Successfully identified regression in *linux* in CI configuration tcwg_kernel/llvm-master-aarch64-mainline-allyesconfig. So far, this commit has regressed CI configurations: - tcwg_kernel/llvm-master-aarch64-mainline-allyesconfig
Culprit: <cut> commit 342f43af70dbc74f8629381998f92c060e1763a2 Author: Maurizio Lombardi mlombard@redhat.com Date: Thu Jul 29 15:52:50 2021 +0200
iscsi_ibft: fix crash due to KASLR physical memory remapping
Starting with commit a799c2bd29d1 ("x86/setup: Consolidate early memory reservations") memory reservations have been moved earlier during the boot process, before the execution of the Kernel Address Space Layout Randomization code.
setup_arch() calls the iscsi_ibft's find_ibft_region() function to find and reserve the memory dedicated to the iBFT and this function also saves a virtual pointer to the iBFT table for later use.
The problem is that if KALSR is active, the physical memory gets remapped somewhere else in the virtual address space and the pointer is no longer valid, this will cause a kernel panic when the iscsi driver tries to dereference it.
iBFT detected. BUG: unable to handle page fault for address: ffff888000099fd8 #PF: supervisor read access in kernel mode #PF: error_code(0x0000) - not-present page PGD 0 P4D 0 Oops: 0000 [#1] SMP PTI
..snip..
Call Trace: ? ibft_create_kobject+0x1d2/0x1d2 [iscsi_ibft] do_one_initcall+0x44/0x1d0 ? kmem_cache_alloc_trace+0x119/0x220 do_init_module+0x5c/0x270 __do_sys_init_module+0x12e/0x1b0 do_syscall_64+0x40/0x80 entry_SYSCALL_64_after_hwframe+0x44/0xae
Fix this bug by saving the address of the physical location of the ibft; later the driver will use isa_bus_to_virt() to get the correct virtual address.
N.B. On each reboot KASLR randomizes the virtual addresses so assuming phys_to_virt before KASLR does its deed is incorrect.
Simplify the code by renaming find_ibft_region() to reserve_ibft_region() and remove all the wrappers.
Signed-off-by: Maurizio Lombardi mlombard@redhat.com Reviewed-by: Mike Rapoport rppt@linux.ibm.com Signed-off-by: Konrad Rzeszutek Wilk konrad@kernel.org </cut>
Results regressed to (for first_bad == 342f43af70dbc74f8629381998f92c060e1763a2) # reset_artifacts: -10 # build_abe binutils: -9 # build_llvm: -5 # build_abe qemu: -2 # linux_n_obj: 19722 # First few build errors in logs:
from (for last_good == 62fb9874f5da54fdb243003b386128037319b219) # reset_artifacts: -10 # build_abe binutils: -9 # build_llvm: -5 # build_abe qemu: -2 # linux_n_obj: 19795 # linux build successful: all
Artifacts of last_good build: https://ci.linaro.org/job/tcwg_kernel-llvm-bisect-llvm-master-aarch64-mainli... Artifacts of first_bad build: https://ci.linaro.org/job/tcwg_kernel-llvm-bisect-llvm-master-aarch64-mainli... Build top page/logs: https://ci.linaro.org/job/tcwg_kernel-llvm-bisect-llvm-master-aarch64-mainli...
Configuration details:
Reproduce builds: <cut> mkdir investigate-linux-342f43af70dbc74f8629381998f92c060e1763a2 cd investigate-linux-342f43af70dbc74f8629381998f92c060e1763a2
git clone https://git.linaro.org/toolchain/jenkins-scripts
mkdir -p artifacts/manifests curl -o artifacts/manifests/build-baseline.sh https://ci.linaro.org/job/tcwg_kernel-llvm-bisect-llvm-master-aarch64-mainli... --fail curl -o artifacts/manifests/build-parameters.sh https://ci.linaro.org/job/tcwg_kernel-llvm-bisect-llvm-master-aarch64-mainli... --fail curl -o artifacts/test.sh https://ci.linaro.org/job/tcwg_kernel-llvm-bisect-llvm-master-aarch64-mainli... --fail chmod +x artifacts/test.sh
# Reproduce the baseline build (build all pre-requisites) ./jenkins-scripts/tcwg_kernel-build.sh @@ artifacts/manifests/build-baseline.sh
# Save baseline build state (which is then restored in artifacts/test.sh) mkdir -p ./bisect rsync -a --del --delete-excluded --exclude /bisect/ --exclude /artifacts/ --exclude /linux/ ./ ./bisect/baseline/
cd linux
# Reproduce first_bad build git checkout --detach 342f43af70dbc74f8629381998f92c060e1763a2 ../artifacts/test.sh
# Reproduce last_good build git checkout --detach 62fb9874f5da54fdb243003b386128037319b219 ../artifacts/test.sh
cd .. </cut>
History of pending regressions and results: https://git.linaro.org/toolchain/ci/base-artifacts.git/log/?h=linaro-local/c...
Artifacts: https://ci.linaro.org/job/tcwg_kernel-llvm-bisect-llvm-master-aarch64-mainli... Build log: https://ci.linaro.org/job/tcwg_kernel-llvm-bisect-llvm-master-aarch64-mainli...
Full commit (up to 1000 lines): <cut> commit 342f43af70dbc74f8629381998f92c060e1763a2 Author: Maurizio Lombardi mlombard@redhat.com Date: Thu Jul 29 15:52:50 2021 +0200
iscsi_ibft: fix crash due to KASLR physical memory remapping
Starting with commit a799c2bd29d1 ("x86/setup: Consolidate early memory reservations") memory reservations have been moved earlier during the boot process, before the execution of the Kernel Address Space Layout Randomization code.
setup_arch() calls the iscsi_ibft's find_ibft_region() function to find and reserve the memory dedicated to the iBFT and this function also saves a virtual pointer to the iBFT table for later use.
The problem is that if KALSR is active, the physical memory gets remapped somewhere else in the virtual address space and the pointer is no longer valid, this will cause a kernel panic when the iscsi driver tries to dereference it.
iBFT detected. BUG: unable to handle page fault for address: ffff888000099fd8 #PF: supervisor read access in kernel mode #PF: error_code(0x0000) - not-present page PGD 0 P4D 0 Oops: 0000 [#1] SMP PTI
..snip..
Call Trace: ? ibft_create_kobject+0x1d2/0x1d2 [iscsi_ibft] do_one_initcall+0x44/0x1d0 ? kmem_cache_alloc_trace+0x119/0x220 do_init_module+0x5c/0x270 __do_sys_init_module+0x12e/0x1b0 do_syscall_64+0x40/0x80 entry_SYSCALL_64_after_hwframe+0x44/0xae
Fix this bug by saving the address of the physical location of the ibft; later the driver will use isa_bus_to_virt() to get the correct virtual address.
N.B. On each reboot KASLR randomizes the virtual addresses so assuming phys_to_virt before KASLR does its deed is incorrect.
Simplify the code by renaming find_ibft_region() to reserve_ibft_region() and remove all the wrappers.
Signed-off-by: Maurizio Lombardi mlombard@redhat.com Reviewed-by: Mike Rapoport rppt@linux.ibm.com Signed-off-by: Konrad Rzeszutek Wilk konrad@kernel.org --- arch/x86/kernel/setup.c | 10 -------- drivers/firmware/iscsi_ibft.c | 10 +++++--- drivers/firmware/iscsi_ibft_find.c | 48 ++++++++++++++------------------------ include/linux/iscsi_ibft.h | 18 ++++++-------- 4 files changed, 32 insertions(+), 54 deletions(-)
diff --git a/arch/x86/kernel/setup.c b/arch/x86/kernel/setup.c index 1e720626069a..b6a62af06a9f 100644 --- a/arch/x86/kernel/setup.c +++ b/arch/x86/kernel/setup.c @@ -571,16 +571,6 @@ void __init reserve_standard_io_resources(void)
}
-static __init void reserve_ibft_region(void) -{ - unsigned long addr, size = 0; - - addr = find_ibft_region(&size); - - if (size) - memblock_reserve(addr, size); -} - static bool __init snb_gfx_workaround_needed(void) { #ifdef CONFIG_PCI diff --git a/drivers/firmware/iscsi_ibft.c b/drivers/firmware/iscsi_ibft.c index 7127a04bca19..612a59e213df 100644 --- a/drivers/firmware/iscsi_ibft.c +++ b/drivers/firmware/iscsi_ibft.c @@ -84,8 +84,10 @@ MODULE_DESCRIPTION("sysfs interface to BIOS iBFT information"); MODULE_LICENSE("GPL"); MODULE_VERSION(IBFT_ISCSI_VERSION);
+static struct acpi_table_ibft *ibft_addr; + #ifndef CONFIG_ISCSI_IBFT_FIND -struct acpi_table_ibft *ibft_addr; +phys_addr_t ibft_phys_addr; #endif
struct ibft_hdr { @@ -858,11 +860,13 @@ static int __init ibft_init(void) int rc = 0;
/* - As on UEFI systems the setup_arch()/find_ibft_region() + As on UEFI systems the setup_arch()/reserve_ibft_region() is called before ACPI tables are parsed and it only does legacy finding. */ - if (!ibft_addr) + if (ibft_phys_addr) + ibft_addr = isa_bus_to_virt(ibft_phys_addr); + else acpi_find_ibft_region();
if (ibft_addr) { diff --git a/drivers/firmware/iscsi_ibft_find.c b/drivers/firmware/iscsi_ibft_find.c index 64bb94523281..a0594590847d 100644 --- a/drivers/firmware/iscsi_ibft_find.c +++ b/drivers/firmware/iscsi_ibft_find.c @@ -31,8 +31,8 @@ /* * Physical location of iSCSI Boot Format Table. */ -struct acpi_table_ibft *ibft_addr; -EXPORT_SYMBOL_GPL(ibft_addr); +phys_addr_t ibft_phys_addr; +EXPORT_SYMBOL_GPL(ibft_phys_addr);
static const struct { char *sign; @@ -47,13 +47,24 @@ static const struct { #define VGA_MEM 0xA0000 /* VGA buffer */ #define VGA_SIZE 0x20000 /* 128kB */
-static int __init find_ibft_in_mem(void) +/* + * Routine used to find and reserve the iSCSI Boot Format Table + */ +void __init reserve_ibft_region(void) { unsigned long pos; unsigned int len = 0; void *virt; int i;
+ ibft_phys_addr = 0; + + /* iBFT 1.03 section 1.4.3.1 mandates that UEFI machines will + * only use ACPI for this + */ + if (efi_enabled(EFI_BOOT)) + return; + for (pos = IBFT_START; pos < IBFT_END; pos += 16) { /* The table can't be inside the VGA BIOS reserved space, * so skip that area */ @@ -70,35 +81,12 @@ static int __init find_ibft_in_mem(void) /* if the length of the table extends past 1M, * the table cannot be valid. */ if (pos + len <= (IBFT_END-1)) { - ibft_addr = (struct acpi_table_ibft *)virt; - pr_info("iBFT found at 0x%lx.\n", pos); - goto done; + ibft_phys_addr = pos; + memblock_reserve(ibft_phys_addr, PAGE_ALIGN(len)); + pr_info("iBFT found at 0x%lx.\n", ibft_phys_addr); + return; } } } } -done: - return len; -} -/* - * Routine used to find the iSCSI Boot Format Table. The logical - * kernel address is set in the ibft_addr global variable. - */ -unsigned long __init find_ibft_region(unsigned long *sizep) -{ - ibft_addr = NULL; - - /* iBFT 1.03 section 1.4.3.1 mandates that UEFI machines will - * only use ACPI for this */ - - if (!efi_enabled(EFI_BOOT)) - find_ibft_in_mem(); - - if (ibft_addr) { - *sizep = PAGE_ALIGN(ibft_addr->header.length); - return (u64)virt_to_phys(ibft_addr); - } - - *sizep = 0; - return 0; } diff --git a/include/linux/iscsi_ibft.h b/include/linux/iscsi_ibft.h index b7b45ca82bea..790e7fcfc1a6 100644 --- a/include/linux/iscsi_ibft.h +++ b/include/linux/iscsi_ibft.h @@ -13,26 +13,22 @@ #ifndef ISCSI_IBFT_H #define ISCSI_IBFT_H
-#include <linux/acpi.h> +#include <linux/types.h>
/* - * Logical location of iSCSI Boot Format Table. - * If the value is NULL there is no iBFT on the machine. + * Physical location of iSCSI Boot Format Table. + * If the value is 0 there is no iBFT on the machine. */ -extern struct acpi_table_ibft *ibft_addr; +extern phys_addr_t ibft_phys_addr;
/* * Routine used to find and reserve the iSCSI Boot Format Table. The - * mapped address is set in the ibft_addr variable. + * physical address is set in the ibft_phys_addr variable. */ #ifdef CONFIG_ISCSI_IBFT_FIND -unsigned long find_ibft_region(unsigned long *sizep); +void reserve_ibft_region(void); #else -static inline unsigned long find_ibft_region(unsigned long *sizep) -{ - *sizep = 0; - return 0; -} +static inline void reserve_ibft_region(void) {} #endif
#endif /* ISCSI_IBFT_H */ </cut>
On Fri, Sep 03, 2021 at 11:54:23PM +0000, ci_notify@linaro.org wrote:
Successfully identified regression in *linux* in CI configuration tcwg_kernel/llvm-master-aarch64-mainline-allyesconfig. So far, this commit has regressed CI configurations:
- tcwg_kernel/llvm-master-aarch64-mainline-allyesconfig
Culprit:
<cut> commit 342f43af70dbc74f8629381998f92c060e1763a2 Author: Maurizio Lombardi <mlombard@redhat.com> Date: Thu Jul 29 15:52:50 2021 +0200
iscsi_ibft: fix crash due to KASLR physical memory remapping
Starting with commit a799c2bd29d1 ("x86/setup: Consolidate early memory reservations") memory reservations have been moved earlier during the boot process, before the execution of the Kernel Address Space Layout Randomization code. setup_arch() calls the iscsi_ibft's find_ibft_region() function to find and reserve the memory dedicated to the iBFT and this function also saves a virtual pointer to the iBFT table for later use. The problem is that if KALSR is active, the physical memory gets remapped somewhere else in the virtual address space and the pointer is no longer valid, this will cause a kernel panic when the iscsi driver tries to dereference it. iBFT detected. BUG: unable to handle page fault for address: ffff888000099fd8 #PF: supervisor read access in kernel mode #PF: error_code(0x0000) - not-present page PGD 0 P4D 0 Oops: 0000 [#1] SMP PTI ..snip.. Call Trace: ? ibft_create_kobject+0x1d2/0x1d2 [iscsi_ibft] do_one_initcall+0x44/0x1d0 ? kmem_cache_alloc_trace+0x119/0x220 do_init_module+0x5c/0x270 __do_sys_init_module+0x12e/0x1b0 do_syscall_64+0x40/0x80 entry_SYSCALL_64_after_hwframe+0x44/0xae Fix this bug by saving the address of the physical location of the ibft; later the driver will use isa_bus_to_virt() to get the correct virtual address. N.B. On each reboot KASLR randomizes the virtual addresses so assuming phys_to_virt before KASLR does its deed is incorrect. Simplify the code by renaming find_ibft_region() to reserve_ibft_region() and remove all the wrappers. Signed-off-by: Maurizio Lombardi mlombard@redhat.com Reviewed-by: Mike Rapoport rppt@linux.ibm.com Signed-off-by: Konrad Rzeszutek Wilk konrad@kernel.org
</cut>
Results regressed to (for first_bad == 342f43af70dbc74f8629381998f92c060e1763a2) # reset_artifacts: -10 # build_abe binutils: -9 # build_llvm: -5 # build_abe qemu: -2 # linux_n_obj: 19722 # First few build errors in logs:
from (for last_good == 62fb9874f5da54fdb243003b386128037319b219) # reset_artifacts: -10 # build_abe binutils: -9 # build_llvm: -5 # build_abe qemu: -2 # linux_n_obj: 19795 # linux build successful: all
Artifacts of last_good build: https://ci.linaro.org/job/tcwg_kernel-llvm-bisect-llvm-master-aarch64-mainli... Artifacts of first_bad build: https://ci.linaro.org/job/tcwg_kernel-llvm-bisect-llvm-master-aarch64-mainli... Build top page/logs: https://ci.linaro.org/job/tcwg_kernel-llvm-bisect-llvm-master-aarch64-mainli...
00:04:11 drivers/firmware/iscsi_ibft.c:868:15: error: implicit declaration of function 'isa_bus_to_virt' [-Werror,-Wimplicit-function-declaration] 00:04:11 ibft_addr = isa_bus_to_virt(ibft_phys_addr); 00:04:11 ^ 00:04:11 drivers/firmware/iscsi_ibft.c:868:13: warning: incompatible integer to pointer conversion assigning to 'struct acpi_table_ibft *' from 'int' [-Wint-conversion] 00:04:11 ibft_addr = isa_bus_to_virt(ibft_phys_addr); 00:04:11 ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Should be fixed as of commit 799206c1302e ("iscsi_ibft: Fix isa_bus_to_virt not working under ARM").
https://lore.kernel.org/r/20210901164722.GA2100627@roeck-us.net/ https://lore.kernel.org/r/YTJ5jFf9iBC50pMs@localhost.localdomain/
Is there any way to get these emails shifted from clang-built-linux@googlegroups.com to llvm@lists.linux.dev? They are very helpful but we'd like to have history of them on lore.kernel.org :)
Cheers, Nathan
linaro-toolchain@lists.linaro.org