl/resctrl.h | 6 ++-- 4 files changed, 66 insertions(+), 44 deletions(-)
diff --git a/tools/testing/selftests/resctrl/cache.c b/tools/testing/selftests/resctrl/cache.c index 129d1c293518..493e0e924123 100644 --- a/tools/testing/selftests/resctrl/cache.c +++ b/tools/testing/selftests/resctrl/cache.c @@ -281,43 +281,17 @@ int cat_val(struct resctrl_val_param *param, size_t span) }
/* - * show_cache_info: show cache test result information - * @sum_llc_val: sum of LLC cache result data - * @no_of_bits: number of bits - * @cache_span: cache span in bytes for CMT or in lines for CAT - * @max_diff: max difference - * @max_diff_percent: max difference percentage - * @num_of_runs: number of runs - * @platform: show test information on this platform - * @cmt: CMT test or CAT test - * - * Return: 0 on success. non-zero on failure. + * show_cache_info - Show generic cache test information + * @no_of_bits: Number of bits + * @avg_llc_val: Average of LLC cache result data + * @cache_span: Cache span + * @lines: @cache_span in lines or bytes */ -int show_cache_info(unsigned long sum_llc_val, int no_of_bits, - size_t cache_span, unsigned long max_diff, - unsigned long max_diff_percent, unsigned long num_of_runs, - bool platform, bool cmt) +void show_cache_info(int no_of_bits, unsigned long avg_llc_val, + size_t cache_span, bool lines) { - unsigned long avg_llc_val = 0; - float diff_percent; - long avg_diff = 0; - int ret; - - avg_llc_val = sum_llc_val / num_of_runs; - avg_diff = (long)abs(cache_span - avg_llc_val); - diff_percent = ((float)cache_span - avg_llc_val) / cache_span * 100; - - ret = platform && abs((int)diff_percent) > max_diff_percent && - (cmt ? (abs(avg_diff) > max_diff) : true); - - ksft_print_msg("%s Check cache miss rate within %lu%%\n", - ret ? "Fail:" : "Pass:", max_diff_percent); - - ksft_print_msg("Percent diff=%d\n", abs((int)diff_percent)); ksft_print_msg("Number of bits: %d\n", no_of_bits); ksft_print_msg("Average LLC val: %lu\n", avg_llc_val); - ksft_print_msg("Cache span (%s): %zu\n", cmt ? "bytes" : "lines", + ksft_print_msg("Cache span (%s): %zu\n", lines ? "lines" : "bytes", cache_span); - - return ret; } diff --git a/tools/testing/selftests/resctrl/cat_test.c b/tools/testing/selftests/resctrl/cat_test.c index 2c8b37c0a179..2842dff292f0 100644 --- a/tools/testing/selftests/resctrl/cat_test.c +++ b/tools/testing/selftests/resctrl/cat_test.c @@ -41,6 +41,30 @@ static int cat_setup(struct resctrl_val_param *p) return ret; }
+static int show_results_info(unsigned long sum_llc_val, int no_of_bits, + unsigned long cache_span, unsigned long max_diff, + unsigned long max_diff_percent, unsigned long num_of_runs, + bool platform) +{ + unsigned long avg_llc_val = 0; + float diff_percent; + int ret; + + avg_llc_val = sum_llc_val / num_of_runs; + diff_percent = ((float)cache_span - avg_llc_val) / cache_span * 100; + + ret = platform && abs((int)diff_percent) > max_diff_percent; + + ksft_print_msg("%s Check cache miss rate within %lu%%\n", + ret ? "Fail:" : "Pass:", max_diff_percent); + + ksft_print_msg("Percent diff=%d\n", abs((int)diff_percent)); + + show_cache_info(no_of_bits, avg_llc_val, cache_span, true); + + return ret; +} + static int check_results(struct resctrl_val_param *param, size_t span) { char *token_array[8], temp[512]; @@ -76,9 +100,9 @@ static int check_results(struct resctrl_val_param *param, size_t span) fclose(fp); no_of_bits = count_bits(param->mask);
- return show_cache_info(sum_llc_perf_miss, no_of_bits, span / 64, - MAX_DIFF, MAX_DIFF_PERCENT, runs - 1, - get_vendor() == ARCH_INTEL, false); + return show_results_info(sum_llc_perf_miss, no_of_bits, span / 64, + MAX_DIFF, MAX_DIFF_PERCENT, runs - 1, + get_vendor() == ARCH_INTEL); }
void cat_test_cleanup(void) diff --git a/tools/testing/selftests/resctrl/cmt_test.c b/tools/testing/selftests/resctrl/cmt_test.c index 51cbfc392f66..4573a99d87b5 100644 --- a/tools/testing/selftests/resctrl/cmt_test.c +++ b/tools/testing/selftests/resctrl/cmt_test.c @@ -27,6 +27,33 @@ static int cmt_setup(struct resctrl_val_param *p) return 0; }
+static int show_results_info(unsigned long sum_llc_val, int no_of_bits, + unsigned long cache_span, unsigned long max_diff, + unsigned long max_diff_percent, unsigned long num_of_runs, + bool platform) +{ + unsigned long avg_llc_val = 0; + float diff_percent; + long avg_diff = 0; + int ret; + + avg_llc_val = sum_llc_val / num_of_runs; + avg_diff = (long)abs(cache_span - avg_llc_val); + diff_percent = ((float)cache_span - avg_llc_val) / cache_span * 100; + + ret = platform && abs((int)diff_percent) > max_diff_percent && + abs(avg_diff) > max_diff; + + ksft_print_msg("%s Check cache miss rate within %lu%%\n", + ret ? "Fail:" : "Pass:", max_diff_percent); + + ksft_print_msg("Percent diff=%d\n", abs((int)diff_percent)); + + show_cache_info(no_of_bits, avg_llc_val, cache_span, false); + + return ret; +} + static int check_results(struct resctrl_val_param *param, size_t span, int no_of_bits) { char *token_array[8], temp[512]; @@ -58,9 +85,8 @@ static int check_results(struct resctrl_val_param *param, size_t span, int no_of } fclose(fp);
- return show_cache_info(sum_llc_occu_resc, no_of_bits, span, - MAX_DIFF, MAX_DIFF_PERCENT, runs - 1, - true, true); + return show_results_info(sum_llc_occu_resc, no_of_bits, span, + MAX_DIFF, MAX_DIFF_PERCENT, runs - 1, true); }
void cmt_test_cleanup(void) diff --git a/tools/testing/selftests/resctrl/resctrl.h b/tools/testing/selftests/resctrl/resctrl.h index d35e3ba4dfa2..4a05efccab35 100644 --- a/tools/testing/selftests/resctrl/resctrl.h +++ b/tools/testing/selftests/resctrl/resctrl.h @@ -115,10 +115,8 @@ unsigned int count_bits(unsigned long n); void cmt_test_cleanup(void); int get_core_sibling(int cpu_no); int measure_llc_resctrl(const char *filename, int bm_pid); -int show_cache_info(unsigned long sum_llc_val, int no_of_bits, - size_t cache_span, unsigned long max_diff, - unsigned long max_diff_percent, unsigned long num_of_runs, - bool platform, bool cmt); +void show_cache_info(int no_of_bits, unsigned long avg_llc_val, + size_t cache_span, bool lines);
/* * cache_portion_size - Calculate the size of a cache portion
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
show_cache_info() calculates results and provides generic cache information. This makes it hard to alter pass/fail conditions.
Separate the test specific checks into CAT and CMT test files and leave only the generic information part into show_cache_info().
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
Reviewed-by: Reinette Chatre reinette.chatre@intel.com
Reinette
Perf counters are __u64 but the code converts them to unsigned long before printing them out.
Remove unnecessary type conversion and retain the value as __u64 whole the time.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/cache.c | 21 ++++++++------------- tools/testing/selftests/resctrl/cat_test.c | 8 ++++---- tools/testing/selftests/resctrl/resctrl.h | 3 +-- 3 files changed, 13 insertions(+), 19 deletions(-)
diff --git a/tools/testing/selftests/resctrl/cache.c b/tools/testing/selftests/resctrl/cache.c index 493e0e924123..6a754ea376c9 100644 --- a/tools/testing/selftests/resctrl/cache.c +++ b/tools/testing/selftests/resctrl/cache.c @@ -83,9 +83,8 @@ static int reset_enable_llc_perf(pid_t pid, int cpu_no) * * Return: =0 on success. <0 on failure. */ -static int get_llc_perf(unsigned long *llc_perf_miss) +static int get_llc_perf(__u64 *llc_perf_miss) { - __u64 total_misses; int ret;
/* Stop counters after one span to get miss rate */ @@ -98,8 +97,7 @@ static int get_llc_perf(unsigned long *llc_perf_miss) return -1; }
- total_misses = rf_cqm.values[0].value; - *llc_perf_miss = total_misses; + *llc_perf_miss = rf_cqm.values[0].value;
return 0; } @@ -147,14 +145,12 @@ static int get_llc_occu_resctrl(unsigned long *llc_occupancy) * * Return: 0 on success. non-zero on failure. */ -static int print_results_cache(const char *filename, int bm_pid, - unsigned long llc_value) +static int print_results_cache(const char *filename, int bm_pid, __u64 llc_value) { FILE *fp;
if (strcmp(filename, "stdio") == 0 || strcmp(filename, "stderr") == 0) { - printf("Pid: %d \t LLC_value: %lu\n", bm_pid, - llc_value); + printf("Pid: %d \t LLC_value: %llu\n", bm_pid, llc_value); } else { fp = fopen(filename, "a"); if (!fp) { @@ -162,7 +158,7 @@ static int print_results_cache(const char *filename, int bm_pid,
return errno; } - fprintf(fp, "Pid: %d \t llc_value: %lu\n", bm_pid, llc_value); + fprintf(fp, "Pid: %d \t llc_value: %llu\n", bm_pid, llc_value); fclose(fp); }
@@ -182,7 +178,7 @@ static int print_results_cache(const char *filename, int bm_pid, */ static int perf_event_measure(const char *filename, int bm_pid) { - unsigned long llc_perf_miss = 0; + __u64 llc_perf_miss = 0; int ret;
ret = get_llc_perf(&llc_perf_miss); @@ -287,11 +283,10 @@ int cat_val(struct resctrl_val_param *param, size_t span) * @cache_span: Cache span * @lines: @cache_span in lines or bytes */ -void show_cache_info(int no_of_bits, unsigned long avg_llc_val, - size_t cache_span, bool lines) +void show_cache_info(int no_of_bits, __u64 avg_llc_val, size_t cache_span, bool lines) { ksft_print_msg("Number of bits: %d\n", no_of_bits); - ksft_print_msg("Average LLC val: %lu\n", avg_llc_val); + ksft_print_msg("Average LLC val: %llu\n", avg_llc_val); ksft_print_msg("Cache span (%s): %zu\n", lines ? "lines" : "bytes", cache_span); } diff --git a/tools/testing/selftests/resctrl/cat_test.c b/tools/testing/selftests/resctrl/cat_test.c index 2842dff292f0..ca2dd46ec733 100644 --- a/tools/testing/selftests/resctrl/cat_test.c +++ b/tools/testing/selftests/resctrl/cat_test.c @@ -41,12 +41,12 @@ static int cat_setup(struct resctrl_val_param *p) return ret; }
-static int show_results_info(unsigned long sum_llc_val, int no_of_bits, +static int show_results_info(__u64 sum_llc_val, int no_of_bits, unsigned long cache_span, unsigned long max_diff, unsigned long max_diff_percent, unsigned long num_of_runs, bool platform) { - unsigned long avg_llc_val = 0; + __u64 avg_llc_val = 0; float diff_percent; int ret;
@@ -68,8 +68,8 @@ static int show_results_info(unsigned long sum_llc_val, int no_of_bits, static int check_results(struct resctrl_val_param *param, size_t span) { char *token_array[8], temp[512]; - unsigned long sum_llc_perf_miss = 0; int runs = 0, no_of_bits = 0; + __u64 sum_llc_perf_miss = 0; FILE *fp;
ksft_print_msg("Checking for pass/fail\n"); @@ -93,7 +93,7 @@ static int check_results(struct resctrl_val_param *param, size_t span) * setup transition phase. */ if (runs > 0) - sum_llc_perf_miss += strtoul(token_array[3], NULL, 0); + sum_llc_perf_miss += strtoull(token_array[3], NULL, 0); runs++; }
diff --git a/tools/testing/selftests/resctrl/resctrl.h b/tools/testing/selftests/resctrl/resctrl.h index 4a05efccab35..e5a7b02a6a54 100644 --- a/tools/testing/selftests/resctrl/resctrl.h +++ b/tools/testing/selftests/resctrl/resctrl.h @@ -115,8 +115,7 @@ unsigned int count_bits(unsigned long n); void cmt_test_cleanup(void); int get_core_sibling(int cpu_no); int measure_llc_resctrl(const char *filename, int bm_pid); -void show_cache_info(int no_of_bits, unsigned long avg_llc_val, - size_t cache_span, bool lines); +void show_cache_info(int no_of_bits, __u64 avg_llc_val, size_t cache_span, bool lines);
/* * cache_portion_size - Calculate the size of a cache portion
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
Perf counters are __u64 but the code converts them to unsigned long before printing them out.
Remove unnecessary type conversion and retain the value as __u64 whole the time.
"whole the time" -> "the whole time"
Even so, this does not seem quite accurate since not all callers of show_cache_info() and print_results_cache() use __u64. This seems ok but it would be nicer if this was highlighted and not use "whole time".
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
Reviewed-by: Reinette Chatre reinette.chatre@intel.com
Reinette
Perf event handling has functions that are the sole caller of another perf event handling related function: - reset_enable_llc_perf() calls perf_event_open_llc_miss() - perf_event_measure() calls get_llc_perf()
Remove the extra layer of calls to make the code easier to follow by moving the code into the calling function.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/cache.c | 65 ++++++------------------- 1 file changed, 14 insertions(+), 51 deletions(-)
diff --git a/tools/testing/selftests/resctrl/cache.c b/tools/testing/selftests/resctrl/cache.c index 6a754ea376c9..aea4f54d32bd 100644 --- a/tools/testing/selftests/resctrl/cache.c +++ b/tools/testing/selftests/resctrl/cache.c @@ -29,24 +29,13 @@ static void initialize_perf_event_attr(void) pea_llc_miss.disabled = 1; }
+/* Start counters to log values */ static void ioctl_perf_event_ioc_reset_enable(void) { ioctl(fd_lm, PERF_EVENT_IOC_RESET, 0); ioctl(fd_lm, PERF_EVENT_IOC_ENABLE, 0); }
-static int perf_event_open_llc_miss(pid_t pid, int cpu_no) -{ - fd_lm = perf_event_open(&pea_llc_miss, pid, cpu_no, -1, - PERF_FLAG_FD_CLOEXEC); - if (fd_lm == -1) { - perror("Error opening leader"); - return -1; - } - - return 0; -} - static void initialize_llc_perf(void) { memset(&pea_llc_miss, 0, sizeof(struct perf_event_attr)); @@ -62,42 +51,13 @@ static void initialize_llc_perf(void)
static int reset_enable_llc_perf(pid_t pid, int cpu_no) { - int ret = 0; - - ret = perf_event_open_llc_miss(pid, cpu_no); - if (ret < 0) - return ret; - - /* Start counters to log values */ - ioctl_perf_event_ioc_reset_enable(); - - return 0; -} - -/* - * get_llc_perf: llc cache miss through perf events - * @llc_perf_miss: LLC miss counter that is filled on success - * - * Perf events like HW_CACHE_MISSES could be used to validate number of - * cache lines allocated. - * - * Return: =0 on success. <0 on failure. - */ -static int get_llc_perf(__u64 *llc_perf_miss) -{ - int ret; - - /* Stop counters after one span to get miss rate */ - - ioctl(fd_lm, PERF_EVENT_IOC_DISABLE, 0); - - ret = read(fd_lm, &rf_cqm, sizeof(struct read_format)); - if (ret == -1) { - perror("Could not get llc misses through perf"); + fd_lm = perf_event_open(&pea_llc_miss, pid, cpu_no, -1, PERF_FLAG_FD_CLOEXEC); + if (fd_lm == -1) { + perror("Error opening leader"); return -1; }
- *llc_perf_miss = rf_cqm.values[0].value; + ioctl_perf_event_ioc_reset_enable();
return 0; } @@ -178,15 +138,18 @@ static int print_results_cache(const char *filename, int bm_pid, __u64 llc_value */ static int perf_event_measure(const char *filename, int bm_pid) { - __u64 llc_perf_miss = 0; int ret;
- ret = get_llc_perf(&llc_perf_miss); - if (ret < 0) - return ret; + /* Stop counters after one span to get miss rate */ + ioctl(fd_lm, PERF_EVENT_IOC_DISABLE, 0);
- ret = print_results_cache(filename, bm_pid, llc_perf_miss); - return ret; + ret = read(fd_lm, &rf_cqm, sizeof(struct read_format)); + if (ret == -1) { + perror("Could not get perf value"); + return -1; + } + + return print_results_cache(filename, bm_pid, rf_cqm.values[0].value); }
/*
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
Perf event handling has functions that are the sole caller of another perf event handling related function:
- reset_enable_llc_perf() calls perf_event_open_llc_miss()
- perf_event_measure() calls get_llc_perf()
Remove the extra layer of calls to make the code easier to follow by moving the code into the calling function.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
Reviewed-by: Reinette Chatre reinette.chatre@intel.com
Reinette
Naming for perf event related functions, types, and variables is currently inconsistent.
Make struct read_format and all functions related to perf events start with "perf_". Adjust variable names towards the same direction but use shorter names for variables where appropriate (pe prefix).
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/cache.c | 42 ++++++++++++------------- 1 file changed, 21 insertions(+), 21 deletions(-)
diff --git a/tools/testing/selftests/resctrl/cache.c b/tools/testing/selftests/resctrl/cache.c index aea4f54d32bd..52d37d1c380f 100644 --- a/tools/testing/selftests/resctrl/cache.c +++ b/tools/testing/selftests/resctrl/cache.c @@ -3,7 +3,7 @@ #include <stdint.h> #include "resctrl.h"
-struct read_format { +struct perf_event_read { __u64 nr; /* The number of events */ struct { __u64 value; /* The value of the event */ @@ -11,11 +11,11 @@ struct read_format { };
static struct perf_event_attr pea_llc_miss; -static struct read_format rf_cqm; -static int fd_lm; +static struct perf_event_read pe_read; +static int pe_fd; char llc_occup_path[1024];
-static void initialize_perf_event_attr(void) +static void perf_event_attr_initialize(void) { pea_llc_miss.type = PERF_TYPE_HARDWARE; pea_llc_miss.size = sizeof(struct perf_event_attr); @@ -30,34 +30,34 @@ static void initialize_perf_event_attr(void) }
/* Start counters to log values */ -static void ioctl_perf_event_ioc_reset_enable(void) +static void perf_event_reset_enable(void) { - ioctl(fd_lm, PERF_EVENT_IOC_RESET, 0); - ioctl(fd_lm, PERF_EVENT_IOC_ENABLE, 0); + ioctl(pe_fd, PERF_EVENT_IOC_RESET, 0); + ioctl(pe_fd, PERF_EVENT_IOC_ENABLE, 0); }
-static void initialize_llc_perf(void) +static void perf_event_initialize(void) { memset(&pea_llc_miss, 0, sizeof(struct perf_event_attr)); - memset(&rf_cqm, 0, sizeof(struct read_format)); + memset(&pe_read, 0, sizeof(struct perf_event_read));
/* Initialize perf_event_attr structures for HW_CACHE_MISSES */ - initialize_perf_event_attr(); + perf_event_attr_initialize();
pea_llc_miss.config = PERF_COUNT_HW_CACHE_MISSES;
- rf_cqm.nr = 1; + pe_read.nr = 1; }
-static int reset_enable_llc_perf(pid_t pid, int cpu_no) +static int perf_open(pid_t pid, int cpu_no) { - fd_lm = perf_event_open(&pea_llc_miss, pid, cpu_no, -1, PERF_FLAG_FD_CLOEXEC); - if (fd_lm == -1) { + pe_fd = perf_event_open(&pea_llc_miss, pid, cpu_no, -1, PERF_FLAG_FD_CLOEXEC); + if (pe_fd == -1) { perror("Error opening leader"); return -1; }
- ioctl_perf_event_ioc_reset_enable(); + perf_event_reset_enable();
return 0; } @@ -141,15 +141,15 @@ static int perf_event_measure(const char *filename, int bm_pid) int ret;
/* Stop counters after one span to get miss rate */ - ioctl(fd_lm, PERF_EVENT_IOC_DISABLE, 0); + ioctl(pe_fd, PERF_EVENT_IOC_DISABLE, 0);
- ret = read(fd_lm, &rf_cqm, sizeof(struct read_format)); + ret = read(pe_fd, &pe_read, sizeof(struct perf_event_read)); if (ret == -1) { perror("Could not get perf value"); return -1; }
- return print_results_cache(filename, bm_pid, rf_cqm.values[0].value); + return print_results_cache(filename, bm_pid, pe_read.values[0].value); }
/* @@ -205,7 +205,7 @@ int cat_val(struct resctrl_val_param *param, size_t span) if (ret) return ret;
- initialize_llc_perf(); + perf_event_initialize();
/* Test runs until the callback setup() tells the test to stop. */ while (1) { @@ -216,7 +216,7 @@ int cat_val(struct resctrl_val_param *param, size_t span) } if (ret < 0) break; - ret = reset_enable_llc_perf(bm_pid, param->cpu_no); + ret = perf_open(bm_pid, param->cpu_no); if (ret) break;
@@ -235,7 +235,7 @@ int cat_val(struct resctrl_val_param *param, size_t span) return ret;
pe_close: - close(fd_lm); + close(pe_fd); return ret; }
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
Naming for perf event related functions, types, and variables is currently inconsistent.
(nitpick ... instead if "currently inconsistent" this can just be "inconsistent")
Make struct read_format and all functions related to perf events start with "perf_". Adjust variable names towards the same direction but use shorter names for variables where appropriate (pe prefix).
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
Reviewed-by: Reinette Chatre reinette.chatre@intel.com
Reinette
struct perf_event_attr initialization is spread into perf_event_initialize() and perf_event_attr_initialize() and setting ->config is hardcoded by the deepest level.
perf_event_attr init belongs to perf_event_attr_initialize() so move it entirely there. Rename the other function perf_event_initialized_read_format().
Call each init function directly from the test as they will take different parameters (especially true after the perf related global variables are moved to local variables).
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/cache.c | 16 ++++++---------- 1 file changed, 6 insertions(+), 10 deletions(-)
diff --git a/tools/testing/selftests/resctrl/cache.c b/tools/testing/selftests/resctrl/cache.c index 52d37d1c380f..315f30f23371 100644 --- a/tools/testing/selftests/resctrl/cache.c +++ b/tools/testing/selftests/resctrl/cache.c @@ -15,8 +15,9 @@ static struct perf_event_read pe_read; static int pe_fd; char llc_occup_path[1024];
-static void perf_event_attr_initialize(void) +static void perf_event_attr_initialize(__u64 config) { + memset(&pea_llc_miss, 0, sizeof(struct perf_event_attr)); pea_llc_miss.type = PERF_TYPE_HARDWARE; pea_llc_miss.size = sizeof(struct perf_event_attr); pea_llc_miss.read_format = PERF_FORMAT_GROUP; @@ -27,6 +28,7 @@ static void perf_event_attr_initialize(void) pea_llc_miss.inherit = 1; pea_llc_miss.exclude_guest = 1; pea_llc_miss.disabled = 1; + pea_llc_miss.config = config; }
/* Start counters to log values */ @@ -36,16 +38,9 @@ static void perf_event_reset_enable(void) ioctl(pe_fd, PERF_EVENT_IOC_ENABLE, 0); }
-static void perf_event_initialize(void) +static void perf_event_initialize_read_format(void) { - memset(&pea_llc_miss, 0, sizeof(struct perf_event_attr)); memset(&pe_read, 0, sizeof(struct perf_event_read)); - - /* Initialize perf_event_attr structures for HW_CACHE_MISSES */ - perf_event_attr_initialize(); - - pea_llc_miss.config = PERF_COUNT_HW_CACHE_MISSES; - pe_read.nr = 1; }
@@ -205,7 +200,8 @@ int cat_val(struct resctrl_val_param *param, size_t span) if (ret) return ret;
- perf_event_initialize(); + perf_event_attr_initialize(PERF_COUNT_HW_CACHE_MISSES); + perf_event_initialize_read_format();
/* Test runs until the callback setup() tells the test to stop. */ while (1) {
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
struct perf_event_attr initialization is spread into perf_event_initialize() and perf_event_attr_initialize() and setting ->config is hardcoded by the deepest level.
perf_event_attr init belongs to perf_event_attr_initialize() so move it entirely there. Rename the other function perf_event_initialized_read_format().
Call each init function directly from the test as they will take different parameters (especially true after the perf related global variables are moved to local variables).
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
Reviewed-by: Reinette Chatre reinette.chatre@intel.com
Reinette
Perf related variables pea_llc_miss, pe_read, and pe_fd are globals in cache.c.
Convert them to locals for better scoping and make pea_llc_miss simpler by renaming it to pea. Make close(pe_fd) handling easier to understand by doing it inside cat_val().
Make also sizeof()s use safer way to determine the right struct.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/cache.c | 72 ++++++++++++++----------- 1 file changed, 40 insertions(+), 32 deletions(-)
diff --git a/tools/testing/selftests/resctrl/cache.c b/tools/testing/selftests/resctrl/cache.c index 315f30f23371..a69e7151db94 100644 --- a/tools/testing/selftests/resctrl/cache.c +++ b/tools/testing/selftests/resctrl/cache.c @@ -10,51 +10,50 @@ struct perf_event_read { } values[2]; };
-static struct perf_event_attr pea_llc_miss; -static struct perf_event_read pe_read; -static int pe_fd; char llc_occup_path[1024];
-static void perf_event_attr_initialize(__u64 config) +static void perf_event_attr_initialize(struct perf_event_attr *pea, __u64 config) { - memset(&pea_llc_miss, 0, sizeof(struct perf_event_attr)); - pea_llc_miss.type = PERF_TYPE_HARDWARE; - pea_llc_miss.size = sizeof(struct perf_event_attr); - pea_llc_miss.read_format = PERF_FORMAT_GROUP; - pea_llc_miss.exclude_kernel = 1; - pea_llc_miss.exclude_hv = 1; - pea_llc_miss.exclude_idle = 1; - pea_llc_miss.exclude_callchain_kernel = 1; - pea_llc_miss.inherit = 1; - pea_llc_miss.exclude_guest = 1; - pea_llc_miss.disabled = 1; - pea_llc_miss.config = config; + memset(pea, 0, sizeof(*pea)); + pea->type = PERF_TYPE_HARDWARE; + pea->size = sizeof(*pea); + pea->read_format = PERF_FORMAT_GROUP; + pea->exclude_kernel = 1; + pea->exclude_hv = 1; + pea->exclude_idle = 1; + pea->exclude_callchain_kernel = 1; + pea->inherit = 1; + pea->exclude_guest = 1; + pea->disabled = 1; + pea->config = config; }
/* Start counters to log values */ -static void perf_event_reset_enable(void) +static void perf_event_reset_enable(int pe_fd) { ioctl(pe_fd, PERF_EVENT_IOC_RESET, 0); ioctl(pe_fd, PERF_EVENT_IOC_ENABLE, 0); }
-static void perf_event_initialize_read_format(void) +static void perf_event_initialize_read_format(struct perf_event_read *pe_read) { - memset(&pe_read, 0, sizeof(struct perf_event_read)); - pe_read.nr = 1; + memset(pe_read, 0, sizeof(*pe_read)); + pe_read->nr = 1; }
-static int perf_open(pid_t pid, int cpu_no) +static int perf_open(struct perf_event_attr *pea, pid_t pid, int cpu_no) { - pe_fd = perf_event_open(&pea_llc_miss, pid, cpu_no, -1, PERF_FLAG_FD_CLOEXEC); + int pe_fd; + + pe_fd = perf_event_open(pea, pid, cpu_no, -1, PERF_FLAG_FD_CLOEXEC); if (pe_fd == -1) { perror("Error opening leader"); return -1; }
- perf_event_reset_enable(); + perf_event_reset_enable(pe_fd);
- return 0; + return pe_fd; }
/* @@ -131,20 +130,21 @@ static int print_results_cache(const char *filename, int bm_pid, __u64 llc_value * * Return: =0 on success. <0 on failure. */ -static int perf_event_measure(const char *filename, int bm_pid) +static int perf_event_measure(int pe_fd, struct perf_event_read *pe_read, + const char *filename, int bm_pid) { int ret;
/* Stop counters after one span to get miss rate */ ioctl(pe_fd, PERF_EVENT_IOC_DISABLE, 0);
- ret = read(pe_fd, &pe_read, sizeof(struct perf_event_read)); + ret = read(pe_fd, pe_read, sizeof(*pe_read)); if (ret == -1) { perror("Could not get perf value"); return -1; }
- return print_results_cache(filename, bm_pid, pe_read.values[0].value); + return print_results_cache(filename, bm_pid, pe_read->values[0].value); }
/* @@ -182,7 +182,10 @@ int cat_val(struct resctrl_val_param *param, size_t span) { int memflush = 1, operation = 0, ret = 0; char *resctrl_val = param->resctrl_val; + struct perf_event_read pe_read; + struct perf_event_attr pea; pid_t bm_pid; + int pe_fd;
if (strcmp(param->filename, "") == 0) sprintf(param->filename, "stdio"); @@ -200,8 +203,8 @@ int cat_val(struct resctrl_val_param *param, size_t span) if (ret) return ret;
- perf_event_attr_initialize(PERF_COUNT_HW_CACHE_MISSES); - perf_event_initialize_read_format(); + perf_event_attr_initialize(&pea, PERF_COUNT_HW_CACHE_MISSES); + perf_event_initialize_read_format(&pe_read);
/* Test runs until the callback setup() tells the test to stop. */ while (1) { @@ -212,9 +215,12 @@ int cat_val(struct resctrl_val_param *param, size_t span) } if (ret < 0) break; - ret = perf_open(bm_pid, param->cpu_no); - if (ret) + + pe_fd = perf_open(&pea, bm_pid, param->cpu_no); + if (pe_fd < 0) { + ret = -1; break; + }
if (run_fill_buf(span, memflush, operation, true)) { fprintf(stderr, "Error-running fill buffer\n"); @@ -223,9 +229,11 @@ int cat_val(struct resctrl_val_param *param, size_t span) }
sleep(1); - ret = perf_event_measure(param->filename, bm_pid); + ret = perf_event_measure(pe_fd, &pe_read, param->filename, bm_pid); if (ret) goto pe_close; + + close(pe_fd); }
return ret;
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
Perf related variables pea_llc_miss, pe_read, and pe_fd are globals in cache.c.
Convert them to locals for better scoping and make pea_llc_miss simpler by renaming it to pea. Make close(pe_fd) handling easier to understand by doing it inside cat_val().
Make also sizeof()s use safer way to determine the right struct.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
Reviewed-by: Reinette Chatre reinette.chatre@intel.com
Reinette
The main CAT test function is called cat_val() and resides in cache.c which is illogical.
Rename the function to cat_test() and move it into cat_test.c.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/cache.c | 90 ++-------------------- tools/testing/selftests/resctrl/cat_test.c | 73 +++++++++++++++++- tools/testing/selftests/resctrl/resctrl.h | 14 +++- 3 files changed, 90 insertions(+), 87 deletions(-)
diff --git a/tools/testing/selftests/resctrl/cache.c b/tools/testing/selftests/resctrl/cache.c index a69e7151db94..29d70fc11f4e 100644 --- a/tools/testing/selftests/resctrl/cache.c +++ b/tools/testing/selftests/resctrl/cache.c @@ -3,16 +3,9 @@ #include <stdint.h> #include "resctrl.h"
-struct perf_event_read { - __u64 nr; /* The number of events */ - struct { - __u64 value; /* The value of the event */ - } values[2]; -}; - char llc_occup_path[1024];
-static void perf_event_attr_initialize(struct perf_event_attr *pea, __u64 config) +void perf_event_attr_initialize(struct perf_event_attr *pea, __u64 config) { memset(pea, 0, sizeof(*pea)); pea->type = PERF_TYPE_HARDWARE; @@ -35,13 +28,13 @@ static void perf_event_reset_enable(int pe_fd) ioctl(pe_fd, PERF_EVENT_IOC_ENABLE, 0); }
-static void perf_event_initialize_read_format(struct perf_event_read *pe_read) +void perf_event_initialize_read_format(struct perf_event_read *pe_read) { memset(pe_read, 0, sizeof(*pe_read)); pe_read->nr = 1; }
-static int perf_open(struct perf_event_attr *pea, pid_t pid, int cpu_no) +int perf_open(struct perf_event_attr *pea, pid_t pid, int cpu_no) { int pe_fd;
@@ -130,8 +123,8 @@ static int print_results_cache(const char *filename, int bm_pid, __u64 llc_value * * Return: =0 on success. <0 on failure. */ -static int perf_event_measure(int pe_fd, struct perf_event_read *pe_read, - const char *filename, int bm_pid) +int perf_event_measure(int pe_fd, struct perf_event_read *pe_read, + const char *filename, int bm_pid) { int ret;
@@ -170,79 +163,6 @@ int measure_llc_resctrl(const char *filename, int bm_pid) return ret; }
-/* - * cache_val: execute benchmark and measure LLC occupancy resctrl - * and perf cache miss for the benchmark - * @param: parameters passed to cache_val() - * @span: buffer size for the benchmark - * - * Return: 0 on success. non-zero on failure. - */ -int cat_val(struct resctrl_val_param *param, size_t span) -{ - int memflush = 1, operation = 0, ret = 0; - char *resctrl_val = param->resctrl_val; - struct perf_event_read pe_read; - struct perf_event_attr pea; - pid_t bm_pid; - int pe_fd; - - if (strcmp(param->filename, "") == 0) - sprintf(param->filename, "stdio"); - - bm_pid = getpid(); - - /* Taskset benchmark to specified cpu */ - ret = taskset_benchmark(bm_pid, param->cpu_no); - if (ret) - return ret; - - /* Write benchmark to specified con_mon grp, mon_grp in resctrl FS*/ - ret = write_bm_pid_to_resctrl(bm_pid, param->ctrlgrp, param->mongrp, - resctrl_val); - if (ret) - return ret; - - perf_event_attr_initialize(&pea, PERF_COUNT_HW_CACHE_MISSES); - perf_event_initialize_read_format(&pe_read); - - /* Test runs until the callback setup() tells the test to stop. */ - while (1) { - ret = param->setup(param); - if (ret == END_OF_TESTS) { - ret = 0; - break; - } - if (ret < 0) - break; - - pe_fd = perf_open(&pea, bm_pid, param->cpu_no); - if (pe_fd < 0) { - ret = -1; - break; - } - - if (run_fill_buf(span, memflush, operation, true)) { - fprintf(stderr, "Error-running fill buffer\n"); - ret = -1; - goto pe_close; - } - - sleep(1); - ret = perf_event_measure(pe_fd, &pe_read, param->filename, bm_pid); - if (ret) - goto pe_close; - - close(pe_fd); - } - - return ret; - -pe_close: - close(pe_fd); - return ret; -} - /* * show_cache_info - Show generic cache test information * @no_of_bits: Number of bits diff --git a/tools/testing/selftests/resctrl/cat_test.c b/tools/testing/selftests/resctrl/cat_test.c index ca2dd46ec733..7e1c383f9119 100644 --- a/tools/testing/selftests/resctrl/cat_test.c +++ b/tools/testing/selftests/resctrl/cat_test.c @@ -111,6 +111,77 @@ void cat_test_cleanup(void) remove(RESULT_FILE_NAME2); }
+/* + * cat_test - Execute CAT benchmark and measure cache misses + * @param: Parameters passed to cat_test() + * @span: Buffer size for the benchmark + * + * Return: 0 on success. Non-zero on failure. + */ +static int cat_test(struct resctrl_val_param *param, size_t span) +{ + int memflush = 1, operation = 0, ret = 0; + char *resctrl_val = param->resctrl_val; + struct perf_event_read pe_read; + struct perf_event_attr pea; + pid_t bm_pid; + int pe_fd; + + if (strcmp(param->filename, "") == 0) + sprintf(param->filename, "stdio"); + + bm_pid = getpid(); + + /* Taskset benchmark to specified cpu */ + ret = taskset_benchmark(bm_pid, param->cpu_no); + if (ret) + return ret; + + /* Write benchmark to specified con_mon grp, mon_grp in resctrl FS*/ + ret = write_bm_pid_to_resctrl(bm_pid, param->ctrlgrp, param->mongrp, + resctrl_val); + if (ret) + return ret; + + perf_event_attr_initialize(&pea, PERF_COUNT_HW_CACHE_MISSES); + perf_event_initialize_read_format(&pe_read); + + /* Test runs until the callback setup() tells the test to stop. */ + while (1) { + ret = param->setup(param); + if (ret == END_OF_TESTS) { + ret = 0; + break; + } + if (ret < 0) + break; + pe_fd = perf_open(&pea, bm_pid, param->cpu_no); + if (pe_fd < 0) { + ret = -1; + break; + } + + if (run_fill_buf(span, memflush, operation, true)) { + fprintf(stderr, "Error-running fill buffer\n"); + ret = -1; + goto pe_close; + } + + sleep(1); + ret = perf_event_measure(pe_fd, &pe_read, param->filename, bm_pid); + if (ret) + goto pe_close; + + close(pe_fd); + } + + return ret; + +pe_close: + close(pe_fd); + return ret; +} + int cat_perf_miss_val(int cpu_no, int n, char *cache_type) { unsigned long full_cache_mask, long_mask; @@ -194,7 +265,7 @@ int cat_perf_miss_val(int cpu_no, int n, char *cache_type)
remove(param.filename);
- ret = cat_val(¶m, span); + ret = cat_test(¶m, span); if (ret == 0) ret = check_results(¶m, span);
diff --git a/tools/testing/selftests/resctrl/resctrl.h b/tools/testing/selftests/resctrl/resctrl.h index e5a7b02a6a54..5626482b41cf 100644 --- a/tools/testing/selftests/resctrl/resctrl.h +++ b/tools/testing/selftests/resctrl/resctrl.h @@ -67,6 +67,13 @@ struct resctrl_val_param { int (*setup)(struct resctrl_val_param *param); };
+struct perf_event_read { + __u64 nr; /* The number of events */ + struct { + __u64 value; /* The value of the event */ + } values[2]; +}; + #define MBM_STR "mbm" #define MBA_STR "mba" #define CMT_STR "cmt" @@ -107,13 +114,18 @@ int get_cache_size(int cpu_no, const char *cache_type, unsigned long *cache_size void ctrlc_handler(int signum, siginfo_t *info, void *ptr); int signal_handler_register(void); void signal_handler_unregister(void); -int cat_val(struct resctrl_val_param *param, size_t span); void cat_test_cleanup(void); int cat_perf_miss_val(int cpu_no, int no_of_bits, char *cache_type); int cmt_resctrl_val(int cpu_no, int n, const char * const *benchmark_cmd); unsigned int count_bits(unsigned long n); void cmt_test_cleanup(void); int get_core_sibling(int cpu_no); + +void perf_event_attr_initialize(struct perf_event_attr *pea, __u64 config); +void perf_event_initialize_read_format(struct perf_event_read *pe_read); +int perf_open(struct perf_event_attr *pea, pid_t pid, int cpu_no); +int perf_event_measure(int pe_fd, struct perf_event_read *pe_read, + const char *filename, int bm_pid); int measure_llc_resctrl(const char *filename, int bm_pid); void show_cache_info(int no_of_bits, __u64 avg_llc_val, size_t cache_span, bool lines);
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
The main CAT test function is called cat_val() and resides in cache.c which is illogical.
Rename the function to cat_test() and move it into cat_test.c.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
Reviewed-by: Reinette Chatre reinette.chatre@intel.com
Reinette
Perf fd (pe_fd) is opened, reset, and enabled during every test the CAT selftest runs. Also, ioctl(pe_fd, ...) calls are not error checked even if ioctl() could return an error.
Open perf fd only once before the tests and only reset and enable the counter within the test loop. Add error checking to pe_fd ioctl() calls.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/cache.c | 19 +++++++++++++++---- tools/testing/selftests/resctrl/cat_test.c | 14 +++++++------- tools/testing/selftests/resctrl/resctrl.h | 1 + 3 files changed, 23 insertions(+), 11 deletions(-)
diff --git a/tools/testing/selftests/resctrl/cache.c b/tools/testing/selftests/resctrl/cache.c index 29d70fc11f4e..012da5952cd6 100644 --- a/tools/testing/selftests/resctrl/cache.c +++ b/tools/testing/selftests/resctrl/cache.c @@ -22,10 +22,19 @@ void perf_event_attr_initialize(struct perf_event_attr *pea, __u64 config) }
/* Start counters to log values */ -static void perf_event_reset_enable(int pe_fd) +int perf_event_reset_enable(int pe_fd) { - ioctl(pe_fd, PERF_EVENT_IOC_RESET, 0); - ioctl(pe_fd, PERF_EVENT_IOC_ENABLE, 0); + int ret; + + ret = ioctl(pe_fd, PERF_EVENT_IOC_RESET, 0); + if (ret < 0) + return ret; + + ret = ioctl(pe_fd, PERF_EVENT_IOC_ENABLE, 0); + if (ret < 0) + return ret; + + return 0; }
void perf_event_initialize_read_format(struct perf_event_read *pe_read) @@ -129,7 +138,9 @@ int perf_event_measure(int pe_fd, struct perf_event_read *pe_read, int ret;
/* Stop counters after one span to get miss rate */ - ioctl(pe_fd, PERF_EVENT_IOC_DISABLE, 0); + ret = ioctl(pe_fd, PERF_EVENT_IOC_DISABLE, 0); + if (ret < 0) + return ret;
ret = read(pe_fd, pe_read, sizeof(*pe_read)); if (ret == -1) { diff --git a/tools/testing/selftests/resctrl/cat_test.c b/tools/testing/selftests/resctrl/cat_test.c index 7e1c383f9119..cfda87667b46 100644 --- a/tools/testing/selftests/resctrl/cat_test.c +++ b/tools/testing/selftests/resctrl/cat_test.c @@ -145,6 +145,9 @@ static int cat_test(struct resctrl_val_param *param, size_t span)
perf_event_attr_initialize(&pea, PERF_COUNT_HW_CACHE_MISSES); perf_event_initialize_read_format(&pe_read); + pe_fd = perf_open(&pea, bm_pid, param->cpu_no); + if (pe_fd < 0) + return pe_fd;
/* Test runs until the callback setup() tells the test to stop. */ while (1) { @@ -155,11 +158,10 @@ static int cat_test(struct resctrl_val_param *param, size_t span) } if (ret < 0) break; - pe_fd = perf_open(&pea, bm_pid, param->cpu_no); - if (pe_fd < 0) { - ret = -1; - break; - } + + ret = perf_event_reset_enable(pe_fd); + if (ret) + goto pe_close;
if (run_fill_buf(span, memflush, operation, true)) { fprintf(stderr, "Error-running fill buffer\n"); @@ -171,8 +173,6 @@ static int cat_test(struct resctrl_val_param *param, size_t span) ret = perf_event_measure(pe_fd, &pe_read, param->filename, bm_pid); if (ret) goto pe_close; - - close(pe_fd); }
return ret; diff --git a/tools/testing/selftests/resctrl/resctrl.h b/tools/testing/selftests/resctrl/resctrl.h index 5626482b41cf..fe07b732cbd9 100644 --- a/tools/testing/selftests/resctrl/resctrl.h +++ b/tools/testing/selftests/resctrl/resctrl.h @@ -124,6 +124,7 @@ int get_core_sibling(int cpu_no); void perf_event_attr_initialize(struct perf_event_attr *pea, __u64 config); void perf_event_initialize_read_format(struct perf_event_read *pe_read); int perf_open(struct perf_event_attr *pea, pid_t pid, int cpu_no); +int perf_event_reset_enable(int pe_fd); int perf_event_measure(int pe_fd, struct perf_event_read *pe_read, const char *filename, int bm_pid); int measure_llc_resctrl(const char *filename, int bm_pid);
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
Perf fd (pe_fd) is opened, reset, and enabled during every test the CAT selftest runs. Also, ioctl(pe_fd, ...) calls are not error checked even if ioctl() could return an error.
Open perf fd only once before the tests and only reset and enable the counter within the test loop. Add error checking to pe_fd ioctl() calls.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
Reviewed-by: Reinette Chatre reinette.chatre@intel.com
Reinette
The fill_buf code prevents compiler optimizating the entire read loop away by writing the final value of the variable into a file. While it achieves the goal, writing into a file requires significant amount of work within the innermost test loop and also error handling.
A simpler approach is to take advantage of volatile. Writing to a variable through a volatile pointer is enough to prevent compiler from optimizing the write away, and therefore compiler cannot remove the read loop either.
Add a volatile 'value_sink' into resctrl_tests.c and make fill_buf to write into it. As a result, the error handling in fill_buf.c can be simplified.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/fill_buf.c | 26 ++++--------------- tools/testing/selftests/resctrl/resctrl.h | 7 +++++ .../testing/selftests/resctrl/resctrl_tests.c | 4 +++ 3 files changed, 16 insertions(+), 21 deletions(-)
diff --git a/tools/testing/selftests/resctrl/fill_buf.c b/tools/testing/selftests/resctrl/fill_buf.c index b9303a9d819b..8fe9574db9d8 100644 --- a/tools/testing/selftests/resctrl/fill_buf.c +++ b/tools/testing/selftests/resctrl/fill_buf.c @@ -78,10 +78,9 @@ static void fill_one_span_write(unsigned char *buf, size_t buf_size) } }
-static int fill_cache_read(unsigned char *buf, size_t buf_size, bool once) +static void fill_cache_read(unsigned char *buf, size_t buf_size, bool once) { int ret = 0; - FILE *fp;
while (1) { ret = fill_one_span_read(buf, buf_size); @@ -90,26 +89,16 @@ static int fill_cache_read(unsigned char *buf, size_t buf_size, bool once) }
/* Consume read result so that reading memory is not optimized out. */ - fp = fopen("/dev/null", "w"); - if (!fp) { - perror("Unable to write to /dev/null"); - return -1; - } - fprintf(fp, "Sum: %d ", ret); - fclose(fp); - - return 0; + *value_sink = ret; }
-static int fill_cache_write(unsigned char *buf, size_t buf_size, bool once) +static void fill_cache_write(unsigned char *buf, size_t buf_size, bool once) { while (1) { fill_one_span_write(buf, buf_size); if (once) break; } - - return 0; }
static unsigned char *alloc_buffer(size_t buf_size, int memflush) @@ -143,21 +132,16 @@ static unsigned char *alloc_buffer(size_t buf_size, int memflush) int run_fill_buf(size_t buf_size, int memflush, int op, bool once) { unsigned char *buf; - int ret;
buf = alloc_buffer(buf_size, memflush); if (!buf) return -1;
if (op == 0) - ret = fill_cache_read(buf, buf_size, once); + fill_cache_read(buf, buf_size, once); else - ret = fill_cache_write(buf, buf_size, once); + fill_cache_write(buf, buf_size, once); free(buf); - if (ret) { - printf("\n Error in fill cache\n"); - return -1; - }
return 0; } diff --git a/tools/testing/selftests/resctrl/resctrl.h b/tools/testing/selftests/resctrl/resctrl.h index fe07b732cbd9..68eb1f1bac3a 100644 --- a/tools/testing/selftests/resctrl/resctrl.h +++ b/tools/testing/selftests/resctrl/resctrl.h @@ -79,6 +79,13 @@ struct perf_event_read { #define CMT_STR "cmt" #define CAT_STR "cat"
+/* + * Memory location that consumes values compiler must not optimize away. + * Volatile ensures writes to this location cannot be optimized away by + * compiler. + */ +extern volatile int *value_sink; + extern pid_t bm_pid, ppid;
extern char llc_occup_path[1024]; diff --git a/tools/testing/selftests/resctrl/resctrl_tests.c b/tools/testing/selftests/resctrl/resctrl_tests.c index 2bbe3045a018..f0d573aa363b 100644 --- a/tools/testing/selftests/resctrl/resctrl_tests.c +++ b/tools/testing/selftests/resctrl/resctrl_tests.c @@ -10,6 +10,10 @@ */ #include "resctrl.h"
+/* Volatile memory sink to prevent compiler optimizations */ +static volatile int sink_target; +volatile int *value_sink = &sink_target; + static int detect_vendor(void) { FILE *inf = fopen("/proc/cpuinfo", "r");
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
The fill_buf code prevents compiler optimizating the entire read loop away by writing the final value of the variable into a file. While it achieves the goal, writing into a file requires significant amount of work within the innermost test loop and also error handling.
A simpler approach is to take advantage of volatile. Writing to a variable through a volatile pointer is enough to prevent compiler from optimizing the write away, and therefore compiler cannot remove the read loop either.
Add a volatile 'value_sink' into resctrl_tests.c and make fill_buf to write into it. As a result, the error handling in fill_buf.c can be simplified.
The subject and changelog describes the need for a volatile variable. The patch introduces two volatile variables. Could you please elaborate why two volatile variables are needed?
Reinette
On Tue, 28 Nov 2023, Reinette Chatre wrote:
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
The fill_buf code prevents compiler optimizating the entire read loop away by writing the final value of the variable into a file. While it achieves the goal, writing into a file requires significant amount of work within the innermost test loop and also error handling.
A simpler approach is to take advantage of volatile. Writing to a variable through a volatile pointer is enough to prevent compiler from optimizing the write away, and therefore compiler cannot remove the read loop either.
Add a volatile 'value_sink' into resctrl_tests.c and make fill_buf to write into it. As a result, the error handling in fill_buf.c can be simplified.
The subject and changelog describes the need for a volatile variable. The patch introduces two volatile variables. Could you please elaborate why two volatile variables are needed?
Well, the other "volatile variable" is a pointer to a volatile variable.
I've seen gcc to kill a static volatile int so I prefer to not take change with its optimizer. Thus, I placed the sink into different compilation unit and just present a pointer to the actual "volatile" variable.
I guess the sink could be marked as plain int instead but this being trickery to begin with I don't see much value either way. It's still a trick.
I'll alter the changelog's wording though, "a volatile variable" is not accurate as it's "a pointer to a volatile variable".
When reading memory in order, HW prefetching optimizations will interfere with measuring how caches and memory are being accessed. This adds noise into the results.
Change the fill_buf reading loop to not use an obvious in-order access using multiply by a prime and modulo.
Using a prime multiplier with modulo ensures the entire buffer is eventually read. 23 is small enough that the reads are spread out but wrapping does not occur very frequently (wrapping too often can trigger L2 hits more frequently which causes noise to the test because getting the data from LLC is not required).
It was discovered that not all primes work equally well and some can cause wildly unstable results (e.g., in an earlier version of this patch, the reads were done in reversed order and 59 was used as the prime resulting in unacceptably high and unstable results in MBA and MBM test on some architectures).
Link: https://lore.kernel.org/linux-kselftest/TYAPR01MB6330025B5E6537F94DA49ACB8B4... Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/fill_buf.c | 38 +++++++++++++++++----- 1 file changed, 30 insertions(+), 8 deletions(-)
diff --git a/tools/testing/selftests/resctrl/fill_buf.c b/tools/testing/selftests/resctrl/fill_buf.c index 8fe9574db9d8..93a3d408339c 100644 --- a/tools/testing/selftests/resctrl/fill_buf.c +++ b/tools/testing/selftests/resctrl/fill_buf.c @@ -51,16 +51,38 @@ static void mem_flush(unsigned char *buf, size_t buf_size) sb(); }
+/* + * Buffer index step advance to workaround HW prefetching interfering with + * the measurements. + * + * Must be a prime to step through all indexes of the buffer. + * + * Some primes work better than others on some architectures (from MBA/MBM + * result stability point of view). + */ +#define FILL_IDX_MULT 23 + static int fill_one_span_read(unsigned char *buf, size_t buf_size) { - unsigned char *end_ptr = buf + buf_size; - unsigned char sum, *p; - - sum = 0; - p = buf; - while (p < end_ptr) { - sum += *p; - p += (CL_SIZE / 2); + unsigned int size = buf_size / (CL_SIZE / 2); + unsigned int i, idx = 0; + unsigned char sum = 0; + + /* + * Read the buffer in an order that is unexpected by HW prefetching + * optimizations to prevent them interfering with the caching pattern. + * + * The read order is (in terms of halves of cachelines): + * i * FILL_IDX_MULT % size + * The formula is open-coded below to avoiding modulo inside the loop + * as it improves MBA/MBM result stability on some architectures. + */ + for (i = 0; i < size; i++) { + sum += buf[idx * (CL_SIZE / 2)]; + + idx += FILL_IDX_MULT; + while (idx >= size) + idx -= size; }
return sum;
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
When reading memory in order, HW prefetching optimizations will interfere with measuring how caches and memory are being accessed. This adds noise into the results.
Change the fill_buf reading loop to not use an obvious in-order access using multiply by a prime and modulo.
Using a prime multiplier with modulo ensures the entire buffer is eventually read. 23 is small enough that the reads are spread out but wrapping does not occur very frequently (wrapping too often can trigger L2 hits more frequently which causes noise to the test because getting the data from LLC is not required).
It was discovered that not all primes work equally well and some can cause wildly unstable results (e.g., in an earlier version of this patch, the reads were done in reversed order and 59 was used as the prime resulting in unacceptably high and unstable results in MBA and MBM test on some architectures).
Link: https://lore.kernel.org/linux-kselftest/TYAPR01MB6330025B5E6537F94DA49ACB8B4... Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
I am not very comfortable with all the uncertainty involved in this patch. A consolation is that this is surely an improvement.
Reviewed-by: Reinette Chatre reinette.chatre@intel.com
Reinette
CAT test spawns two processes into two different control groups with exclusive schemata. Both the processes alloc a buffer from memory matching their allocated LLC block size and flush the entire buffer out of caches. Since the processes are reading through the buffer only once during the measurement and initially all the buffer was flushed, the test isn't testing CAT.
Rewrite the CAT test to allocate a buffer sized to half of LLC. Then perform a sequence of tests with different LLC alloc sizes starting from half of the CBM bits down to 1-bit CBM. Flush the buffer before each test and read the buffer twice. Observe the LLC misses on the second read through the buffer. As the allocated LLC block gets smaller and smaller, the LLC misses will become larger and larger giving a strong signal on CAT working properly.
The new CAT test is using only a single process because it relies on measured effect against another run of itself rather than another process adding noise. The rest of the system is set to use the CBM bits not used by the CAT test to keep the test isolated.
Replace count_bits() with count_contiguous_bits() to get the first bit position in order to be able to calculate masks based on it.
This change has been tested with a number of systems from different generations.
Suggested-by: Reinette Chatre reinette.chatre@intel.com Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/cat_test.c | 282 +++++++++----------- tools/testing/selftests/resctrl/fill_buf.c | 6 +- tools/testing/selftests/resctrl/resctrl.h | 5 +- tools/testing/selftests/resctrl/resctrlfs.c | 44 +-- 4 files changed, 138 insertions(+), 199 deletions(-)
diff --git a/tools/testing/selftests/resctrl/cat_test.c b/tools/testing/selftests/resctrl/cat_test.c index cfda87667b46..4169b17b8f91 100644 --- a/tools/testing/selftests/resctrl/cat_test.c +++ b/tools/testing/selftests/resctrl/cat_test.c @@ -11,65 +11,69 @@ #include "resctrl.h" #include <unistd.h>
-#define RESULT_FILE_NAME1 "result_cat1" -#define RESULT_FILE_NAME2 "result_cat2" +#define RESULT_FILE_NAME "result_cat" #define NUM_OF_RUNS 5 -#define MAX_DIFF_PERCENT 4 -#define MAX_DIFF 1000000
/* - * Change schemata. Write schemata to specified - * con_mon grp, mon_grp in resctrl FS. - * Run 5 times in order to get average values. + * Minimum difference in LLC misses between a test with n+1 bits CBM to the + * test with n bits is MIN_DIFF_PERCENT_PER_BIT * (n - 1). With e.g. 5 vs 4 + * bits in the CBM mask, the minimum difference must be at least + * MIN_DIFF_PERCENT_PER_BIT * (4 - 1) = 3 percent. + * + * The relationship between number of used CBM bits and difference in LLC + * misses is not expected to be linear. With a small number of bits, the + * margin is smaller than with larger number of bits. For selftest purposes, + * however, linear approach is enough because ultimately only pass/fail + * decision has to be made and distinction between strong and stronger + * signal is irrelevant. */ -static int cat_setup(struct resctrl_val_param *p) -{ - char schemata[64]; - int ret = 0; - - /* Run NUM_OF_RUNS times */ - if (p->num_of_runs >= NUM_OF_RUNS) - return END_OF_TESTS; - - if (p->num_of_runs == 0) { - sprintf(schemata, "%lx", p->mask); - ret = write_schemata(p->ctrlgrp, schemata, p->cpu_no, - p->resctrl_val); - } - p->num_of_runs++; - - return ret; -} +#define MIN_DIFF_PERCENT_PER_BIT 1
static int show_results_info(__u64 sum_llc_val, int no_of_bits, - unsigned long cache_span, unsigned long max_diff, - unsigned long max_diff_percent, unsigned long num_of_runs, - bool platform) + unsigned long cache_span, long min_diff_percent, + unsigned long num_of_runs, bool platform, + __s64 *prev_avg_llc_val) { __u64 avg_llc_val = 0; - float diff_percent; - int ret; + float avg_diff; + int ret = 0;
avg_llc_val = sum_llc_val / num_of_runs; - diff_percent = ((float)cache_span - avg_llc_val) / cache_span * 100; + if (*prev_avg_llc_val) { + float delta = (__s64)(avg_llc_val - *prev_avg_llc_val);
- ret = platform && abs((int)diff_percent) > max_diff_percent; + avg_diff = delta / *prev_avg_llc_val; + ret = platform && (avg_diff * 100) < (float)min_diff_percent;
- ksft_print_msg("%s Check cache miss rate within %lu%%\n", - ret ? "Fail:" : "Pass:", max_diff_percent); + ksft_print_msg("%s Check cache miss rate changed more than %.1f%%\n", + ret ? "Fail:" : "Pass:", (float)min_diff_percent);
- ksft_print_msg("Percent diff=%d\n", abs((int)diff_percent)); + ksft_print_msg("Percent diff=%.1f\n", avg_diff * 100); + } + *prev_avg_llc_val = avg_llc_val;
show_cache_info(no_of_bits, avg_llc_val, cache_span, true);
return ret; }
-static int check_results(struct resctrl_val_param *param, size_t span) +/* Remove one bit from the consecutive cbm mask */ +static unsigned long next_mask(unsigned long current_mask) +{ + return current_mask & (current_mask >> 1); +} + +static int check_results(struct resctrl_val_param *param, const char *cache_type, + unsigned long cache_total_size, unsigned long full_cache_mask, + unsigned long current_mask) { char *token_array[8], temp[512]; - int runs = 0, no_of_bits = 0; __u64 sum_llc_perf_miss = 0; + __s64 prev_avg_llc_val = 0; + unsigned long alloc_size; + int runs = 0; + int fail = 0; + int ret; FILE *fp;
ksft_print_msg("Checking for pass/fail\n"); @@ -83,49 +87,78 @@ static int check_results(struct resctrl_val_param *param, size_t span) while (fgets(temp, sizeof(temp), fp)) { char *token = strtok(temp, ":\t"); int fields = 0; + int bits;
while (token) { token_array[fields++] = token; token = strtok(NULL, ":\t"); } - /* - * Discard the first value which is inaccurate due to monitoring - * setup transition phase. - */ - if (runs > 0) - sum_llc_perf_miss += strtoull(token_array[3], NULL, 0); + + sum_llc_perf_miss += strtoull(token_array[3], NULL, 0); runs++; + + if (runs < NUM_OF_RUNS) + continue; + + if (!current_mask) { + ksft_print_msg("Unexpected empty cache mask\n"); + break; + } + + alloc_size = cache_portion_size(cache_total_size, current_mask, full_cache_mask); + + bits = count_bits(current_mask); + + ret = show_results_info(sum_llc_perf_miss, bits, + alloc_size / 64, + MIN_DIFF_PERCENT_PER_BIT * (bits - 1), runs, + get_vendor() == ARCH_INTEL, + &prev_avg_llc_val); + if (ret) + fail = 1; + + runs = 0; + sum_llc_perf_miss = 0; + current_mask = next_mask(current_mask); }
fclose(fp); - no_of_bits = count_bits(param->mask);
- return show_results_info(sum_llc_perf_miss, no_of_bits, span / 64, - MAX_DIFF, MAX_DIFF_PERCENT, runs - 1, - get_vendor() == ARCH_INTEL); + return fail; }
void cat_test_cleanup(void) { - remove(RESULT_FILE_NAME1); - remove(RESULT_FILE_NAME2); + remove(RESULT_FILE_NAME); }
/* * cat_test - Execute CAT benchmark and measure cache misses * @param: Parameters passed to cat_test() * @span: Buffer size for the benchmark + * @current_mask Start mask for the first iteration + * + * Run CAT selftest by varying the allocated cache portion and comparing the + * impact on cache misses (the result analysis is done in check_results() + * and show_results_info(), not in this function). + * + * One bit is removed from the CAT allocation bit mask (in current_mask) for + * each subsequent test which keeps reducing the size of the allocated cache + * portion. A single test flushes the buffer, reads it to warm up the cache, + * and reads the buffer again. The cache misses are measured during the last + * read pass. * * Return: 0 on success. Non-zero on failure. */ -static int cat_test(struct resctrl_val_param *param, size_t span) +static int cat_test(struct resctrl_val_param *param, size_t span, unsigned long current_mask) { - int memflush = 1, operation = 0, ret = 0; char *resctrl_val = param->resctrl_val; struct perf_event_read pe_read; struct perf_event_attr pea; + unsigned char *buf; + char schemata[64]; + int ret, i, pe_fd; pid_t bm_pid; - int pe_fd;
if (strcmp(param->filename, "") == 0) sprintf(param->filename, "stdio"); @@ -149,48 +182,55 @@ static int cat_test(struct resctrl_val_param *param, size_t span) if (pe_fd < 0) return pe_fd;
- /* Test runs until the callback setup() tells the test to stop. */ - while (1) { - ret = param->setup(param); - if (ret == END_OF_TESTS) { - ret = 0; - break; - } - if (ret < 0) - break; + buf = alloc_buffer(span, 1); + if (!buf) { + ret = -1; + goto pe_close; + }
- ret = perf_event_reset_enable(pe_fd); + while (current_mask) { + snprintf(schemata, sizeof(schemata), "%lx", param->mask & ~current_mask); + ret = write_schemata("", schemata, param->cpu_no, param->resctrl_val); if (ret) - goto pe_close; + goto free_buf; + snprintf(schemata, sizeof(schemata), "%lx", current_mask); + ret = write_schemata(param->ctrlgrp, schemata, param->cpu_no, param->resctrl_val); + if (ret) + goto free_buf;
- if (run_fill_buf(span, memflush, operation, true)) { - fprintf(stderr, "Error-running fill buffer\n"); - ret = -1; - goto pe_close; - } + for (i = 0; i < NUM_OF_RUNS; i++) { + mem_flush(buf, span); + fill_cache_read(buf, span, true);
- sleep(1); - ret = perf_event_measure(pe_fd, &pe_read, param->filename, bm_pid); - if (ret) - goto pe_close; - } + ret = perf_event_reset_enable(pe_fd); + if (ret) + goto free_buf;
- return ret; + fill_cache_read(buf, span, true);
+ ret = perf_event_measure(pe_fd, &pe_read, param->filename, bm_pid); + if (ret) + goto free_buf; + } + current_mask = next_mask(current_mask); + } + +free_buf: + free(buf); pe_close: close(pe_fd); + return ret; }
int cat_perf_miss_val(int cpu_no, int n, char *cache_type) { - unsigned long full_cache_mask, long_mask; - unsigned long l_mask, l_mask_1; - int ret, pipefd[2], sibling_cpu_no; + unsigned long long_mask, start_mask, full_cache_mask; unsigned long cache_total_size = 0; + unsigned int start; int count_of_bits; - char pipe_message; size_t span; + int ret;
ret = get_full_cbm(cache_type, &full_cache_mask); if (ret) @@ -206,7 +246,7 @@ int cat_perf_miss_val(int cpu_no, int n, char *cache_type) return ret; ksft_print_msg("Cache size :%lu\n", cache_total_size);
- count_of_bits = count_bits(long_mask); + count_of_bits = count_contiguous_bits(long_mask, &start);
if (!n) n = count_of_bits / 2; @@ -217,88 +257,26 @@ int cat_perf_miss_val(int cpu_no, int n, char *cache_type) count_of_bits - 1); return -1; } - - /* Get core id from same socket for running another thread */ - sibling_cpu_no = get_core_sibling(cpu_no); - if (sibling_cpu_no < 0) - return -1; + start_mask = create_bit_mask(start, n);
struct resctrl_val_param param = { .resctrl_val = CAT_STR, .cpu_no = cpu_no, - .setup = cat_setup, + .ctrlgrp = "c1", + .filename = RESULT_FILE_NAME, + .num_of_runs = 0, }; - - l_mask = long_mask >> n; - l_mask_1 = ~l_mask & long_mask; - - /* Set param values for parent thread which will be allocated bitmask - * with (max_bits - n) bits - */ - span = cache_portion_size(cache_total_size, l_mask, full_cache_mask); - strcpy(param.ctrlgrp, "c2"); - strcpy(param.mongrp, "m2"); - strcpy(param.filename, RESULT_FILE_NAME2); - param.mask = l_mask; - param.num_of_runs = 0; - - if (pipe(pipefd)) { - perror("# Unable to create pipe"); - return errno; - } - - fflush(stdout); - bm_pid = fork(); - - /* Set param values for child thread which will be allocated bitmask - * with n bits - */ - if (bm_pid == 0) { - param.mask = l_mask_1; - strcpy(param.ctrlgrp, "c1"); - strcpy(param.mongrp, "m1"); - span = cache_portion_size(cache_total_size, l_mask_1, full_cache_mask); - strcpy(param.filename, RESULT_FILE_NAME1); - param.num_of_runs = 0; - param.cpu_no = sibling_cpu_no; - } + param.mask = long_mask; + span = cache_portion_size(cache_total_size, start_mask, full_cache_mask);
remove(param.filename);
- ret = cat_test(¶m, span); - if (ret == 0) - ret = check_results(¶m, span); - - if (bm_pid == 0) { - /* Tell parent that child is ready */ - close(pipefd[0]); - pipe_message = 1; - if (write(pipefd[1], &pipe_message, sizeof(pipe_message)) < - sizeof(pipe_message)) - /* - * Just print the error message. - * Let while(1) run and wait for itself to be killed. - */ - perror("# failed signaling parent process"); - - close(pipefd[1]); - while (1) - ; - } else { - /* Parent waits for child to be ready. */ - close(pipefd[1]); - pipe_message = 0; - while (pipe_message != 1) { - if (read(pipefd[0], &pipe_message, - sizeof(pipe_message)) < sizeof(pipe_message)) { - perror("# failed reading from child process"); - break; - } - } - close(pipefd[0]); - kill(bm_pid, SIGKILL); - } + ret = cat_test(¶m, span, start_mask); + if (ret) + goto out;
+ ret = check_results(¶m, cache_type, cache_total_size, full_cache_mask, start_mask); +out: cat_test_cleanup();
return ret; diff --git a/tools/testing/selftests/resctrl/fill_buf.c b/tools/testing/selftests/resctrl/fill_buf.c index 93a3d408339c..ae120f1735c0 100644 --- a/tools/testing/selftests/resctrl/fill_buf.c +++ b/tools/testing/selftests/resctrl/fill_buf.c @@ -38,7 +38,7 @@ static void cl_flush(void *p) #endif }
-static void mem_flush(unsigned char *buf, size_t buf_size) +void mem_flush(unsigned char *buf, size_t buf_size) { unsigned char *cp = buf; size_t i = 0; @@ -100,7 +100,7 @@ static void fill_one_span_write(unsigned char *buf, size_t buf_size) } }
-static void fill_cache_read(unsigned char *buf, size_t buf_size, bool once) +void fill_cache_read(unsigned char *buf, size_t buf_size, bool once) { int ret = 0;
@@ -123,7 +123,7 @@ static void fill_cache_write(unsigned char *buf, size_t buf_size, bool once) } }
-static unsigned char *alloc_buffer(size_t buf_size, int memflush) +unsigned char *alloc_buffer(size_t buf_size, int memflush) { void *buf = NULL; uint64_t *p64; diff --git a/tools/testing/selftests/resctrl/resctrl.h b/tools/testing/selftests/resctrl/resctrl.h index 68eb1f1bac3a..0f072a0346d7 100644 --- a/tools/testing/selftests/resctrl/resctrl.h +++ b/tools/testing/selftests/resctrl/resctrl.h @@ -106,6 +106,9 @@ int write_bm_pid_to_resctrl(pid_t bm_pid, char *ctrlgrp, char *mongrp, char *resctrl_val); int perf_event_open(struct perf_event_attr *hw_event, pid_t pid, int cpu, int group_fd, unsigned long flags); +unsigned char *alloc_buffer(size_t buf_size, int memflush); +void mem_flush(unsigned char *buf, size_t buf_size); +void fill_cache_read(unsigned char *buf, size_t buf_size, bool once); int run_fill_buf(size_t buf_size, int memflush, int op, bool once); int resctrl_val(const char * const *benchmark_cmd, struct resctrl_val_param *param); int mbm_bw_change(int cpu_no, const char * const *benchmark_cmd); @@ -114,6 +117,7 @@ void mbm_test_cleanup(void); int mba_schemata_change(int cpu_no, const char * const *benchmark_cmd); void mba_test_cleanup(void); unsigned long create_bit_mask(unsigned int start, unsigned int len); +unsigned int count_contiguous_bits(unsigned long val, unsigned int *start); int get_full_cbm(const char *cache_type, unsigned long *mask); int get_shareable_mask(const char *cache_type, unsigned long *shareable_mask); int get_mask_no_shareable(const char *cache_type, unsigned long *mask); @@ -126,7 +130,6 @@ int cat_perf_miss_val(int cpu_no, int no_of_bits, char *cache_type); int cmt_resctrl_val(int cpu_no, int n, const char * const *benchmark_cmd); unsigned int count_bits(unsigned long n); void cmt_test_cleanup(void); -int get_core_sibling(int cpu_no);
void perf_event_attr_initialize(struct perf_event_attr *pea, __u64 config); void perf_event_initialize_read_format(struct perf_event_read *pe_read); diff --git a/tools/testing/selftests/resctrl/resctrlfs.c b/tools/testing/selftests/resctrl/resctrlfs.c index 3a9ed918d657..9b09210c5ece 100644 --- a/tools/testing/selftests/resctrl/resctrlfs.c +++ b/tools/testing/selftests/resctrl/resctrlfs.c @@ -245,7 +245,7 @@ unsigned long create_bit_mask(unsigned int start, unsigned int len) * * Return: The length of the contiguous bits in the longest train of bits */ -static unsigned int count_contiguous_bits(unsigned long val, unsigned int *start) +unsigned int count_contiguous_bits(unsigned long val, unsigned int *start) { unsigned long last_val; unsigned int count = 0; @@ -343,48 +343,6 @@ int get_mask_no_shareable(const char *cache_type, unsigned long *mask) return 0; }
-/* - * get_core_sibling - Get sibling core id from the same socket for given CPU - * @cpu_no: CPU number - * - * Return: > 0 on success, < 0 on failure. - */ -int get_core_sibling(int cpu_no) -{ - char core_siblings_path[1024], cpu_list_str[64]; - int sibling_cpu_no = -1; - FILE *fp; - - sprintf(core_siblings_path, "%s%d/topology/core_siblings_list", - CORE_SIBLINGS_PATH, cpu_no); - - fp = fopen(core_siblings_path, "r"); - if (!fp) { - perror("Failed to open core siblings path"); - - return -1; - } - if (fscanf(fp, "%s", cpu_list_str) <= 0) { - perror("Could not get core_siblings list"); - fclose(fp); - - return -1; - } - fclose(fp); - - char *token = strtok(cpu_list_str, "-,"); - - while (token) { - sibling_cpu_no = atoi(token); - /* Skipping core 0 as we don't want to run test on core 0 */ - if (sibling_cpu_no != 0 && sibling_cpu_no != cpu_no) - break; - token = strtok(NULL, "-,"); - } - - return sibling_cpu_no; -} - /* * taskset_benchmark - Taskset PID (i.e. benchmark) to a specified cpu * @bm_pid: PID that should be binded
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
CAT test spawns two processes into two different control groups with exclusive schemata. Both the processes alloc a buffer from memory matching their allocated LLC block size and flush the entire buffer out of caches. Since the processes are reading through the buffer only once during the measurement and initially all the buffer was flushed, the test isn't testing CAT.
Rewrite the CAT test to allocate a buffer sized to half of LLC. Then perform a sequence of tests with different LLC alloc sizes starting from half of the CBM bits down to 1-bit CBM. Flush the buffer before each test and read the buffer twice. Observe the LLC misses on the second read through the buffer. As the allocated LLC block gets smaller and smaller, the LLC misses will become larger and larger giving a strong signal on CAT working properly.
The new CAT test is using only a single process because it relies on measured effect against another run of itself rather than another process adding noise. The rest of the system is set to use the CBM bits not used by the CAT test to keep the test isolated.
Replace count_bits() with count_contiguous_bits() to get the first bit position in order to be able to calculate masks based on it.
This change has been tested with a number of systems from different generations.
Suggested-by: Reinette Chatre reinette.chatre@intel.com Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
tools/testing/selftests/resctrl/cat_test.c | 282 +++++++++----------- tools/testing/selftests/resctrl/fill_buf.c | 6 +- tools/testing/selftests/resctrl/resctrl.h | 5 +- tools/testing/selftests/resctrl/resctrlfs.c | 44 +-- 4 files changed, 138 insertions(+), 199 deletions(-)
diff --git a/tools/testing/selftests/resctrl/cat_test.c b/tools/testing/selftests/resctrl/cat_test.c index cfda87667b46..4169b17b8f91 100644 --- a/tools/testing/selftests/resctrl/cat_test.c +++ b/tools/testing/selftests/resctrl/cat_test.c @@ -11,65 +11,69 @@ #include "resctrl.h" #include <unistd.h> -#define RESULT_FILE_NAME1 "result_cat1" -#define RESULT_FILE_NAME2 "result_cat2" +#define RESULT_FILE_NAME "result_cat" #define NUM_OF_RUNS 5 -#define MAX_DIFF_PERCENT 4 -#define MAX_DIFF 1000000 /*
- Change schemata. Write schemata to specified
- con_mon grp, mon_grp in resctrl FS.
- Run 5 times in order to get average values.
- Minimum difference in LLC misses between a test with n+1 bits CBM to the
- test with n bits is MIN_DIFF_PERCENT_PER_BIT * (n - 1). With e.g. 5 vs 4
- bits in the CBM mask, the minimum difference must be at least
- MIN_DIFF_PERCENT_PER_BIT * (4 - 1) = 3 percent.
- The relationship between number of used CBM bits and difference in LLC
- misses is not expected to be linear. With a small number of bits, the
- margin is smaller than with larger number of bits. For selftest purposes,
- however, linear approach is enough because ultimately only pass/fail
- decision has to be made and distinction between strong and stronger
*/
- signal is irrelevant.
-static int cat_setup(struct resctrl_val_param *p) -{
- char schemata[64];
- int ret = 0;
- /* Run NUM_OF_RUNS times */
- if (p->num_of_runs >= NUM_OF_RUNS)
return END_OF_TESTS;- if (p->num_of_runs == 0) {
sprintf(schemata, "%lx", p->mask);ret = write_schemata(p->ctrlgrp, schemata, p->cpu_no,p->resctrl_val);- }
- p->num_of_runs++;
- return ret;
-} +#define MIN_DIFF_PERCENT_PER_BIT 1 static int show_results_info(__u64 sum_llc_val, int no_of_bits,
unsigned long cache_span, unsigned long max_diff,unsigned long max_diff_percent, unsigned long num_of_runs,bool platform)
unsigned long cache_span, long min_diff_percent,
With all care taken in unsigned use I wonder why min_diff_percent is just long?
unsigned long num_of_runs, bool platform,__s64 *prev_avg_llc_val){ __u64 avg_llc_val = 0;
- float diff_percent;
- int ret;
- float avg_diff;
- int ret = 0;
avg_llc_val = sum_llc_val / num_of_runs;
- diff_percent = ((float)cache_span - avg_llc_val) / cache_span * 100;
- if (*prev_avg_llc_val) {
float delta = (__s64)(avg_llc_val - *prev_avg_llc_val);
- ret = platform && abs((int)diff_percent) > max_diff_percent;
avg_diff = delta / *prev_avg_llc_val;ret = platform && (avg_diff * 100) < (float)min_diff_percent;
- ksft_print_msg("%s Check cache miss rate within %lu%%\n",
ret ? "Fail:" : "Pass:", max_diff_percent);
ksft_print_msg("%s Check cache miss rate changed more than %.1f%%\n",ret ? "Fail:" : "Pass:", (float)min_diff_percent);
- ksft_print_msg("Percent diff=%d\n", abs((int)diff_percent));
ksft_print_msg("Percent diff=%.1f\n", avg_diff * 100);- }
- *prev_avg_llc_val = avg_llc_val;
show_cache_info(no_of_bits, avg_llc_val, cache_span, true); return ret; } -static int check_results(struct resctrl_val_param *param, size_t span) +/* Remove one bit from the consecutive cbm mask */
The use of "consecutive" is not clear. In this usage it removes one bit from the previous mask in order to create the consecutive mask, no?
+static unsigned long next_mask(unsigned long current_mask) +{
- return current_mask & (current_mask >> 1);
+}
The new test looks very good to me. Thank you very much for creating it.
It looks to me as though this test impacts the affinity of main program since it is only one process, changes its affinity, but never change it back.
Reinette
On Tue, 28 Nov 2023, Reinette Chatre wrote:
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
CAT test spawns two processes into two different control groups with exclusive schemata. Both the processes alloc a buffer from memory matching their allocated LLC block size and flush the entire buffer out of caches. Since the processes are reading through the buffer only once during the measurement and initially all the buffer was flushed, the test isn't testing CAT.
Rewrite the CAT test to allocate a buffer sized to half of LLC. Then perform a sequence of tests with different LLC alloc sizes starting from half of the CBM bits down to 1-bit CBM. Flush the buffer before each test and read the buffer twice. Observe the LLC misses on the second read through the buffer. As the allocated LLC block gets smaller and smaller, the LLC misses will become larger and larger giving a strong signal on CAT working properly.
The new CAT test is using only a single process because it relies on measured effect against another run of itself rather than another process adding noise. The rest of the system is set to use the CBM bits not used by the CAT test to keep the test isolated.
Replace count_bits() with count_contiguous_bits() to get the first bit position in order to be able to calculate masks based on it.
This change has been tested with a number of systems from different generations.
Suggested-by: Reinette Chatre reinette.chatre@intel.com Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
tools/testing/selftests/resctrl/cat_test.c | 282 +++++++++----------- tools/testing/selftests/resctrl/fill_buf.c | 6 +- tools/testing/selftests/resctrl/resctrl.h | 5 +- tools/testing/selftests/resctrl/resctrlfs.c | 44 +-- 4 files changed, 138 insertions(+), 199 deletions(-)
diff --git a/tools/testing/selftests/resctrl/cat_test.c b/tools/testing/selftests/resctrl/cat_test.c index cfda87667b46..4169b17b8f91 100644 --- a/tools/testing/selftests/resctrl/cat_test.c +++ b/tools/testing/selftests/resctrl/cat_test.c @@ -11,65 +11,69 @@ #include "resctrl.h" #include <unistd.h> -#define RESULT_FILE_NAME1 "result_cat1" -#define RESULT_FILE_NAME2 "result_cat2" +#define RESULT_FILE_NAME "result_cat" #define NUM_OF_RUNS 5 -#define MAX_DIFF_PERCENT 4 -#define MAX_DIFF 1000000 /*
- Change schemata. Write schemata to specified
- con_mon grp, mon_grp in resctrl FS.
- Run 5 times in order to get average values.
- Minimum difference in LLC misses between a test with n+1 bits CBM to the
- test with n bits is MIN_DIFF_PERCENT_PER_BIT * (n - 1). With e.g. 5 vs 4
- bits in the CBM mask, the minimum difference must be at least
- MIN_DIFF_PERCENT_PER_BIT * (4 - 1) = 3 percent.
- The relationship between number of used CBM bits and difference in LLC
- misses is not expected to be linear. With a small number of bits, the
- margin is smaller than with larger number of bits. For selftest purposes,
- however, linear approach is enough because ultimately only pass/fail
- decision has to be made and distinction between strong and stronger
*/
- signal is irrelevant.
-static int cat_setup(struct resctrl_val_param *p) -{
- char schemata[64];
- int ret = 0;
- /* Run NUM_OF_RUNS times */
- if (p->num_of_runs >= NUM_OF_RUNS)
return END_OF_TESTS;- if (p->num_of_runs == 0) {
sprintf(schemata, "%lx", p->mask);ret = write_schemata(p->ctrlgrp, schemata, p->cpu_no,p->resctrl_val);- }
- p->num_of_runs++;
- return ret;
-} +#define MIN_DIFF_PERCENT_PER_BIT 1 static int show_results_info(__u64 sum_llc_val, int no_of_bits,
unsigned long cache_span, unsigned long max_diff,unsigned long max_diff_percent, unsigned long num_of_runs,bool platform)
unsigned long cache_span, long min_diff_percent,With all care taken in unsigned use I wonder why min_diff_percent is just long?
This was a leftover from the time when I still wasn't using floats so the compare typing was easier. But's it's long gone now so I'll make that unsigned long.
It looks to me as though this test impacts the affinity of main program since it is only one process, changes its affinity, but never change it back.
Ah, right.
It looks pre-existing problem though as despite more than one process in the old CAT test, it altered affinity of both of them. I'll need to look into fixing this.
resctrl_tests reads a set of parameters and passes them individually for each tests which causes variations in the call signature between the tests.
Add struct input_params to hold all input parameters. It can be easily passed to every test without varying the call signature.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/cat_test.c | 20 +++--- tools/testing/selftests/resctrl/cmt_test.c | 16 ++--- tools/testing/selftests/resctrl/mba_test.c | 9 ++- tools/testing/selftests/resctrl/mbm_test.c | 9 ++- tools/testing/selftests/resctrl/resctrl.h | 28 ++++++--- .../testing/selftests/resctrl/resctrl_tests.c | 61 +++++++++++-------- tools/testing/selftests/resctrl/resctrl_val.c | 21 ++++--- 7 files changed, 93 insertions(+), 71 deletions(-)
diff --git a/tools/testing/selftests/resctrl/cat_test.c b/tools/testing/selftests/resctrl/cat_test.c index 4169b17b8f91..3cc75ca74036 100644 --- a/tools/testing/selftests/resctrl/cat_test.c +++ b/tools/testing/selftests/resctrl/cat_test.c @@ -134,6 +134,7 @@ void cat_test_cleanup(void)
/* * cat_test - Execute CAT benchmark and measure cache misses + * @uparams: User supplied parameters * @param: Parameters passed to cat_test() * @span: Buffer size for the benchmark * @current_mask Start mask for the first iteration @@ -150,7 +151,8 @@ void cat_test_cleanup(void) * * Return: 0 on success. Non-zero on failure. */ -static int cat_test(struct resctrl_val_param *param, size_t span, unsigned long current_mask) +static int cat_test(const struct user_params *uparams, struct resctrl_val_param *param, + size_t span, unsigned long current_mask) { char *resctrl_val = param->resctrl_val; struct perf_event_read pe_read; @@ -166,7 +168,7 @@ static int cat_test(struct resctrl_val_param *param, size_t span, unsigned long bm_pid = getpid();
/* Taskset benchmark to specified cpu */ - ret = taskset_benchmark(bm_pid, param->cpu_no); + ret = taskset_benchmark(bm_pid, uparams->cpu); if (ret) return ret;
@@ -178,7 +180,7 @@ static int cat_test(struct resctrl_val_param *param, size_t span, unsigned long
perf_event_attr_initialize(&pea, PERF_COUNT_HW_CACHE_MISSES); perf_event_initialize_read_format(&pe_read); - pe_fd = perf_open(&pea, bm_pid, param->cpu_no); + pe_fd = perf_open(&pea, bm_pid, uparams->cpu); if (pe_fd < 0) return pe_fd;
@@ -190,11 +192,11 @@ static int cat_test(struct resctrl_val_param *param, size_t span, unsigned long
while (current_mask) { snprintf(schemata, sizeof(schemata), "%lx", param->mask & ~current_mask); - ret = write_schemata("", schemata, param->cpu_no, param->resctrl_val); + ret = write_schemata("", schemata, uparams->cpu, param->resctrl_val); if (ret) goto free_buf; snprintf(schemata, sizeof(schemata), "%lx", current_mask); - ret = write_schemata(param->ctrlgrp, schemata, param->cpu_no, param->resctrl_val); + ret = write_schemata(param->ctrlgrp, schemata, uparams->cpu, param->resctrl_val); if (ret) goto free_buf;
@@ -223,10 +225,11 @@ static int cat_test(struct resctrl_val_param *param, size_t span, unsigned long return ret; }
-int cat_perf_miss_val(int cpu_no, int n, char *cache_type) +int cat_perf_miss_val(const struct user_params *uparams, char *cache_type) { unsigned long long_mask, start_mask, full_cache_mask; unsigned long cache_total_size = 0; + int n = uparams->bits; unsigned int start; int count_of_bits; size_t span; @@ -241,7 +244,7 @@ int cat_perf_miss_val(int cpu_no, int n, char *cache_type) return ret;
/* Get L3/L2 cache size */ - ret = get_cache_size(cpu_no, cache_type, &cache_total_size); + ret = get_cache_size(uparams->cpu, cache_type, &cache_total_size); if (ret) return ret; ksft_print_msg("Cache size :%lu\n", cache_total_size); @@ -261,7 +264,6 @@ int cat_perf_miss_val(int cpu_no, int n, char *cache_type)
struct resctrl_val_param param = { .resctrl_val = CAT_STR, - .cpu_no = cpu_no, .ctrlgrp = "c1", .filename = RESULT_FILE_NAME, .num_of_runs = 0, @@ -271,7 +273,7 @@ int cat_perf_miss_val(int cpu_no, int n, char *cache_type)
remove(param.filename);
- ret = cat_test(¶m, span, start_mask); + ret = cat_test(uparams, ¶m, span, start_mask); if (ret) goto out;
diff --git a/tools/testing/selftests/resctrl/cmt_test.c b/tools/testing/selftests/resctrl/cmt_test.c index 4573a99d87b5..73c8f3013235 100644 --- a/tools/testing/selftests/resctrl/cmt_test.c +++ b/tools/testing/selftests/resctrl/cmt_test.c @@ -16,7 +16,7 @@ #define MAX_DIFF 2000000 #define MAX_DIFF_PERCENT 15
-static int cmt_setup(struct resctrl_val_param *p) +static int cmt_setup(const struct user_params *uparams, struct resctrl_val_param *p) { /* Run NUM_OF_RUNS times */ if (p->num_of_runs >= NUM_OF_RUNS) @@ -94,11 +94,12 @@ void cmt_test_cleanup(void) remove(RESULT_FILE_NAME); }
-int cmt_resctrl_val(int cpu_no, int n, const char * const *benchmark_cmd) +int cmt_resctrl_val(const struct user_params *uparams) { - const char * const *cmd = benchmark_cmd; + const char * const *cmd = uparams->benchmark_cmd; const char *new_cmd[BENCHMARK_ARGS]; unsigned long cache_total_size = 0; + int n = uparams->bits ? : 5; unsigned long long_mask; char *span_str = NULL; int count_of_bits; @@ -109,7 +110,7 @@ int cmt_resctrl_val(int cpu_no, int n, const char * const *benchmark_cmd) if (ret) return ret;
- ret = get_cache_size(cpu_no, "L3", &cache_total_size); + ret = get_cache_size(uparams->cpu, "L3", &cache_total_size); if (ret) return ret; ksft_print_msg("Cache size :%lu\n", cache_total_size); @@ -126,7 +127,6 @@ int cmt_resctrl_val(int cpu_no, int n, const char * const *benchmark_cmd) .resctrl_val = CMT_STR, .ctrlgrp = "c1", .mongrp = "m1", - .cpu_no = cpu_no, .filename = RESULT_FILE_NAME, .mask = ~(long_mask << n) & long_mask, .num_of_runs = 0, @@ -137,8 +137,8 @@ int cmt_resctrl_val(int cpu_no, int n, const char * const *benchmark_cmd)
if (strcmp(cmd[0], "fill_buf") == 0) { /* Duplicate the command to be able to replace span in it */ - for (i = 0; benchmark_cmd[i]; i++) - new_cmd[i] = benchmark_cmd[i]; + for (i = 0; uparams->benchmark_cmd[i]; i++) + new_cmd[i] = uparams->benchmark_cmd[i]; new_cmd[i] = NULL;
ret = asprintf(&span_str, "%zu", span); @@ -150,7 +150,7 @@ int cmt_resctrl_val(int cpu_no, int n, const char * const *benchmark_cmd)
remove(RESULT_FILE_NAME);
- ret = resctrl_val(cmd, ¶m); + ret = resctrl_val(uparams, cmd, ¶m); if (ret) goto out;
diff --git a/tools/testing/selftests/resctrl/mba_test.c b/tools/testing/selftests/resctrl/mba_test.c index d3bf4368341e..f669a0d97de8 100644 --- a/tools/testing/selftests/resctrl/mba_test.c +++ b/tools/testing/selftests/resctrl/mba_test.c @@ -22,7 +22,7 @@ * con_mon grp, mon_grp in resctrl FS. * For each allocation, run 5 times in order to get average values. */ -static int mba_setup(struct resctrl_val_param *p) +static int mba_setup(const struct user_params *uparams, struct resctrl_val_param *p) { static int runs_per_allocation, allocation = 100; char allocation_str[64]; @@ -40,7 +40,7 @@ static int mba_setup(struct resctrl_val_param *p)
sprintf(allocation_str, "%d", allocation);
- ret = write_schemata(p->ctrlgrp, allocation_str, p->cpu_no, + ret = write_schemata(p->ctrlgrp, allocation_str, uparams->cpu, p->resctrl_val); if (ret < 0) return ret; @@ -141,13 +141,12 @@ void mba_test_cleanup(void) remove(RESULT_FILE_NAME); }
-int mba_schemata_change(int cpu_no, const char * const *benchmark_cmd) +int mba_schemata_change(const struct user_params *uparams) { struct resctrl_val_param param = { .resctrl_val = MBA_STR, .ctrlgrp = "c1", .mongrp = "m1", - .cpu_no = cpu_no, .filename = RESULT_FILE_NAME, .bw_report = "reads", .setup = mba_setup @@ -156,7 +155,7 @@ int mba_schemata_change(int cpu_no, const char * const *benchmark_cmd)
remove(RESULT_FILE_NAME);
- ret = resctrl_val(benchmark_cmd, ¶m); + ret = resctrl_val(uparams, uparams->benchmark_cmd, ¶m); if (ret) goto out;
diff --git a/tools/testing/selftests/resctrl/mbm_test.c b/tools/testing/selftests/resctrl/mbm_test.c index 741533f2b075..7d25ca98c61a 100644 --- a/tools/testing/selftests/resctrl/mbm_test.c +++ b/tools/testing/selftests/resctrl/mbm_test.c @@ -86,7 +86,7 @@ static int check_results(size_t span) return ret; }
-static int mbm_setup(struct resctrl_val_param *p) +static int mbm_setup(const struct user_params *uparams, struct resctrl_val_param *p) { int ret = 0;
@@ -96,7 +96,7 @@ static int mbm_setup(struct resctrl_val_param *p)
/* Set up shemata with 100% allocation on the first run. */ if (p->num_of_runs == 0 && validate_resctrl_feature_request("MB", NULL)) - ret = write_schemata(p->ctrlgrp, "100", p->cpu_no, + ret = write_schemata(p->ctrlgrp, "100", uparams->cpu, p->resctrl_val);
p->num_of_runs++; @@ -109,13 +109,12 @@ void mbm_test_cleanup(void) remove(RESULT_FILE_NAME); }
-int mbm_bw_change(int cpu_no, const char * const *benchmark_cmd) +int mbm_bw_change(const struct user_params *uparams) { struct resctrl_val_param param = { .resctrl_val = MBM_STR, .ctrlgrp = "c1", .mongrp = "m1", - .cpu_no = cpu_no, .filename = RESULT_FILE_NAME, .bw_report = "reads", .setup = mbm_setup @@ -124,7 +123,7 @@ int mbm_bw_change(int cpu_no, const char * const *benchmark_cmd)
remove(RESULT_FILE_NAME);
- ret = resctrl_val(benchmark_cmd, ¶m); + ret = resctrl_val(uparams, uparams->benchmark_cmd, ¶m); if (ret) goto out;
diff --git a/tools/testing/selftests/resctrl/resctrl.h b/tools/testing/selftests/resctrl/resctrl.h index 0f072a0346d7..e61e11ba6a55 100644 --- a/tools/testing/selftests/resctrl/resctrl.h +++ b/tools/testing/selftests/resctrl/resctrl.h @@ -45,12 +45,23 @@ exit(EXIT_FAILURE); \ } while (0)
+/* + * user_params: User supplied parameters + * @cpu: CPU number to which the benchmark will be bound to + * @bits: Number of bits used for cache allocation size + * @benchmark_cmd: Benchmark command to run during (some of the) tests + */ +struct user_params { + int cpu; + int bits; + const char *benchmark_cmd[BENCHMARK_ARGS]; +}; + /* * resctrl_val_param: resctrl test parameters * @resctrl_val: Resctrl feature (Eg: mbm, mba.. etc) * @ctrlgrp: Name of the control monitor group (con_mon grp) * @mongrp: Name of the monitor group (mon grp) - * @cpu_no: CPU number to which the benchmark would be binded * @filename: Name of file to which the o/p should be written * @bw_report: Bandwidth report type (reads vs writes) * @setup: Call back function to setup test environment @@ -59,12 +70,12 @@ struct resctrl_val_param { char *resctrl_val; char ctrlgrp[64]; char mongrp[64]; - int cpu_no; char filename[64]; char *bw_report; unsigned long mask; int num_of_runs; - int (*setup)(struct resctrl_val_param *param); + int (*setup)(const struct user_params *uparams, + struct resctrl_val_param *param); };
struct perf_event_read { @@ -110,11 +121,12 @@ unsigned char *alloc_buffer(size_t buf_size, int memflush); void mem_flush(unsigned char *buf, size_t buf_size); void fill_cache_read(unsigned char *buf, size_t buf_size, bool once); int run_fill_buf(size_t buf_size, int memflush, int op, bool once); -int resctrl_val(const char * const *benchmark_cmd, struct resctrl_val_param *param); -int mbm_bw_change(int cpu_no, const char * const *benchmark_cmd); +int resctrl_val(const struct user_params *uparams, const char * const *benchmark_cmd, + struct resctrl_val_param *param); +int mbm_bw_change(const struct user_params *uparams); void tests_cleanup(void); void mbm_test_cleanup(void); -int mba_schemata_change(int cpu_no, const char * const *benchmark_cmd); +int mba_schemata_change(const struct user_params *uparams); void mba_test_cleanup(void); unsigned long create_bit_mask(unsigned int start, unsigned int len); unsigned int count_contiguous_bits(unsigned long val, unsigned int *start); @@ -126,8 +138,8 @@ void ctrlc_handler(int signum, siginfo_t *info, void *ptr); int signal_handler_register(void); void signal_handler_unregister(void); void cat_test_cleanup(void); -int cat_perf_miss_val(int cpu_no, int no_of_bits, char *cache_type); -int cmt_resctrl_val(int cpu_no, int n, const char * const *benchmark_cmd); +int cat_perf_miss_val(const struct user_params *uparams, char *cache_type); +int cmt_resctrl_val(const struct user_params *uparams); unsigned int count_bits(unsigned long n); void cmt_test_cleanup(void);
diff --git a/tools/testing/selftests/resctrl/resctrl_tests.c b/tools/testing/selftests/resctrl/resctrl_tests.c index f0d573aa363b..f7f11415cc85 100644 --- a/tools/testing/selftests/resctrl/resctrl_tests.c +++ b/tools/testing/selftests/resctrl/resctrl_tests.c @@ -96,7 +96,7 @@ static void test_cleanup(void) signal_handler_unregister(); }
-static void run_mbm_test(const char * const *benchmark_cmd, int cpu_no) +static void run_mbm_test(const struct user_params *uparams) { int res;
@@ -114,7 +114,7 @@ static void run_mbm_test(const char * const *benchmark_cmd, int cpu_no) goto cleanup; }
- res = mbm_bw_change(cpu_no, benchmark_cmd); + res = mbm_bw_change(uparams); ksft_test_result(!res, "MBM: bw change\n"); if ((get_vendor() == ARCH_INTEL) && res) ksft_print_msg("Intel MBM may be inaccurate when Sub-NUMA Clustering is enabled. Check BIOS configuration.\n"); @@ -123,7 +123,7 @@ static void run_mbm_test(const char * const *benchmark_cmd, int cpu_no) test_cleanup(); }
-static void run_mba_test(const char * const *benchmark_cmd, int cpu_no) +static void run_mba_test(const struct user_params *uparams) { int res;
@@ -141,14 +141,14 @@ static void run_mba_test(const char * const *benchmark_cmd, int cpu_no) goto cleanup; }
- res = mba_schemata_change(cpu_no, benchmark_cmd); + res = mba_schemata_change(uparams); ksft_test_result(!res, "MBA: schemata change\n");
cleanup: test_cleanup(); }
-static void run_cmt_test(const char * const *benchmark_cmd, int cpu_no) +static void run_cmt_test(const struct user_params *uparams) { int res;
@@ -165,7 +165,7 @@ static void run_cmt_test(const char * const *benchmark_cmd, int cpu_no) goto cleanup; }
- res = cmt_resctrl_val(cpu_no, 5, benchmark_cmd); + res = cmt_resctrl_val(uparams); ksft_test_result(!res, "CMT: test\n"); if ((get_vendor() == ARCH_INTEL) && res) ksft_print_msg("Intel CMT may be inaccurate when Sub-NUMA Clustering is enabled. Check BIOS configuration.\n"); @@ -174,7 +174,7 @@ static void run_cmt_test(const char * const *benchmark_cmd, int cpu_no) test_cleanup(); }
-static void run_cat_test(int cpu_no, int no_of_bits) +static void run_cat_test(const struct user_params *uparams) { int res;
@@ -190,22 +190,29 @@ static void run_cat_test(int cpu_no, int no_of_bits) goto cleanup; }
- res = cat_perf_miss_val(cpu_no, no_of_bits, "L3"); + res = cat_perf_miss_val(uparams, "L3"); ksft_test_result(!res, "CAT: test\n");
cleanup: test_cleanup(); }
+static void init_user_params(struct user_params *uparams) +{ + uparams->cpu = 1; + uparams->bits = 0; +} + int main(int argc, char **argv) { bool mbm_test = true, mba_test = true, cmt_test = true; - const char *benchmark_cmd[BENCHMARK_ARGS] = {}; - int c, cpu_no = 1, i, no_of_bits = 0; + struct user_params uparams = {}; char *span_str = NULL; bool cat_test = true; int tests = 0; - int ret; + int ret, c, i; + + init_user_params(&uparams);
while ((c = getopt(argc, argv, "ht:b:n:p:")) != -1) { char *token; @@ -223,8 +230,8 @@ int main(int argc, char **argv)
/* Extract benchmark command from command line. */ for (i = 0; i < argc - optind; i++) - benchmark_cmd[i] = argv[i + optind]; - benchmark_cmd[i] = NULL; + uparams.benchmark_cmd[i] = argv[i + optind]; + uparams.benchmark_cmd[i] = NULL;
goto last_arg; case 't': @@ -256,11 +263,11 @@ int main(int argc, char **argv) } break; case 'p': - cpu_no = atoi(optarg); + uparams.cpu = atoi(optarg); break; case 'n': - no_of_bits = atoi(optarg); - if (no_of_bits <= 0) { + uparams.bits = atoi(optarg); + if (uparams.bits <= 0) { printf("Bail out! invalid argument for no_of_bits\n"); return -1; } @@ -295,32 +302,32 @@ int main(int argc, char **argv)
filter_dmesg();
- if (!benchmark_cmd[0]) { + if (!uparams.benchmark_cmd[0]) { /* If no benchmark is given by "-b" argument, use fill_buf. */ - benchmark_cmd[0] = "fill_buf"; + uparams.benchmark_cmd[0] = "fill_buf"; ret = asprintf(&span_str, "%u", DEFAULT_SPAN); if (ret < 0) ksft_exit_fail_msg("Out of memory!\n"); - benchmark_cmd[1] = span_str; - benchmark_cmd[2] = "1"; - benchmark_cmd[3] = "0"; - benchmark_cmd[4] = "false"; - benchmark_cmd[5] = NULL; + uparams.benchmark_cmd[1] = span_str; + uparams.benchmark_cmd[2] = "1"; + uparams.benchmark_cmd[3] = "0"; + uparams.benchmark_cmd[4] = "false"; + uparams.benchmark_cmd[5] = NULL; }
ksft_set_plan(tests ? : 4);
if (mbm_test) - run_mbm_test(benchmark_cmd, cpu_no); + run_mbm_test(&uparams);
if (mba_test) - run_mba_test(benchmark_cmd, cpu_no); + run_mba_test(&uparams);
if (cmt_test) - run_cmt_test(benchmark_cmd, cpu_no); + run_cmt_test(&uparams);
if (cat_test) - run_cat_test(cpu_no, no_of_bits); + run_cat_test(&uparams);
free(span_str); ksft_finished(); diff --git a/tools/testing/selftests/resctrl/resctrl_val.c b/tools/testing/selftests/resctrl/resctrl_val.c index a07ba336db48..24bf91cfdb21 100644 --- a/tools/testing/selftests/resctrl/resctrl_val.c +++ b/tools/testing/selftests/resctrl/resctrl_val.c @@ -593,8 +593,9 @@ static void initialize_llc_occu_resctrl(const char *ctrlgrp, const char *mongrp, set_cmt_path(ctrlgrp, mongrp, resource_id); }
-static int -measure_vals(struct resctrl_val_param *param, unsigned long *bw_resc_start) +static int measure_vals(const struct user_params *uparams, + struct resctrl_val_param *param, + unsigned long *bw_resc_start) { unsigned long bw_resc, bw_resc_end; float bw_imc; @@ -607,7 +608,7 @@ measure_vals(struct resctrl_val_param *param, unsigned long *bw_resc_start) * Compare the two values to validate resctrl value. * It takes 1sec to measure the data. */ - ret = get_mem_bw_imc(param->cpu_no, param->bw_report, &bw_imc); + ret = get_mem_bw_imc(uparams->cpu, param->bw_report, &bw_imc); if (ret < 0) return ret;
@@ -678,12 +679,14 @@ static void run_benchmark(int signum, siginfo_t *info, void *ucontext) /* * resctrl_val: execute benchmark and measure memory bandwidth on * the benchmark + * @uparams: user supplied parameters * @benchmark_cmd: benchmark command and its arguments * @param: parameters passed to resctrl_val() * * Return: 0 on success. non-zero on failure. */ -int resctrl_val(const char * const *benchmark_cmd, struct resctrl_val_param *param) +int resctrl_val(const struct user_params *uparams, const char * const *benchmark_cmd, + struct resctrl_val_param *param) { char *resctrl_val = param->resctrl_val; unsigned long bw_resc_start = 0; @@ -769,7 +772,7 @@ int resctrl_val(const char * const *benchmark_cmd, struct resctrl_val_param *par value.sival_ptr = (void *)benchmark_cmd;
/* Taskset benchmark to specified cpu */ - ret = taskset_benchmark(bm_pid, param->cpu_no); + ret = taskset_benchmark(bm_pid, uparams->cpu); if (ret) goto out;
@@ -786,10 +789,10 @@ int resctrl_val(const char * const *benchmark_cmd, struct resctrl_val_param *par goto out;
initialize_mem_bw_resctrl(param->ctrlgrp, param->mongrp, - param->cpu_no, resctrl_val); + uparams->cpu, resctrl_val); } else if (!strncmp(resctrl_val, CMT_STR, sizeof(CMT_STR))) initialize_llc_occu_resctrl(param->ctrlgrp, param->mongrp, - param->cpu_no, resctrl_val); + uparams->cpu, resctrl_val);
/* Parent waits for child to be ready. */ close(pipefd[1]); @@ -815,7 +818,7 @@ int resctrl_val(const char * const *benchmark_cmd, struct resctrl_val_param *par
/* Test runs until the callback setup() tells the test to stop. */ while (1) { - ret = param->setup(param); + ret = param->setup(uparams, param); if (ret == END_OF_TESTS) { ret = 0; break; @@ -825,7 +828,7 @@ int resctrl_val(const char * const *benchmark_cmd, struct resctrl_val_param *par
if (!strncmp(resctrl_val, MBM_STR, sizeof(MBM_STR)) || !strncmp(resctrl_val, MBA_STR, sizeof(MBA_STR))) { - ret = measure_vals(param, &bw_resc_start); + ret = measure_vals(uparams, param, &bw_resc_start); if (ret) break; } else if (!strncmp(resctrl_val, CMT_STR, sizeof(CMT_STR))) {
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
...
+static void init_user_params(struct user_params *uparams) +{
- uparams->cpu = 1;
- uparams->bits = 0;
+}
int main(int argc, char **argv) { bool mbm_test = true, mba_test = true, cmt_test = true;
- const char *benchmark_cmd[BENCHMARK_ARGS] = {};
- int c, cpu_no = 1, i, no_of_bits = 0;
- struct user_params uparams = {}; char *span_str = NULL; bool cat_test = true; int tests = 0;
- int ret;
- int ret, c, i;
- init_user_params(&uparams);
It is unexpected that this uses a combination of designated initializer (via {}) as well as an explicit init function to initialize the struct. I think it would be simpler to do all initialization in the init function.
Reinette
Each test currently has a "run test" function in per test file and another resctrl_tests.c. The functions in resctrl_tests.c are almost identical.
Generalize the one in resctrl_tests.c such that it can be shared between all of the tests. It makes adding new tests easier and removes the per test if () forests.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/cat_test.c | 18 +- tools/testing/selftests/resctrl/cmt_test.c | 17 +- tools/testing/selftests/resctrl/mba_test.c | 16 +- tools/testing/selftests/resctrl/mbm_test.c | 18 +- tools/testing/selftests/resctrl/resctrl.h | 29 +++- .../testing/selftests/resctrl/resctrl_tests.c | 160 ++++++------------ tools/testing/selftests/resctrl/resctrlfs.c | 5 + 7 files changed, 142 insertions(+), 121 deletions(-)
diff --git a/tools/testing/selftests/resctrl/cat_test.c b/tools/testing/selftests/resctrl/cat_test.c index 3cc75ca74036..4b034850603e 100644 --- a/tools/testing/selftests/resctrl/cat_test.c +++ b/tools/testing/selftests/resctrl/cat_test.c @@ -225,7 +225,7 @@ static int cat_test(const struct user_params *uparams, struct resctrl_val_param return ret; }
-int cat_perf_miss_val(const struct user_params *uparams, char *cache_type) +static int cat_run_test(const struct resctrl_test *test, const struct user_params *uparams) { unsigned long long_mask, start_mask, full_cache_mask; unsigned long cache_total_size = 0; @@ -235,16 +235,16 @@ int cat_perf_miss_val(const struct user_params *uparams, char *cache_type) size_t span; int ret;
- ret = get_full_cbm(cache_type, &full_cache_mask); + ret = get_full_cbm(test->resource, &full_cache_mask); if (ret) return ret; /* Get the exclusive portion of the cache */ - ret = get_mask_no_shareable(cache_type, &long_mask); + ret = get_mask_no_shareable(test->resource, &long_mask); if (ret) return ret;
/* Get L3/L2 cache size */ - ret = get_cache_size(uparams->cpu, cache_type, &cache_total_size); + ret = get_cache_size(uparams->cpu, test->resource, &cache_total_size); if (ret) return ret; ksft_print_msg("Cache size :%lu\n", cache_total_size); @@ -277,9 +277,17 @@ int cat_perf_miss_val(const struct user_params *uparams, char *cache_type) if (ret) goto out;
- ret = check_results(¶m, cache_type, cache_total_size, full_cache_mask, start_mask); + ret = check_results(¶m, test->resource, + cache_total_size, full_cache_mask, start_mask); out: cat_test_cleanup();
return ret; } + +struct resctrl_test l3_cat_test = { + .name = "CAT", + .resource = "L3", + .feature_check = test_resource_feature_check, + .run_test = cat_run_test, +}; diff --git a/tools/testing/selftests/resctrl/cmt_test.c b/tools/testing/selftests/resctrl/cmt_test.c index 73c8f3013235..89e1928761f3 100644 --- a/tools/testing/selftests/resctrl/cmt_test.c +++ b/tools/testing/selftests/resctrl/cmt_test.c @@ -94,7 +94,7 @@ void cmt_test_cleanup(void) remove(RESULT_FILE_NAME); }
-int cmt_resctrl_val(const struct user_params *uparams) +static int cmt_run_test(const struct resctrl_test *test, const struct user_params *uparams) { const char * const *cmd = uparams->benchmark_cmd; const char *new_cmd[BENCHMARK_ARGS]; @@ -155,6 +155,8 @@ int cmt_resctrl_val(const struct user_params *uparams) goto out;
ret = check_results(¶m, span, n); + if (ret && (get_vendor() == ARCH_INTEL)) + ksft_print_msg("Intel CMT may be inaccurate when Sub-NUMA Clustering is enabled. Check BIOS configuration.\n");
out: cmt_test_cleanup(); @@ -162,3 +164,16 @@ int cmt_resctrl_val(const struct user_params *uparams)
return ret; } + +static bool cmt_feature_check(const struct resctrl_test *test) +{ + return validate_resctrl_feature_request("L3_MON", "llc_occupancy") && + validate_resctrl_feature_request("L3", NULL); +} + +struct resctrl_test cmt_test = { + .name = "CMT", + .resource = "L3", + .feature_check = cmt_feature_check, + .run_test = cmt_run_test, +}; diff --git a/tools/testing/selftests/resctrl/mba_test.c b/tools/testing/selftests/resctrl/mba_test.c index f669a0d97de8..e4870f96d7ea 100644 --- a/tools/testing/selftests/resctrl/mba_test.c +++ b/tools/testing/selftests/resctrl/mba_test.c @@ -141,7 +141,7 @@ void mba_test_cleanup(void) remove(RESULT_FILE_NAME); }
-int mba_schemata_change(const struct user_params *uparams) +static int mba_run_test(const struct resctrl_test *test, const struct user_params *uparams) { struct resctrl_val_param param = { .resctrl_val = MBA_STR, @@ -166,3 +166,17 @@ int mba_schemata_change(const struct user_params *uparams)
return ret; } + +static bool mba_feature_check(const struct resctrl_test *test) +{ + return test_resource_feature_check(test) && + validate_resctrl_feature_request("L3_MON", "mbm_local_bytes"); +} + +struct resctrl_test mba_test = { + .name = "MBA", + .resource = "MB", + .vendor_specific = ARCH_INTEL, + .feature_check = mba_feature_check, + .run_test = mba_run_test, +}; diff --git a/tools/testing/selftests/resctrl/mbm_test.c b/tools/testing/selftests/resctrl/mbm_test.c index 7d25ca98c61a..2274a5eebbca 100644 --- a/tools/testing/selftests/resctrl/mbm_test.c +++ b/tools/testing/selftests/resctrl/mbm_test.c @@ -109,7 +109,7 @@ void mbm_test_cleanup(void) remove(RESULT_FILE_NAME); }
-int mbm_bw_change(const struct user_params *uparams) +static int mbm_run_test(const struct resctrl_test *test, const struct user_params *uparams) { struct resctrl_val_param param = { .resctrl_val = MBM_STR, @@ -128,9 +128,25 @@ int mbm_bw_change(const struct user_params *uparams) goto out;
ret = check_results(DEFAULT_SPAN); + if (ret && (get_vendor() == ARCH_INTEL)) + ksft_print_msg("Intel MBM may be inaccurate when Sub-NUMA Clustering is enabled. Check BIOS configuration.\n");
out: mbm_test_cleanup();
return ret; } + +static bool mbm_feature_check(const struct resctrl_test *test) +{ + return validate_resctrl_feature_request("L3_MON", "mbm_total_bytes") && + validate_resctrl_feature_request("L3_MON", "mbm_local_bytes"); +} + +struct resctrl_test mbm_test = { + .name = "MBM", + .resource = "MB", + .vendor_specific = ARCH_INTEL, + .feature_check = mbm_feature_check, + .run_test = mbm_run_test, +}; diff --git a/tools/testing/selftests/resctrl/resctrl.h b/tools/testing/selftests/resctrl/resctrl.h index e61e11ba6a55..d92663593850 100644 --- a/tools/testing/selftests/resctrl/resctrl.h +++ b/tools/testing/selftests/resctrl/resctrl.h @@ -57,6 +57,25 @@ struct user_params { const char *benchmark_cmd[BENCHMARK_ARGS]; };
+/* + * resctrl_test: resctrl test definition + * @name: Test name + * @resource: Resource to test (e.g., MB, L3, L2, etc.) + * @vendor_specific: Bitmask for vendor-specific tests (can be 0 for universal tests) + * @disabled: Test is disabled + * @feature_check: Callback to check required resctrl features + * @run_test: Callback to run the test + */ +struct resctrl_test { + const char *name; + const char *resource; + unsigned int vendor_specific; + bool disabled; + bool (*feature_check)(const struct resctrl_test *test); + int (*run_test)(const struct resctrl_test *test, + const struct user_params *uparams); +}; + /* * resctrl_val_param: resctrl test parameters * @resctrl_val: Resctrl feature (Eg: mbm, mba.. etc) @@ -109,6 +128,7 @@ int mount_resctrlfs(void); int umount_resctrlfs(void); int validate_bw_report_request(char *bw_report); bool validate_resctrl_feature_request(const char *resource, const char *feature); +bool test_resource_feature_check(const struct resctrl_test *test); char *fgrep(FILE *inf, const char *str); int taskset_benchmark(pid_t bm_pid, int cpu_no); int write_schemata(char *ctrlgrp, char *schemata, int cpu_no, @@ -123,10 +143,8 @@ void fill_cache_read(unsigned char *buf, size_t buf_size, bool once); int run_fill_buf(size_t buf_size, int memflush, int op, bool once); int resctrl_val(const struct user_params *uparams, const char * const *benchmark_cmd, struct resctrl_val_param *param); -int mbm_bw_change(const struct user_params *uparams); void tests_cleanup(void); void mbm_test_cleanup(void); -int mba_schemata_change(const struct user_params *uparams); void mba_test_cleanup(void); unsigned long create_bit_mask(unsigned int start, unsigned int len); unsigned int count_contiguous_bits(unsigned long val, unsigned int *start); @@ -138,8 +156,6 @@ void ctrlc_handler(int signum, siginfo_t *info, void *ptr); int signal_handler_register(void); void signal_handler_unregister(void); void cat_test_cleanup(void); -int cat_perf_miss_val(const struct user_params *uparams, char *cache_type); -int cmt_resctrl_val(const struct user_params *uparams); unsigned int count_bits(unsigned long n); void cmt_test_cleanup(void);
@@ -167,4 +183,9 @@ static inline int cache_portion_size(unsigned long cache_size, return cache_size * count_bits(portion_mask) / count_bits(full_cache_mask); }
+extern struct resctrl_test mbm_test; +extern struct resctrl_test mba_test; +extern struct resctrl_test cmt_test; +extern struct resctrl_test l3_cat_test; + #endif /* RESCTRL_H */ diff --git a/tools/testing/selftests/resctrl/resctrl_tests.c b/tools/testing/selftests/resctrl/resctrl_tests.c index f7f11415cc85..fec677d41cf4 100644 --- a/tools/testing/selftests/resctrl/resctrl_tests.c +++ b/tools/testing/selftests/resctrl/resctrl_tests.c @@ -14,6 +14,13 @@ static volatile int sink_target; volatile int *value_sink = &sink_target;
+static struct resctrl_test *resctrl_tests[] = { + &mbm_test, + &mba_test, + &cmt_test, + &l3_cat_test, +}; + static int detect_vendor(void) { FILE *inf = fopen("/proc/cpuinfo", "r"); @@ -53,11 +60,16 @@ int get_vendor(void)
static void cmd_help(void) { + int i; + printf("usage: resctrl_tests [-h] [-t test list] [-n no_of_bits] [-b benchmark_cmd [option]...]\n"); printf("\t-b benchmark_cmd [option]...: run specified benchmark for MBM, MBA and CMT\n"); printf("\t default benchmark is builtin fill_buf\n"); printf("\t-t test list: run tests specified in the test list, "); printf("e.g. -t mbm,mba,cmt,cat\n"); + printf("\t\tSupported tests:\n"); + for (i = 0; i < ARRAY_SIZE(resctrl_tests); i++) + printf("\t\t\t%s\n", resctrl_tests[i]->name); printf("\t-n no_of_bits: run cache tests using specified no of bits in cache bit mask\n"); printf("\t-p cpu_no: specify CPU number to run the test. 1 is default\n"); printf("\t-h: help\n"); @@ -96,102 +108,41 @@ static void test_cleanup(void) signal_handler_unregister(); }
-static void run_mbm_test(const struct user_params *uparams) +static bool test_vendor_specific_check(const struct resctrl_test *test) { - int res; - - ksft_print_msg("Starting MBM BW change ...\n"); - - if (test_prepare()) { - ksft_exit_fail_msg("Abnormal failure when preparing for the test\n"); - return; - } - - if (!validate_resctrl_feature_request("L3_MON", "mbm_total_bytes") || - !validate_resctrl_feature_request("L3_MON", "mbm_local_bytes") || - (get_vendor() != ARCH_INTEL)) { - ksft_test_result_skip("Hardware does not support MBM or MBM is disabled\n"); - goto cleanup; - } - - res = mbm_bw_change(uparams); - ksft_test_result(!res, "MBM: bw change\n"); - if ((get_vendor() == ARCH_INTEL) && res) - ksft_print_msg("Intel MBM may be inaccurate when Sub-NUMA Clustering is enabled. Check BIOS configuration.\n"); + if (!test->vendor_specific) + return true;
-cleanup: - test_cleanup(); + return get_vendor() & test->vendor_specific; }
-static void run_mba_test(const struct user_params *uparams) +static void run_single_test(const struct resctrl_test *test, const struct user_params *uparams) { - int res; - - ksft_print_msg("Starting MBA Schemata change ...\n"); + int ret;
- if (test_prepare()) { - ksft_exit_fail_msg("Abnormal failure when preparing for the test\n"); + if (test->disabled) return; - }
- if (!validate_resctrl_feature_request("MB", NULL) || - !validate_resctrl_feature_request("L3_MON", "mbm_local_bytes") || - (get_vendor() != ARCH_INTEL)) { - ksft_test_result_skip("Hardware does not support MBA or MBA is disabled\n"); - goto cleanup; - } - - res = mba_schemata_change(uparams); - ksft_test_result(!res, "MBA: schemata change\n"); - -cleanup: - test_cleanup(); -} - -static void run_cmt_test(const struct user_params *uparams) -{ - int res; - - ksft_print_msg("Starting CMT test ...\n"); - - if (test_prepare()) { - ksft_exit_fail_msg("Abnormal failure when preparing for the test\n"); + if (!test_vendor_specific_check(test)) { + ksft_test_result_skip("Hardware does not support %s\n", test->name); return; }
- if (!validate_resctrl_feature_request("L3_MON", "llc_occupancy") || - !validate_resctrl_feature_request("L3", NULL)) { - ksft_test_result_skip("Hardware does not support CMT or CMT is disabled\n"); - goto cleanup; - } - - res = cmt_resctrl_val(uparams); - ksft_test_result(!res, "CMT: test\n"); - if ((get_vendor() == ARCH_INTEL) && res) - ksft_print_msg("Intel CMT may be inaccurate when Sub-NUMA Clustering is enabled. Check BIOS configuration.\n"); - -cleanup: - test_cleanup(); -} - -static void run_cat_test(const struct user_params *uparams) -{ - int res; - - ksft_print_msg("Starting CAT test ...\n"); + ksft_print_msg("Starting %s test ...\n", test->name);
if (test_prepare()) { ksft_exit_fail_msg("Abnormal failure when preparing for the test\n"); return; }
- if (!validate_resctrl_feature_request("L3", NULL)) { - ksft_test_result_skip("Hardware does not support CAT or CAT is disabled\n"); + if (!test->feature_check(test)) { + ksft_test_result_skip("Hardware does not support %s or %s is disabled\n", + test->name, test->name); goto cleanup; }
- res = cat_perf_miss_val(uparams, "L3"); - ksft_test_result(!res, "CAT: test\n"); + ret = test->run_test(test, uparams); + ksft_test_result(!ret, "%s: test\n", test->name);
cleanup: test_cleanup(); @@ -205,11 +156,10 @@ static void init_user_params(struct user_params *uparams)
int main(int argc, char **argv) { - bool mbm_test = true, mba_test = true, cmt_test = true; + int tests = ARRAY_SIZE(resctrl_tests); struct user_params uparams = {}; + bool test_param_seen = false; char *span_str = NULL; - bool cat_test = true; - int tests = 0; int ret, c, i;
init_user_params(&uparams); @@ -237,25 +187,26 @@ int main(int argc, char **argv) case 't': token = strtok(optarg, ",");
- mbm_test = false; - mba_test = false; - cmt_test = false; - cat_test = false; + if (!test_param_seen) { + for (i = 0; i < ARRAY_SIZE(resctrl_tests); i++) + resctrl_tests[i]->disabled = true; + tests = 0; + test_param_seen = true; + } while (token) { - if (!strncmp(token, MBM_STR, sizeof(MBM_STR))) { - mbm_test = true; - tests++; - } else if (!strncmp(token, MBA_STR, sizeof(MBA_STR))) { - mba_test = true; - tests++; - } else if (!strncmp(token, CMT_STR, sizeof(CMT_STR))) { - cmt_test = true; - tests++; - } else if (!strncmp(token, CAT_STR, sizeof(CAT_STR))) { - cat_test = true; - tests++; - } else { - printf("invalid argument\n"); + bool found = false; + + for (i = 0; i < ARRAY_SIZE(resctrl_tests); i++) { + if (!strcasecmp(token, resctrl_tests[i]->name)) { + if (resctrl_tests[i]->disabled) + tests++; + resctrl_tests[i]->disabled = false; + found = true; + } + } + + if (!found) { + printf("invalid test: %s\n", token);
return -1; } @@ -315,19 +266,10 @@ int main(int argc, char **argv) uparams.benchmark_cmd[5] = NULL; }
- ksft_set_plan(tests ? : 4); - - if (mbm_test) - run_mbm_test(&uparams); - - if (mba_test) - run_mba_test(&uparams); - - if (cmt_test) - run_cmt_test(&uparams); + ksft_set_plan(tests);
- if (cat_test) - run_cat_test(&uparams); + for (i = 0; i < ARRAY_SIZE(resctrl_tests); i++) + run_single_test(resctrl_tests[i], &uparams);
free(span_str); ksft_finished(); diff --git a/tools/testing/selftests/resctrl/resctrlfs.c b/tools/testing/selftests/resctrl/resctrlfs.c index 9b09210c5ece..b711326b2141 100644 --- a/tools/testing/selftests/resctrl/resctrlfs.c +++ b/tools/testing/selftests/resctrl/resctrlfs.c @@ -679,6 +679,11 @@ bool validate_resctrl_feature_request(const char *resource, const char *feature) return !!res; }
+bool test_resource_feature_check(const struct resctrl_test *test) +{ + return validate_resctrl_feature_request(test->resource, NULL); +} + int filter_dmesg(void) { char line[1024];
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote: ...
+static bool cmt_feature_check(const struct resctrl_test *test) +{
- return validate_resctrl_feature_request("L3_MON", "llc_occupancy") &&
validate_resctrl_feature_request("L3", NULL);+}
...
+static bool mba_feature_check(const struct resctrl_test *test) +{
- return test_resource_feature_check(test) &&
validate_resctrl_feature_request("L3_MON", "mbm_local_bytes");+}
Could cmt_feature_check() not also use test_resource_feature_check(test)? Why are cmt_feature_check() and mba_feature_check() different in this regard?
...
+/*
- resctrl_test: resctrl test definition
- @name: Test name
- @resource: Resource to test (e.g., MB, L3, L2, etc.)
- @vendor_specific: Bitmask for vendor-specific tests (can be 0 for universal tests)
I do not think these values were originally intended to be used in a bitmask. The current values do make this possible but I would like to suggest that their definition gets a comment to highlight how those values are used.
The rest looks good to me. This is a good addition. Thank you.
Reinette
write_schemata() takes the test name as an argument and determines the relevant resource based on the test name. Such mapping from name to resource does not really belong to resctrlfs.c that should provide only generic, test-independent functions.
Pass the resource stored in the test information structure to write_schemata() instead of the test name. The new API is also more flexible as it enables to use write_schemata() for more than one resource within a test.
While touching the sprintf(), move the unnecessary %c that is always '=' directly into the format string.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/cat_test.c | 11 ++++++---- tools/testing/selftests/resctrl/cmt_test.c | 6 +++-- tools/testing/selftests/resctrl/mba_test.c | 9 ++++---- tools/testing/selftests/resctrl/mbm_test.c | 9 ++++---- tools/testing/selftests/resctrl/resctrl.h | 10 +++++---- tools/testing/selftests/resctrl/resctrl_val.c | 7 ++++-- tools/testing/selftests/resctrl/resctrlfs.c | 22 +++++-------------- 7 files changed, 37 insertions(+), 37 deletions(-)
diff --git a/tools/testing/selftests/resctrl/cat_test.c b/tools/testing/selftests/resctrl/cat_test.c index 4b034850603e..293b42ccd39f 100644 --- a/tools/testing/selftests/resctrl/cat_test.c +++ b/tools/testing/selftests/resctrl/cat_test.c @@ -134,6 +134,7 @@ void cat_test_cleanup(void)
/* * cat_test - Execute CAT benchmark and measure cache misses + * @test: Test information structure * @uparams: User supplied parameters * @param: Parameters passed to cat_test() * @span: Buffer size for the benchmark @@ -151,7 +152,9 @@ void cat_test_cleanup(void) * * Return: 0 on success. Non-zero on failure. */ -static int cat_test(const struct user_params *uparams, struct resctrl_val_param *param, +static int cat_test(const struct resctrl_test *test, + const struct user_params *uparams, + struct resctrl_val_param *param, size_t span, unsigned long current_mask) { char *resctrl_val = param->resctrl_val; @@ -192,11 +195,11 @@ static int cat_test(const struct user_params *uparams, struct resctrl_val_param
while (current_mask) { snprintf(schemata, sizeof(schemata), "%lx", param->mask & ~current_mask); - ret = write_schemata("", schemata, uparams->cpu, param->resctrl_val); + ret = write_schemata("", schemata, uparams->cpu, test->resource); if (ret) goto free_buf; snprintf(schemata, sizeof(schemata), "%lx", current_mask); - ret = write_schemata(param->ctrlgrp, schemata, uparams->cpu, param->resctrl_val); + ret = write_schemata(param->ctrlgrp, schemata, uparams->cpu, test->resource); if (ret) goto free_buf;
@@ -273,7 +276,7 @@ static int cat_run_test(const struct resctrl_test *test, const struct user_param
remove(param.filename);
- ret = cat_test(uparams, ¶m, span, start_mask); + ret = cat_test(test, uparams, ¶m, span, start_mask); if (ret) goto out;
diff --git a/tools/testing/selftests/resctrl/cmt_test.c b/tools/testing/selftests/resctrl/cmt_test.c index 89e1928761f3..312dd93f8ec7 100644 --- a/tools/testing/selftests/resctrl/cmt_test.c +++ b/tools/testing/selftests/resctrl/cmt_test.c @@ -16,7 +16,9 @@ #define MAX_DIFF 2000000 #define MAX_DIFF_PERCENT 15
-static int cmt_setup(const struct user_params *uparams, struct resctrl_val_param *p) +static int cmt_setup(const struct resctrl_test *test, + const struct user_params *uparams, + struct resctrl_val_param *p) { /* Run NUM_OF_RUNS times */ if (p->num_of_runs >= NUM_OF_RUNS) @@ -150,7 +152,7 @@ static int cmt_run_test(const struct resctrl_test *test, const struct user_param
remove(RESULT_FILE_NAME);
- ret = resctrl_val(uparams, cmd, ¶m); + ret = resctrl_val(test, uparams, cmd, ¶m); if (ret) goto out;
diff --git a/tools/testing/selftests/resctrl/mba_test.c b/tools/testing/selftests/resctrl/mba_test.c index e4870f96d7ea..85d3c24e4976 100644 --- a/tools/testing/selftests/resctrl/mba_test.c +++ b/tools/testing/selftests/resctrl/mba_test.c @@ -22,7 +22,9 @@ * con_mon grp, mon_grp in resctrl FS. * For each allocation, run 5 times in order to get average values. */ -static int mba_setup(const struct user_params *uparams, struct resctrl_val_param *p) +static int mba_setup(const struct resctrl_test *test, + const struct user_params *uparams, + struct resctrl_val_param *p) { static int runs_per_allocation, allocation = 100; char allocation_str[64]; @@ -40,8 +42,7 @@ static int mba_setup(const struct user_params *uparams, struct resctrl_val_param
sprintf(allocation_str, "%d", allocation);
- ret = write_schemata(p->ctrlgrp, allocation_str, uparams->cpu, - p->resctrl_val); + ret = write_schemata(p->ctrlgrp, allocation_str, uparams->cpu, test->resource); if (ret < 0) return ret;
@@ -155,7 +156,7 @@ static int mba_run_test(const struct resctrl_test *test, const struct user_param
remove(RESULT_FILE_NAME);
- ret = resctrl_val(uparams, uparams->benchmark_cmd, ¶m); + ret = resctrl_val(test, uparams, uparams->benchmark_cmd, ¶m); if (ret) goto out;
diff --git a/tools/testing/selftests/resctrl/mbm_test.c b/tools/testing/selftests/resctrl/mbm_test.c index 2274a5eebbca..e3c3fc8fed8d 100644 --- a/tools/testing/selftests/resctrl/mbm_test.c +++ b/tools/testing/selftests/resctrl/mbm_test.c @@ -86,7 +86,9 @@ static int check_results(size_t span) return ret; }
-static int mbm_setup(const struct user_params *uparams, struct resctrl_val_param *p) +static int mbm_setup(const struct resctrl_test *test, + const struct user_params *uparams, + struct resctrl_val_param *p) { int ret = 0;
@@ -96,8 +98,7 @@ static int mbm_setup(const struct user_params *uparams, struct resctrl_val_param
/* Set up shemata with 100% allocation on the first run. */ if (p->num_of_runs == 0 && validate_resctrl_feature_request("MB", NULL)) - ret = write_schemata(p->ctrlgrp, "100", uparams->cpu, - p->resctrl_val); + ret = write_schemata(p->ctrlgrp, "100", uparams->cpu, test->resource);
p->num_of_runs++;
@@ -123,7 +124,7 @@ static int mbm_run_test(const struct resctrl_test *test, const struct user_param
remove(RESULT_FILE_NAME);
- ret = resctrl_val(uparams, uparams->benchmark_cmd, ¶m); + ret = resctrl_val(test, uparams, uparams->benchmark_cmd, ¶m); if (ret) goto out;
diff --git a/tools/testing/selftests/resctrl/resctrl.h b/tools/testing/selftests/resctrl/resctrl.h index d92663593850..ed7a69578daf 100644 --- a/tools/testing/selftests/resctrl/resctrl.h +++ b/tools/testing/selftests/resctrl/resctrl.h @@ -93,7 +93,8 @@ struct resctrl_val_param { char *bw_report; unsigned long mask; int num_of_runs; - int (*setup)(const struct user_params *uparams, + int (*setup)(const struct resctrl_test *test, + const struct user_params *uparams, struct resctrl_val_param *param); };
@@ -131,8 +132,7 @@ bool validate_resctrl_feature_request(const char *resource, const char *feature) bool test_resource_feature_check(const struct resctrl_test *test); char *fgrep(FILE *inf, const char *str); int taskset_benchmark(pid_t bm_pid, int cpu_no); -int write_schemata(char *ctrlgrp, char *schemata, int cpu_no, - char *resctrl_val); +int write_schemata(char *ctrlgrp, char *schemata, int cpu_no, const char *resource); int write_bm_pid_to_resctrl(pid_t bm_pid, char *ctrlgrp, char *mongrp, char *resctrl_val); int perf_event_open(struct perf_event_attr *hw_event, pid_t pid, int cpu, @@ -141,7 +141,9 @@ unsigned char *alloc_buffer(size_t buf_size, int memflush); void mem_flush(unsigned char *buf, size_t buf_size); void fill_cache_read(unsigned char *buf, size_t buf_size, bool once); int run_fill_buf(size_t buf_size, int memflush, int op, bool once); -int resctrl_val(const struct user_params *uparams, const char * const *benchmark_cmd, +int resctrl_val(const struct resctrl_test *test, + const struct user_params *uparams, + const char * const *benchmark_cmd, struct resctrl_val_param *param); void tests_cleanup(void); void mbm_test_cleanup(void); diff --git a/tools/testing/selftests/resctrl/resctrl_val.c b/tools/testing/selftests/resctrl/resctrl_val.c index 24bf91cfdb21..4e0c356b7f18 100644 --- a/tools/testing/selftests/resctrl/resctrl_val.c +++ b/tools/testing/selftests/resctrl/resctrl_val.c @@ -679,13 +679,16 @@ static void run_benchmark(int signum, siginfo_t *info, void *ucontext) /* * resctrl_val: execute benchmark and measure memory bandwidth on * the benchmark + * @test: test information structure * @uparams: user supplied parameters * @benchmark_cmd: benchmark command and its arguments * @param: parameters passed to resctrl_val() * * Return: 0 on success. non-zero on failure. */ -int resctrl_val(const struct user_params *uparams, const char * const *benchmark_cmd, +int resctrl_val(const struct resctrl_test *test, + const struct user_params *uparams, + const char * const *benchmark_cmd, struct resctrl_val_param *param) { char *resctrl_val = param->resctrl_val; @@ -818,7 +821,7 @@ int resctrl_val(const struct user_params *uparams, const char * const *benchmark
/* Test runs until the callback setup() tells the test to stop. */ while (1) { - ret = param->setup(uparams, param); + ret = param->setup(test, uparams, param); if (ret == END_OF_TESTS) { ret = 0; break; diff --git a/tools/testing/selftests/resctrl/resctrlfs.c b/tools/testing/selftests/resctrl/resctrlfs.c index b711326b2141..fda5ad812faa 100644 --- a/tools/testing/selftests/resctrl/resctrlfs.c +++ b/tools/testing/selftests/resctrl/resctrlfs.c @@ -502,23 +502,17 @@ int write_bm_pid_to_resctrl(pid_t bm_pid, char *ctrlgrp, char *mongrp, * @ctrlgrp: Name of the con_mon grp * @schemata: Schemata that should be updated to * @cpu_no: CPU number that the benchmark PID is binded to - * @resctrl_val: Resctrl feature (Eg: mbm, mba.. etc) + * @resource: Resctrl resource (Eg: MB, L3, L2, etc.) * * Update schemata of a con_mon grp *only* if requested resctrl feature is * allocation type * * Return: 0 on success, non-zero on failure */ -int write_schemata(char *ctrlgrp, char *schemata, int cpu_no, char *resctrl_val) +int write_schemata(char *ctrlgrp, char *schemata, int cpu_no, const char *resource) { char controlgroup[1024], reason[128], schema[1024] = {}; - int resource_id, fd, schema_len = -1, ret = 0; - - if (strncmp(resctrl_val, MBA_STR, sizeof(MBA_STR)) && - strncmp(resctrl_val, MBM_STR, sizeof(MBM_STR)) && - strncmp(resctrl_val, CAT_STR, sizeof(CAT_STR)) && - strncmp(resctrl_val, CMT_STR, sizeof(CMT_STR))) - return -ENOENT; + int resource_id, fd, schema_len, ret = 0;
if (!schemata) { ksft_print_msg("Skipping empty schemata update\n"); @@ -538,14 +532,8 @@ int write_schemata(char *ctrlgrp, char *schemata, int cpu_no, char *resctrl_val) else sprintf(controlgroup, "%s/schemata", RESCTRL_PATH);
- if (!strncmp(resctrl_val, CAT_STR, sizeof(CAT_STR)) || - !strncmp(resctrl_val, CMT_STR, sizeof(CMT_STR))) - schema_len = snprintf(schema, sizeof(schema), "%s%d%c%s\n", - "L3:", resource_id, '=', schemata); - if (!strncmp(resctrl_val, MBA_STR, sizeof(MBA_STR)) || - !strncmp(resctrl_val, MBM_STR, sizeof(MBM_STR))) - schema_len = snprintf(schema, sizeof(schema), "%s%d%c%s\n", - "MB:", resource_id, '=', schemata); + schema_len = snprintf(schema, sizeof(schema), "%s:%d=%s\n", + resource, resource_id, schemata); if (schema_len < 0 || schema_len >= sizeof(schema)) { snprintf(reason, sizeof(reason), "snprintf() failed with return value : %d", schema_len);
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
write_schemata() takes the test name as an argument and determines the relevant resource based on the test name. Such mapping from name to resource does not really belong to resctrlfs.c that should provide only generic, test-independent functions.
Pass the resource stored in the test information structure to write_schemata() instead of the test name. The new API is also more flexible as it enables to use write_schemata() for more than one resource within a test.
While touching the sprintf(), move the unnecessary %c that is always '=' directly into the format string.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
...
break;diff --git a/tools/testing/selftests/resctrl/resctrlfs.c b/tools/testing/selftests/resctrl/resctrlfs.c index b711326b2141..fda5ad812faa 100644 --- a/tools/testing/selftests/resctrl/resctrlfs.c +++ b/tools/testing/selftests/resctrl/resctrlfs.c @@ -502,23 +502,17 @@ int write_bm_pid_to_resctrl(pid_t bm_pid, char *ctrlgrp, char *mongrp,
- @ctrlgrp: Name of the con_mon grp
- @schemata: Schemata that should be updated to
- @cpu_no: CPU number that the benchmark PID is binded to
- @resctrl_val: Resctrl feature (Eg: mbm, mba.. etc)
- @resource: Resctrl resource (Eg: MB, L3, L2, etc.)
- Update schemata of a con_mon grp *only* if requested resctrl feature is
- allocation type
Note above there remains a usage of "feature" that has just been changed to "resource".
Reviewed-by: Reinette Chatre reinette.chatre@intel.com
Reinette
"L2"/"L3" conversion to integer is embedded into get_cache_size() which prevents reuse.
Create a helper for the cache string to integer conversion to make it reusable.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/resctrlfs.c | 28 +++++++++++++++------ 1 file changed, 20 insertions(+), 8 deletions(-)
diff --git a/tools/testing/selftests/resctrl/resctrlfs.c b/tools/testing/selftests/resctrl/resctrlfs.c index fda5ad812faa..38ca3ae562e9 100644 --- a/tools/testing/selftests/resctrl/resctrlfs.c +++ b/tools/testing/selftests/resctrl/resctrlfs.c @@ -94,6 +94,23 @@ int umount_resctrlfs(void) return 0; }
+/* + * get_cache_level - Convert cache level from string to integer + * @cache_type: Cache level as string + * + * Return: cache level as integer or -1 if @cache_type is invalid. + */ +static int get_cache_level(const char *cache_type) +{ + if (!strcmp(cache_type, "L3")) + return 3; + if (!strcmp(cache_type, "L2")) + return 2; + + perror("Invalid cache level"); + return -1; +} + /* * get_resource_id - Get socket number/l3 id for a specified CPU * @cpu_no: CPU number @@ -144,14 +161,9 @@ int get_cache_size(int cpu_no, const char *cache_type, unsigned long *cache_size int length, i, cache_num; FILE *fp;
- if (!strcmp(cache_type, "L3")) { - cache_num = 3; - } else if (!strcmp(cache_type, "L2")) { - cache_num = 2; - } else { - perror("Invalid cache level"); - return -1; - } + cache_num = get_cache_level(cache_type); + if (cache_num < 0) + return cache_num;
sprintf(cache_path, "/sys/bus/cpu/devices/cpu%d/cache/index%d/size", cpu_no, cache_num);
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
"L2"/"L3" conversion to integer is embedded into get_cache_size() which prevents reuse.
Create a helper for the cache string to integer conversion to make it reusable.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
tools/testing/selftests/resctrl/resctrlfs.c | 28 +++++++++++++++------ 1 file changed, 20 insertions(+), 8 deletions(-)
diff --git a/tools/testing/selftests/resctrl/resctrlfs.c b/tools/testing/selftests/resctrl/resctrlfs.c index fda5ad812faa..38ca3ae562e9 100644 --- a/tools/testing/selftests/resctrl/resctrlfs.c +++ b/tools/testing/selftests/resctrl/resctrlfs.c @@ -94,6 +94,23 @@ int umount_resctrlfs(void) return 0; } +/*
- get_cache_level - Convert cache level from string to integer
- @cache_type: Cache level as string
- Return: cache level as integer or -1 if @cache_type is invalid.
- */
+static int get_cache_level(const char *cache_type) +{
- if (!strcmp(cache_type, "L3"))
return 3;- if (!strcmp(cache_type, "L2"))
return 2;- perror("Invalid cache level");
I know that you are just copying code here but this usage of perror() does not look right. strcmp() does not set errno, it does not "fail".
Reinette
Kernel-side calls the instances of a resource domains.
Change the resource_id naming in the selftest code to domain_id to match the kernel side better.
Suggested-by: Maciej Wieczór-Retman maciej.wieczor-retman@intel.com Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/resctrl.h | 2 +- tools/testing/selftests/resctrl/resctrl_val.c | 30 +++++++++---------- tools/testing/selftests/resctrl/resctrlfs.c | 18 +++++------ 3 files changed, 25 insertions(+), 25 deletions(-)
diff --git a/tools/testing/selftests/resctrl/resctrl.h b/tools/testing/selftests/resctrl/resctrl.h index ed7a69578daf..280ab6b6bd07 100644 --- a/tools/testing/selftests/resctrl/resctrl.h +++ b/tools/testing/selftests/resctrl/resctrl.h @@ -124,7 +124,7 @@ extern char llc_occup_path[1024]; int get_vendor(void); bool check_resctrlfs_support(void); int filter_dmesg(void); -int get_resource_id(int cpu_no, int *resource_id); +int get_domain_id(int cpu_no, int *domain_id); int mount_resctrlfs(void); int umount_resctrlfs(void); int validate_bw_report_request(char *bw_report); diff --git a/tools/testing/selftests/resctrl/resctrl_val.c b/tools/testing/selftests/resctrl/resctrl_val.c index 4e0c356b7f18..f278f5c9195c 100644 --- a/tools/testing/selftests/resctrl/resctrl_val.c +++ b/tools/testing/selftests/resctrl/resctrl_val.c @@ -387,20 +387,20 @@ static int get_mem_bw_imc(int cpu_no, char *bw_report, float *bw_imc) return 0; }
-void set_mbm_path(const char *ctrlgrp, const char *mongrp, int resource_id) +void set_mbm_path(const char *ctrlgrp, const char *mongrp, int domain_id) { if (ctrlgrp && mongrp) sprintf(mbm_total_path, CON_MON_MBM_LOCAL_BYTES_PATH, - RESCTRL_PATH, ctrlgrp, mongrp, resource_id); + RESCTRL_PATH, ctrlgrp, mongrp, domain_id); else if (!ctrlgrp && mongrp) sprintf(mbm_total_path, MON_MBM_LOCAL_BYTES_PATH, RESCTRL_PATH, - mongrp, resource_id); + mongrp, domain_id); else if (ctrlgrp && !mongrp) sprintf(mbm_total_path, CON_MBM_LOCAL_BYTES_PATH, RESCTRL_PATH, - ctrlgrp, resource_id); + ctrlgrp, domain_id); else if (!ctrlgrp && !mongrp) sprintf(mbm_total_path, MBM_LOCAL_BYTES_PATH, RESCTRL_PATH, - resource_id); + domain_id); }
/* @@ -413,23 +413,23 @@ void set_mbm_path(const char *ctrlgrp, const char *mongrp, int resource_id) static void initialize_mem_bw_resctrl(const char *ctrlgrp, const char *mongrp, int cpu_no, char *resctrl_val) { - int resource_id; + int domain_id;
- if (get_resource_id(cpu_no, &resource_id) < 0) { - perror("Could not get resource_id"); + if (get_domain_id(cpu_no, &domain_id) < 0) { + perror("Could not get domain ID"); return; }
if (!strncmp(resctrl_val, MBM_STR, sizeof(MBM_STR))) - set_mbm_path(ctrlgrp, mongrp, resource_id); + set_mbm_path(ctrlgrp, mongrp, domain_id);
if (!strncmp(resctrl_val, MBA_STR, sizeof(MBA_STR))) { if (ctrlgrp) sprintf(mbm_total_path, CON_MBM_LOCAL_BYTES_PATH, - RESCTRL_PATH, ctrlgrp, resource_id); + RESCTRL_PATH, ctrlgrp, domain_id); else sprintf(mbm_total_path, MBM_LOCAL_BYTES_PATH, - RESCTRL_PATH, resource_id); + RESCTRL_PATH, domain_id); } }
@@ -582,15 +582,15 @@ static void set_cmt_path(const char *ctrlgrp, const char *mongrp, char sock_num) static void initialize_llc_occu_resctrl(const char *ctrlgrp, const char *mongrp, int cpu_no, char *resctrl_val) { - int resource_id; + int domain_id;
- if (get_resource_id(cpu_no, &resource_id) < 0) { - perror("# Unable to resource_id"); + if (get_domain_id(cpu_no, &domain_id) < 0) { + perror("# Could not get domain ID"); return; }
if (!strncmp(resctrl_val, CMT_STR, sizeof(CMT_STR))) - set_cmt_path(ctrlgrp, mongrp, resource_id); + set_cmt_path(ctrlgrp, mongrp, domain_id); }
static int measure_vals(const struct user_params *uparams, diff --git a/tools/testing/selftests/resctrl/resctrlfs.c b/tools/testing/selftests/resctrl/resctrlfs.c index 38ca3ae562e9..c173e0fa5c94 100644 --- a/tools/testing/selftests/resctrl/resctrlfs.c +++ b/tools/testing/selftests/resctrl/resctrlfs.c @@ -112,13 +112,13 @@ static int get_cache_level(const char *cache_type) }
/* - * get_resource_id - Get socket number/l3 id for a specified CPU + * get_domain_id - Get resctrl domain ID for a specified CPU * @cpu_no: CPU number - * @resource_id: Socket number or l3_id + * @domain_id: domain ID (cache ID; for MB, L3 cache ID) * * Return: >= 0 on success, < 0 on failure. */ -int get_resource_id(int cpu_no, int *resource_id) +int get_domain_id(int cpu_no, int *domain_id) { char phys_pkg_path[1024]; FILE *fp; @@ -136,8 +136,8 @@ int get_resource_id(int cpu_no, int *resource_id)
return -1; } - if (fscanf(fp, "%d", resource_id) <= 0) { - perror("Could not get socket number or l3 id"); + if (fscanf(fp, "%d", domain_id) <= 0) { + perror("Could not get domain ID"); fclose(fp);
return -1; @@ -524,7 +524,7 @@ int write_bm_pid_to_resctrl(pid_t bm_pid, char *ctrlgrp, char *mongrp, int write_schemata(char *ctrlgrp, char *schemata, int cpu_no, const char *resource) { char controlgroup[1024], reason[128], schema[1024] = {}; - int resource_id, fd, schema_len, ret = 0; + int domain_id, fd, schema_len, ret = 0;
if (!schemata) { ksft_print_msg("Skipping empty schemata update\n"); @@ -532,8 +532,8 @@ int write_schemata(char *ctrlgrp, char *schemata, int cpu_no, const char *resour return -1; }
- if (get_resource_id(cpu_no, &resource_id) < 0) { - sprintf(reason, "Failed to get resource id"); + if (get_domain_id(cpu_no, &domain_id) < 0) { + sprintf(reason, "Failed to get domain ID"); ret = -1;
goto out; @@ -545,7 +545,7 @@ int write_schemata(char *ctrlgrp, char *schemata, int cpu_no, const char *resour sprintf(controlgroup, "%s/schemata", RESCTRL_PATH);
schema_len = snprintf(schema, sizeof(schema), "%s:%d=%s\n", - resource, resource_id, schemata); + resource, domain_id, schemata); if (schema_len < 0 || schema_len >= sizeof(schema)) { snprintf(reason, sizeof(reason), "snprintf() failed with return value : %d", schema_len);
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
Kernel-side calls the instances of a resource domains.
Change the resource_id naming in the selftest code to domain_id to match the kernel side better.
Suggested-by: Maciej Wieczór-Retman maciej.wieczor-retman@intel.com Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
tools/testing/selftests/resctrl/resctrl.h | 2 +- tools/testing/selftests/resctrl/resctrl_val.c | 30 +++++++++---------- tools/testing/selftests/resctrl/resctrlfs.c | 18 +++++------ 3 files changed, 25 insertions(+), 25 deletions(-)
diff --git a/tools/testing/selftests/resctrl/resctrl.h b/tools/testing/selftests/resctrl/resctrl.h index ed7a69578daf..280ab6b6bd07 100644 --- a/tools/testing/selftests/resctrl/resctrl.h +++ b/tools/testing/selftests/resctrl/resctrl.h @@ -124,7 +124,7 @@ extern char llc_occup_path[1024]; int get_vendor(void); bool check_resctrlfs_support(void); int filter_dmesg(void); -int get_resource_id(int cpu_no, int *resource_id); +int get_domain_id(int cpu_no, int *domain_id); int mount_resctrlfs(void); int umount_resctrlfs(void); int validate_bw_report_request(char *bw_report); diff --git a/tools/testing/selftests/resctrl/resctrl_val.c b/tools/testing/selftests/resctrl/resctrl_val.c index 4e0c356b7f18..f278f5c9195c 100644 --- a/tools/testing/selftests/resctrl/resctrl_val.c +++ b/tools/testing/selftests/resctrl/resctrl_val.c @@ -387,20 +387,20 @@ static int get_mem_bw_imc(int cpu_no, char *bw_report, float *bw_imc) return 0; } -void set_mbm_path(const char *ctrlgrp, const char *mongrp, int resource_id) +void set_mbm_path(const char *ctrlgrp, const char *mongrp, int domain_id) { if (ctrlgrp && mongrp) sprintf(mbm_total_path, CON_MON_MBM_LOCAL_BYTES_PATH,
RESCTRL_PATH, ctrlgrp, mongrp, resource_id);
RESCTRL_PATH, ctrlgrp, mongrp, domain_id);
mongrp, resource_id);
mongrp, domain_id);
ctrlgrp, resource_id);
ctrlgrp, domain_id);
resource_id);
domain_id);} /* @@ -413,23 +413,23 @@ void set_mbm_path(const char *ctrlgrp, const char *mongrp, int resource_id) static void initialize_mem_bw_resctrl(const char *ctrlgrp, const char *mongrp, int cpu_no, char *resctrl_val) {
- int resource_id;
- int domain_id;
- if (get_resource_id(cpu_no, &resource_id) < 0) {
perror("Could not get resource_id");
- if (get_domain_id(cpu_no, &domain_id) < 0) {
perror("Could not get domain ID");
I do not think perror() is appropriate here. It is too far away from failing system call (for which perror() was already called anyway).
return;} if (!strncmp(resctrl_val, MBM_STR, sizeof(MBM_STR)))
set_mbm_path(ctrlgrp, mongrp, resource_id);
set_mbm_path(ctrlgrp, mongrp, domain_id);if (!strncmp(resctrl_val, MBA_STR, sizeof(MBA_STR))) { if (ctrlgrp) sprintf(mbm_total_path, CON_MBM_LOCAL_BYTES_PATH,
RESCTRL_PATH, ctrlgrp, resource_id);
RESCTRL_PATH, ctrlgrp, domain_id);
RESCTRL_PATH, resource_id);
RESCTRL_PATH, domain_id);} @@ -582,15 +582,15 @@ static void set_cmt_path(const char *ctrlgrp, const char *mongrp, char sock_num) static void initialize_llc_occu_resctrl(const char *ctrlgrp, const char *mongrp, int cpu_no, char *resctrl_val) {
- int resource_id;
- int domain_id;
- if (get_resource_id(cpu_no, &resource_id) < 0) {
perror("# Unable to resource_id");
- if (get_domain_id(cpu_no, &domain_id) < 0) {
perror("# Could not get domain ID");
here also
return;} if (!strncmp(resctrl_val, CMT_STR, sizeof(CMT_STR)))
set_cmt_path(ctrlgrp, mongrp, resource_id);
set_cmt_path(ctrlgrp, mongrp, domain_id);} static int measure_vals(const struct user_params *uparams,
Reinette
Domain id is acquired differently depending on CPU. AMD tests use id from L3 cache, whereas CPUs from other vendors base the id on topology package id. In order to support L2 CAT test, this has to be generalized.
The driver side code seems to get the domain ids from cache ids so the approach used by the AMD branch seems to match the kernel-side code. It will also work with L2 domain IDs as long as the cache level is generalized.
Using the topology id was always fragile due to mismatch with the kernel-side way to acquire the domain id. It got incorrect domain id, e.g., when Cluster-on-Die (CoD) is enabled for CPU (but CoD is not well suited for resctrl in the first place so it has not been a big issue if tests don't work correctly with it).
Taking all the above into account, generalize acquiring the domain id by taking it from the cache id and do not hard-code the cache level.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/resctrl.h | 2 +- tools/testing/selftests/resctrl/resctrl_val.c | 4 +-- tools/testing/selftests/resctrl/resctrlfs.c | 27 ++++++++++++------- 3 files changed, 21 insertions(+), 12 deletions(-)
diff --git a/tools/testing/selftests/resctrl/resctrl.h b/tools/testing/selftests/resctrl/resctrl.h index 280ab6b6bd07..9b49065cc6da 100644 --- a/tools/testing/selftests/resctrl/resctrl.h +++ b/tools/testing/selftests/resctrl/resctrl.h @@ -124,7 +124,7 @@ extern char llc_occup_path[1024]; int get_vendor(void); bool check_resctrlfs_support(void); int filter_dmesg(void); -int get_domain_id(int cpu_no, int *domain_id); +int get_domain_id(const char *resource, int cpu_no, int *domain_id); int mount_resctrlfs(void); int umount_resctrlfs(void); int validate_bw_report_request(char *bw_report); diff --git a/tools/testing/selftests/resctrl/resctrl_val.c b/tools/testing/selftests/resctrl/resctrl_val.c index f278f5c9195c..5a072adfe8d6 100644 --- a/tools/testing/selftests/resctrl/resctrl_val.c +++ b/tools/testing/selftests/resctrl/resctrl_val.c @@ -415,7 +415,7 @@ static void initialize_mem_bw_resctrl(const char *ctrlgrp, const char *mongrp, { int domain_id;
- if (get_domain_id(cpu_no, &domain_id) < 0) { + if (get_domain_id("MB", cpu_no, &domain_id) < 0) { perror("Could not get domain ID"); return; } @@ -584,7 +584,7 @@ static void initialize_llc_occu_resctrl(const char *ctrlgrp, const char *mongrp, { int domain_id;
- if (get_domain_id(cpu_no, &domain_id) < 0) { + if (get_domain_id("L3", cpu_no, &domain_id) < 0) { perror("# Could not get domain ID"); return; } diff --git a/tools/testing/selftests/resctrl/resctrlfs.c b/tools/testing/selftests/resctrl/resctrlfs.c index c173e0fa5c94..5ca3022d8aa6 100644 --- a/tools/testing/selftests/resctrl/resctrlfs.c +++ b/tools/testing/selftests/resctrl/resctrlfs.c @@ -111,28 +111,37 @@ static int get_cache_level(const char *cache_type) return -1; }
+static int get_resource_cache_level(const char *resource) +{ + /* "MB" use L3 (LLC) as resource */ + if (!strcmp(resource, "MB")) + return 3; + return get_cache_level(resource); +} + /* * get_domain_id - Get resctrl domain ID for a specified CPU + * @resource: resource name * @cpu_no: CPU number * @domain_id: domain ID (cache ID; for MB, L3 cache ID) * * Return: >= 0 on success, < 0 on failure. */ -int get_domain_id(int cpu_no, int *domain_id) +int get_domain_id(const char *resource, int cpu_no, int *domain_id) { char phys_pkg_path[1024]; + int cache_num; FILE *fp;
- if (get_vendor() == ARCH_AMD) - sprintf(phys_pkg_path, "%s%d/cache/index3/id", - PHYS_ID_PATH, cpu_no); - else - sprintf(phys_pkg_path, "%s%d/topology/physical_package_id", - PHYS_ID_PATH, cpu_no); + cache_num = get_resource_cache_level(resource); + if (cache_num < 0) + return cache_num; + + sprintf(phys_pkg_path, "%s%d/cache/index%d/id", PHYS_ID_PATH, cpu_no, cache_num);
fp = fopen(phys_pkg_path, "r"); if (!fp) { - perror("Failed to open physical_package_id"); + perror("Failed to open cache id file");
return -1; } @@ -532,7 +541,7 @@ int write_schemata(char *ctrlgrp, char *schemata, int cpu_no, const char *resour return -1; }
- if (get_domain_id(cpu_no, &domain_id) < 0) { + if (get_domain_id(resource, cpu_no, &domain_id) < 0) { sprintf(reason, "Failed to get domain ID"); ret = -1;
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
Domain id is acquired differently depending on CPU. AMD tests use id from L3 cache, whereas CPUs from other vendors base the id on topology package id. In order to support L2 CAT test, this has to be generalized.
The driver side code seems to get the domain ids from cache ids so the approach used by the AMD branch seems to match the kernel-side code. It will also work with L2 domain IDs as long as the cache level is generalized.
Using the topology id was always fragile due to mismatch with the kernel-side way to acquire the domain id. It got incorrect domain id, e.g., when Cluster-on-Die (CoD) is enabled for CPU (but CoD is not well suited for resctrl in the first place so it has not been a big issue if tests don't work correctly with it).
Taking all the above into account, generalize acquiring the domain id by taking it from the cache id and do not hard-code the cache level.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
Reviewed-by: Reinette Chatre reinette.chatre@intel.com
Reinette
To select test to run -t parameter can be used. However, -t cat currently maps to L3 CAT test which is confusing after more CAT related tests are added.
Allow selecting tests as groups and call L3 CAT test "L3_CAT", "CAT" group will enable all CAT related tests.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com --- tools/testing/selftests/resctrl/cat_test.c | 3 ++- tools/testing/selftests/resctrl/resctrl.h | 2 ++ tools/testing/selftests/resctrl/resctrl_tests.c | 16 +++++++++++----- 3 files changed, 15 insertions(+), 6 deletions(-)
diff --git a/tools/testing/selftests/resctrl/cat_test.c b/tools/testing/selftests/resctrl/cat_test.c index 293b42ccd39f..869fd280c993 100644 --- a/tools/testing/selftests/resctrl/cat_test.c +++ b/tools/testing/selftests/resctrl/cat_test.c @@ -289,7 +289,8 @@ static int cat_run_test(const struct resctrl_test *test, const struct user_param }
struct resctrl_test l3_cat_test = { - .name = "CAT", + .name = "L3_CAT", + .group = "CAT", .resource = "L3", .feature_check = test_resource_feature_check, .run_test = cat_run_test, diff --git a/tools/testing/selftests/resctrl/resctrl.h b/tools/testing/selftests/resctrl/resctrl.h index 9b49065cc6da..2f0fa8a604e6 100644 --- a/tools/testing/selftests/resctrl/resctrl.h +++ b/tools/testing/selftests/resctrl/resctrl.h @@ -60,6 +60,7 @@ struct user_params { /* * resctrl_test: resctrl test definition * @name: Test name + * @group: Test group (e.g., L2 and L3 CAT test belong to CAT group), can be NULL * @resource: Resource to test (e.g., MB, L3, L2, etc.) * @vendor_specific: Bitmask for vendor-specific tests (can be 0 for universal tests) * @disabled: Test is disabled @@ -68,6 +69,7 @@ struct user_params { */ struct resctrl_test { const char *name; + const char *group; const char *resource; unsigned int vendor_specific; bool disabled; diff --git a/tools/testing/selftests/resctrl/resctrl_tests.c b/tools/testing/selftests/resctrl/resctrl_tests.c index fec677d41cf4..3d8fd1d3ca43 100644 --- a/tools/testing/selftests/resctrl/resctrl_tests.c +++ b/tools/testing/selftests/resctrl/resctrl_tests.c @@ -65,11 +65,15 @@ static void cmd_help(void) printf("usage: resctrl_tests [-h] [-t test list] [-n no_of_bits] [-b benchmark_cmd [option]...]\n"); printf("\t-b benchmark_cmd [option]...: run specified benchmark for MBM, MBA and CMT\n"); printf("\t default benchmark is builtin fill_buf\n"); - printf("\t-t test list: run tests specified in the test list, "); + printf("\t-t test list: run tests/groups specified by the list, "); printf("e.g. -t mbm,mba,cmt,cat\n"); - printf("\t\tSupported tests:\n"); - for (i = 0; i < ARRAY_SIZE(resctrl_tests); i++) - printf("\t\t\t%s\n", resctrl_tests[i]->name); + printf("\t\tSupported tests (group):\n"); + for (i = 0; i < ARRAY_SIZE(resctrl_tests); i++) { + if (resctrl_tests[i]->group) + printf("\t\t\t%s (%s)\n", resctrl_tests[i]->name, resctrl_tests[i]->group); + else + printf("\t\t\t%s\n", resctrl_tests[i]->name); + } printf("\t-n no_of_bits: run cache tests using specified no of bits in cache bit mask\n"); printf("\t-p cpu_no: specify CPU number to run the test. 1 is default\n"); printf("\t-h: help\n"); @@ -197,7 +201,9 @@ int main(int argc, char **argv) bool found = false;
for (i = 0; i < ARRAY_SIZE(resctrl_tests); i++) { - if (!strcasecmp(token, resctrl_tests[i]->name)) { + if (!strcasecmp(token, resctrl_tests[i]->name) || + (resctrl_tests[i]->group && + !strcasecmp(token, resctrl_tests[i]->group))) { if (resctrl_tests[i]->disabled) tests++; resctrl_tests[i]->disabled = false;
Hi Ilpo,
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
To select test to run -t parameter can be used. However, -t cat currently maps to L3 CAT test which is confusing after more CAT related tests are added.
Allow selecting tests as groups and call L3 CAT test "L3_CAT", "CAT" group will enable all CAT related tests.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
Could you please defer this patch to accompany the series that introduces other CAT related tests?
Reinette
On Tue, 28 Nov 2023, Reinette Chatre wrote:
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
To select test to run -t parameter can be used. However, -t cat currently maps to L3 CAT test which is confusing after more CAT related tests are added.
Allow selecting tests as groups and call L3 CAT test "L3_CAT", "CAT" group will enable all CAT related tests.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
Could you please defer this patch to accompany the series that introduces other CAT related tests?
Can be done but it implies Maciej non-continuous CAT test series would still be blocked by the lack of this feature which is why I included it here.
But perhaps Maciej wants to take this patch and submit it as a part of his series. Maciej, are you okay with that change of patch ownership?
On 2023-11-29 at 13:11:44 +0200, Ilpo Järvinen wrote:
On Tue, 28 Nov 2023, Reinette Chatre wrote:
On 11/20/2023 3:13 AM, Ilpo Järvinen wrote:
To select test to run -t parameter can be used. However, -t cat currently maps to L3 CAT test which is confusing after more CAT related tests are added.
Allow selecting tests as groups and call L3 CAT test "L3_CAT", "CAT" group will enable all CAT related tests.
Signed-off-by: Ilpo Järvinen ilpo.jarvinen@linux.intel.com
Could you please defer this patch to accompany the series that introduces other CAT related tests?
Can be done but it implies Maciej non-continuous CAT test series would still be blocked by the lack of this feature which is why I included it here.
But perhaps Maciej wants to take this patch and submit it as a part of his series. Maciej, are you okay with that change of patch ownership?
Sure, I'll try to send a new version of my series on monday and I'll add this patch there.
linux-kselftest-mirror@lists.linaro.org
-
Ilpo Järvinen -
Maciej Wieczór-Retman -
Reinette Chatre