acpi: zero-initialize acpi_object union structure
PATCH 1/1 acpi: zero-initialize acpi_object union structure
The way in which acpi_object union is being initialized varies based on compiler type, version and flags used. Some will zero-initialize the entire union structure and some will only initialize the first N-bytes of the union structure. This could lead to uninitialized union members.
This bug was confirmed by observing non-zero value for object->processor structure variables.
non-zero initialized members of acpi_object union structure causes incorrect error reporting by the driver.
If a BIOS is using "Device" statement as opposed to "Processor" statement, object variable may contain uninitialized members causing the driver to report "Invalid PBLK length" incorrectly.
Using memset to zero-initialize the union structure fixes this issue and also removes the dependency of this function on compiler versions and flags being used.
Tested: Tested on ARM64 hardware that was printing this error and confirmed the prints were gone.
Also confirmed this does not cause regression on ARM64 and X86 machines.
--- drivers/acpi/acpi_processor.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/acpi/acpi_processor.c +3 -1
@@ -275,7 +275,7 @@ static inline int acpi_processor_hotadd_init(struct acpi_processor *pr,static int acpi_processor_get_info (struct acpi_device *device) { -union acpi_object object = {0 }; +union acpi_object object;struct acpi_buffer buffer = {sizeof (union acpi_object ), &object };struct acpi_processor *pr =acpi_driver_data (device);int device_declaration =0 ;
@@ -284,6 +284,8 @@ static int acpi_processor_get_info(struct acpi_device *device)unsigned long long value;int ret; +memset (&object,0 ,sizeof (union acpi_object )); +acpi_processor_errata ();/*
Signature · 1 line
2.47.0.rc1.288.g06298d1525-goog
Rafael J. Wysocki
maintainerQuoted 5 lines · Payam Moradshahi
The way in which
acpi_objectunion is being initialized varies based on compiler type, version and flags used. Some will zero-initialize the entire union structure and some will only initialize the first N-bytes of the union structure.
Any details?
Quoted 1 line
This could lead to uninitialized union members.
So this is working around a compiler bug AFAICS.
If the compiler has this bug, is it guaranteed to compile the rest of the kernel correctly?
Quoted 2 lines
This bug was confirmed by observing non-zero value for
object->processorstructure variables.
Where has it been observed? What compiler version(s)? etc.
Quoted 45 lines
non-zero initialized members of
acpi_objectunion structure causes incorrect error reporting by the driver.If a BIOS is using "Device" statement as opposed to "Processor" statement, object variable may contain uninitialized members causing the driver to report "Invalid PBLK length" incorrectly.
Using memset to zero-initialize the union structure fixes this issue and also removes the dependency of this function on compiler versions and flags being used.
Tested: Tested on ARM64 hardware that was printing this error and confirmed the prints were gone.
Also confirmed this does not cause regression on ARM64 and X86 machines.
Signed-off-by Payam Moradshahi <[email protected]>--- drivers/acpi/acpi_processor.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)drivers/acpi/acpi_processor.c +3
-2@@ -275,7 +275,7 @@ static inline int acpi_processor_hotadd_init(struct acpi_processor *pr,static int acpi_processor_get_info (struct acpi_device *device) { -union acpi_object object = {0 }; +union acpi_object object;struct acpi_buffer buffer = {sizeof (union acpi_object ), &object };struct acpi_processor *pr =acpi_driver_data (device);int device_declaration =0 ;@@ -284,6 +284,8 @@ static int acpi_processor_get_info(struct acpi_device *device)unsigned long long value;int ret; +memset (&object,0 ,sizeof (union acpi_object )); +acpi_processor_errata ();/* --
Hi Rafael,
Thank you for your response. Please see below for my response.
Payam
Quoted 8 lines · Rafael J. Wysocki
Quoted 5 lines · Payam Moradshahi
The way in which
acpi_objectunion is being initialized varies based on compiler type, version and flags used. Some will zero-initialize the entire union structure and some will only initialize the first N-bytes of the union structure.Any details?
I dumped acpi_object after initialization and noticed either the entire structure was zero-initialized or just the first 8 bytes depending on compiler type and version.
gcc 13.2.0: bad
CLANG_CL=362121269: good
CLANG_CL=609443126: bad
CLANG_CL=684773960: good
Quoted 6 lines
Quoted 1 line
This could lead to uninitialized union members.
So this is working around a compiler bug AFAICS.
If the compiler has this bug, is it guaranteed to compile the rest of the kernel correctly?
This is not a compiler bug. The way acpi_object union is being initialized is not c-spec compliant.
Based on C Standard ISO/IEC 9899:202x (E):
Section 6.2.6.1 (7) says:
When a value is stored in a member of an object of union type, the bytes of the object representation that do not correspond to that member but do correspond to other members take unspecified values
Section 6.7.9 says:
If an object that has automatic storage duration is not initialized explicitly, its value is indeterminate.
If an object that has static or thread storage duration is not initialized explicitly, then:
- if it is a union, the first named member is initialized (recursively) according to these rules, and any padding is initialized to zero bits;
The above guarantees zero only in the case of static or thread storage, which is not the case here.
Quoted 4 lines
Quoted 2 lines
This bug was confirmed by observing non-zero value for
object->processorstructure variables.Where has it been observed? What compiler version(s)? etc.
See above for some examples
Quoted 49 lines
Quoted 17 lines
non-zero initialized members of
acpi_objectunion structure causes incorrect error reporting by the driver.If a BIOS is using "Device" statement as opposed to "Processor" statement, object variable may contain uninitialized members causing the driver to report "Invalid PBLK length" incorrectly.
Using memset to zero-initialize the union structure fixes this issue and also removes the dependency of this function on compiler versions and flags being used.
Tested: Tested on ARM64 hardware that was printing this error and confirmed the prints were gone.
Also confirmed this does not cause regression on ARM64 and X86 machines.
Review
Quoted 6 lines
Signed-off-by Payam Moradshahi <[email protected]>--- drivers/acpi/acpi_processor.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)(quoted diff) +0
-0
b/drivers/acpi/acpi_processor.cdrivers/acpi/acpi_processor.c static inline int acpi_processor_hotadd_init(struct
Quoted 4 lines
index
7cf6101cb4c73..6696ad4937d21100644drivers/acpi/acpi_processor.c +0
-0@@ -275,7 +275,7 @@ static inline int acpi_processor_hotadd_init(struct
acpi_processor*pr,drivers/acpi/acpi_processor.c static inline int acpi_processor_hotadd_init(struct acpi_processor *pr,
Quoted 6 lines
drivers/acpi/acpi_processor.c +1
-1static int acpi_processor_get_info (struct acpi_device *device) { -union acpi_object object = {0 }; +union acpi_object object;struct acpi_buffer buffer = {sizeof (union acpi_object ),&object };
drivers/acpi/acpi_processor.c static int acpi_processor_get_info(struct
Quoted 3 lines
drivers/acpi/acpi_processor.c +0
-0struct acpi_processor *pr =acpi_driver_data (device);int device_declaration =0 ;@@ -284,6 +284,8 @@ static int acpi_processor_get_info(struct
acpi_device*device)Quoted 9 lines
drivers/acpi/acpi_processor.c +2
-1unsigned long long value;int ret; +memset (&object,0 ,sizeof (union acpi_object )); +acpi_processor_errata ();/* --
Hi Rafael,
Quoted 17 lines · Payam Moradshahi
Hi Rafael,
Thank you for your response. Please see below for my response.
Payam
Quoted 4 lines · Rafael J. Wysocki
Quoted 2 lines · Payam Moradshahi
The way in which
acpi_objectunion is being initialized varies basedon
Quoted 5 lines
Quoted 3 lines
compiler type, version and flags used. Some will zero-initialize the entire union structure and some will only initialize the first N-bytes of the union structure.
Any details?
Quoted 8 lines
I dumped
acpi_objectafter initialization and noticed either the entire structure was zero-initialized or just the first 8 bytes depending on compiler type and version.gcc 13.2.0: bad
CLANG_CL=362121269: good
CLANG_CL=609443126: bad
CLANG_CL=684773960: good
For reference the latter ones are Google internal builds of clang, but we've played around in godbolt with this in a variety of versions and some work and some don't. Payam has the relevant section from the C standard below.
Quoted 8 lines
Quoted 6 lines
Quoted 1 line
This could lead to uninitialized union members.
So this is working around a compiler bug AFAICS.
If the compiler has this bug, is it guaranteed to compile the rest of the kernel correctly?
This is not a compiler bug. The way
acpi_objectunion is being initialized is not c-spec compliant.
I think in past versions we might've gotten lucky with this, recent compilers might've tightened this up some or changed behavior.
Quoted 25 lines
Based on C Standard ISO/IEC 9899:202x (E):
Section 6.2.6.1 (7) says:
When a value is stored in a member of an object of union type, the bytes of the
object representation that do not correspond to that member but do correspond
to other members take unspecified valuesSection 6.7.9 says:
If an object that has automatic storage duration is not initialized explicitly,
its value is indeterminate.If an object that has static or thread storage duration is not initialized explicitly, then:
- if it is a union, the first named member is initialized (recursively)according to these rules, and any padding is initialized to zero bits;The above guarantees zero only in the case of static or thread storage, which is not the case here.
Quoted 6 lines
Quoted 1 line
Quoted 1 line
This bug was confirmed by observing non-zero value for
object->processorQuoted 3 lines
Quoted 1 line
structure variables.
Where has it been observed? What compiler version(s)? etc.
See above for some examples
This manifests like this on Google Axion arm64 builds using Arm's prebuilt GCC 13.2.0:
[ 1.844508] acpi ACPI0007:08: Invalid PBLK length [271170112] [ 1.850253] acpi ACPI0007:09: Invalid PBLK length [271170112] [ 1.855992] acpi ACPI0007:0a: Invalid PBLK length [271170112] [ 1.861730] acpi ACPI0007:0b: Invalid PBLK length [271170112] [ 1.867470] acpi ACPI0007:0c: Invalid PBLK length [271170112] [ 1.873208] acpi ACPI0007:0d: Invalid PBLK length [271170112] [ 1.878947] acpi ACPI0007:0e: Invalid PBLK length [271170112] [ 1.884686] acpi ACPI0007:0f: Invalid PBLK length [271170112] [ 1.890424] acpi ACPI0007:10: Invalid PBLK length [271170112] [ 1.896161] acpi ACPI0007:11: Invalid PBLK length [271170112] [ 1.901898] acpi ACPI0007:12: Invalid PBLK length [271170112] [ 1.907636] acpi ACPI0007:13: Invalid PBLK length [271170112] [ 1.913374] acpi ACPI0007:14: Invalid PBLK length [271170112] [ 1.919113] acpi ACPI0007:15: Invalid PBLK length [271170112] [ 1.924851] acpi ACPI0007:16: Invalid PBLK length [271170112]
Quoted 51 lines
Quoted 5 lines
Quoted 5 lines
non-zero initialized members of
acpi_objectunion structure causes incorrect error reporting by the driver.If a BIOS is using "Device" statement as opposed to "Processor" statement, object variable may contain uninitialized members causing
the
Quoted 3 lines
Quoted 3 lines
driver to report "Invalid PBLK length" incorrectly.
Using memset to zero-initialize the union structure fixes this issue
and
Quoted 41 lines
Quoted 9 lines
also removes the dependency of this function on compiler versions and flags being used.
Tested: Tested on ARM64 hardware that was printing this error and confirmed the prints were gone.
Also confirmed this does not cause regression on ARM64 and X86 machines.
Review
Quoted 6 lines
Signed-off-by Payam Moradshahi <[email protected]>--- drivers/acpi/acpi_processor.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)(quoted diff) +0
-0
b/drivers/acpi/acpi_processor.cdrivers/acpi/acpi_processor.c static inline int
Quoted 4 lines
index
7cf6101cb4c73..6696ad4937d21100644drivers/acpi/acpi_processor.c +0
-0@@ -275,7 +275,7 @@ static inline intacpi_processor_hotadd_init(struct
acpi_processor*pr,drivers/acpi/acpi_processor.c static inline int acpi_processor_hotadd_init(struct acpi_processor *pr,
Quoted 6 lines
drivers/acpi/acpi_processor.c +1
-1static int acpi_processor_get_info (struct acpi_device *device) { -union acpi_object object = {0 }; +union acpi_object object;struct acpi_buffer buffer = {sizeof (union acpi_object ),&object };
drivers/acpi/acpi_processor.c static int acpi_processor_get_info(struct
Quoted 3 lines
drivers/acpi/acpi_processor.c +0
-0struct acpi_processor *pr =acpi_driver_data (device);int device_declaration =0 ;@@ -284,6 +284,8 @@ static int acpi_processor_get_info(struct
acpi_device*device)Quoted 9 lines
drivers/acpi/acpi_processor.c +2
-1unsigned long long value;int ret; +memset (&object,0 ,sizeof (union acpi_object )); +acpi_processor_errata ();/* --
Payam, it might be good to add a log snippet and references to the relevant spec sections to your v2 commit message.
Moreover, for v2 of the patch you probably want to add a
Cc: [email protected]
to your patch to make sure it gets picked up once it gets picked up for mainline. See [1] for more info on the stable process.
Cheers,
Moritz
[1] https://www.kernel.org/doc/Documentation/process/stable-kernel-rules.rst