x86: efi: avoid BUILD_BUG_ON() for non-constant p4d_index
Subsystems (3)
PATCH 1/1 x86: efi: avoid BUILD_BUG_ON() for non-constant p4d_index
From: Arnd Bergmann <[email protected]>
When 5-level page tables are enabled, clang triggers a BUILD_BUG_ON():
x86_64-linux-ld: arch/x86/platform/efi/efi_64.o: in function `efi_sync_low_kernel_mappings':
efi_64.c:(.text+0x22c): undefined reference to `__compiletime_assert_354'
Use the same method as in commit c65e774fb3f6 ("x86/mm: Make PGDIR_SHIFT and PTRS_PER_P4D variable") and change it to MAYBE_BUILD_BUG_ON(), so it only triggers for constant input.
--- arch/x86/platform/efi/efi_64.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
arch/x86/platform/efi/efi_64.c +2 -2
@@ -137,8 +137,8 @@ void efi_sync_low_kernel_mappings(void)* As with PGDs, we share all P4D entries apart from the one entry * that covers the EFI runtime mapping space. */ -BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); -BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); +MAYBE_BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); +MAYBE_BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); pgd_efi = efi_pgd +pgd_index (EFI_VA_END ); pgd_k =pgd_offset_k (EFI_VA_END );
Signature · 1 line
2.29.2
Quoted 7 lines · Arnd Bergmann
From: Arnd Bergmann <[email protected]>
When 5-level page tables are enabled, clang triggers a
BUILD_BUG_ON():x86_64-linux-ld:
arch/x86/platform/efi/efi_64.o: in function `efi_sync_low_kernel_mappings':
efi_64.c:(.text+0x22c): undefined reference to `__compiletime_assert_354'
Quoted 6 lines
Use the same method as in commit
c65e774fb3f6("x86/mm: MakePGDIR_SHIFTandPTRS_PER_P4Dvariable") and change it toMAYBE_BUILD_BUG_ON(), so it only triggers for constant input.
Quoted 22 lines
--- arch/x86/platform/efi/efi_64.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)arch/x86/platform/efi/efi_64.c +2
-2@@ -137,8 +137,8 @@ void efi_sync_low_kernel_mappings(void)* As with PGDs, we share all P4D entries apart from the one entry * that covers the EFI runtime mapping space. */ -BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); -BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); +MAYBE_BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); +MAYBE_BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); pgd_efi = efi_pgd +pgd_index (EFI_VA_END ); pgd_k =pgd_offset_k (EFI_VA_END );Signature · 1 line
2.29.2
Quoted 8 lines · Arnd Bergmann
From: Arnd Bergmann <[email protected]>
When 5-level page tables are enabled, clang triggers a
BUILD_BUG_ON():x86_64-linux-ld:
arch/x86/platform/efi/efi_64.o: in function `efi_sync_low_kernel_mappings':
efi_64.c:(.text+0x22c): undefined reference to `__compiletime_assert_354'
Quoted 6 lines
Use the same method as in commit
c65e774fb3f6("x86/mm: MakePGDIR_SHIFTandPTRS_PER_P4Dvariable") and change it toMAYBE_BUILD_BUG_ON(), so it only triggers for constant input.
This can go via the x86 tree directly, IMO
Quoted 22 lines
--- arch/x86/platform/efi/efi_64.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)arch/x86/platform/efi/efi_64.c +2
-3@@ -137,8 +137,8 @@ void efi_sync_low_kernel_mappings(void)* As with PGDs, we share all P4D entries apart from the one entry * that covers the EFI runtime mapping space. */ -BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); -BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); +MAYBE_BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); +MAYBE_BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); pgd_efi = efi_pgd +pgd_index (EFI_VA_END ); pgd_k =pgd_offset_k (EFI_VA_END ); --2.29.2
Borislav Petkov
maintainerQuoted 3 lines · Arnd Bergmann
From: Arnd Bergmann <[email protected]>
When 5-level page tables are enabled, clang triggers a
BUILD_BUG_ON():
I have CONFIG_X86_5LEVEL=y, CONFIG_EFI=y and am using Debian clang version 10.0.1-8+b1 but my .config builds just fine.
How do you trigger this?
Thx.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteQuoted 9 lines · Borislav Petkov
Quoted 3 lines · Arnd Bergmann
From: Arnd Bergmann <[email protected]>
When 5-level page tables are enabled, clang triggers a
BUILD_BUG_ON():I have
CONFIG_X86_5LEVEL=y,CONFIG_EFI=yand am using Debian clang version 10.0.1-8+b1 but my .config builds just fine.How do you trigger this?
I triggered it with CONFIG_UBSAN=y + CONFIG_UBSAN_UNSIGNED_OVERFLOW=y (it can be exposed with an allyesconfig/allmodconfig on mainline currently).
Cheers,
Nathan
Borislav Petkov
maintainerQuoted 3 lines · Nathan Chancellor
I triggered it with
CONFIG_UBSAN=y+CONFIG_UBSAN_UNSIGNED_OVERFLOW=y(it can be exposed with an allyesconfig/allmodconfig on mainline currently).
Yah, I can trigger with that, thanks.
But I'll be damned, check this out:
clang preprocesses to this:
do { extern void __compiletime_assert_332(void) ; if (!(!(p4d_index((-68 * ((1UL) << 30))) != p4d_index((0xffffffffff000000UL))))) __compiletime_assert_332(); } while (0);
The resulting asm is:
.LBB1_32:
movabsq $-73014444032, %r13 # imm = 0xFFFFFFEF00000000
testb $1, %al
jne .LBB1_33
.LBB1_34:
xorl %r14d, %ebx
testl $33554431, %ebx # imm = 0x1FFFFFF
je .LBB1_36# %bb.35:
callq __compiletime_assert_332
so the undefined symbol is there, leading to:
ld: arch/x86/platform/efi/efi_64.o: in function `efi_sync_low_kernel_mappings':/home/boris/kernel/linux/arch/x86/platform/efi/efi_64.c:140: undefined reference to `__compiletime_assert_332'
Now look at gcc:
It preprocesses to:
do { extern void __compiletime_assert_332(void) __attribute__((__error__("BUILD_BUG_ON failed: " "p4d_index(EFI_VA_END) != p4d_index(MODULES_END)"))); if (!(!(p4d_index((-68 * ((1UL) << 30))) != p4d_index((0xffffffffff000000UL))))) __compiletime_assert_332(); } while (0);
Resulting asm:
$ grep __compiletime_assert_332 arch/x86/platform/efi/efi_64.s $
That thing has been optimized away!
Which means, those build assertions are gone on gcc and they don't catch diddly squat. I sure hope I'm missing something here...
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteQuoted 40 lines · Borislav Petkov
Quoted 3 lines · Nathan Chancellor
I triggered it with
CONFIG_UBSAN=y+CONFIG_UBSAN_UNSIGNED_OVERFLOW=y(it can be exposed with an allyesconfig/allmodconfig on mainline currently).Yah, I can trigger with that, thanks.
But I'll be damned, check this out:
clang preprocesses to this:
do { extern void
__compiletime_assert_332(void); if (!(!(p4d_index((-68 * ((1UL) << 30))) != p4d_index((0xffffffffff000000UL)))))__compiletime_assert_332(); } while (0);The resulting asm is:
.LBB1_32:
movabsq $-73014444032, %r13 # imm = 0xFFFFFFEF00000000 testb $1, %al jne .LBB1_33 .LBB1_34: xorl %r14d, %ebx testl $33554431, %ebx # imm = 0x1FFFFFF je .LBB1_36# %bb.35:
callq__compiletime_assert_332so the undefined symbol is there, leading to:
ld:
arch/x86/platform/efi/efi_64.o: in function `efi_sync_low_kernel_mappings':/home/boris/kernel/linux/arch/x86/platform/efi/efi_64.c:140: undefined reference to `__compiletime_assert_332'Now look at gcc:
It preprocesses to:
do { extern void
__compiletime_assert_332(void)__attribute__((__error__("BUILD_BUG_ON failed: " "p4d_index(EFI_VA_END) != p4d_index(MODULES_END)"))); if (!(!(p4d_index((-68 * ((1UL) << 30))) != p4d_index((0xffffffffff000000UL)))))__compiletime_assert_332(); } while (0);Resulting asm:
Quoted 7 lines
$ grep __compiletime_assert_332 arch/x86/platform/efi/efi_64.s $That thing has been optimized away!
Which means, those build assertions are gone on gcc and they don't catch diddly squat. I sure hope I'm missing something here...
That's how build-time assertions work: they are _supposed_ to be optimized away completely when the assertion is true. If they're _not_ optimized away, the build will fail.
Borislav Petkov
maintainerQuoted 3 lines · Arvind Sankar
That's how build-time assertions work: they are _supposed_ to be optimized away completely when the assertion is true. If they're _not_ optimized away, the build will fail.
Yah, that I know, thanks.
If gcc really inlines p4d_index() and does a lot more aggressive optimization to determine that the condition is false and thus optimize everything away (and clang doesn't), then that would explain the observation.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteQuoted 6 lines · Borislav Petkov
Quoted 3 lines · Arvind Sankar
That's how build-time assertions work: they are _supposed_ to be optimized away completely when the assertion is true. If they're _not_ optimized away, the build will fail.
Quoted 6 lines
Yah, that I know, thanks.
If gcc really inlines
p4d_index()and does a lot more aggressive optimization to determine that the condition is false and thus optimize everything away (and clang doesn't), then that would explain the observation.
One difference is that gcc does not have -fsanitize=unsigned-integer-overflow at all, and I don't see the assertion without that on clang either, so it's possible that clang behaves as designed here.
The description is:
-fsanitize=unsigned-integer-overflow: Unsigned integer overflow, where
the result of an unsigned integer computation cannot be represented in its type. Unlike signed integer overflow, this is not undefined behavior, but it is often unintentional. This sanitizer does not check for
lossy implicit
conversions performed before such a computation (see
-fsanitize=implicit-conversion).
The "-68 * ((1UL) << 30" computation does overflow an unsigned long as intended, right? Maybe this is enough for the ubsan code in clang to just disable some of the optimization steps that the assertion relies on.
Arnd
Quoted 28 lines · Arnd Bergmann
Quoted 6 lines · Borislav Petkov
Quoted 3 lines · Arvind Sankar
That's how build-time assertions work: they are _supposed_ to be optimized away completely when the assertion is true. If they're _not_ optimized away, the build will fail.
Quoted 6 lines
Yah, that I know, thanks.
If gcc really inlines
p4d_index()and does a lot more aggressive optimization to determine that the condition is false and thus optimize everything away (and clang doesn't), then that would explain the observation.One difference is that gcc does not have -fsanitize=unsigned-integer-overflow at all, and I don't see the assertion without that on clang either, so it's possible that clang behaves as designed here.
The description is:
-fsanitize=unsigned-integer-overflow: Unsigned integer overflow, where
the result of an unsigned integer computation cannot be represented in its type. Unlike signed integer overflow, this is not undefined behavior, but it is often unintentional. This sanitizer does not check for
lossy implicit
conversions performed before such a computation (see
-fsanitize=implicit-conversion).
Quoted 5 lines
The "-68 * ((1UL) << 30" computation does overflow an unsigned long as intended, right? Maybe this is enough for the ubsan code in clang to just disable some of the optimization steps that the assertion relies on.
Arnd
That does seem to be an overflow, but that happens at compile-time. Maybe
AC(-68, UL) << 30
would be a better definition to avoid overflow.
The real issue might be (ptrs_per_p4d - 1) which can overflow at run-time, and maybe the added ubsan code makes p4d_index() not worth inlining according to clang?
Quoted 33 lines · Arvind Sankar
Quoted 32 lines · Arnd Bergmann
Quoted 6 lines · Borislav Petkov
Quoted 3 lines · Arvind Sankar
That's how build-time assertions work: they are _supposed_ to be optimized away completely when the assertion is true. If they're _not_ optimized away, the build will fail.
Quoted 6 lines
Yah, that I know, thanks.
If gcc really inlines
p4d_index()and does a lot more aggressive optimization to determine that the condition is false and thus optimize everything away (and clang doesn't), then that would explain the observation.One difference is that gcc does not have -fsanitize=unsigned-integer-overflow at all, and I don't see the assertion without that on clang either, so it's possible that clang behaves as designed here.
The description is:
-fsanitize=unsigned-integer-overflow: Unsigned integer overflow, where
the result of an unsigned integer computation cannot be represented in its type. Unlike signed integer overflow, this is not undefined behavior, but it is often unintentional. This sanitizer does not check for
lossy implicit
conversions performed before such a computation (see
-fsanitize=implicit-conversion).The "-68 * ((1UL) << 30" computation does overflow an unsigned long as intended, right? Maybe this is enough for the ubsan code in clang to just disable some of the optimization steps that the assertion relies on.
Quoted 6 lines
Quoted 1 line
Arnd
That does seem to be an overflow, but that happens at compile-time. Maybe
AC(-68, UL) << 30would be a better definition to avoid overflow.
Eh, that's an overflow too, isn't it :( Is this option really useful with kernel code -- this sort of thing is probably done all over the place?
Quoted 4 lines
The real issue might be (
ptrs_per_p4d- 1) which can overflow at run-time, and maybe the added ubsan code makesp4d_index()not worth inlining according to clang?
Quoted 35 lines · Arnd Bergmann
From: Arnd Bergmann <[email protected]>
When 5-level page tables are enabled, clang triggers a
BUILD_BUG_ON():x86_64-linux-ld:
arch/x86/platform/efi/efi_64.o: in function `efi_sync_low_kernel_mappings':
efi_64.c:(.text+0x22c): undefined reference to `__compiletime_assert_354'Use the same method as in commit
c65e774fb3f6("x86/mm: MakePGDIR_SHIFTandPTRS_PER_P4Dvariable") and change it toMAYBE_BUILD_BUG_ON(), so it only triggers for constant input.Signed-off-by Arnd Bergmann <[email protected]>--- arch/x86/platform/efi/efi_64.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)arch/x86/platform/efi/efi_64.c +2
-2@@ -137,8 +137,8 @@ void efi_sync_low_kernel_mappings(void)* As with PGDs, we share all P4D entries apart from the one entry * that covers the EFI runtime mapping space. */ -BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); -BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); +MAYBE_BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); +MAYBE_BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); pgd_efi = efi_pgd +pgd_index (EFI_VA_END ); pgd_k =pgd_offset_k (EFI_VA_END );Signature · 1 line
2.29.2
I think this needs more explanation as to why clang is triggering this. The issue mentions clang not inline p4d_index(), and I guess not performing inter-procedural analysis either?
For the second assertion there, everything is always constant AFAICT:EFI_VA_START, EFI_VA_END and P4D_MASK are all constants regardless of CONFIG_5LEVEL.
For the first assertion, it isn't technically constant, but if p4d_index() gets inlined, the compiler should be able to see that the two are always equal, even though ptrs_per_p4d is not constant:
EFI_VA_END >> 39 == MODULES_END >> 39
so the masking with ptrs_per_p4d-1 doesn't matter for the comparison.
As a matter of fact, it seems like the four assertions could be combined into:
BUILD_BUG_ON((EFI_VA_END & P4D_MASK) != (MODULES_END & P4D_MASK)); BUILD_BUG_ON((EFI_VA_START & P4D_MASK) != (EFI_VA_END & P4D_MASK));
instead of separately asserting they're the same PGD entry and the same P4D entry.
Thanks.
Quoted 51 lines · Arvind Sankar
Quoted 35 lines · Arnd Bergmann
From: Arnd Bergmann <[email protected]>
When 5-level page tables are enabled, clang triggers a
BUILD_BUG_ON():x86_64-linux-ld:
arch/x86/platform/efi/efi_64.o: in function `efi_sync_low_kernel_mappings':
efi_64.c:(.text+0x22c): undefined reference to `__compiletime_assert_354'Use the same method as in commit
c65e774fb3f6("x86/mm: MakePGDIR_SHIFTandPTRS_PER_P4Dvariable") and change it toMAYBE_BUILD_BUG_ON(), so it only triggers for constant input.Signed-off-by Arnd Bergmann <[email protected]>--- arch/x86/platform/efi/efi_64.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)arch/x86/platform/efi/efi_64.c +2
-2@@ -137,8 +137,8 @@ void efi_sync_low_kernel_mappings(void)* As with PGDs, we share all P4D entries apart from the one entry * that covers the EFI runtime mapping space. */ -BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); -BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); +MAYBE_BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); +MAYBE_BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); pgd_efi = efi_pgd +pgd_index (EFI_VA_END ); pgd_k =pgd_offset_k (EFI_VA_END );Signature · 1 line
2.29.2I think this needs more explanation as to why clang is triggering this. The issue mentions clang not inline
p4d_index(), and I guess not performing inter-procedural analysis either?For the second assertion there, everything is always constant AFAICT:
EFI_VA_START,EFI_VA_ENDandP4D_MASKare all constants regardless ofCONFIG_5LEVEL.For the first assertion, it isn't technically constant, but if
p4d_index()gets inlined, the compiler should be able to see that the two are always equal, even thoughptrs_per_p4dis not constant:EFI_VA_END >> 39 == MODULES_END >> 39so the masking with ptrs_per_p4d-1 doesn't matter for the comparison.
Quoted 8 lines
As a matter of fact, it seems like the four assertions could be combined into:
BUILD_BUG_ON((EFI_VA_END & P4D_MASK) != (MODULES_END & P4D_MASK)); BUILD_BUG_ON((EFI_VA_START & P4D_MASK) != (EFI_VA_END & P4D_MASK));instead of separately asserting they're the same PGD entry and the same P4D entry.
Thanks.
I actually don't quite get the MODULES_END check -- Ard, do you know what that's for?
What we really should be checking is that EFI_VA_START is in the top-most PGD entry and the top-most P4D entry, since we only copy PGD/P4D entries before EFI_VA_END, but not after EFI_VA_START. So the checks should really be
BUILD_BUG_ON(((EFI_VA_START - 1) & P4D_MASK) != (-1ul & P4D_MASK)); BUILD_BUG_ON(((EFI_VA_START - 1) & P4D_MASK) != (EFI_VA_END & P4D_MASK));
imo. I guess that's what using MODULES_END is effectively checking, but it would be clearer to check it directly.
Ard Biesheuvel
maintainerQuoted 60 lines · Arvind Sankar
Quoted 58 lines · Arvind Sankar
Quoted 33 lines · Arnd Bergmann
From: Arnd Bergmann <[email protected]>
When 5-level page tables are enabled, clang triggers a
BUILD_BUG_ON():x86_64-linux-ld:
arch/x86/platform/efi/efi_64.o: in function `efi_sync_low_kernel_mappings':
efi_64.c:(.text+0x22c): undefined reference to `__compiletime_assert_354'Use the same method as in commit
c65e774fb3f6("x86/mm: MakePGDIR_SHIFTandPTRS_PER_P4Dvariable") and change it toMAYBE_BUILD_BUG_ON(), so it only triggers for constant input.Signed-off-by Arnd Bergmann <[email protected]>--- arch/x86/platform/efi/efi_64.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)arch/x86/platform/efi/efi_64.c +2
-3@@ -137,8 +137,8 @@ void efi_sync_low_kernel_mappings(void)* As with PGDs, we share all P4D entries apart from the one entry * that covers the EFI runtime mapping space. */ -BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); -BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); +MAYBE_BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); +MAYBE_BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); pgd_efi = efi_pgd +pgd_index (EFI_VA_END ); pgd_k =pgd_offset_k (EFI_VA_END ); --Quoted 2 lines
2.29.2
I think this needs more explanation as to why clang is triggering this. The issue mentions clang not inline
p4d_index(), and I guess not performing inter-procedural analysis either?For the second assertion there, everything is always constant AFAICT:
EFI_VA_START,EFI_VA_ENDandP4D_MASKare all constants regardless ofCONFIG_5LEVEL.For the first assertion, it isn't technically constant, but if
p4d_index()gets inlined, the compiler should be able to see that the two are always equal, even thoughptrs_per_p4dis not constant:EFI_VA_END >> 39 == MODULES_END >> 39so the masking with ptrs_per_p4d-1 doesn't matter for the comparison.
As a matter of fact, it seems like the four assertions could be combined into:
BUILD_BUG_ON((EFI_VA_END & P4D_MASK) != (MODULES_END & P4D_MASK)); BUILD_BUG_ON((EFI_VA_START & P4D_MASK) != (EFI_VA_END & P4D_MASK));instead of separately asserting they're the same PGD entry and the same P4D entry.
Quoted 5 lines
Quoted 1 line
Thanks.
I actually don't quite get the
MODULES_ENDcheck -- Ard, do you know what that's for?
Maybe Boris remembers? He wrote the original code for the 'new' EFI page table layout.
Quoted 8 lines
What we really should be checking is that
EFI_VA_STARTis in the top-most PGD entry and the top-most P4D entry, since we only copy PGD/P4D entries beforeEFI_VA_END, but not afterEFI_VA_START. So the checks should really beBUILD_BUG_ON(((EFI_VA_START - 1) & P4D_MASK) != (-1ul & P4D_MASK)); BUILD_BUG_ON(((EFI_VA_START - 1) & P4D_MASK) != (EFI_VA_END & P4D_MASK));imo. I guess that's what using
MODULES_ENDis effectively checking, but it would be clearer to check it directly.
This obviously needs a comment, but checking that everything lives in the top 512 GB of the kernel VA space seems sufficient to me,
Borislav Petkov
maintainerQuoted 61 lines · Ard Biesheuvel
Quoted 60 lines · Arvind Sankar
Quoted 58 lines · Arvind Sankar
Quoted 33 lines · Arnd Bergmann
From: Arnd Bergmann <[email protected]>
When 5-level page tables are enabled, clang triggers a
BUILD_BUG_ON():x86_64-linux-ld:
arch/x86/platform/efi/efi_64.o: in function `efi_sync_low_kernel_mappings':
efi_64.c:(.text+0x22c): undefined reference to `__compiletime_assert_354'Use the same method as in commit
c65e774fb3f6("x86/mm: MakePGDIR_SHIFTandPTRS_PER_P4Dvariable") and change it toMAYBE_BUILD_BUG_ON(), so it only triggers for constant input.Signed-off-by Arnd Bergmann <[email protected]>--- arch/x86/platform/efi/efi_64.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)arch/x86/platform/efi/efi_64.c +2
-3@@ -137,8 +137,8 @@ void efi_sync_low_kernel_mappings(void)* As with PGDs, we share all P4D entries apart from the one entry * that covers the EFI runtime mapping space. */ -BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); -BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); +MAYBE_BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); +MAYBE_BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); pgd_efi = efi_pgd +pgd_index (EFI_VA_END ); pgd_k =pgd_offset_k (EFI_VA_END ); --Quoted 2 lines
2.29.2
I think this needs more explanation as to why clang is triggering this. The issue mentions clang not inline
p4d_index(), and I guess not performing inter-procedural analysis either?For the second assertion there, everything is always constant AFAICT:
EFI_VA_START,EFI_VA_ENDandP4D_MASKare all constants regardless ofCONFIG_5LEVEL.For the first assertion, it isn't technically constant, but if
p4d_index()gets inlined, the compiler should be able to see that the two are always equal, even thoughptrs_per_p4dis not constant:EFI_VA_END >> 39 == MODULES_END >> 39so the masking with ptrs_per_p4d-1 doesn't matter for the comparison.
As a matter of fact, it seems like the four assertions could be combined into:
BUILD_BUG_ON((EFI_VA_END & P4D_MASK) != (MODULES_END & P4D_MASK)); BUILD_BUG_ON((EFI_VA_START & P4D_MASK) != (EFI_VA_END & P4D_MASK));instead of separately asserting they're the same PGD entry and the same P4D entry.
Quoted 8 lines
Quoted 5 lines
Quoted 1 line
Thanks.
I actually don't quite get the
MODULES_ENDcheck -- Ard, do you know what that's for?Maybe Boris remembers? He wrote the original code for the 'new' EFI page table layout.
That was added by Kirill for 5-level pgtables:
e981316f5604 ("x86/efi: Add 5-level paging support")
Documentation/x86/x86_64/mm.rst should explain the pagetable layout:
ffffff8000000000 | -512 GB | ffffffeeffffffff | 444 GB | ... unused hole
ffffffef00000000 | -68 GB | fffffffeffffffff | 64 GB | EFI region mapping space
ffffffff00000000 | -4 GB | ffffffff7fffffff | 2 GB | ... unused hole
ffffffff80000000 | -2 GB | ffffffff9fffffff | 512 MB | kernel text mapping, mapped to physical address 0
ffffffff80000000 |-2048 MB | | |
ffffffffa0000000 |-1536 MB | fffffffffeffffff | 1520 MB | module mapping space
ffffffffff000000 | -16 MB | | |
FIXADDR_START | ~-11 MB | ffffffffff5fffff | ~0.5 MB | kernel-internal fixmap range, variable size and offsetThat thing which starts at -512 GB above is the last PGD on the pagetable. In it, between -4G and -68G there are 64G which are the EFI region mapping space for runtime services.
Frankly I'm not sure what this thing is testing because the EFI VA range is hardcoded and I can't imagine it being somewhere else *except* in the last PGD.
Lemme add Kirill for clarification.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteQuoted 13 lines · Borislav Petkov
Quoted 13 lines
Quoted 12 lines
Quoted 8 lines
As a matter of fact, it seems like the four assertions could be combined into:
BUILD_BUG_ON((EFI_VA_END & P4D_MASK) != (MODULES_END & P4D_MASK)); BUILD_BUG_ON((EFI_VA_START & P4D_MASK) != (EFI_VA_END & P4D_MASK));instead of separately asserting they're the same PGD entry and the same P4D entry.
Thanks.
I actually don't quite get the
MODULES_ENDcheck -- Ard, do you know what that's for?
Quoted 6 lines
Quoted 2 lines
Maybe Boris remembers? He wrote the original code for the 'new' EFI page table layout.
That was added by Kirill for 5-level pgtables:
e981316f5604("x86/efi: Add 5-level paging support")
That just duplicates the existing pgd_index() check for the p4d_index() as well. It looks like the original commit adding efi_sync_low_kernel_mappings() used to copy upto the PGD entry including MODULES_END:
d2f7cbe7b26a7 ("x86/efi: Runtime services virtual mapping")
and then Matt changed that when creating efi_mm:
67a9108ed4313 ("x86/efi: Build our own page table structures")
to use EFI_VA_END instead but have a check that EFI_VA_END is in the same entry as MODULES_END.
AFAICT, MODULES_END is only relevant as being something that happens to be in the top 512GiB, and -1ul would be clearer.
Quoted 12 lines
Documentation/x86/x86_64/mm.rstshould explain the pagetable layout:ffffff8000000000 | -512 GB | ffffffeeffffffff | 444 GB | ... unused hole ffffffef00000000 | -68 GB | fffffffeffffffff | 64 GB | EFI region mapping space ffffffff00000000 | -4 GB | ffffffff7fffffff | 2 GB | ... unused hole ffffffff80000000 | -2 GB | ffffffff9fffffff | 512 MB | kernel text mapping, mapped to physical address 0 ffffffff80000000 |-2048 MB | | | ffffffffa0000000 |-1536 MB | fffffffffeffffff | 1520 MB | module mapping space ffffffffff000000 | -16 MB | | | FIXADDR_START | ~-11 MB | ffffffffff5fffff | ~0.5 MB | kernel-internal fixmap range, variable size and offset
Quoted 7 lines
That thing which starts at -512 GB above is the last PGD on the pagetable. In it, between -4G and -68G there are 64G which are the EFI region mapping space for runtime services.
Frankly I'm not sure what this thing is testing because the EFI VA range is hardcoded and I can't imagine it being somewhere else *except* in the last PGD.
It's just so that someone doesn't just change the #define's for EFI_VA_END/START and think that it will work, I guess.
Another reasonable option, for example, would be to reserve an entire PGD entry, allowing everything but the PGD level to be shared, and adding the EFI PGD to the pgd_list and getting rid of efi_sync_low_kernel_mappings() altogether. There aren't that many PGD entries still unused though, so this is probably not worth it.
Quoted 8 lines
Lemme add Kirill for clarification.
Signature · 4 lines
Regards/Gruss, Boris. https://people.kernel.org/tglx/notes-about-netiquette
Ard Biesheuvel
maintainerQuoted 55 lines · Arvind Sankar
Quoted 13 lines · Borislav Petkov
Quoted 13 lines
Quoted 12 lines
Quoted 8 lines
As a matter of fact, it seems like the four assertions could be combined into:
BUILD_BUG_ON((EFI_VA_END & P4D_MASK) != (MODULES_END & P4D_MASK)); BUILD_BUG_ON((EFI_VA_START & P4D_MASK) != (EFI_VA_END & P4D_MASK));instead of separately asserting they're the same PGD entry and the same P4D entry.
Thanks.
I actually don't quite get the
MODULES_ENDcheck -- Ard, do you know what that's for?Quoted 6 lines
Quoted 2 lines
Maybe Boris remembers? He wrote the original code for the 'new' EFI page table layout.
That was added by Kirill for 5-level pgtables:
e981316f5604("x86/efi: Add 5-level paging support")That just duplicates the existing
pgd_index()check for thep4d_index()as well. It looks like the original commit addingefi_sync_low_kernel_mappings()used to copy upto the PGD entry includingMODULES_END:
d2f7cbe7b26a7("x86/efi: Runtime services virtual mapping")
and then Matt changed that when creatingefi_mm:
67a9108ed4313("x86/efi: Build our own page table structures")
to useEFI_VA_ENDinstead but have a check thatEFI_VA_ENDis in the same entry asMODULES_END.AFAICT,
MODULES_ENDis only relevant as being something that happens to be in the top 512GiB, and -1ul would be clearer.Quoted 12 lines
Documentation/x86/x86_64/mm.rstshould explain the pagetable layout:ffffff8000000000 | -512 GB | ffffffeeffffffff | 444 GB | ... unused hole ffffffef00000000 | -68 GB | fffffffeffffffff | 64 GB | EFI region mapping space ffffffff00000000 | -4 GB | ffffffff7fffffff | 2 GB | ... unused hole ffffffff80000000 | -2 GB | ffffffff9fffffff | 512 MB | kernel text mapping, mapped to physical address 0 ffffffff80000000 |-2048 MB | | | ffffffffa0000000 |-1536 MB | fffffffffeffffff | 1520 MB | module mapping space ffffffffff000000 | -16 MB | | | FIXADDR_START | ~-11 MB | ffffffffff5fffff | ~0.5 MB | kernel-internal fixmap range, variable size and offsetQuoted 7 lines
That thing which starts at -512 GB above is the last PGD on the pagetable. In it, between -4G and -68G there are 64G which are the EFI region mapping space for runtime services.
Frankly I'm not sure what this thing is testing because the EFI VA range is hardcoded and I can't imagine it being somewhere else *except* in the last PGD.
Quoted 9 lines
It's just so that someone doesn't just change the #define's for EFI_VA_END/START and think that it will work, I guess.
Another reasonable option, for example, would be to reserve an entire PGD entry, allowing everything but the PGD level to be shared, and adding the EFI PGD to the
pgd_listand getting rid ofefi_sync_low_kernel_mappings()altogether. There aren't that many PGD entries still unused though, so this is probably not worth it.
The churn doesn't seem to be worth it, tbh.
So could we get rid of the complexity here, and only build_bug() on the start address of the EFI region being outside the topmost p4d? That should make the PGD test redundant as well.
Borislav Petkov
maintainerQuoted 5 lines · Ard Biesheuvel
The churn doesn't seem to be worth it, tbh.
So could we get rid of the complexity here, and only
build_bug()on the start address of the EFI region being outside the topmost p4d? That should make the PGD test redundant as well.
Yah, makes sense to me.
Thx.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteQuoted 66 lines · Ard Biesheuvel
Quoted 55 lines · Arvind Sankar
Quoted 13 lines · Borislav Petkov
Quoted 13 lines
Quoted 12 lines
Quoted 8 lines
As a matter of fact, it seems like the four assertions could be combined into:
BUILD_BUG_ON((EFI_VA_END & P4D_MASK) != (MODULES_END & P4D_MASK)); BUILD_BUG_ON((EFI_VA_START & P4D_MASK) != (EFI_VA_END & P4D_MASK));instead of separately asserting they're the same PGD entry and the same P4D entry.
Thanks.
I actually don't quite get the
MODULES_ENDcheck -- Ard, do you know what that's for?Quoted 6 lines
Quoted 2 lines
Maybe Boris remembers? He wrote the original code for the 'new' EFI page table layout.
That was added by Kirill for 5-level pgtables:
e981316f5604("x86/efi: Add 5-level paging support")That just duplicates the existing
pgd_index()check for thep4d_index()as well. It looks like the original commit addingefi_sync_low_kernel_mappings()used to copy upto the PGD entry includingMODULES_END:
d2f7cbe7b26a7("x86/efi: Runtime services virtual mapping")
and then Matt changed that when creatingefi_mm:
67a9108ed4313("x86/efi: Build our own page table structures")
to useEFI_VA_ENDinstead but have a check thatEFI_VA_ENDis in the same entry asMODULES_END.AFAICT,
MODULES_ENDis only relevant as being something that happens to be in the top 512GiB, and -1ul would be clearer.Quoted 12 lines
Documentation/x86/x86_64/mm.rstshould explain the pagetable layout:ffffff8000000000 | -512 GB | ffffffeeffffffff | 444 GB | ... unused hole ffffffef00000000 | -68 GB | fffffffeffffffff | 64 GB | EFI region mapping space ffffffff00000000 | -4 GB | ffffffff7fffffff | 2 GB | ... unused hole ffffffff80000000 | -2 GB | ffffffff9fffffff | 512 MB | kernel text mapping, mapped to physical address 0 ffffffff80000000 |-2048 MB | | | ffffffffa0000000 |-1536 MB | fffffffffeffffff | 1520 MB | module mapping space ffffffffff000000 | -16 MB | | | FIXADDR_START | ~-11 MB | ffffffffff5fffff | ~0.5 MB | kernel-internal fixmap range, variable size and offsetQuoted 7 lines
That thing which starts at -512 GB above is the last PGD on the pagetable. In it, between -4G and -68G there are 64G which are the EFI region mapping space for runtime services.
Frankly I'm not sure what this thing is testing because the EFI VA range is hardcoded and I can't imagine it being somewhere else *except* in the last PGD.
Quoted 9 lines
It's just so that someone doesn't just change the #define's for EFI_VA_END/START and think that it will work, I guess.
Another reasonable option, for example, would be to reserve an entire PGD entry, allowing everything but the PGD level to be shared, and adding the EFI PGD to the
pgd_listand getting rid ofefi_sync_low_kernel_mappings()altogether. There aren't that many PGD entries still unused though, so this is probably not worth it.
Quoted 5 lines
The churn doesn't seem to be worth it, tbh.
So could we get rid of the complexity here, and only
build_bug()on the start address of the EFI region being outside the topmost p4d? That should make the PGD test redundant as well.
Was there ever a resolution to this conversation or a patch sent? I am still seeing the build failure that Arnd initially sent the patch for. x86_64 all{mod,yes}config with clang are going to ship broken in 5.11.
Cheers,
Nathan
Ard Biesheuvel
maintainerQuoted 70 lines · Nathan Chancellor
Quoted 68 lines · Ard Biesheuvel
Quoted 55 lines · Arvind Sankar
Quoted 13 lines · Borislav Petkov
Quoted 13 lines
Quoted 12 lines
Quoted 8 lines
As a matter of fact, it seems like the four assertions could be combined into:
BUILD_BUG_ON((EFI_VA_END & P4D_MASK) != (MODULES_END & P4D_MASK)); BUILD_BUG_ON((EFI_VA_START & P4D_MASK) != (EFI_VA_END & P4D_MASK));instead of separately asserting they're the same PGD entry and the same P4D entry.
Thanks.
I actually don't quite get the
MODULES_ENDcheck -- Ard, do you know what that's for?Quoted 6 lines
Quoted 2 lines
Maybe Boris remembers? He wrote the original code for the 'new' EFI page table layout.
That was added by Kirill for 5-level pgtables:
e981316f5604("x86/efi: Add 5-level paging support")That just duplicates the existing
pgd_index()check for thep4d_index()as well. It looks like the original commit addingefi_sync_low_kernel_mappings()used to copy upto the PGD entry includingMODULES_END:
d2f7cbe7b26a7("x86/efi: Runtime services virtual mapping")
and then Matt changed that when creatingefi_mm:
67a9108ed4313("x86/efi: Build our own page table structures")
to useEFI_VA_ENDinstead but have a check thatEFI_VA_ENDis in the same entry asMODULES_END.AFAICT,
MODULES_ENDis only relevant as being something that happens to be in the top 512GiB, and -1ul would be clearer.Quoted 12 lines
Documentation/x86/x86_64/mm.rstshould explain the pagetable layout:ffffff8000000000 | -512 GB | ffffffeeffffffff | 444 GB | ... unused hole ffffffef00000000 | -68 GB | fffffffeffffffff | 64 GB | EFI region mapping space ffffffff00000000 | -4 GB | ffffffff7fffffff | 2 GB | ... unused hole ffffffff80000000 | -2 GB | ffffffff9fffffff | 512 MB | kernel text mapping, mapped to physical address 0 ffffffff80000000 |-2048 MB | | | ffffffffa0000000 |-1536 MB | fffffffffeffffff | 1520 MB | module mapping space ffffffffff000000 | -16 MB | | | FIXADDR_START | ~-11 MB | ffffffffff5fffff | ~0.5 MB | kernel-internal fixmap range, variable size and offsetQuoted 7 lines
That thing which starts at -512 GB above is the last PGD on the pagetable. In it, between -4G and -68G there are 64G which are the EFI region mapping space for runtime services.
Frankly I'm not sure what this thing is testing because the EFI VA range is hardcoded and I can't imagine it being somewhere else *except* in the last PGD.
Quoted 9 lines
It's just so that someone doesn't just change the #define's for EFI_VA_END/START and think that it will work, I guess.
Another reasonable option, for example, would be to reserve an entire PGD entry, allowing everything but the PGD level to be shared, and adding the EFI PGD to the
pgd_listand getting rid ofefi_sync_low_kernel_mappings()altogether. There aren't that many PGD entries still unused though, so this is probably not worth it.The churn doesn't seem to be worth it, tbh.
Quoted 8 lines
Quoted 3 lines
So could we get rid of the complexity here, and only
build_bug()on the start address of the EFI region being outside the topmost p4d? That should make the PGD test redundant as well.Was there ever a resolution to this conversation or a patch sent? I am still seeing the build failure that Arnd initially sent the patch for.
x86_64all{mod,yes}config with clang are going to ship broken in 5.11.
I think we have agreement on the approach but it is unclear who is going to write the patch.
Borislav Petkov
maintainerQuoted 2 lines · Ard Biesheuvel
I think we have agreement on the approach but it is unclear who is going to write the patch.
How's that below?
And frankly, I'd even vote for removing those assertions altogether. If somehow the EFI pgd lands somewhere else, the kernel will crash'n'burn spectacularly and quickly so it's not like we won't catch it...
---
arch/x86/include/asm/pgtable_64_types.h +2 -2
@@ -156,8 +156,8 @@ extern unsigned int ptrs_per_p4d;#define CPU_ENTRY_AREA_PGD _AC(-4, UL)#define CPU_ENTRY_AREA_BASE (CPU_ENTRY_AREA_PGD << P4D_SHIFT) -#define EFI_VA_START ( -4 * (_AC(1, UL) << 30)) -#define EFI_VA_END (-68 * (_AC(1, UL) << 30)) +#define EFI_VA_START ( -4UL * (_AC(1, UL) << 30)) +#define EFI_VA_END (-68UL * (_AC(1, UL) << 30))#define EARLY_DYNAMIC_PAGE_TABLES 64
arch/x86/platform/efi/efi_64.c +2 -5
@@ -123,9 +123,7 @@ void efi_sync_low_kernel_mappings(void)* only span a single PGD entry and that the entry also maps * other important kernel regions. */ -MAYBE_BUILD_BUG_ON (pgd_index (EFI_VA_END ) !=pgd_index (MODULES_END )); -MAYBE_BUILD_BUG_ON ((EFI_VA_START &PGDIR_MASK ) != - (EFI_VA_END &PGDIR_MASK )); +MAYBE_BUILD_BUG_ON ((EFI_VA_START &PGDIR_MASK ) !=PGDIR_MASK ); pgd_efi = efi_pgd +pgd_index (PAGE_OFFSET ); pgd_k =pgd_offset_k (PAGE_OFFSET );
@@ -137,8 +135,7 @@ void efi_sync_low_kernel_mappings(void)* As with PGDs, we share all P4D entries apart from the one entry * that covers the EFI runtime mapping space. */ -BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); -BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); +BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) !=P4D_MASK ); pgd_efi = efi_pgd +pgd_index (EFI_VA_END ); pgd_k =pgd_offset_k (EFI_VA_END );
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteArd Biesheuvel
maintainerQuoted 11 lines · Borislav Petkov
Quoted 2 lines · Ard Biesheuvel
I think we have agreement on the approach but it is unclear who is going to write the patch.
How's that below?
And frankly, I'd even vote for removing those assertions altogether. If somehow the EFI pgd lands somewhere else, the kernel will crash'n'burn spectacularly and quickly so it's not like we won't catch it...
+1 for just removing them
Quoted 48 lines
---
arch/x86/include/asm/pgtable_64_types.h +2
-2@@ -156,8 +156,8 @@ extern unsigned int ptrs_per_p4d;#define CPU_ENTRY_AREA_PGD _AC(-4, UL)#define CPU_ENTRY_AREA_BASE (CPU_ENTRY_AREA_PGD << P4D_SHIFT) -#define EFI_VA_START ( -4 * (_AC(1, UL) << 30)) -#define EFI_VA_END (-68 * (_AC(1, UL) << 30)) +#define EFI_VA_START ( -4UL * (_AC(1, UL) << 30)) +#define EFI_VA_END (-68UL * (_AC(1, UL) << 30))#define EARLY_DYNAMIC_PAGE_TABLES 64arch/x86/platform/efi/efi_64.c +2
-6@@ -123,9 +123,7 @@ void efi_sync_low_kernel_mappings(void)* only span a single PGD entry and that the entry also maps * other important kernel regions. */ -MAYBE_BUILD_BUG_ON (pgd_index (EFI_VA_END ) !=pgd_index (MODULES_END )); -MAYBE_BUILD_BUG_ON ((EFI_VA_START &PGDIR_MASK ) != - (EFI_VA_END &PGDIR_MASK )); +MAYBE_BUILD_BUG_ON ((EFI_VA_START &PGDIR_MASK ) !=PGDIR_MASK ); pgd_efi = efi_pgd +pgd_index (PAGE_OFFSET ); pgd_k =pgd_offset_k (PAGE_OFFSET );@@ -137,8 +135,7 @@ void efi_sync_low_kernel_mappings(void)* As with PGDs, we share all P4D entries apart from the one entry * that covers the EFI runtime mapping space. */ -BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); -BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); +BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) !=P4D_MASK ); pgd_efi = efi_pgd +pgd_index (EFI_VA_END ); pgd_k =pgd_offset_k (EFI_VA_END ); --Regards/Gruss,
Boris.
Nathan Chancellor
Tested-byQuoted 11 lines · Borislav Petkov
Quoted 2 lines · Ard Biesheuvel
I think we have agreement on the approach but it is unclear who is going to write the patch.
How's that below?
And frankly, I'd even vote for removing those assertions altogether. If somehow the EFI pgd lands somewhere else, the kernel will crash'n'burn spectacularly and quickly so it's not like we won't catch it...
---
This resolves the issue initially reported in this thread. Obviously removing the assertions will do that as well. Feel free to carry forward
on a patch if you send it out.
Quoted 47 lines
arch/x86/include/asm/pgtable_64_types.h +2
-2@@ -156,8 +156,8 @@ extern unsigned int ptrs_per_p4d;#define CPU_ENTRY_AREA_PGD _AC(-4, UL)#define CPU_ENTRY_AREA_BASE (CPU_ENTRY_AREA_PGD << P4D_SHIFT) -#define EFI_VA_START ( -4 * (_AC(1, UL) << 30)) -#define EFI_VA_END (-68 * (_AC(1, UL) << 30)) +#define EFI_VA_START ( -4UL * (_AC(1, UL) << 30)) +#define EFI_VA_END (-68UL * (_AC(1, UL) << 30))#define EARLY_DYNAMIC_PAGE_TABLES 64arch/x86/platform/efi/efi_64.c +2
-5@@ -123,9 +123,7 @@ void efi_sync_low_kernel_mappings(void)* only span a single PGD entry and that the entry also maps * other important kernel regions. */ -MAYBE_BUILD_BUG_ON (pgd_index (EFI_VA_END ) !=pgd_index (MODULES_END )); -MAYBE_BUILD_BUG_ON ((EFI_VA_START &PGDIR_MASK ) != - (EFI_VA_END &PGDIR_MASK )); +MAYBE_BUILD_BUG_ON ((EFI_VA_START &PGDIR_MASK ) !=PGDIR_MASK ); pgd_efi = efi_pgd +pgd_index (PAGE_OFFSET ); pgd_k =pgd_offset_k (PAGE_OFFSET );@@ -137,8 +135,7 @@ void efi_sync_low_kernel_mappings(void)* As with PGDs, we share all P4D entries apart from the one entry * that covers the EFI runtime mapping space. */ -BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); -BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); +BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) !=P4D_MASK ); pgd_efi = efi_pgd +pgd_index (EFI_VA_END ); pgd_k =pgd_offset_k (EFI_VA_END );Signature · 5 lines
-- Regards/Gruss, Boris. https://people.kernel.org/tglx/notes-about-netiquette
Quoted 9 lines · Borislav Petkov
Quoted 2 lines · Ard Biesheuvel
I think we have agreement on the approach but it is unclear who is going to write the patch.
How's that below?
And frankly, I'd even vote for removing those assertions altogether. If somehow the EFI pgd lands somewhere else, the kernel will crash'n'burn spectacularly and quickly so it's not like we won't catch it...
Removing altogether should be fine, but see below if we don't.
Quoted 7 lines
---
arch/x86/include/asm/pgtable_64_types.h +0
-0
arch/x86/include/asm/pgtable_64_types.h extern unsigned int ptrs_per_p4d; · line 160
Quoted 7 lines
arch/x86/include/asm/pgtable_64_types.h +2
-2@@ -156,8 +156,8 @@ extern unsigned int ptrs_per_p4d;#define CPU_ENTRY_AREA_PGD _AC(-4, UL)#define CPU_ENTRY_AREA_BASE (CPU_ENTRY_AREA_PGD << P4D_SHIFT) -#define EFI_VA_START ( -4 * (_AC(1, UL) << 30)) -#define EFI_VA_END (-68 * (_AC(1, UL) << 30)) +#define EFI_VA_START ( -4UL * (_AC(1, UL) << 30)) +#define EFI_VA_END (-68UL * (_AC(1, UL) << 30))
This doesn't have any effect right? And the reason for the _AC() stuff in there is to allow the #define to be used in assembler -- this particular one isn't, but it makes no sense to use the UL suffix as well as _AC() in the same macro.
Quoted 8 lines
arch/x86/include/asm/pgtable_64_types.h +0
-0#define EARLY_DYNAMIC_PAGE_TABLES 64arch/x86/platform/efi/efi_64.c +0
-0
arch/x86/platform/efi/efi_64.c void efi_sync_low_kernel_mappings(void) · line 126
Quoted 7 lines
arch/x86/platform/efi/efi_64.c +1
-3@@ -123,9 +123,7 @@ void efi_sync_low_kernel_mappings(void)* only span a single PGD entry and that the entry also maps * other important kernel regions. */ -MAYBE_BUILD_BUG_ON (pgd_index (EFI_VA_END ) !=pgd_index (MODULES_END )); -MAYBE_BUILD_BUG_ON ((EFI_VA_START &PGDIR_MASK ) != - (EFI_VA_END &PGDIR_MASK )); +MAYBE_BUILD_BUG_ON ((EFI_VA_START &PGDIR_MASK ) !=PGDIR_MASK );
This check is superfluous. Just do the P4D one.
arch/x86/platform/efi/efi_64.c void efi_sync_low_kernel_mappings(void) · line 138
Quoted 10 lines
arch/x86/platform/efi/efi_64.c +1
-2pgd_efi = efi_pgd +pgd_index (PAGE_OFFSET ); pgd_k =pgd_offset_k (PAGE_OFFSET );@@ -137,8 +135,7 @@ void efi_sync_low_kernel_mappings(void)* As with PGDs, we share all P4D entries apart from the one entry * that covers the EFI runtime mapping space. */ -BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); -BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); +BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) !=P4D_MASK );
This should check EFI_VA_END instead of EFI_VA_START, and maybe throw in a BUG_ON if EFI_VA_END >= EFI_VA_START.
Quoted 10 lines
arch/x86/platform/efi/efi_64.c +0
-0pgd_efi = efi_pgd +pgd_index (EFI_VA_END ); pgd_k =pgd_offset_k (EFI_VA_END );Signature · 5 lines
-- Regards/Gruss, Boris. https://people.kernel.org/tglx/notes-about-netiquette
Borislav Petkov
maintainerQuoted 2 lines · Arvind Sankar
This should check
EFI_VA_ENDinstead ofEFI_VA_START, and maybe throw in aBUG_ONifEFI_VA_END>=EFI_VA_START.
No need:
if (efi_va < EFI_VA_END) {
pr_warn(FW_WARN "VA address range overflow!\n");
return;
}We already check we're not going over at map time. And our runtime services range is hardcoded. And we're switching to that PGD on each runtime services call.
So I don't see the point for keeping any of the assertions.
Unless you have other valid arguments for keeping them...
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteQuoted 11 lines · Borislav Petkov
Quoted 2 lines · Arvind Sankar
This should check
EFI_VA_ENDinstead ofEFI_VA_START, and maybe throw in aBUG_ONifEFI_VA_END>=EFI_VA_START.No need:
if (efi_va < EFI_VA_END) { pr_warn(FW_WARN "VA address range overflow!\n"); return; }
Quoted 8 lines
We already check we're not going over at map time. And our runtime services range is hardcoded. And we're switching to that PGD on each runtime services call.
So I don't see the point for keeping any of the assertions.
Unless you have other valid arguments for keeping them...
No, I don't have any objections to removing them altogether. All the comments other than the one about changing the #define's only apply if it's decided to keep them.
Borislav Petkov
maintainerPATCH 1/1 x86/efi: Remove EFI PGD build time checks
From: Borislav Petkov <[email protected]>
With CONFIG_X86_5LEVEL, CONFIG_UBSAN and CONFIG_UBSAN_UNSIGNED_OVERFLOW enabled, clang fails the build with
x86_64-linux-ld: arch/x86/platform/efi/efi_64.o: in function `efi_sync_low_kernel_mappings':
efi_64.c:(.text+0x22c): undefined reference to `__compiletime_assert_354'
which happens due to -fsanitize=unsigned-integer-overflow being enabled:
-fsanitize=unsigned-integer-overflow: Unsigned integer overflow, where the result of an unsigned integer computation cannot be represented in its type. Unlike signed integer overflow, this is not undefined behavior, but it is often unintentional. This sanitizer does not check for lossy implicit conversions performed before such a computation (see -fsanitize=implicit-conversion).
and that fires when the (intentional) EFI_VA_START/END defines overflow an unsigned long, leading to the assertion expressions not getting optimized away (on GCC they do)...
However, those checks are superfluous: the runtime services mapping code already makes sure the ranges don't overshoot EFI_VA_END as the EFI mapping range is hardcoded. On each runtime services call, it is switched to the EFI-specific PGD and even if mappings manage to escape that last PGD, this won't remain unnoticed for long.
So rip them out.
See https://github.com/ClangBuiltLinux/linux/issues/256 for more info.
--- arch/x86/platform/efi/efi_64.c | 19 ------------------- 1 file changed, 19 deletions(-)
arch/x86/platform/efi/efi_64.c +0 -19
@@ -115,31 +115,12 @@ void efi_sync_low_kernel_mappings(void)pud_t *pud_k, *pud_efi;pgd_t *efi_pgd = efi_mm.pgd; -/* -* We can share all PGD entries apart from the one entry that -* covers the EFI runtime mapping space. -* -* Make sure the EFI runtime region mappings are guaranteed to -* only span a single PGD entry and that the entry also maps -* other important kernel regions. -*/ -MAYBE_BUILD_BUG_ON (pgd_index (EFI_VA_END ) !=pgd_index (MODULES_END )); -MAYBE_BUILD_BUG_ON ((EFI_VA_START &PGDIR_MASK ) != - (EFI_VA_END &PGDIR_MASK )); - pgd_efi = efi_pgd +pgd_index (PAGE_OFFSET ); pgd_k =pgd_offset_k (PAGE_OFFSET ); num_entries =pgd_index (EFI_VA_END ) -pgd_index (PAGE_OFFSET );memcpy (pgd_efi, pgd_k,sizeof (pgd_t) * num_entries); -/* -* As with PGDs, we share all P4D entries apart from the one entry -* that covers the EFI runtime mapping space. -*/ -BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); -BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); - pgd_efi = efi_pgd +pgd_index (EFI_VA_END ); pgd_k =pgd_offset_k (EFI_VA_END ); p4d_efi =p4d_offset (pgd_efi,0 );
Signature · 7 lines
2.29.2
--
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteRe: [PATCH] x86/efi: Remove EFI PGD build time checks
Quoted 28 lines · Borislav Petkov
From: Borislav Petkov <[email protected]>
With
CONFIG_X86_5LEVEL,CONFIG_UBSANandCONFIG_UBSAN_UNSIGNED_OVERFLOWenabled, clang fails the build withx86_64-linux-ld:
arch/x86/platform/efi/efi_64.o: in function `efi_sync_low_kernel_mappings':
efi_64.c:(.text+0x22c): undefined reference to `__compiletime_assert_354'which happens due to -fsanitize=unsigned-integer-overflow being enabled:
-fsanitize=unsigned-integer-overflow: Unsigned integer overflow, where the result of an unsigned integer computation cannot be represented in its type. Unlike signed integer overflow, this is not undefined behavior, but it is often unintentional. This sanitizer does not check for lossy implicit conversions performed before such a computation (see -fsanitize=implicit-conversion).
and that fires when the (intentional) EFI_VA_START/END defines overflow an unsigned long, leading to the assertion expressions not getting optimized away (on GCC they do)...
However, those checks are superfluous: the runtime services mapping code already makes sure the ranges don't overshoot
EFI_VA_ENDas the EFI mapping range is hardcoded. On each runtime services call, it is switched to the EFI-specific PGD and even if mappings manage to escape that last PGD, this won't remain unnoticed for long.
Quoted 7 lines
So rip them out.
See https://github.com/ClangBuiltLinux/linux/issues/256 for more info.
Reported-by Arnd Bergmann <[email protected]>Signed-off-by Borislav Petkov <[email protected]>
Quoted 48 lines
--- arch/x86/platform/efi/efi_64.c | 19 ------------------- 1 file changed, 19 deletions(-)arch/x86/platform/efi/efi_64.c +0
-20@@ -115,31 +115,12 @@ void efi_sync_low_kernel_mappings(void)pud_t *pud_k, *pud_efi;pgd_t *efi_pgd = efi_mm.pgd; -/* -* We can share all PGD entries apart from the one entry that -* covers the EFI runtime mapping space. -* -* Make sure the EFI runtime region mappings are guaranteed to -* only span a single PGD entry and that the entry also maps -* other important kernel regions. -*/ -MAYBE_BUILD_BUG_ON (pgd_index (EFI_VA_END ) !=pgd_index (MODULES_END )); -MAYBE_BUILD_BUG_ON ((EFI_VA_START &PGDIR_MASK ) != - (EFI_VA_END &PGDIR_MASK )); - pgd_efi = efi_pgd +pgd_index (PAGE_OFFSET ); pgd_k =pgd_offset_k (PAGE_OFFSET ); num_entries =pgd_index (EFI_VA_END ) -pgd_index (PAGE_OFFSET );memcpy (pgd_efi, pgd_k,sizeof (pgd_t) * num_entries); -/* -* As with PGDs, we share all P4D entries apart from the one entry -* that covers the EFI runtime mapping space. -*/ -BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); -BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); - pgd_efi = efi_pgd +pgd_index (EFI_VA_END ); pgd_k =pgd_offset_k (EFI_VA_END ); p4d_efi =p4d_offset (pgd_efi,0 ); --2.29.2
--
Regards/Gruss,
Boris.
Re: [PATCH] x86/efi: Remove EFI PGD build time checks
In reply to Borislav PetkovQuoted 28 lines · Borislav Petkov
From: Borislav Petkov <[email protected]>
With
CONFIG_X86_5LEVEL,CONFIG_UBSANandCONFIG_UBSAN_UNSIGNED_OVERFLOWenabled, clang fails the build withx86_64-linux-ld:
arch/x86/platform/efi/efi_64.o: in function `efi_sync_low_kernel_mappings':
efi_64.c:(.text+0x22c): undefined reference to `__compiletime_assert_354'which happens due to -fsanitize=unsigned-integer-overflow being enabled:
-fsanitize=unsigned-integer-overflow: Unsigned integer overflow, where the result of an unsigned integer computation cannot be represented in its type. Unlike signed integer overflow, this is not undefined behavior, but it is often unintentional. This sanitizer does not check for lossy implicit conversions performed before such a computation (see -fsanitize=implicit-conversion).
and that fires when the (intentional) EFI_VA_START/END defines overflow an unsigned long, leading to the assertion expressions not getting optimized away (on GCC they do)...
However, those checks are superfluous: the runtime services mapping code already makes sure the ranges don't overshoot
EFI_VA_ENDas the EFI mapping range is hardcoded. On each runtime services call, it is switched to the EFI-specific PGD and even if mappings manage to escape that last PGD, this won't remain unnoticed for long.
Quoted 7 lines
So rip them out.
See https://github.com/ClangBuiltLinux/linux/issues/256 for more info.
Reported-by Arnd Bergmann <[email protected]>Signed-off-by Borislav Petkov <[email protected]>
Thanks, this fixes the failed assertion for me.
(https://lore.kernel.org/lkml/[email protected]/
is needed to finish a build of that configuration; going to chase that next)
(consider applying Arvind's+Ard's suggested by tag)
Quoted 48 lines
--- arch/x86/platform/efi/efi_64.c | 19 ------------------- 1 file changed, 19 deletions(-)arch/x86/platform/efi/efi_64.c +0
-20@@ -115,31 +115,12 @@ void efi_sync_low_kernel_mappings(void)pud_t *pud_k, *pud_efi;pgd_t *efi_pgd = efi_mm.pgd; -/* -* We can share all PGD entries apart from the one entry that -* covers the EFI runtime mapping space. -* -* Make sure the EFI runtime region mappings are guaranteed to -* only span a single PGD entry and that the entry also maps -* other important kernel regions. -*/ -MAYBE_BUILD_BUG_ON (pgd_index (EFI_VA_END ) !=pgd_index (MODULES_END )); -MAYBE_BUILD_BUG_ON ((EFI_VA_START &PGDIR_MASK ) != - (EFI_VA_END &PGDIR_MASK )); - pgd_efi = efi_pgd +pgd_index (PAGE_OFFSET ); pgd_k =pgd_offset_k (PAGE_OFFSET ); num_entries =pgd_index (EFI_VA_END ) -pgd_index (PAGE_OFFSET );memcpy (pgd_efi, pgd_k,sizeof (pgd_t) * num_entries); -/* -* As with PGDs, we share all P4D entries apart from the one entry -* that covers the EFI runtime mapping space. -*/ -BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); -BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); - pgd_efi = efi_pgd +pgd_index (EFI_VA_END ); pgd_k =pgd_offset_k (EFI_VA_END ); p4d_efi =p4d_offset (pgd_efi,0 ); --2.29.2
--
Regards/Gruss,
Boris.
Signature · 2 lines
Thanks, ~Nick Desaulniers
Re: [PATCH] x86/efi: Remove EFI PGD build time checks
In reply to Borislav PetkovQuoted 27 lines · Borislav Petkov
From: Borislav Petkov <[email protected]>
With
CONFIG_X86_5LEVEL,CONFIG_UBSANandCONFIG_UBSAN_UNSIGNED_OVERFLOWenabled, clang fails the build withx86_64-linux-ld:
arch/x86/platform/efi/efi_64.o: in function `efi_sync_low_kernel_mappings':
efi_64.c:(.text+0x22c): undefined reference to `__compiletime_assert_354'which happens due to -fsanitize=unsigned-integer-overflow being enabled:
-fsanitize=unsigned-integer-overflow: Unsigned integer overflow, where the result of an unsigned integer computation cannot be represented in its type. Unlike signed integer overflow, this is not undefined behavior, but it is often unintentional. This sanitizer does not check for lossy implicit conversions performed before such a computation (see -fsanitize=implicit-conversion).
and that fires when the (intentional) EFI_VA_START/END defines overflow an unsigned long, leading to the assertion expressions not getting optimized away (on GCC they do)...
However, those checks are superfluous: the runtime services mapping code already makes sure the ranges don't overshoot
EFI_VA_ENDas the EFI mapping range is hardcoded. On each runtime services call, it is switched to the EFI-specific PGD and even if mappings manage to escape that last PGD, this won't remain unnoticed for long.
Quoted 7 lines
So rip them out.
See https://github.com/ClangBuiltLinux/linux/issues/256 for more info.
Reported-by Arnd Bergmann <[email protected]>Signed-off-by Borislav Petkov <[email protected]>
Quoted 48 lines
--- arch/x86/platform/efi/efi_64.c | 19 ------------------- 1 file changed, 19 deletions(-)arch/x86/platform/efi/efi_64.c +0
-19@@ -115,31 +115,12 @@ void efi_sync_low_kernel_mappings(void)pud_t *pud_k, *pud_efi;pgd_t *efi_pgd = efi_mm.pgd; -/* -* We can share all PGD entries apart from the one entry that -* covers the EFI runtime mapping space. -* -* Make sure the EFI runtime region mappings are guaranteed to -* only span a single PGD entry and that the entry also maps -* other important kernel regions. -*/ -MAYBE_BUILD_BUG_ON (pgd_index (EFI_VA_END ) !=pgd_index (MODULES_END )); -MAYBE_BUILD_BUG_ON ((EFI_VA_START &PGDIR_MASK ) != - (EFI_VA_END &PGDIR_MASK )); - pgd_efi = efi_pgd +pgd_index (PAGE_OFFSET ); pgd_k =pgd_offset_k (PAGE_OFFSET ); num_entries =pgd_index (EFI_VA_END ) -pgd_index (PAGE_OFFSET );memcpy (pgd_efi, pgd_k,sizeof (pgd_t) * num_entries); -/* -* As with PGDs, we share all P4D entries apart from the one entry -* that covers the EFI runtime mapping space. -*/ -BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); -BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); - pgd_efi = efi_pgd +pgd_index (EFI_VA_END ); pgd_k =pgd_offset_k (EFI_VA_END ); p4d_efi =p4d_offset (pgd_efi,0 );Signature · 7 lines
2.29.2 -- Regards/Gruss, Boris. https://people.kernel.org/tglx/notes-about-netiquette
Borislav Petkov
maintainerQuoted 1 line · Nathan Chancellor
x86_64all{mod,yes}config with clang are going to ship broken in 5.11.
Dunno, it is still broken here even with those build assertions removed. And it ain't even an all{mod,yes}config - just my machine's config with
CONFIG_ARCH_HAS_UBSAN_SANITIZE_ALL=yCONFIG_UBSAN=y
# CONFIG_UBSAN_TRAP is not setCONFIG_CC_HAS_UBSAN_BOUNDS=yCONFIG_CC_HAS_UBSAN_ARRAY_BOUNDS=yCONFIG_UBSAN_BOUNDS=yCONFIG_UBSAN_ARRAY_BOUNDS=yCONFIG_UBSAN_SHIFT=yCONFIG_UBSAN_DIV_ZERO=yCONFIG_UBSAN_SIGNED_OVERFLOW=yCONFIG_UBSAN_UNSIGNED_OVERFLOW=yCONFIG_UBSAN_OBJECT_SIZE=yCONFIG_UBSAN_BOOL=yCONFIG_UBSAN_ENUM=yCONFIG_UBSAN_ALIGNMENT=yCONFIG_UBSAN_SANITIZE_ALL=y
# CONFIG_TEST_UBSAN is not set
and clang-10:
lib/strncpy_from_user.o: warning: objtool: strncpy_from_user()+0x253: call to __ubsan_handle_add_overflow() with UACCESS enabled lib/strnlen_user.o: warning: objtool: strnlen_user()+0x244: call to __ubsan_handle_add_overflow() with UACCESS enabled
ld: init/main.o: in function `kmalloc':/home/boris/kernel/linux/./include/linux/slab.h:557: undefined reference to `__ubsan_handle_alignment_assumption'
ld: init/initramfs.o: in function `kmalloc':/home/boris/kernel/linux/./include/linux/slab.h:552: undefined reference to `__ubsan_handle_alignment_assumption'
ld: init/initramfs.o: in function `kmalloc_large':/home/boris/kernel/linux/./include/linux/slab.h:481: undefined reference to `__ubsan_handle_alignment_assumption'
ld: init/initramfs.o: in function `kmalloc':/home/boris/kernel/linux/./include/linux/slab.h:552: undefined reference to `__ubsan_handle_alignment_assumption'
ld: /home/boris/kernel/linux/./include/linux/slab.h:552: undefined reference to `__ubsan_handle_alignment_assumption'
ld: init/initramfs.o:/home/boris/kernel/linux/./include/linux/slab.h:552: more undefined references to `__ubsan_handle_alignment_assumption' follow
ld: mm/mremap.o: in function `get_extent':/home/boris/kernel/linux/mm/mremap.c:355: undefined reference to `__compiletime_assert_327'
ld: mm/rmap.o: in function `anon_vma_chain_alloc':/home/boris/kernel/linux/mm/rmap.c:136: undefined reference to `__ubsan_handle_alignment_assumption'
ld: mm/rmap.o: in function `anon_vma_alloc':/home/boris/kernel/linux/mm/rmap.c:89: undefined reference to `__ubsan_handle_alignment_assumption'
ld: mm/rmap.o: in function `anon_vma_chain_alloc':/home/boris/kernel/linux/mm/rmap.c:136: undefined reference to `__ubsan_handle_alignment_assumption'
ld: /home/boris/kernel/linux/mm/rmap.c:136: undefined reference to `__ubsan_handle_alignment_assumption'
ld: /home/boris/kernel/linux/mm/rmap.c:136: undefined reference to `__ubsan_handle_alignment_assumption'
ld: mm/vmalloc.o:/home/boris/kernel/linux/mm/vmalloc.c:1213: more undefined references to `__ubsan_handle_alignment_assumption' follow
make: *** [Makefile:1164: vmlinux] Error 1
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteQuoted 27 lines · Borislav Petkov
Quoted 1 line · Nathan Chancellor
x86_64all{mod,yes}config with clang are going to ship broken in 5.11.Dunno, it is still broken here even with those build assertions removed. And it ain't even an all{mod,yes}config - just my machine's config with
CONFIG_ARCH_HAS_UBSAN_SANITIZE_ALL=yCONFIG_UBSAN=y
#CONFIG_UBSAN_TRAPis not setCONFIG_CC_HAS_UBSAN_BOUNDS=yCONFIG_CC_HAS_UBSAN_ARRAY_BOUNDS=yCONFIG_UBSAN_BOUNDS=yCONFIG_UBSAN_ARRAY_BOUNDS=yCONFIG_UBSAN_SHIFT=yCONFIG_UBSAN_DIV_ZERO=yCONFIG_UBSAN_SIGNED_OVERFLOW=yCONFIG_UBSAN_UNSIGNED_OVERFLOW=yCONFIG_UBSAN_OBJECT_SIZE=yCONFIG_UBSAN_BOOL=yCONFIG_UBSAN_ENUM=yCONFIG_UBSAN_ALIGNMENT=yCONFIG_UBSAN_SANITIZE_ALL=y
#CONFIG_TEST_UBSANis not setand clang-10:
Quoted 14 lines
lib/strncpy_from_user.o: warning: objtool: strncpy_from_user()+0x253: call to__ubsan_handle_add_overflow()with UACCESS enabledlib/strnlen_user.o: warning: objtool: strnlen_user()+0x244: call to__ubsan_handle_add_overflow()with UACCESS enabled
ld:init/main.o: in function `kmalloc':/home/boris/kernel/linux/./include/linux/slab.h:557: undefined reference to `__ubsan_handle_alignment_assumption'
ld:init/initramfs.o: in function `kmalloc':/home/boris/kernel/linux/./include/linux/slab.h:552: undefined reference to `__ubsan_handle_alignment_assumption'
ld:init/initramfs.o: in function `kmalloc_large':/home/boris/kernel/linux/./include/linux/slab.h:481: undefined reference to `__ubsan_handle_alignment_assumption'
ld:init/initramfs.o: in function `kmalloc':/home/boris/kernel/linux/./include/linux/slab.h:552: undefined reference to `__ubsan_handle_alignment_assumption'
ld:/home/boris/kernel/linux/./include/linux/slab.h:552: undefined reference to `__ubsan_handle_alignment_assumption'
ld: init/initramfs.o:/home/boris/kernel/linux/./include/linux/slab.h:552: more undefined references to `__ubsan_handle_alignment_assumption' follow
ld:mm/mremap.o: in function `get_extent':/home/boris/kernel/linux/mm/mremap.c:355: undefined reference to `__compiletime_assert_327'
^ this one is https://lore.kernel.org/lkml/[email protected]/.
Trying to get the last of these tracked down. I think there were some changes to UBSAN configs that weren't tested with clang before merged.
Quoted 16 lines
ld:
mm/rmap.o: in function `anon_vma_chain_alloc':/home/boris/kernel/linux/mm/rmap.c:136: undefined reference to `__ubsan_handle_alignment_assumption'
ld:mm/rmap.o: in function `anon_vma_alloc':/home/boris/kernel/linux/mm/rmap.c:89: undefined reference to `__ubsan_handle_alignment_assumption'
ld:mm/rmap.o: in function `anon_vma_chain_alloc':/home/boris/kernel/linux/mm/rmap.c:136: undefined reference to `__ubsan_handle_alignment_assumption'
ld:/home/boris/kernel/linux/mm/rmap.c:136: undefined reference to `__ubsan_handle_alignment_assumption'
ld:/home/boris/kernel/linux/mm/rmap.c:136: undefined reference to `__ubsan_handle_alignment_assumption'
ld: mm/vmalloc.o:/home/boris/kernel/linux/mm/vmalloc.c:1213: more undefined references to `__ubsan_handle_alignment_assumption' follow
make: *** [Makefile:1164: vmlinux] Error 1--
Regards/Gruss,
Boris.
Signature · 2 lines
Thanks, ~Nick Desaulniers
Quoted 43 lines · Nick Desaulniers
Quoted 27 lines · Borislav Petkov
Quoted 1 line · Nathan Chancellor
x86_64all{mod,yes}config with clang are going to ship broken in 5.11.Dunno, it is still broken here even with those build assertions removed. And it ain't even an all{mod,yes}config - just my machine's config with
CONFIG_ARCH_HAS_UBSAN_SANITIZE_ALL=yCONFIG_UBSAN=y
#CONFIG_UBSAN_TRAPis not setCONFIG_CC_HAS_UBSAN_BOUNDS=yCONFIG_CC_HAS_UBSAN_ARRAY_BOUNDS=yCONFIG_UBSAN_BOUNDS=yCONFIG_UBSAN_ARRAY_BOUNDS=yCONFIG_UBSAN_SHIFT=yCONFIG_UBSAN_DIV_ZERO=yCONFIG_UBSAN_SIGNED_OVERFLOW=yCONFIG_UBSAN_UNSIGNED_OVERFLOW=yCONFIG_UBSAN_OBJECT_SIZE=yCONFIG_UBSAN_BOOL=yCONFIG_UBSAN_ENUM=yCONFIG_UBSAN_ALIGNMENT=yCONFIG_UBSAN_SANITIZE_ALL=y
#CONFIG_TEST_UBSANis not setand clang-10:
Quoted 14 lines
lib/strncpy_from_user.o: warning: objtool: strncpy_from_user()+0x253: call to__ubsan_handle_add_overflow()with UACCESS enabledlib/strnlen_user.o: warning: objtool: strnlen_user()+0x244: call to__ubsan_handle_add_overflow()with UACCESS enabled
ld:init/main.o: in function `kmalloc':/home/boris/kernel/linux/./include/linux/slab.h:557: undefined reference to `__ubsan_handle_alignment_assumption'
ld:init/initramfs.o: in function `kmalloc':/home/boris/kernel/linux/./include/linux/slab.h:552: undefined reference to `__ubsan_handle_alignment_assumption'
ld:init/initramfs.o: in function `kmalloc_large':/home/boris/kernel/linux/./include/linux/slab.h:481: undefined reference to `__ubsan_handle_alignment_assumption'
ld:init/initramfs.o: in function `kmalloc':/home/boris/kernel/linux/./include/linux/slab.h:552: undefined reference to `__ubsan_handle_alignment_assumption'
ld:/home/boris/kernel/linux/./include/linux/slab.h:552: undefined reference to `__ubsan_handle_alignment_assumption'
ld: init/initramfs.o:/home/boris/kernel/linux/./include/linux/slab.h:552: more undefined references to `__ubsan_handle_alignment_assumption' follow
ld:mm/mremap.o: in function `get_extent':/home/boris/kernel/linux/mm/mremap.c:355: undefined reference to `__compiletime_assert_327'
Quoted 3 lines
^ this one is https://lore.kernel.org/lkml/[email protected]/.
Trying to get the last of these tracked down. I think there were some changes to UBSAN configs that weren't tested with clang before merged.
The rest of these should be resolved by
https://lore.kernel.org/r/20210205023257.NJnJdyyZk%[email protected]/,
which is currently on its way to Linus.
Cheers,
Nathan
Quoted 2 lines · Arvind Sankar
AFAICT,
MODULES_ENDis only relevant as being something that happens to be in the top 512GiB, and -1ul would be clearer.
I think you are right. But -1UL is not very self-descriptive. :/
Signature · 1 line
Kirill A. Shutemov
Quoted 41 lines · Arvind Sankar
Quoted 35 lines · Arnd Bergmann
From: Arnd Bergmann <[email protected]>
When 5-level page tables are enabled, clang triggers a
BUILD_BUG_ON():x86_64-linux-ld:
arch/x86/platform/efi/efi_64.o: in function `efi_sync_low_kernel_mappings':
efi_64.c:(.text+0x22c): undefined reference to `__compiletime_assert_354'Use the same method as in commit
c65e774fb3f6("x86/mm: MakePGDIR_SHIFTandPTRS_PER_P4Dvariable") and change it toMAYBE_BUILD_BUG_ON(), so it only triggers for constant input.Signed-off-by Arnd Bergmann <[email protected]>--- arch/x86/platform/efi/efi_64.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)arch/x86/platform/efi/efi_64.c +2
-2@@ -137,8 +137,8 @@ void efi_sync_low_kernel_mappings(void)* As with PGDs, we share all P4D entries apart from the one entry * that covers the EFI runtime mapping space. */ -BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); -BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); +MAYBE_BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); +MAYBE_BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); pgd_efi = efi_pgd +pgd_index (EFI_VA_END ); pgd_k =pgd_offset_k (EFI_VA_END );Signature · 1 line
2.29.2I think this needs more explanation as to why clang is triggering this. The issue mentions clang not inline
p4d_index(), and I guess not performing inter-procedural analysis either?
Quoted 3 lines
For the second assertion there, everything is always constant AFAICT:
EFI_VA_START,EFI_VA_ENDandP4D_MASKare all constants regardless ofCONFIG_5LEVEL.
Back in the days BUILD_BUG_ON() false-triggered on GCC-4.8 as well.
Signature · 1 line
Kirill A. Shutemov
[tip: x86/urgent] x86/efi: Remove EFI PGD build time checks
In reply to Arnd BergmannThe following commit has been merged into the x86/urgent branch of tip:
The following commit has been merged into the x86/urgent branch of tip:
Commit-ID: 816ef8d7a2c4182e19bc06ab65751cb9e3951e94 Gitweb: https://git.kernel.org/tip/816ef8d7a2c4182e19bc06ab65751cb9e3951e94 Author: Borislav Petkov <[email protected]> AuthorDate: Fri, 05 Feb 2021 11:31:31 +01:00 Committer: Borislav Petkov <[email protected]> CommitterDate: Sat, 06 Feb 2021 13:54:14 +01:00
x86/efi: Remove EFI PGD build time checks
With CONFIG_X86_5LEVEL, CONFIG_UBSAN and CONFIG_UBSAN_UNSIGNED_OVERFLOW enabled, clang fails the build with
x86_64-linux-ld: arch/x86/platform/efi/efi_64.o: in function `efi_sync_low_kernel_mappings':
efi_64.c:(.text+0x22c): undefined reference to `__compiletime_assert_354'
which happens due to -fsanitize=unsigned-integer-overflow being enabled:
-fsanitize=unsigned-integer-overflow: Unsigned integer overflow, where the result of an unsigned integer computation cannot be represented in its type. Unlike signed integer overflow, this is not undefined behavior, but it is often unintentional. This sanitizer does not check for lossy implicit conversions performed before such a computation (see -fsanitize=implicit-conversion).
and that fires when the (intentional) EFI_VA_START/END defines overflow an unsigned long, leading to the assertion expressions not getting optimized away (on GCC they do)...
However, those checks are superfluous: the runtime services mapping code already makes sure the ranges don't overshoot EFI_VA_END as the EFI mapping range is hardcoded. On each runtime services call, it is switched to the EFI-specific PGD and even if mappings manage to escape that last PGD, this won't remain unnoticed for long.
So rip them out.
See https://github.com/ClangBuiltLinux/linux/issues/256 for more info.
--- arch/x86/platform/efi/efi_64.c | 19 ------------------- 1 file changed, 19 deletions(-)
arch/x86/platform/efi/efi_64.c +0 -19
@@ -115,31 +115,12 @@ void efi_sync_low_kernel_mappings(void)pud_t *pud_k, *pud_efi;pgd_t *efi_pgd = efi_mm.pgd; -/* -* We can share all PGD entries apart from the one entry that -* covers the EFI runtime mapping space. -* -* Make sure the EFI runtime region mappings are guaranteed to -* only span a single PGD entry and that the entry also maps -* other important kernel regions. -*/ -MAYBE_BUILD_BUG_ON (pgd_index (EFI_VA_END ) !=pgd_index (MODULES_END )); -MAYBE_BUILD_BUG_ON ((EFI_VA_START &PGDIR_MASK ) != - (EFI_VA_END &PGDIR_MASK )); - pgd_efi = efi_pgd +pgd_index (PAGE_OFFSET ); pgd_k =pgd_offset_k (PAGE_OFFSET ); num_entries =pgd_index (EFI_VA_END ) -pgd_index (PAGE_OFFSET );memcpy (pgd_efi, pgd_k,sizeof (pgd_t) * num_entries); -/* -* As with PGDs, we share all P4D entries apart from the one entry -* that covers the EFI runtime mapping space. -*/ -BUILD_BUG_ON (p4d_index (EFI_VA_END ) !=p4d_index (MODULES_END )); -BUILD_BUG_ON ((EFI_VA_START &P4D_MASK ) != (EFI_VA_END &P4D_MASK )); - pgd_efi = efi_pgd +pgd_index (EFI_VA_END ); pgd_k =pgd_offset_k (EFI_VA_END ); p4d_efi =p4d_offset (pgd_efi,0 );