Hi, Willy
As we discussed in v5, I have proposed a my_syscall() macro, it can convert all of the sys_* functions to macros, and such macros can simply preserve input types from library routines and inherit the 'long' return type from my_syscall<N>. As a result, our __sysret() helper will only require to accept integer types and therefore we can simply revert it to our old sign comparison version (but as macro).
I have already prepared a series of my_syscall() patchset but it includes several not that simple patches, before sending it for review, to directly solve the __sysret() issue at first, it is better to only convert the current three sys_* functions to return 'long' instead of pointer, which will make things easier.
Here is the testing result on all archs (except loongarch) with Arnd's gcc 13.2.0, before testing it, we'd better apply the CROSS_COMPILE patchset [1] manually:
// before $ ARCHS="i386 x86_64 arm64 arm mips ppc ppc64 ppc64le riscv s390" $ for arch in ${ARCHS[@]}; do printf "%9s: " $arch; make run-user XARCH=$arch 2>/dev/null | grep status | tr '\n' ' '; \ size nolibc-test | tail -1 | tr '\t' ' ' | tr -s ' ' | cut -d ' ' -f2; done i386: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 19654 x86_64: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 22337 arm64: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 26292 arm: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 23140 mips: 160 test(s): 157 passed, 3 skipped, 0 failed => status: warning 23164 ppc: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 26812 ppc64: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 27380 ppc64le: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 28004 riscv: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 22062 s390: 160 test(s): 157 passed, 3 skipped, 0 failed => status: warning 22592
// after $ for arch in ${ARCHS[@]}; do printf "%9s: " $arch; make run-user XARCH=$arch 2>/dev/null | grep status | tr '\n' ' '; \ size nolibc-test | tail -1 | tr '\t' ' ' | tr -s ' ' | cut -d ' ' -f2; done i386: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 19502 x86_64: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 22000 arm64: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 25860 arm: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 23108 mips: 160 test(s): 157 passed, 3 skipped, 0 failed => status: warning 22908 ppc: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 26616 ppc64: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 27192 ppc64le: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 27816 riscv: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 21790 s390: 160 test(s): 157 passed, 3 skipped, 0 failed => status: warning 22184
// compare i386: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 19654 -> 19502 x86_64: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 22337 -> 22000 arm64: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 26292 -> 25860 arm: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 23140 -> 23108 mips: 160 test(s): 157 passed, 3 skipped, 0 failed => status: warning 23164 -> 22908 ppc: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 26812 -> 26616 ppc64: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 27380 -> 27192 ppc64le: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 28004 -> 27816 riscv: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 22062 -> 21790 s390: 160 test(s): 157 passed, 3 skipped, 0 failed => status: warning 22592 -> 22184
After these two patches, will send the proposed my_syscall() patchset tomorrow, it can even further reduce more type conversions and therefore reduce more binary bytes, here is a preview of the testing result:
// with the coming my_syscall() patchset, sys_* from functionsn to macros i386: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 19250 x86_64: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 21733 arm64: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 25804 arm: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 22828 mips: 160 test(s): 157 passed, 3 skipped, 0 failed => status: warning 22740 ppc: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 26376 ppc64: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 26752 ppc64le: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 27360 riscv: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 21746 s390: 160 test(s): 157 passed, 3 skipped, 0 failed => status: warning 21928
// compare: __sysret() function -> __sysret() macro -> sys_* macros i386: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 19654 -> 19502 -> 19250 x86_64: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 22337 -> 22000 -> 21733 arm64: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 26292 -> 25860 -> 25804 arm: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 23140 -> 23108 -> 22828 mips: 160 test(s): 157 passed, 3 skipped, 0 failed => status: warning 23164 -> 22908 -> 22740 ppc: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 26812 -> 26616 -> 26376 ppc64: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 27380 -> 27192 -> 26752 ppc64le: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 28004 -> 27816 -> 27360 riscv: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 22062 -> 21790 -> 21746 s390: 160 test(s): 157 passed, 3 skipped, 0 failed => status: warning 22592 -> 22184 -> 21928
It can also shrink the whole sys.h from 1171 lines to around 738 lines.
Changes from v5 --> v6:
* The method introduced in v5 works but it is too complex ;-)
* Convert the return type of sys_brk/mmap/mmap2 from pointer to 'long' (like my_syscall<N> does), after this, all of the sys_* functions return integer.
* Restore __sysret() helper to sign comparison as originally, but also use macro instead of inline function to avoid useless input type and return type conversion.
Changes from v4 --> v5:
* Use __typeof__((arg) + 0) to lose the 'const' flag for old gcc versions.
* Import the famous __is_constexpr() macro from kernel side and add a __is_pointer() macro based on it. (David, to avoid introduce extra discuss on the prove-in-use __is_constexpr macro, this patch uses the original version instead of your suggested version, more info here: https://lore.kernel.org/lkml/20220131204357.1133674-1-keescook@chromium.org/)
* Use __builtin_choose_expr() to merge two comparisons to share the same errno setting code and the -1L assignment code.
Changes from v3 --> v4:
* fix up a new warning about 'ret < 0' when the input arg type is (void *)
Changes from v2 --> v3:
* define a __GXX_HAS_AUTO_TYPE_WITH_CONST_SUPPORT for gcc >= 11.0 (ABI_VERSION >= 1016) * split __sysret() to two versions by the macro instead of a mixed unified and unreadable version * use shorter __ret instead of __sysret_arg
Changes from v1 --> v2:
* fix up argument with 'const' in the type * support "void *" argument
Best regards, Zhangjin Wu ---
v5: https://lore.kernel.org/lkml/b6ff2684f557f6ce00151905990643e651391614.169143... v4: https://lore.kernel.org/lkml/a4084f7fac7a89f861b5582774bc7a98634d1e76.169139... v3: https://lore.kernel.org/lkml/8eaab5da2dcbba42e3f3efc2ae686a22c95f84f0.169138... v2: https://lore.kernel.org/lkml/95fe3e732f455fab653fe1427118d905e4d04257.169133... v1: https://lore.kernel.org/lkml/20230806131921.52453-1-falcon@tinylab.org/
[1]: https://lore.kernel.org/lkml/cover.1691783604.git.falcon@tinylab.org/
Zhangjin Wu (2): tools/nolibc: let sys_brk, sys_mmap and sys_mmap2 return long tools/nolibc: fix up size inflate regression
tools/include/nolibc/arch-s390.h | 4 +-- tools/include/nolibc/sys.h | 43 +++++++++++++------------------- 2 files changed, 20 insertions(+), 27 deletions(-)
Firstly, since the sys_* functions are internally used by our library routines, it is ok to let them preserve the 'long' return type of my_syscall<N> macros, that means not necessary to return pointer like their library routines do.
Secondly, in order to avoid the size inflating issues introduced by the sign extension, it is better to let __sysret() only accept integer input types, to do so, we must let all of the sys_* functions not return pointers.
There are only three sys_* functions which return pointer, let's make them return 'long' instead of pointer.
Link: https://lore.kernel.org/lkml/20230809221743.83107-1-falcon@tinylab.org/ Signed-off-by: Zhangjin Wu falcon@tinylab.org --- tools/include/nolibc/arch-s390.h | 4 ++-- tools/include/nolibc/sys.h | 16 ++++++++-------- 2 files changed, 10 insertions(+), 10 deletions(-)
diff --git a/tools/include/nolibc/arch-s390.h b/tools/include/nolibc/arch-s390.h index 5d60fd43f883..6396c2a6bc3a 100644 --- a/tools/include/nolibc/arch-s390.h +++ b/tools/include/nolibc/arch-s390.h @@ -160,7 +160,7 @@ struct s390_mmap_arg_struct { };
static __attribute__((unused)) -void *sys_mmap(void *addr, size_t length, int prot, int flags, int fd, +long sys_mmap(void *addr, size_t length, int prot, int flags, int fd, off_t offset) { struct s390_mmap_arg_struct args = { @@ -172,7 +172,7 @@ void *sys_mmap(void *addr, size_t length, int prot, int flags, int fd, .offset = (unsigned long)offset };
- return (void *)my_syscall1(__NR_mmap, &args); + return my_syscall1(__NR_mmap, &args); } #define sys_mmap sys_mmap
diff --git a/tools/include/nolibc/sys.h b/tools/include/nolibc/sys.h index 833d6c5e86dc..a28e7fbff448 100644 --- a/tools/include/nolibc/sys.h +++ b/tools/include/nolibc/sys.h @@ -74,9 +74,9 @@ long __sysret(unsigned long ret) */
static __attribute__((unused)) -void *sys_brk(void *addr) +long sys_brk(void *addr) { - return (void *)my_syscall1(__NR_brk, addr); + return my_syscall1(__NR_brk, addr); }
static __attribute__((unused)) @@ -89,12 +89,12 @@ static __attribute__((unused)) void *sbrk(intptr_t inc) { /* first call to find current end */ - void *ret = sys_brk(0); + void *ret = (void *)sys_brk(0);
- if (ret && sys_brk(ret + inc) == ret + inc) + if (ret && (void *)sys_brk(ret + inc) == ret + inc) return ret + inc;
- return (void *)__sysret(-ENOMEM); + return (void *)__sysret((long)-ENOMEM); }
@@ -658,7 +658,7 @@ int mknod(const char *path, mode_t mode, dev_t dev)
#ifndef sys_mmap static __attribute__((unused)) -void *sys_mmap(void *addr, size_t length, int prot, int flags, int fd, +long sys_mmap(void *addr, size_t length, int prot, int flags, int fd, off_t offset) { int n; @@ -670,7 +670,7 @@ void *sys_mmap(void *addr, size_t length, int prot, int flags, int fd, n = __NR_mmap; #endif
- return (void *)my_syscall6(n, addr, length, prot, flags, fd, offset); + return my_syscall6(n, addr, length, prot, flags, fd, offset); } #endif
@@ -682,7 +682,7 @@ void *sys_mmap(void *addr, size_t length, int prot, int flags, int fd, static __attribute__((unused)) void *mmap(void *addr, size_t length, int prot, int flags, int fd, off_t offset) { - return (void *)__sysret((unsigned long)sys_mmap(addr, length, prot, flags, fd, offset)); + return (void *)__sysret(sys_mmap(addr, length, prot, flags, fd, offset)); }
static __attribute__((unused))
As reported and suggested by Willy, the inline __sysret() helper introduces three types of conversions and increases the size:
(1) the "unsigned long" argument to __sysret() forces a sign extension from all sys_* functions that used to return 'int'
(2) the comparison with the error range now has to be performed on a 'unsigned long' instead of an 'int'
(3) the return value from __sysret() is a 'long' (note, a signed long) which then has to be turned back to an 'int' before being returned by the caller to satisfy the caller's prototype.
To fix up this, firstly, let's use macro instead of inline function to preserves the input type and avoids these useless conversions (1), (3).
Secondly, since all of the sys_* functions have been converted to return integer, now, it is able to remove comparison to a 'unsigned long' -MAX_ERRNO (2) and restore the simple sign comparison as before.
Suggested-by: Willy Tarreau w@1wt.eu Link: https://lore.kernel.org/lkml/20230806095846.GB10627@1wt.eu/ Signed-off-by: Zhangjin Wu falcon@tinylab.org --- tools/include/nolibc/sys.h | 27 ++++++++++----------------- 1 file changed, 10 insertions(+), 17 deletions(-)
diff --git a/tools/include/nolibc/sys.h b/tools/include/nolibc/sys.h index a28e7fbff448..e0b68d3532b6 100644 --- a/tools/include/nolibc/sys.h +++ b/tools/include/nolibc/sys.h @@ -27,23 +27,16 @@ #include "errno.h" #include "types.h"
- -/* Syscall return helper for library routines, set errno as -ret when ret is in - * range of [-MAX_ERRNO, -1] - * - * Note, No official reference states the errno range here aligns with musl - * (src/internal/syscall_ret.c) and glibc (sysdeps/unix/sysv/linux/sysdep.h) - */ - -static __inline__ __attribute__((unused, always_inline)) -long __sysret(unsigned long ret) -{ - if (ret >= (unsigned long)-MAX_ERRNO) { - SET_ERRNO(-(long)ret); - return -1; - } - return ret; -} +/* Syscall return helper, set errno as -ret when ret < 0 */ +#define __sysret(arg) \ +({ \ + __typeof__(arg) __ret = (arg); \ + if (__ret < 0) { \ + SET_ERRNO(-__ret); \ + __ret = -1L; \ + } \ + __ret; \ +})
/* Functions in this file only describe syscalls. They're declared static so * that the compiler usually decides to inline them while still being allowed
On Sat, Aug 12, 2023 at 05:51:53AM +0800, Zhangjin Wu wrote:
As reported and suggested by Willy, the inline __sysret() helper introduces three types of conversions and increases the size:
(1) the "unsigned long" argument to __sysret() forces a sign extension from all sys_* functions that used to return 'int'
(2) the comparison with the error range now has to be performed on a 'unsigned long' instead of an 'int'
(3) the return value from __sysret() is a 'long' (note, a signed long) which then has to be turned back to an 'int' before being returned by the caller to satisfy the caller's prototype.
To fix up this, firstly, let's use macro instead of inline function to preserves the input type and avoids these useless conversions (1), (3).
Secondly, since all of the sys_* functions have been converted to return integer, now, it is able to remove comparison to a 'unsigned long' -MAX_ERRNO (2) and restore the simple sign comparison as before.
(...)
+/* Syscall return helper, set errno as -ret when ret < 0 */ +#define __sysret(arg) \ +({ \
- __typeof__(arg) __ret = (arg); \
- if (__ret < 0) { \
SET_ERRNO(-__ret); \__ret = -1L; \- } \
- __ret; \
+})
Except that this now breaks brk(), mmap() and sbrk() by taking any value with MSB set as an error. Also you've re-introduced the problem you've faced with const. See my simplification in the other thread by using "?:" which does avoids any assignment.
Let's just roll brk(), mmap() and sbrk() to their original, working, definition:
static __attribute__((unused)) void *mmap(void *addr, size_t length, int prot, int flags, int fd, off_t offset) { void *ret = sys_mmap(addr, length, prot, flags, fd, offset);
if ((unsigned long)ret >= -MAX_ERRNO) { SET_ERRNO(-(long)ret); ret = MAP_FAILED; } return ret; }
And we're done, you can then keep the simplified __sysret() macro for all other call places.
Willy
Hi, Willy
On Sat, Aug 12, 2023 at 05:51:53AM +0800, Zhangjin Wu wrote:
As reported and suggested by Willy, the inline __sysret() helper introduces three types of conversions and increases the size:
(1) the "unsigned long" argument to __sysret() forces a sign extension from all sys_* functions that used to return 'int'
(2) the comparison with the error range now has to be performed on a 'unsigned long' instead of an 'int'
(3) the return value from __sysret() is a 'long' (note, a signed long) which then has to be turned back to an 'int' before being returned by the caller to satisfy the caller's prototype.
To fix up this, firstly, let's use macro instead of inline function to preserves the input type and avoids these useless conversions (1), (3).
Secondly, since all of the sys_* functions have been converted to return integer, now, it is able to remove comparison to a 'unsigned long' -MAX_ERRNO (2) and restore the simple sign comparison as before.
(...)
+/* Syscall return helper, set errno as -ret when ret < 0 */ +#define __sysret(arg) \ +({ \
- __typeof__(arg) __ret = (arg); \
- if (__ret < 0) { \
SET_ERRNO(-__ret); \__ret = -1L; \- } \
- __ret; \
+})
Except that this now breaks brk(), mmap() and sbrk() by taking any value with MSB set as an error. Also you've re-introduced the problem you've faced with const. See my simplification in the other thread by using "?:" which does avoids any assignment.
Yeah, thanks for your explanation in this reply [1], the 'const' flag only triggers build error on the second 'assign' (__ret == -1L), the first 'assign' is a definition, it is not problematic. so, your "?:" method is a great idea to simply return without the second 'assign'.
Let's just roll brk(), mmap() and sbrk() to their original, working, definition:
static __attribute__((unused)) void *mmap(void *addr, size_t length, int prot, int flags, int fd, off_t offset) { void *ret = sys_mmap(addr, length, prot, flags, fd, offset); if ((unsigned long)ret >= -MAX_ERRNO) { SET_ERRNO(-(long)ret); ret = MAP_FAILED; } return ret; }
Agree, only left a suggestion here [2] about whether we can apply the 2nd patch instead of rolling them back, let's discuss it in [2] thread.
And we're done, you can then keep the simplified __sysret() macro for all other call places.
Now, this issue is near to the end ;-)
Thanks! Zhangjin ---
[1]: https://lore.kernel.org/lkml/20230813085140.GD8237@1wt.eu/#R [2]: https://lore.kernel.org/lkml/20230813132620.19411-1-falcon@tinylab.org/
Willy
On Sun, Aug 13, 2023 at 09:39:44PM +0800, Zhangjin Wu wrote:
Let's just roll brk(), mmap() and sbrk() to their original, working, definition:
static __attribute__((unused)) void *mmap(void *addr, size_t length, int prot, int flags, int fd, off_t offset) { void *ret = sys_mmap(addr, length, prot, flags, fd, offset); if ((unsigned long)ret >= -MAX_ERRNO) { SET_ERRNO(-(long)ret); ret = MAP_FAILED; } return ret; }
Agree, only left a suggestion here [2] about whether we can apply the 2nd patch instead of rolling them back, let's discuss it in [2] thread.
(...)
I'm sorry but I can't find this "suggestion" in this yet-another-super- long-description-of-another-idea-of-redesign. In addition it's extremely painful to constantly have to go through web links to follow a single conversation. Mail works in threads for a reason. When the same discussion is handled in many parallel threads it becomes impossible to keep it focused on a specific topic. This is also why you should stop systematically responding to a message with yet another redesign suggestion, this is super hard to follow and it literally takes me several hours a week! And at the end we've not addressed the initial problem but discussed plenty of other things.
Thanks, Willy
On Sun, Aug 13, 2023 at 09:39:44PM +0800, Zhangjin Wu wrote:
And we're done, you can then keep the simplified __sysret() macro for all other call places.
Now, this issue is near to the end ;-)
I've now pushed the simplified fix (without changing the SET_ERRNO() macro, enough last minute breaking changes for now) in branch 20230815-for-6.6-2.
The tests pass and riscv/loongarch are even very slightly smaller than before (~8 bytes) but again that doesn't count as it depends on how the compiler decides to arrange if/else branches.
I'll let Shuah know about these late fixes.
Regards, Willy
On Sun, Aug 13, 2023 at 09:39:44PM +0800, Zhangjin Wu wrote:
And we're done, you can then keep the simplified __sysret() macro for all other call places.
Now, this issue is near to the end ;-)
I've now pushed the simplified fix (without changing the SET_ERRNO() macro, enough last minute breaking changes for now) in branch 20230815-for-6.6-2.
The tests pass and riscv/loongarch are even very slightly smaller than before (~8 bytes) but again that doesn't count as it depends on how the compiler decides to arrange if/else branches.
Tested 20230815-for-6.6-2 with latest Arnd's gcc 13.2.0 (left: old, right: new), no warning, no failure:
// run-user $ for arch in ${ARCHS[@]}; do printf "%9s: " $arch; make run-user XARCH=$arch | grep status | tr '\n' ' '; \ size nolibc-test | tail -1 | tr '\t' ' ' | tr -s ' ' | cut -d ' ' -f2; done i386: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 19654 > 19508 x86_64: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 22337 > 22011 arm64: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 26292 > 25868 arm: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 23140 > 23112 mips: 160 test(s): 157 passed, 3 skipped, 0 failed => status: warning 23164 > 22924 // mips-linux- has smaller size, here uses mips64 ppc: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 26812 > 26628 ppc64: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 27380 > 27204 ppc64le: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 28004 > 27828 riscv: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 22062 > 21794 s390: 160 test(s): 157 passed, 3 skipped, 0 failed => status: warning 22592 > 22192
// kernel build + run arch/board | result ------------|------------ arm/vexpress-a9 | 160 test(s): 159 passed, 1 skipped, 0 failed => status: warning. arm/virt | 160 test(s): 156 passed, 4 skipped, 0 failed => status: warning. aarch64/virt | 160 test(s): 159 passed, 1 skipped, 0 failed => status: warning. ppc/g3beige | 160 test(s): 159 passed, 1 skipped, 0 failed => status: warning. ppc/ppce500 | 160 test(s): 159 passed, 1 skipped, 0 failed => status: warning. ppc64le/pseries | 160 test(s): 159 passed, 1 skipped, 0 failed => status: warning. ppc64le/powernv | 160 test(s): 159 passed, 1 skipped, 0 failed => status: warning. ppc64/pseries | 160 test(s): 159 passed, 1 skipped, 0 failed => status: warning. ppc64/powernv | 160 test(s): 159 passed, 1 skipped, 0 failed => status: warning. i386/pc | 160 test(s): 159 passed, 1 skipped, 0 failed => status: warning. x86_64/pc | 160 test(s): 159 passed, 1 skipped, 0 failed => status: warning. mipsel/malta | 160 test(s): 159 passed, 1 skipped, 0 failed => status: warning. loongarch64/virt | 160 test(s): 159 passed, 1 skipped, 0 failed => status: warning. riscv64/virt | 160 test(s): 159 passed, 1 skipped, 0 failed => status: warning. s390x/s390-ccw-virtio | 160 test(s): 159 passed, 1 skipped, 0 failed => status: warning.
Thanks, Zhangjin
I'll let Shuah know about these late fixes.
Regards, Willy
On Sat, Aug 12, 2023 at 05:49:36AM +0800, Zhangjin Wu wrote:
After these two patches, will send the proposed my_syscall() patchset tomorrow, it can even further reduce more type conversions and therefore reduce more binary bytes, here is a preview of the testing result:
// with the coming my_syscall() patchset, sys_* from functionsn to macros i386: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 19250 x86_64: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 21733
(...)
It can also shrink the whole sys.h from 1171 lines to around 738 lines.
Please, Zhangjin, please. Let's stop constantly speaking about potential future improvements when the present is broken. It needlessly adds a lot of noise in the discussion and tends to encourage you to explore areas that are incompatible with what is required to fix the breakage, and very likely steers your approach to fixes in a direction that you think is compatible with such future paths. But as long as existing code is broken you cannot speculate on how better the next iteration will be, because it's built on a broken basis. And I would like to remind that the *only* reason for the current breakage is this attempt to save even more code lines, that was not a requirement at all in the first place! Sure it can be fine to remove code when possible, but not at the cost of trying to force squares to enter round holes like this. The reality is that *some* syscalls are different and *some* archs are different, and these differences have to be taken into account, and if we keep exceptions it's fine.
So let's only speak about this later once the issue is completely solved.
Thanks, Willy
Hi, Willy
On Sat, Aug 12, 2023 at 05:49:36AM +0800, Zhangjin Wu wrote:
After these two patches, will send the proposed my_syscall() patchset tomorrow, it can even further reduce more type conversions and therefore reduce more binary bytes, here is a preview of the testing result:
// with the coming my_syscall() patchset, sys_* from functionsn to macros i386: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 19250 x86_64: 160 test(s): 158 passed, 2 skipped, 0 failed => status: warning 21733
(...)
It can also shrink the whole sys.h from 1171 lines to around 738 lines.
Please, Zhangjin, please. Let's stop constantly speaking about potential future improvements when the present is broken. It needlessly adds a lot of noise in the discussion and tends to encourage you to explore areas that are incompatible with what is required to fix the breakage, and very likely steers your approach to fixes in a direction that you think is compatible with such future paths. But as long as existing code is broken you cannot speculate on how better the next iteration will be, because it's built on a broken basis. And I would like to remind that the *only* reason for the current breakage is this attempt to save even more code lines, that was not a requirement at all in the first place! Sure it can be fine to remove code when possible, but not at the cost of trying to force squares to enter round holes like this. The reality is that *some* syscalls are different and *some* archs are different, and these differences have to be taken into account, and if we keep exceptions it's fine.
Agree very much, that's why I didn't send the new patchset but only send these two ones about size inflate regression, I don't want to discuss more than one issue at a time either (and you also have shared this idea several times) ;-)
The progress and preview data here is only because the patch 1/2 [1] is an important preparation of the new patchset, the data here is more or less providing a selling point why we need patch 1/2, I have explained it in this reply [2]. Of course, we can roll them back directly, and If we do need sys_brk/mmap return 'long', we can revert the rolling-back and apply patch 1/2.
[PATCH v6 1/2] tools/nolibc: let sys_brk, sys_mmap and sys_mmap2 return long
So let's only speak about this later once the issue is completely solved.
Ok, it is the right direction.
Best regards, Zhangjin --- [1]: https://lore.kernel.org/lkml/82b584cbda5cee8d5318986644a2a64ba749a098.169178... [2]: https://lore.kernel.org/lkml/20230813132620.19411-1-falcon@tinylab.org/
Thanks, Willy
linux-kselftest-mirror@lists.linaro.org
-
Willy Tarreau -
Zhangjin Wu