drm/amdgpu: fix stack alignment ABI mismatch
cover drm/amdgpu: fix stack alignment ABI mismatch
The x86 kernel is compiled with an 8B stack alignment via -mpreferred-stack-boundary=3 for GCC since 3.6-rc1 via
commit d9b0cde91c60 ("x86-64, gcc: Use -mpreferred-stack-boundary=3 if supported") or -mstack-alignment=8 for Clang. Parts of the AMDGPU driver are compiled with 16B stack alignment.
Generally, the stack alignment is part of the ABI. Linking together two different translation units with differing stack alignment is dangerous, particularly when the translation unit with the smaller stack alignment makes calls into the translation unit with the larger stack alignment. While 8B aligned stacks are sometimes also 16B aligned, they are not always.
Multiple users have reported General Protection Faults (GPF) when using the AMDGPU driver compiled with Clang. Clang is placing objects in stack slots assuming the stack is 16B aligned, and selecting instructions that require 16B aligned memory operands.
At runtime, syscall handlers with 8B aligned stack call into code that assumes 16B stack alignment. When the stack is a multiple of 8B but not 16B, these instructions result in a GPF.
Remove the code that added compatibility between the differing compiler flags, as it will result in runtime GPFs when built with Clang.
The series is broken into 3 patches, the first is an important fix for Clang for ChromeOS. The rest are attempted cleanups for GCC, but require additional boot testing. The first patch is critical, the rest are nice to have. I've compile tested the series with ToT Clang, GCC 4.9, and GCC 8.3 **but** I do not have hardware to test on, so I need folks with the above compilers and relevant hardware to help test the series.
The first patch is a functional change for Clang only. It does not change anything for any version of GCC. Yuxuan boot tested a previous incarnation on hardware, but I've changed it enough that I think it made sense to drop the previous tested by tag.
The second patch is a functional change for GCC 7.1+ only. It does not affect older versions of GCC or Clang (though if someone wanted to double check with pre-GCC 7.1 it wouldn't hurt). It should be boot tested on GCC 7.1+ on the relevant hardware.
The final patch is also a functional change for GCC 7.1+ only. It does not affect older versions of GCC or Clang. It should be boot tested on GCC 7.1+ on the relevant hardware. Theoretically, there may be an issue with it, and it's ok to drop it. The series was intentional broken into 3 in order to allow them to be incrementally tested and accepted. It's ok to take earlier patches without the later patches.
And finally, I do not condone linking object files of differing stack alignments. Idealistically, we'd mark the driver broken for pre-GCC 7.1. Pragmatically, "if it ain't broke, don't fix it."
Nick Desaulniers (3):
drm/amdgpu: fix stack alignment ABI mismatch for Clang drm/amdgpu: fix stack alignment ABI mismatch for GCC 7.1+ drm/amdgpu: enable -msse2 for GCC 7.1+ users
drivers/gpu/drm/amd/display/dc/calcs/Makefile | 19 ++++++++++++------- drivers/gpu/drm/amd/display/dc/dcn20/Makefile | 19 ++++++++++++------- drivers/gpu/drm/amd/display/dc/dcn21/Makefile | 19 ++++++++++++------- drivers/gpu/drm/amd/display/dc/dml/Makefile | 19 ++++++++++++------- drivers/gpu/drm/amd/display/dc/dsc/Makefile | 19 ++++++++++++------- 5 files changed, 60 insertions(+), 35 deletions(-)
Signature · 1 line
2.23.0.700.g56cf767bdb-goog
PATCH 1/3 drm/amdgpu: fix stack alignment ABI mismatch for Clang
The x86 kernel is compiled with an 8B stack alignment via -mpreferred-stack-boundary=3 for GCC since 3.6-rc1 via
commit d9b0cde91c60 ("x86-64, gcc: Use -mpreferred-stack-boundary=3 if supported") or -mstack-alignment=8 for Clang. Parts of the AMDGPU driver are compiled with 16B stack alignment.
Generally, the stack alignment is part of the ABI. Linking together two different translation units with differing stack alignment is dangerous, particularly when the translation unit with the smaller stack alignment makes calls into the translation unit with the larger stack alignment. While 8B aligned stacks are sometimes also 16B aligned, they are not always.
Multiple users have reported General Protection Faults (GPF) when using the AMDGPU driver compiled with Clang. Clang is placing objects in stack slots assuming the stack is 16B aligned, and selecting instructions that require 16B aligned memory operands.
At runtime, syscall handlers with 8B aligned stack call into code that assumes 16B stack alignment. When the stack is a multiple of 8B but not 16B, these instructions result in a GPF.
Remove the code that added compatibility between the differing compiler flags, as it will result in runtime GPFs when built with Clang. Cleanups for GCC will be sent in later patches in the series.
--- drivers/gpu/drm/amd/display/dc/calcs/Makefile | 10 ++++------ drivers/gpu/drm/amd/display/dc/dcn20/Makefile | 10 ++++------ drivers/gpu/drm/amd/display/dc/dcn21/Makefile | 10 ++++------ drivers/gpu/drm/amd/display/dc/dml/Makefile | 10 ++++------ drivers/gpu/drm/amd/display/dc/dsc/Makefile | 10 ++++------ 5 files changed, 20 insertions(+), 30 deletions(-)
drivers/gpu/drm/amd/display/dc/calcs/Makefile +4 -6
@@ -24,13 +24,11 @@ # It calculates Bandwidth and Watermarks values for HW programming # -ifneq ($(call cc-option, -mpreferred-stack-boundary=4),) - cc_stack_align := -mpreferred-stack-boundary=4 -else ifneq ($(call cc-option, -mstack-alignment=16),) - cc_stack_align := -mstack-alignment=16 -endif +calcs_ccflags := -mhard-float -msse -calcs_ccflags := -mhard-float -msse $(cc_stack_align) +ifdef CONFIG_CC_IS_GCC +calcs_ccflags += -mpreferred-stack-boundary=4 +endif ifdef CONFIG_CC_IS_CLANG calcs_ccflags += -msse2
drivers/gpu/drm/amd/display/dc/dcn20/Makefile +4 -6
@@ -10,13 +10,11 @@ ifdef CONFIG_DRM_AMD_DC_DSC_SUPPORT DCN20 += dcn20_dsc.o endif -ifneq ($(call cc-option, -mpreferred-stack-boundary=4),) - cc_stack_align := -mpreferred-stack-boundary=4 -else ifneq ($(call cc-option, -mstack-alignment=16),) - cc_stack_align := -mstack-alignment=16 -endif +CFLAGS_$(AMDDALPATH)/dc/dcn20/dcn20_resource.o := -mhard-float -msse -CFLAGS_$(AMDDALPATH)/dc/dcn20/dcn20_resource.o := -mhard-float -msse $(cc_stack_align) +ifdef CONFIG_CC_IS_GCC +CFLAGS_$(AMDDALPATH)/dc/dcn20/dcn20_resource.o += -mpreferred-stack-boundary=4 +endif ifdef CONFIG_CC_IS_CLANG CFLAGS_$(AMDDALPATH)/dc/dcn20/dcn20_resource.o += -msse2
drivers/gpu/drm/amd/display/dc/dcn21/Makefile +4 -6
@@ -3,13 +3,11 @@ DCN21 = dcn21_hubp.o dcn21_hubbub.o dcn21_resource.o -ifneq ($(call cc-option, -mpreferred-stack-boundary=4),) - cc_stack_align := -mpreferred-stack-boundary=4 -else ifneq ($(call cc-option, -mstack-alignment=16),) - cc_stack_align := -mstack-alignment=16 -endif +CFLAGS_$(AMDDALPATH)/dc/dcn21/dcn21_resource.o := -mhard-float -msse -CFLAGS_$(AMDDALPATH)/dc/dcn21/dcn21_resource.o := -mhard-float -msse $(cc_stack_align) +ifdef CONFIG_CC_IS_GCC +CFLAGS_$(AMDDALPATH)/dc/dcn21/dcn21_resource.o += -mpreferred-stack-boundary=4 +endif ifdef CONFIG_CC_IS_CLANG CFLAGS_$(AMDDALPATH)/dc/dcn21/dcn21_resource.o += -msse2
drivers/gpu/drm/amd/display/dc/dml/Makefile +4 -6
@@ -24,13 +24,11 @@ # It provides the general basic services required by other DAL # subcomponents. -ifneq ($(call cc-option, -mpreferred-stack-boundary=4),) - cc_stack_align := -mpreferred-stack-boundary=4 -else ifneq ($(call cc-option, -mstack-alignment=16),) - cc_stack_align := -mstack-alignment=16 -endif +dml_ccflags := -mhard-float -msse -dml_ccflags := -mhard-float -msse $(cc_stack_align) +ifdef CONFIG_CC_IS_GCC +dml_ccflags += -mpreferred-stack-boundary=4 +endif ifdef CONFIG_CC_IS_CLANG dml_ccflags += -msse2
drivers/gpu/drm/amd/display/dc/dsc/Makefile +4 -6
@@ -1,13 +1,11 @@ # # Makefile for the 'dsc' sub-component of DAL. -ifneq ($(call cc-option, -mpreferred-stack-boundary=4),) - cc_stack_align := -mpreferred-stack-boundary=4 -else ifneq ($(call cc-option, -mstack-alignment=16),) - cc_stack_align := -mstack-alignment=16 -endif +dsc_ccflags := -mhard-float -msse -dsc_ccflags := -mhard-float -msse $(cc_stack_align) +ifdef CONFIG_CC_IS_GCC +dsc_ccflags += -mpreferred-stack-boundary=4 +endif ifdef CONFIG_CC_IS_CLANG dsc_ccflags += -msse2
Signature · 1 line
2.23.0.700.g56cf767bdb-goog
RE: [PATCH 1/3] drm/amdgpu: fix stack alignment ABI mismatch for Clang
Regards,
Shirish S
-----Original Message-----
From: Nick Desaulniers <[email protected]>
Sent: Thursday, October 17, 2019 4:32 AM
To: Wentland, Harry <[email protected]>; Deucher, Alexander <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; S, Shirish <[email protected]>; Zhou, David(ChunMing) <[email protected]>; Koenig, Christian <[email protected]>; [email protected]; [email protected]; Nick Desaulniers <[email protected]>
Subject: [PATCH 1/3] drm/amdgpu: fix stack alignment ABI mismatch for Clang
The x86 kernel is compiled with an 8B stack alignment via -mpreferred-stack-boundary=3 for GCC since 3.6-rc1 via commit d9b0cde91c60 ("x86-64, gcc: Use -mpreferred-stack-boundary=3 if supported") or -mstack-alignment=8 for Clang. Parts of the AMDGPU driver are compiled with 16B stack alignment.
Generally, the stack alignment is part of the ABI. Linking together two different translation units with differing stack alignment is dangerous, particularly when the translation unit with the smaller stack alignment makes calls into the translation unit with the larger stack alignment. While 8B aligned stacks are sometimes also 16B aligned, they are not always.
Multiple users have reported General Protection Faults (GPF) when using the AMDGPU driver compiled with Clang. Clang is placing objects in stack slots assuming the stack is 16B aligned, and selecting instructions that require 16B aligned memory operands.
At runtime, syscall handlers with 8B aligned stack call into code that assumes 16B stack alignment. When the stack is a multiple of 8B but not 16B, these instructions result in a GPF.
Remove the code that added compatibility between the differing compiler flags, as it will result in runtime GPFs when built with Clang. Cleanups for GCC will be sent in later patches in the series.
--- drivers/gpu/drm/amd/display/dc/calcs/Makefile | 10 ++++------ drivers/gpu/drm/amd/display/dc/dcn20/Makefile | 10 ++++------ drivers/gpu/drm/amd/display/dc/dcn21/Makefile | 10 ++++------ drivers/gpu/drm/amd/display/dc/dml/Makefile | 10 ++++------ drivers/gpu/drm/amd/display/dc/dsc/Makefile | 10 ++++------ 5 files changed, 20 insertions(+), 30 deletions(-)
drivers/gpu/drm/amd/display/dc/calcs/Makefile +3 -6
@@ -24,13 +24,11 @@ # It calculates Bandwidth and Watermarks values for HW programming # -ifneq ($(call cc-option, -mpreferred-stack-boundary=4),) - cc_stack_align := -mpreferred-stack-boundary=4 -else ifneq ($(call cc-option, -mstack-alignment=16),) - cc_stack_align := -mstack-alignment=16 -endif +calcs_ccflags := -mhard-float -msse -calcs_ccflags := -mhard-float -msse $(cc_stack_align) +ifdef CONFIG_CC_IS_GCC +calcs_ccflags += -mpreferred-stack-boundary=4 endif ifdef CONFIG_CC_IS_CLANG calcs_ccflags += -msse2
drivers/gpu/drm/amd/display/dc/dcn20/Makefile +4 -6
@@ -10,13 +10,11 @@ ifdef CONFIG_DRM_AMD_DC_DSC_SUPPORT DCN20 += dcn20_dsc.o endif -ifneq ($(call cc-option, -mpreferred-stack-boundary=4),) - cc_stack_align := -mpreferred-stack-boundary=4 -else ifneq ($(call cc-option, -mstack-alignment=16),) - cc_stack_align := -mstack-alignment=16 -endif +CFLAGS_$(AMDDALPATH)/dc/dcn20/dcn20_resource.o := -mhard-float -msse -CFLAGS_$(AMDDALPATH)/dc/dcn20/dcn20_resource.o := -mhard-float -msse $(cc_stack_align) +ifdef CONFIG_CC_IS_GCC +CFLAGS_$(AMDDALPATH)/dc/dcn20/dcn20_resource.o += +-mpreferred-stack-boundary=4 endif ifdef CONFIG_CC_IS_CLANG CFLAGS_$(AMDDALPATH)/dc/dcn20/dcn20_resource.o += -msse2 diff --git a/drivers/gpu/drm/amd/display/dc/dcn21/Makefile b/drivers/gpu/drm/amd/display/dc/dcn21/Makefile
index ef673bffc241..7057e20748b9 100644
drivers/gpu/drm/amd/display/dc/dcn21/Makefile +4 -6
@@ -3,13 +3,11 @@ DCN21 = dcn21_hubp.o dcn21_hubbub.o dcn21_resource.o -ifneq ($(call cc-option, -mpreferred-stack-boundary=4),) - cc_stack_align := -mpreferred-stack-boundary=4 -else ifneq ($(call cc-option, -mstack-alignment=16),) - cc_stack_align := -mstack-alignment=16 -endif +CFLAGS_$(AMDDALPATH)/dc/dcn21/dcn21_resource.o := -mhard-float -msse -CFLAGS_$(AMDDALPATH)/dc/dcn21/dcn21_resource.o := -mhard-float -msse $(cc_stack_align) +ifdef CONFIG_CC_IS_GCC +CFLAGS_$(AMDDALPATH)/dc/dcn21/dcn21_resource.o += +-mpreferred-stack-boundary=4 endif ifdef CONFIG_CC_IS_CLANG CFLAGS_$(AMDDALPATH)/dc/dcn21/dcn21_resource.o += -msse2 diff --git a/drivers/gpu/drm/amd/display/dc/dml/Makefile b/drivers/gpu/drm/amd/display/dc/dml/Makefile
index 5b2a65b42403..1bd6e307b7f8 100644
drivers/gpu/drm/amd/display/dc/dml/Makefile +3 -6
@@ -24,13 +24,11 @@ # It provides the general basic services required by other DAL # subcomponents. -ifneq ($(call cc-option, -mpreferred-stack-boundary=4),) - cc_stack_align := -mpreferred-stack-boundary=4 -else ifneq ($(call cc-option, -mstack-alignment=16),) - cc_stack_align := -mstack-alignment=16 -endif +dml_ccflags := -mhard-float -msse -dml_ccflags := -mhard-float -msse $(cc_stack_align) +ifdef CONFIG_CC_IS_GCC +dml_ccflags += -mpreferred-stack-boundary=4 endif ifdef CONFIG_CC_IS_CLANG dml_ccflags += -msse2
drivers/gpu/drm/amd/display/dc/dsc/Makefile +3 -6
@@ -1,13 +1,11 @@ # # Makefile for the 'dsc' sub-component of DAL. -ifneq ($(call cc-option, -mpreferred-stack-boundary=4),) - cc_stack_align := -mpreferred-stack-boundary=4 -else ifneq ($(call cc-option, -mstack-alignment=16),) - cc_stack_align := -mstack-alignment=16 -endif +dsc_ccflags := -mhard-float -msse -dsc_ccflags := -mhard-float -msse $(cc_stack_align) +ifdef CONFIG_CC_IS_GCC +dsc_ccflags += -mpreferred-stack-boundary=4 endif ifdef CONFIG_CC_IS_CLANG dsc_ccflags += -msse2
--
2.23.0.700.g56cf767bdb-goog
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
PATCH 2/3 drm/amdgpu: fix stack alignment ABI mismatch for GCC 7.1+
In reply to Nick DesaulniersGCC earlier than 7.1 errors when compiling code that makes use of doubles and sets a stack alignment outside of the range of [2^4-2^12]:
$ cat foo.c
double foo(double x, double y) {
return x + y;
}
$ gcc-4.9 -mpreferred-stack-boundary=3 foo.c
error: -mpreferred-stack-boundary=3 is not between 4 and 12This is likely why the AMDGPU driver was ever compiled with a different stack alignment (and thus different ABI) than the rest of the x86 kernel. The kernel uses 8B stack alignment, while the driver was using 16B stack alignment in a few places.
Since GCC 7.1+ doesn't error, fix the ABI mismatch for users of newer versions of GCC.
There was discussion about whether to mark the driver broken or not for users of GCC earlier than 7.1, but since the driver currently is working, don't explicitly break the driver for them here.
Relying on differing stack alignment is unspecified behavior, and brittle, and may break in the future.
This patch is no functional change for GCC users earlier than 7.1. It's been compile tested on GCC 4.9 and 8.3 to check the correct flags. It should be boot tested when built with GCC 7.1+.
-mincoming-stack-boundary= or -mstackrealign may help keep this code building for pre-GCC 7.1 users.
The version check for GCC is broken into two conditionals, both because cc-ifversion is currently GCC specific, and it simplifies a subsequent patch.
--- drivers/gpu/drm/amd/display/dc/calcs/Makefile | 9 +++++++++ drivers/gpu/drm/amd/display/dc/dcn20/Makefile | 9 +++++++++ drivers/gpu/drm/amd/display/dc/dcn21/Makefile | 9 +++++++++ drivers/gpu/drm/amd/display/dc/dml/Makefile | 9 +++++++++ drivers/gpu/drm/amd/display/dc/dsc/Makefile | 9 +++++++++ 5 files changed, 45 insertions(+)
drivers/gpu/drm/amd/display/dc/calcs/Makefile +9 -0
@@ -27,6 +27,15 @@ calcs_ccflags := -mhard-float -msse ifdef CONFIG_CC_IS_GCC +ifeq ($(call cc-ifversion, -lt, 0701, y), y) +IS_OLD_GCC = 1 +endif +endif + +ifdef IS_OLD_GCC +# Stack alignment mismatch, proceed with caution. +# GCC < 7.1 cannot compile code using `double` and -mpreferred-stack-boundary=3 +# (8B stack alignment). calcs_ccflags += -mpreferred-stack-boundary=4 endif
drivers/gpu/drm/amd/display/dc/dcn20/Makefile +9 -0
@@ -13,6 +13,15 @@ endif CFLAGS_$(AMDDALPATH)/dc/dcn20/dcn20_resource.o := -mhard-float -msse ifdef CONFIG_CC_IS_GCC +ifeq ($(call cc-ifversion, -lt, 0701, y), y) +IS_OLD_GCC = 1 +endif +endif + +ifdef IS_OLD_GCC +# Stack alignment mismatch, proceed with caution. +# GCC < 7.1 cannot compile code using `double` and -mpreferred-stack-boundary=3 +# (8B stack alignment). CFLAGS_$(AMDDALPATH)/dc/dcn20/dcn20_resource.o += -mpreferred-stack-boundary=4 endif
drivers/gpu/drm/amd/display/dc/dcn21/Makefile +9 -0
@@ -6,6 +6,15 @@ DCN21 = dcn21_hubp.o dcn21_hubbub.o dcn21_resource.o CFLAGS_$(AMDDALPATH)/dc/dcn21/dcn21_resource.o := -mhard-float -msse ifdef CONFIG_CC_IS_GCC +ifeq ($(call cc-ifversion, -lt, 0701, y), y) +IS_OLD_GCC = 1 +endif +endif + +ifdef IS_OLD_GCC +# Stack alignment mismatch, proceed with caution. +# GCC < 7.1 cannot compile code using `double` and -mpreferred-stack-boundary=3 +# (8B stack alignment). CFLAGS_$(AMDDALPATH)/dc/dcn21/dcn21_resource.o += -mpreferred-stack-boundary=4 endif
drivers/gpu/drm/amd/display/dc/dml/Makefile +9 -0
@@ -27,6 +27,15 @@ dml_ccflags := -mhard-float -msse ifdef CONFIG_CC_IS_GCC +ifeq ($(call cc-ifversion, -lt, 0701, y), y) +IS_OLD_GCC = 1 +endif +endif + +ifdef IS_OLD_GCC +# Stack alignment mismatch, proceed with caution. +# GCC < 7.1 cannot compile code using `double` and -mpreferred-stack-boundary=3 +# (8B stack alignment). dml_ccflags += -mpreferred-stack-boundary=4 endif
drivers/gpu/drm/amd/display/dc/dsc/Makefile +9 -0
@@ -4,6 +4,15 @@ dsc_ccflags := -mhard-float -msse ifdef CONFIG_CC_IS_GCC +ifeq ($(call cc-ifversion, -lt, 0701, y), y) +IS_OLD_GCC = 1 +endif +endif + +ifdef IS_OLD_GCC +# Stack alignment mismatch, proceed with caution. +# GCC < 7.1 cannot compile code using `double` and -mpreferred-stack-boundary=3 +# (8B stack alignment). dsc_ccflags += -mpreferred-stack-boundary=4 endif
Signature · 1 line
2.23.0.700.g56cf767bdb-goog
PATCH 3/3 drm/amdgpu: enable -msse2 for GCC 7.1+ users
In reply to Nick DesaulniersA final attempt at enabling sse2 for GCC users.
Orininally attempted in:
commit 10117450735c ("drm/amd/display: add -msse2 to prevent Clang from emitting libcalls to undefined SW FP routines")
Reverted due to "reported instability" in:
commit 193392ed9f69 ("Revert "drm/amd/display: add -msse2 to prevent Clang from emitting libcalls to undefined SW FP routines"")
Re-added just for Clang in:
commit 0f0727d971f6 ("drm/amd/display: readd -msse2 to prevent Clang from emitting libcalls to undefined SW FP routines")
The original report didn't have enough information to know if the GPF was due to misalignment, but I suspect that it was. (The missing information was the disassembly of the function at the bottom of the trace, to see if the instruction pointer pointed to an instruction with 16B alignment memory operand requirements. The stack trace does show the stack was only 8B but not 16B aligned though, which makes this a strong possibility).
Now that the stack misalignment issue has been fixed for users of GCC 7.1+, reattempt adding -msse2. This matches Clang.
It will likely never be safe to enable this for pre-GCC 7.1 AND use a 16B aligned stack in these translation units.
This is only a functional change for GCC 7.1+ users, and should be boot tested.
--- drivers/gpu/drm/amd/display/dc/calcs/Makefile | 4 +--- drivers/gpu/drm/amd/display/dc/dcn20/Makefile | 4 +--- drivers/gpu/drm/amd/display/dc/dcn21/Makefile | 4 +--- drivers/gpu/drm/amd/display/dc/dml/Makefile | 4 +--- drivers/gpu/drm/amd/display/dc/dsc/Makefile | 4 +--- 5 files changed, 5 insertions(+), 15 deletions(-)
drivers/gpu/drm/amd/display/dc/calcs/Makefile +1 -3
@@ -37,9 +37,7 @@ ifdef IS_OLD_GCC # GCC < 7.1 cannot compile code using `double` and -mpreferred-stack-boundary=3 # (8B stack alignment). calcs_ccflags += -mpreferred-stack-boundary=4 -endif - -ifdef CONFIG_CC_IS_CLANG +else calcs_ccflags += -msse2 endif
drivers/gpu/drm/amd/display/dc/dcn20/Makefile +1 -3
@@ -23,9 +23,7 @@ ifdef IS_OLD_GCC # GCC < 7.1 cannot compile code using `double` and -mpreferred-stack-boundary=3 # (8B stack alignment). CFLAGS_$(AMDDALPATH)/dc/dcn20/dcn20_resource.o += -mpreferred-stack-boundary=4 -endif - -ifdef CONFIG_CC_IS_CLANG +else CFLAGS_$(AMDDALPATH)/dc/dcn20/dcn20_resource.o += -msse2 endif
drivers/gpu/drm/amd/display/dc/dcn21/Makefile +1 -3
@@ -16,9 +16,7 @@ ifdef IS_OLD_GCC # GCC < 7.1 cannot compile code using `double` and -mpreferred-stack-boundary=3 # (8B stack alignment). CFLAGS_$(AMDDALPATH)/dc/dcn21/dcn21_resource.o += -mpreferred-stack-boundary=4 -endif - -ifdef CONFIG_CC_IS_CLANG +else CFLAGS_$(AMDDALPATH)/dc/dcn21/dcn21_resource.o += -msse2 endif
drivers/gpu/drm/amd/display/dc/dml/Makefile +1 -3
@@ -37,9 +37,7 @@ ifdef IS_OLD_GCC # GCC < 7.1 cannot compile code using `double` and -mpreferred-stack-boundary=3 # (8B stack alignment). dml_ccflags += -mpreferred-stack-boundary=4 -endif - -ifdef CONFIG_CC_IS_CLANG +else dml_ccflags += -msse2 endif
drivers/gpu/drm/amd/display/dc/dsc/Makefile +1 -3
@@ -14,9 +14,7 @@ ifdef IS_OLD_GCC # GCC < 7.1 cannot compile code using `double` and -mpreferred-stack-boundary=3 # (8B stack alignment). dsc_ccflags += -mpreferred-stack-boundary=4 -endif - -ifdef CONFIG_CC_IS_CLANG +else dsc_ccflags += -msse2 endif
Signature · 1 line
2.23.0.700.g56cf767bdb-goog
Quoted 33 lines · Nick Desaulniers
The x86 kernel is compiled with an 8B stack alignment via
-mpreferred-stack-boundary=3for GCC since 3.6-rc1 via
commitd9b0cde91c60("x86-64, gcc: Use -mpreferred-stack-boundary=3 if supported") or-mstack-alignment=8for Clang. Parts of the AMDGPU driver are compiled with 16B stack alignment.Generally, the stack alignment is part of the ABI. Linking together two different translation units with differing stack alignment is dangerous, particularly when the translation unit with the smaller stack alignment makes calls into the translation unit with the larger stack alignment. While 8B aligned stacks are sometimes also 16B aligned, they are not always.
Multiple users have reported General Protection Faults (GPF) when using the AMDGPU driver compiled with Clang. Clang is placing objects in stack slots assuming the stack is 16B aligned, and selecting instructions that require 16B aligned memory operands.
At runtime, syscall handlers with 8B aligned stack call into code that assumes 16B stack alignment. When the stack is a multiple of 8B but not 16B, these instructions result in a GPF.
Remove the code that added compatibility between the differing compiler flags, as it will result in runtime GPFs when built with Clang.
The series is broken into 3 patches, the first is an important fix for Clang for ChromeOS. The rest are attempted cleanups for GCC, but require additional boot testing. The first patch is critical, the rest are nice to have. I've compile tested the series with ToT Clang, GCC 4.9, and GCC 8.3 **but** I do not have hardware to test on, so I need folks with the above compilers and relevant hardware to help test the series.
Quoted 4 lines
The first patch is a functional change for Clang only. It does not change anything for any version of GCC. Yuxuan boot tested a previous incarnation on hardware, but I've changed it enough that I think it made sense to drop the previous tested by tag.
Thanks for testing the first patch Shirish. Are you or Yuxuan able to test the rest of the series with any combination of {clang|gcc < 7.1| gcc >= 7.1} on hardware and report your findings?
Quoted 13 lines
The second patch is a functional change for GCC 7.1+ only. It does not affect older versions of GCC or Clang (though if someone wanted to double check with pre-GCC 7.1 it wouldn't hurt). It should be boot tested on GCC 7.1+ on the relevant hardware.
The final patch is also a functional change for GCC 7.1+ only. It does not affect older versions of GCC or Clang. It should be boot tested on GCC 7.1+ on the relevant hardware. Theoretically, there may be an issue with it, and it's ok to drop it. The series was intentional broken into 3 in order to allow them to be incrementally tested and accepted. It's ok to take earlier patches without the later patches.
Quoted 3 lines
And finally, I do not condone linking object files of differing stack alignments. Idealistically, we'd mark the driver broken for pre-GCC 7.1. Pragmatically, "if it ain't broke, don't fix it."
Harry, Alex,
Thoughts on the series? Has AMD been able to stress test these more internally?
Quoted 16 lines
Nick Desaulniers (3):
drm/amdgpu: fix stack alignment ABI mismatch for Clang drm/amdgpu: fix stack alignment ABI mismatch for GCC 7.1+ drm/amdgpu: enable -msse2 for GCC 7.1+ usersdrivers/gpu/drm/amd/display/dc/calcs/Makefile | 19 ++++++++++++------- drivers/gpu/drm/amd/display/dc/dcn20/Makefile | 19 ++++++++++++------- drivers/gpu/drm/amd/display/dc/dcn21/Makefile | 19 ++++++++++++------- drivers/gpu/drm/amd/display/dc/dml/Makefile | 19 ++++++++++++------- drivers/gpu/drm/amd/display/dc/dsc/Makefile | 19 ++++++++++++------- 5 files changed, 60 insertions(+), 35 deletions(-)--
2.23.0.700.g56cf767bdb-goog
Signature · 2 lines
Thanks, ~Nick Desaulniers
Quoted 58 lines · Nick Desaulniers
Quoted 33 lines · Nick Desaulniers
The x86 kernel is compiled with an 8B stack alignment via
-mpreferred-stack-boundary=3for GCC since 3.6-rc1 via
commitd9b0cde91c60("x86-64, gcc: Use -mpreferred-stack-boundary=3 if supported") or-mstack-alignment=8for Clang. Parts of the AMDGPU driver are compiled with 16B stack alignment.Generally, the stack alignment is part of the ABI. Linking together two different translation units with differing stack alignment is dangerous, particularly when the translation unit with the smaller stack alignment makes calls into the translation unit with the larger stack alignment. While 8B aligned stacks are sometimes also 16B aligned, they are not always.
Multiple users have reported General Protection Faults (GPF) when using the AMDGPU driver compiled with Clang. Clang is placing objects in stack slots assuming the stack is 16B aligned, and selecting instructions that require 16B aligned memory operands.
At runtime, syscall handlers with 8B aligned stack call into code that assumes 16B stack alignment. When the stack is a multiple of 8B but not 16B, these instructions result in a GPF.
Remove the code that added compatibility between the differing compiler flags, as it will result in runtime GPFs when built with Clang.
The series is broken into 3 patches, the first is an important fix for Clang for ChromeOS. The rest are attempted cleanups for GCC, but require additional boot testing. The first patch is critical, the rest are nice to have. I've compile tested the series with ToT Clang, GCC 4.9, and GCC 8.3 **but** I do not have hardware to test on, so I need folks with the above compilers and relevant hardware to help test the series.
Quoted 4 lines
The first patch is a functional change for Clang only. It does not change anything for any version of GCC. Yuxuan boot tested a previous incarnation on hardware, but I've changed it enough that I think it made sense to drop the previous tested by tag.
Thanks for testing the first patch Shirish. Are you or Yuxuan able to test the rest of the series with any combination of {clang|gcc < 7.1| gcc >= 7.1} on hardware and report your findings?
Quoted 13 lines
The second patch is a functional change for GCC 7.1+ only. It does not affect older versions of GCC or Clang (though if someone wanted to double check with pre-GCC 7.1 it wouldn't hurt). It should be boot tested on GCC 7.1+ on the relevant hardware.
The final patch is also a functional change for GCC 7.1+ only. It does not affect older versions of GCC or Clang. It should be boot tested on GCC 7.1+ on the relevant hardware. Theoretically, there may be an issue with it, and it's ok to drop it. The series was intentional broken into 3 in order to allow them to be incrementally tested and accepted. It's ok to take earlier patches without the later patches.
Quoted 7 lines
Quoted 3 lines
And finally, I do not condone linking object files of differing stack alignments. Idealistically, we'd mark the driver broken for pre-GCC 7.1. Pragmatically, "if it ain't broke, don't fix it."
Harry, Alex,
Thoughts on the series? Has AMD been able to stress test these more internally?
Was out of the office most of the week. They look good to me. I'll pull them in today and see what happens.
Alex
Quoted 18 lines
Quoted 13 lines
Nick Desaulniers (3):
drm/amdgpu: fix stack alignment ABI mismatch for Clang drm/amdgpu: fix stack alignment ABI mismatch for GCC 7.1+ drm/amdgpu: enable -msse2 for GCC 7.1+ usersdrivers/gpu/drm/amd/display/dc/calcs/Makefile | 19 ++++++++++++------- drivers/gpu/drm/amd/display/dc/dcn20/Makefile | 19 ++++++++++++------- drivers/gpu/drm/amd/display/dc/dcn21/Makefile | 19 ++++++++++++------- drivers/gpu/drm/amd/display/dc/dml/Makefile | 19 ++++++++++++------- drivers/gpu/drm/amd/display/dc/dsc/Makefile | 19 ++++++++++++------- 5 files changed, 60 insertions(+), 35 deletions(-)Quoted 3 lines
--
2.23.0.700.g56cf767bdb-goog
Quoted 7 lines
--
Thanks,
~Nick Desaulniers
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx