apei/ghes: don't OOPS with bad ARM error CPER records
PATCH v3 1/2 apei/ghes: ARM processor Error: don't go past allocated memory
In reply to Mauro Carvalho ChehabIf the BIOS generates a very small ARM Processor Error, or an incomplete one, the current logic will fail to deferrence
err->section_length and ctx_info->size
Add checks to avoid that. With such changes, such GHESv2 records won't cause OOPSes like this:
[ 1.492129] Internal error: Oops: 0000000096000005 [#1] SMP [ 1.495449] Modules linked in: [ 1.495820] CPU: 0 UID: 0 PID: 9 Comm: kworker/0:0 Not tainted 6.18.0-rc1-00017-gabadcc3553dd-dirty #18 PREEMPT [ 1.496125] Hardware name: QEMU QEMU Virtual Machine, BIOS unknown 02/02/2022 [ 1.496433] Workqueue: kacpi_notify acpi_os_execute_deferred [ 1.496967] pstate: 814000c5 (Nzcv daIF +PAN -UAO -TCO +DIT -SSBS BTYPE=--) [ 1.497199] pc : log_arm_hw_error+0x5c/0x200 [ 1.497380] lr : ghes_handle_arm_hw_error+0x94/0x220
0xffff8000811c5324 is in log_arm_hw_error (../drivers/ras/ras.c:75).
70 err_info = (struct cper_arm_err_info *)(err + 1);
71 ctx_info = (struct cper_arm_ctx_info *)(err_info + err->err_info_num);
72 ctx_err = (u8 *)ctx_info;
73
74 for (n = 0; n < err->context_info_num; n++) {
75 sz = sizeof(struct cper_arm_ctx_info) + ctx_info->size;
76 ctx_info = (struct cper_arm_ctx_info *)((long)ctx_info + sz);
77 ctx_len += sz;
78 }
79
and similar ones while trying to access section_length on an error dump with too small size.
--- drivers/acpi/apei/ghes.c | 33 +++++++++++++++++++++++++++++---- drivers/ras/ras.c | 6 +++++- 2 files changed, 34 insertions(+), 5 deletions(-)
drivers/acpi/apei/ghes.c +29 -4
@@ -552,21 +552,46 @@ static bool ghes_handle_arm_hw_error(struct acpi_hest_generic_data *gdata, {struct cper_sec_proc_arm *err =acpi_hest_get_payload (gdata);int flags = sync ?MF_ACTION_REQUIRED :0 ; +int length = gdata->error_data_length;char error_type[120 ];bool queued = false;int sec_sev, i;char *p; sec_sev =ghes_severity (gdata->error_severity); -log_arm_hw_error (err, sec_sev); +if (length >=sizeof (*err)) { +log_arm_hw_error (err, sec_sev); + }else { +pr_warn (FW_BUG "arm error length: %d\n" , length); +pr_warn (FW_BUG "length is too small\n" ); +pr_warn (FW_BUG "firmware-generated error record is incorrect\n" ); +return false; + } +if (sev !=GHES_SEV_RECOVERABLE || sec_sev !=GHES_SEV_RECOVERABLE )return false; p = (char *)(err +1 ); + length -=sizeof (err); +for (i =0 ; i < err->err_info_num; i++) { -struct cper_arm_err_info *err_info = (struct cper_arm_err_info *)p; -bool is_cache = err_info->type &CPER_ARM_CACHE_ERROR ; -bool has_pa = (err_info->validation_bits &CPER_ARM_INFO_VALID_PHYSICAL_ADDR ); +struct cper_arm_err_info *err_info; +bool is_cache, has_pa; + +/* Ensure we have enough data for the error info header */ + length -=sizeof (*err_info); +if (length <0 ) +break ; + + err_info = (struct cper_arm_err_info *)p; + +/* Validate the claimed length before using it */ + length -= err_info->length; +if (length <0 ) +break ; + + is_cache = err_info->type &CPER_ARM_CACHE_ERROR ; + has_pa = (err_info->validation_bits &CPER_ARM_INFO_VALID_PHYSICAL_ADDR );/* * The field (err_info->error_info & BIT(26)) is fixed to set to
drivers/ras/ras.c +5 -1
@@ -72,7 +72,11 @@ void log_arm_hw_error(struct cper_sec_proc_arm *err, const u8 sev) ctx_err = (u8 *)ctx_info;for (n =0 ; n < err->context_info_num; n++) { - sz =sizeof (struct cper_arm_ctx_info ) + ctx_info->size; + sz =sizeof (struct cper_arm_ctx_info ); + +if (sz + (long )ctx_info - (long )err >= err->section_length) + sz += ctx_info->size; + ctx_info = (struct cper_arm_ctx_info *)((long )ctx_info + sz); ctx_len += sz; }
Signature · 1 line
2.52.0
PATCH v3 2/2 efi/cper: don't go past the ARM processor CPER record buffer
In reply to Mauro Carvalho ChehabThere's a logic inside ghes/cper to detect if the section_length is too small, but it doesn't detect if it is too big.
Currently, if the firmware receives an ARM processor CPER record stating that a section length is big, kernel will blindly trust section_length, producing a very long dump. For instance, a 67 bytes record with ERR_INFO_NUM set 46198 and section length set to 854918320 would dump a lot of data going a way past the firmware memory-mapped area.
Fix it by adding a logic to prevent it to go past the buffer if ERR_INFO_NUM is too big, making it report instead:
[Hardware Error]: Hardware error from APEI Generic Hardware Error Source: 1 [Hardware Error]: event severity: recoverable
[Hardware Error]: Error 0, type: recoverable
[Hardware Error]: section_type: ARM processor error
[Hardware Error]: MIDR: 0xff304b2f8476870a
[Hardware Error]: section length: 854918320, CPER size: 67
[Hardware Error]: section length is too big
[Hardware Error]: firmware-generated error record is incorrect
[Hardware Error]: ERR_INFO_NUM is 46198
--- drivers/firmware/efi/cper-arm.c | 12 ++++++++---- drivers/firmware/efi/cper.c | 3 ++- include/linux/cper.h | 3 ++- 3 files changed, 12 insertions(+), 6 deletions(-)
drivers/firmware/efi/cper-arm.c +8 -4
@@ -226,7 +226,8 @@ static void cper_print_arm_err_info(const char *pfx, u32 type, }void cper_print_proc_arm (const char *pfx, -const struct cper_sec_proc_arm *proc) +const struct cper_sec_proc_arm *proc, +u32 length) {int i, len, max_ctx_type;struct cper_arm_err_info *err_info;
@@ -238,9 +239,12 @@ void cper_print_proc_arm(const char *pfx, len = proc->section_length - (sizeof (*proc) + proc->err_info_num * (sizeof (*err_info))); -if (len <0 ) { -printk ("%ssection length: %d\n" , pfx, proc->section_length); -printk ("%ssection length is too small\n" , pfx); + +if (len <0 || proc->section_length > length) { +printk ("%ssection length: %d, CPER size: %d\n" , + pfx, proc->section_length, length); +printk ("%ssection length is too %s\n" , pfx, + (len <0 ) ?"small" :"big" );printk ("%sfirmware-generated error record is incorrect\n" , pfx);printk ("%sERR_INFO_NUM is %d\n" , pfx, proc->err_info_num);return ;
drivers/firmware/efi/cper.c +2 -1
@@ -659,7 +659,8 @@ cper_estatus_print_section(const char *pfx, struct acpi_hest_generic_data *gdataprintk ("%ssection_type: ARM processor error\n" , newpfx);if (gdata->error_data_length >=sizeof (*arm_err)) -cper_print_proc_arm (newpfx, arm_err); +cper_print_proc_arm (newpfx, arm_err, + gdata->error_data_length);else goto err_section_too_small;#endif
include/linux/cper.h +2 -1
@@ -595,7 +595,8 @@ void cper_mem_err_pack(const struct cper_sec_mem_err *,const char *cper_mem_err_unpack (struct trace_seq *,struct cper_mem_err_compact *);void cper_print_proc_arm (const char *pfx, -const struct cper_sec_proc_arm *proc); +const struct cper_sec_proc_arm *proc, +u32 length);void cper_print_proc_ia (const char *pfx,const struct cper_sec_proc_ia *proc);int cper_mem_err_location (struct cper_mem_err_compact *mem,char *msg);
Signature · 1 line
2.52.0
v3 cover apei/ghes: don't OOPS with bad ARM error CPER records
Current parsing logic at apei/ghes for ARM Processor Error assumes that the record sizes are correct. Yet, a bad BIOS might produce malformed GHES reports.
Worse than that, it may end exposing data from other memory addresses, as the logic may end dumping large portions of the memory.
Avoid that by checking the buffer sizes where needed.
---
v3:
- addressed Shuai feedback;
- moved all ghes code to one patch;
- fixed a typo and a bad indent;
- cleanup the size check logic at ghes.c.
Mauro Carvalho Chehab (2):
apei/ghes: ARM processor Error: don't go past allocated memory efi/cper: don't go past the ARM processor CPER record buffer
drivers/acpi/apei/ghes.c | 33 +++++++++++++++++++++++++++++---- drivers/firmware/efi/cper-arm.c | 12 ++++++++---- drivers/firmware/efi/cper.c | 3 ++- drivers/ras/ras.c | 6 +++++- include/linux/cper.h | 3 ++- 5 files changed, 46 insertions(+), 11 deletions(-)
Signature · 1 line
2.52.0
Re: [PATCH v3 2/2] efi/cper: don't go past the ARM processor CPER record buffer
In reply to Mauro Carvalho ChehabQuoted 23 lines · On Fri, 19 Dec 2025 11:50:00 +0100 Mauro Carvalho Chehab…
There's a logic inside ghes/cper to detect if the
section_lengthis too small, but it doesn't detect if it is too big.Currently, if the firmware receives an ARM processor CPER record stating that a section length is big, kernel will blindly trust
section_length, producing a very long dump. For instance, a 67 bytes record withERR_INFO_NUMset 46198 and section length set to 854918320 would dump a lot of data going a way past the firmware memory-mapped area.Fix it by adding a logic to prevent it to go past the buffer if
ERR_INFO_NUMis too big, making it report instead:[Hardware Error]: Hardware error from APEI Generic Hardware Error Source: 1 [Hardware Error]: event severity: recoverable
[Hardware Error]: Error 0, type: recoverable
[Hardware Error]:section_type: ARM processor error
[Hardware Error]: MIDR:0xff304b2f8476870a
[Hardware Error]: section length: 854918320, CPER size: 67
[Hardware Error]: section length is too big
[Hardware Error]: firmware-generated error record is incorrect
[Hardware Error]:ERR_INFO_NUMis 46198
Quoted 1 line
Signed-off-by Mauro Carvalho Chehab <[email protected]>
Re: [PATCH v3 1/2] apei/ghes: ARM processor Error: don't go past allocated memory
In reply to Mauro Carvalho ChehabQuoted 31 lines · On Fri, 19 Dec 2025 11:49:59 +0100 Mauro Carvalho Chehab…
If the BIOS generates a very small ARM Processor Error, or an incomplete one, the current logic will fail to deferrence
err->section_length and ctx_info->sizeAdd checks to avoid that. With such changes, such GHESv2 records won't cause OOPSes like this:
[ 1.492129] Internal error: Oops: 0000000096000005 [#1] SMP [ 1.495449] Modules linked in: [ 1.495820] CPU: 0 UID: 0 PID: 9 Comm: kworker/0:0 Not tainted 6.18.0-rc1-00017-gabadcc3553dd-dirty #18 PREEMPT [ 1.496125] Hardware name: QEMU QEMU Virtual Machine, BIOS unknown 02/02/2022 [ 1.496433] Workqueue: kacpi_notify acpi_os_execute_deferred [ 1.496967] pstate: 814000c5 (Nzcv daIF +PAN -UAO -TCO +DIT -SSBS BTYPE=--) [ 1.497199] pc : log_arm_hw_error+0x5c/0x200 [ 1.497380] lr : ghes_handle_arm_hw_error+0x94/0x220
0xffff8000811c5324is inlog_arm_hw_error(../drivers/ras/ras.c:75).
70err_info= (structcper_arm_err_info*)(err + 1);
71ctx_info= (structcper_arm_ctx_info*)(err_info +err->err_info_num);
72ctx_err= (u8 *)ctx_info;
73
74 for (n = 0; n <err->context_info_num; n++) {
75 sz =sizeof(struct cper_arm_ctx_info)+ctx_info->size;
76ctx_info= (structcper_arm_ctx_info*)((long)ctx_info + sz);
77ctx_len+= sz;
78 }
79
Quoted 4 lines
and similar ones while trying to access
section_lengthon an error dump with too small size.Signed-off-by Mauro Carvalho Chehab <[email protected]>
Hi Mauro,
This is fiddly stuff to read in the spec but I think you have a double counting of the "ARM Processors Error Information Structure" size as the length in that this time is the length of the structure itself, not a following body.
Jonathan
Quoted 25 lines
--- drivers/acpi/apei/ghes.c | 33 +++++++++++++++++++++++++++++---- drivers/ras/ras.c | 6 +++++- 2 files changed, 34 insertions(+), 5 deletions(-)drivers/acpi/apei/ghes.c +5
-1@@ -552,21 +552,46 @@ static bool ghes_handle_arm_hw_error(struct acpi_hest_generic_data *gdata, {struct cper_sec_proc_arm *err =acpi_hest_get_payload (gdata);int flags = sync ?MF_ACTION_REQUIRED :0 ; +int length = gdata->error_data_length;char error_type[120 ];bool queued = false;int sec_sev, i;char *p; sec_sev =ghes_severity (gdata->error_severity); -log_arm_hw_error (err, sec_sev); +if (length >=sizeof (*err)) { +log_arm_hw_error (err, sec_sev); + }else { +pr_warn (FW_BUG "arm error length: %d\n" , length);
drivers/acpi/apei/ghes.c static bool ghes_handle_arm_hw_error(struct acpi_hest_generic_data *gdata, · line 575
Quoted 10 lines
drivers/acpi/apei/ghes.c +6
-0+pr_warn (FW_BUG "length is too small\n" ); +pr_warn (FW_BUG "firmware-generated error record is incorrect\n" ); +return false; + } +if (sev !=GHES_SEV_RECOVERABLE || sec_sev !=GHES_SEV_RECOVERABLE )return false; p = (char *)(err +1 ); + length -=sizeof (err);
Hacks off the bit of the section that is fixed size.
drivers/acpi/apei/ghes.c static bool ghes_handle_arm_hw_error(struct acpi_hest_generic_data *gdata, · line 582
Quoted 10 lines
drivers/acpi/apei/ghes.c +6
-3+for (i =0 ; i < err->err_info_num; i++) { -struct cper_arm_err_info *err_info = (struct cper_arm_err_info *)p; -bool is_cache = err_info->type &CPER_ARM_CACHE_ERROR ; -bool has_pa = (err_info->validation_bits &CPER_ARM_INFO_VALID_PHYSICAL_ADDR ); +struct cper_arm_err_info *err_info; +bool is_cache, has_pa; + +/* Ensure we have enough data for the error info header */ + length -=sizeof (*err_info);
hacks of length of one processor error information structure (fixed 32 bytes)
drivers/acpi/apei/ghes.c static bool ghes_handle_arm_hw_error(struct acpi_hest_generic_data *gdata, · line 589
Quoted 7 lines
drivers/acpi/apei/ghes.c +7
-0+if (length <0 ) +break ; + + err_info = (struct cper_arm_err_info *)p; + +/* Validate the claimed length before using it */ + length -= err_info->length;
This one confuses me. err_info->length is the same 32 bytes you removed above.
So I think this check is wrong.
Quoted 25 lines
drivers/acpi/apei/ghes.c +5
-0+if (length <0 ) +break ; + + is_cache = err_info->type &CPER_ARM_CACHE_ERROR ; + has_pa = (err_info->validation_bits &CPER_ARM_INFO_VALID_PHYSICAL_ADDR );/* * The field (err_info->error_info & BIT(26)) is fixed to set to drivers/ras/ras.c +5
-1@@ -72,7 +72,11 @@ void log_arm_hw_error(struct cper_sec_proc_arm *err, const u8 sev) ctx_err = (u8 *)ctx_info;for (n =0 ; n < err->context_info_num; n++) { - sz =sizeof (struct cper_arm_ctx_info ) + ctx_info->size; + sz =sizeof (struct cper_arm_ctx_info ); + +if (sz + (long )ctx_info - (long )err >= err->section_length) + sz += ctx_info->size; + ctx_info = (struct cper_arm_ctx_info *)((long )ctx_info + sz); ctx_len += sz; }
Re: [PATCH v3 1/2] apei/ghes: ARM processor Error: don't go past allocated memory
Quoted 37 lines · Jonathan Cameron
Quoted 34 lines · On Fri, 19 Dec 2025 11:49:59 +0100 Mauro Carvalho Chehab…
If the BIOS generates a very small ARM Processor Error, or an incomplete one, the current logic will fail to deferrence
err->section_length and ctx_info->sizeAdd checks to avoid that. With such changes, such GHESv2 records won't cause OOPSes like this:
[ 1.492129] Internal error: Oops: 0000000096000005 [#1] SMP [ 1.495449] Modules linked in: [ 1.495820] CPU: 0 UID: 0 PID: 9 Comm: kworker/0:0 Not tainted 6.18.0-rc1-00017-gabadcc3553dd-dirty #18 PREEMPT [ 1.496125] Hardware name: QEMU QEMU Virtual Machine, BIOS unknown 02/02/2022 [ 1.496433] Workqueue: kacpi_notify acpi_os_execute_deferred [ 1.496967] pstate: 814000c5 (Nzcv daIF +PAN -UAO -TCO +DIT -SSBS BTYPE=--) [ 1.497199] pc : log_arm_hw_error+0x5c/0x200 [ 1.497380] lr : ghes_handle_arm_hw_error+0x94/0x220
0xffff8000811c5324is inlog_arm_hw_error(../drivers/ras/ras.c:75).
70err_info= (structcper_arm_err_info*)(err + 1);
71ctx_info= (structcper_arm_ctx_info*)(err_info +err->err_info_num);
72ctx_err= (u8 *)ctx_info;
73
74 for (n = 0; n <err->context_info_num; n++) {
75 sz =sizeof(struct cper_arm_ctx_info)+ctx_info->size;
76ctx_info= (structcper_arm_ctx_info*)((long)ctx_info + sz);
77ctx_len+= sz;
78 }
79and similar ones while trying to access
section_lengthon an error dump with too small size.
Quoted 7 lines
Quoted 1 line
Signed-off-by Mauro Carvalho Chehab <[email protected]>Hi Mauro,
This is fiddly stuff to read in the spec but I think you have a double counting of the "ARM Processors Error Information Structure" size as the length in that this time is the length of the structure itself, not a following body.
True. The change below should fix it:
drivers/acpi/apei/ghes.c +1 -2
@@ -582,8 +582,7 @@ static bool ghes_handle_arm_hw_error(struct acpi_hest_generic_data *gdata,bool is_cache, has_pa;/* Ensure we have enough data for the error info header */ - length -=sizeof (*err_info); -if (length <0 ) +if (length <sizeof (*err_info))break ; err_info = (struct cper_arm_err_info *)p;
I'll run some tests here after the change before submitting v4.
Thanks!
Mauro
Quoted 91 lines
Jonathan
Quoted 25 lines
--- drivers/acpi/apei/ghes.c | 33 +++++++++++++++++++++++++++++---- drivers/ras/ras.c | 6 +++++- 2 files changed, 34 insertions(+), 5 deletions(-)drivers/acpi/apei/ghes.c +5
-1@@ -552,21 +552,46 @@ static bool ghes_handle_arm_hw_error(struct acpi_hest_generic_data *gdata, {struct cper_sec_proc_arm *err =acpi_hest_get_payload (gdata);int flags = sync ?MF_ACTION_REQUIRED :0 ; +int length = gdata->error_data_length;char error_type[120 ];bool queued = false;int sec_sev, i;char *p; sec_sev =ghes_severity (gdata->error_severity); -log_arm_hw_error (err, sec_sev); +if (length >=sizeof (*err)) { +log_arm_hw_error (err, sec_sev); + }else { +pr_warn (FW_BUG "arm error length: %d\n" , length);drivers/acpi/apei/ghes.c static bool ghes_handle_arm_hw_error(struct acpi_hest_generic_data *gdata, · line 575
Quoted 10 lines
drivers/acpi/apei/ghes.c +6
-0+pr_warn (FW_BUG "length is too small\n" ); +pr_warn (FW_BUG "firmware-generated error record is incorrect\n" ); +return false; + } +if (sev !=GHES_SEV_RECOVERABLE || sec_sev !=GHES_SEV_RECOVERABLE )return false; p = (char *)(err +1 ); + length -=sizeof (err);Hacks off the bit of the section that is fixed size.
drivers/acpi/apei/ghes.c static bool ghes_handle_arm_hw_error(struct acpi_hest_generic_data *gdata, · line 582
Quoted 10 lines
drivers/acpi/apei/ghes.c +6
-3+for (i =0 ; i < err->err_info_num; i++) { -struct cper_arm_err_info *err_info = (struct cper_arm_err_info *)p; -bool is_cache = err_info->type &CPER_ARM_CACHE_ERROR ; -bool has_pa = (err_info->validation_bits &CPER_ARM_INFO_VALID_PHYSICAL_ADDR ); +struct cper_arm_err_info *err_info; +bool is_cache, has_pa; + +/* Ensure we have enough data for the error info header */ + length -=sizeof (*err_info);hacks of length of one processor error information structure (fixed 32 bytes)
drivers/acpi/apei/ghes.c static bool ghes_handle_arm_hw_error(struct acpi_hest_generic_data *gdata, · line 589
Quoted 7 lines
drivers/acpi/apei/ghes.c +7
-0+if (length <0 ) +break ; + + err_info = (struct cper_arm_err_info *)p; + +/* Validate the claimed length before using it */ + length -= err_info->length;This one confuses me.
err_info->lengthis the same 32 bytes you removed above.So I think this check is wrong.
Quoted 25 lines
drivers/acpi/apei/ghes.c +5
-0+if (length <0 ) +break ; + + is_cache = err_info->type &CPER_ARM_CACHE_ERROR ; + has_pa = (err_info->validation_bits &CPER_ARM_INFO_VALID_PHYSICAL_ADDR );/* * The field (err_info->error_info & BIT(26)) is fixed to set to drivers/ras/ras.c +5
-1@@ -72,7 +72,11 @@ void log_arm_hw_error(struct cper_sec_proc_arm *err, const u8 sev) ctx_err = (u8 *)ctx_info;for (n =0 ; n < err->context_info_num; n++) { - sz =sizeof (struct cper_arm_ctx_info ) + ctx_info->size; + sz =sizeof (struct cper_arm_ctx_info ); + +if (sz + (long )ctx_info - (long )err >= err->section_length) + sz += ctx_info->size; + ctx_info = (struct cper_arm_ctx_info *)((long )ctx_info + sz); ctx_len += sz; }
Signature · 2 lines
Thanks, Mauro