Skip to contentSkip to conversation

x86: fix bitops.h warning with a moved cast

PATCH Applied Jesse Brandeburg 6 participants 11 messages Mar 10, 2020 – May 5, 2020

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH v6 1/2 x86: fix bitops.h warning with a moved cast

Fix many sparse warnings when building with C=1. These are useless noise from the bitops.h file and getting rid of them helps devs make more use of the tools and possibly find real bugs.

When the kernel is compiled with C=1, there are lots of messages like:
arch/x86/include/asm/bitops.h:77:37: warning: cast truncates bits from constant value (ffffff7f becomes 7f)

CONST_MASK() is using a signed integer "1" to create the mask which is later cast to (u8), in order to yield an 8-bit value for the assembly instructions to use. Simplify the expressions used to clearly indicate they are working on 8-bit values only, which still keeps sparse happy without an accidental promotion to a 32 bit integer.

The warning was occurring because certain bitmasks that end with a bit set next to a natural boundary like 7, 15, 23, 31, end up with a mask like 0x7f, which then results in sign extension due to the integer type promotion rules[1]. It was really only "clear_bit" that was having problems, and it was only on some bit checks that resulted in a mask like 0xffffff7f being generated after the inversion.

Verified with a test module (see next patch) and assembly inspection that the patch doesn't introduce any change in generated code.

[1] https://stackoverflow.com/questions/46073295/implicit-type-promotion-rules

---
v6: reworded commit message, enhanced explanation
v5: changed code to use simple AND and XOR, updated commit message
v4: reverse argument order as suggested by David Laight, added reviewed-by
v3: Clean up the header file changes as per peterz.
v2: use correct CC: list

---
 arch/x86/include/asm/bitops.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/x86/include/asm/bitops.h +2 -2
index 062cdecb2f24..53f246e9df5a 100644
@@ -54,7 +54,7 @@ arch_set_bit(long nr, volatile unsigned long *addr)
 	if (__builtin_constant_p(nr)) {
 		asm volatile(LOCK_PREFIX "orb %1,%0"
 			: CONST_MASK_ADDR(nr, addr)
-			: "iq" ((u8)CONST_MASK(nr))
+			: "iq" (CONST_MASK(nr) & 0xff)
 			: "memory");
 	} else {
 		asm volatile(LOCK_PREFIX __ASM_SIZE(bts) " %1,%0"
@@ -74,7 +74,7 @@ arch_clear_bit(long nr, volatile unsigned long *addr)
 	if (__builtin_constant_p(nr)) {
 		asm volatile(LOCK_PREFIX "andb %1,%0"
 			: CONST_MASK_ADDR(nr, addr)
-			: "iq" ((u8)~CONST_MASK(nr)));
+			: "iq" (CONST_MASK(nr) ^ 0xff));
 	} else {
 		asm volatile(LOCK_PREFIX __ASM_SIZE(btr) " %1,%0"
 			: : RLONG_ADDR(addr), "Ir" (nr) : "memory");

base-commit: 8b614cb8f1dcac8ca77cf4dd85f46ef3055f8238

Signature · 1 line
2.24.1
#1 lore raw

PATCH v6 2/2 lib: make a test module with set/clear bit

Test some bit clears/sets to make sure assembly doesn't change, and that the set_bit and clear_bit functions work and don't cause sparse warnings.

Instruct Kbuild to build this file with extra warning level -Wextra, to catch new issues, and also doesn't hurt to build with C=1.

This was used to test changes to arch/x86/include/asm/bitops.h.

In particular, sparse (C=1) was very concerned when the last bit before a natural boundary, like 7, or 31, was being tested, as this causes sign extension (0xffffff7f) for instance when clearing bit 7.

Recommended usage:
make defconfig
scripts/config -m CONFIG_TEST_BITOPS
make modules_prepare
make C=1 W=1 lib/test_bitops.ko
objdump -S -d lib/test_bitops.ko
insmod lib/test_bitops.ko
rmmod lib/test_bitops.ko
<check dmesg>, there should be no compiler/sparse warnings and no error messages in log.

---
v5: no change
v4: Slight change to bitops_fun last member, as suggested by Andy, added his reviewed-by too. Added module load/unload to usage. Added copyright.
v3: Update the test to fail if bits aren't cleared, and make the test reproduce the original issue without patch 1/2, showing that the issue is fixed in patch 1/2. Thanks PeterZ!
v2: Correct CC: list

---
 lib/Kconfig.debug                  | 13 +++++++
 lib/Makefile                       |  2 +
 lib/test_bitops.c                  | 60 ++++++++++++++++++++++++++++++
 tools/testing/selftests/lib/config |  1 +
 4 files changed, 76 insertions(+)
 create mode 100644 lib/test_bitops.c
lib/Kconfig.debug +13 -0
index 69def4a9df00..61a5d00ea064 100644
@@ -1947,6 +1947,19 @@ config TEST_LKM
 
 	  If unsure, say N.
 
+config TEST_BITOPS
+	tristate "Test module for compilation of clear_bit/set_bit operations"
+	depends on m
+	help
+	  This builds the "test_bitops" module that is much like the
+	  TEST_LKM module except that it does a basic exercise of the
+	  clear_bit and set_bit macros to make sure there are no compiler
+	  warnings from C=1 sparse checker or -Wextra compilations. It has
+	  no dependencies and doesn't run or load unless explicitly requested
+	  by name.  for example: modprobe test_bitops.
+
+	  If unsure, say N.
+
 config TEST_VMALLOC
 	tristate "Test module for stress/performance analysis of vmalloc allocator"
 	default n
lib/Makefile +2 -0
index 611872c06926..b18db565b355 100644
@@ -89,6 +89,8 @@ obj-$(CONFIG_TEST_OBJAGG) += test_objagg.o
 obj-$(CONFIG_TEST_STACKINIT) += test_stackinit.o
 obj-$(CONFIG_TEST_BLACKHOLE_DEV) += test_blackhole_dev.o
 obj-$(CONFIG_TEST_MEMINIT) += test_meminit.o
+obj-$(CONFIG_TEST_BITOPS) += test_bitops.o
+CFLAGS_test_bitops.o += -Werror
 
 obj-$(CONFIG_TEST_LIVEPATCH) += livepatch/
 
lib/test_bitops.c new file +60 -0

60 lines not on this page · raw message

tools/testing/selftests/lib/config +1 -0
index 14a77ea4a8da..b80ee3f6e265 100644
@@ -2,3 +2,4 @@ CONFIG_TEST_PRINTF=m
 CONFIG_TEST_BITMAP=m
 CONFIG_PRIME_NUMBERS=m
 CONFIG_TEST_STRSCPY=m
+CONFIG_TEST_BITOPS=m
Signature · 1 line
2.24.1
#2 lore raw

Re: [PATCH v6 2/2] lib: make a test module with set/clear bit

Quoted 13 lines · Jesse Brandeburg

Test some bit clears/sets to make sure assembly doesn't change, and that the set_bit and clear_bit functions work and don't cause sparse warnings.

Instruct Kbuild to build this file with extra warning level -Wextra, to catch new issues, and also doesn't hurt to build with C=1.

This was used to test changes to arch/x86/include/asm/bitops.h.

In particular, sparse (C=1) was very concerned when the last bit before a natural boundary, like 7, or 31, was being tested, as this causes sign extension (0xffffff7f) for instance when clearing bit 7.

Quoted 10 lines

Recommended usage:
make defconfig
scripts/config -m CONFIG_TEST_BITOPS
make modules_prepare
make C=1 W=1 lib/test_bitops.ko
objdump -S -d lib/test_bitops.ko
insmod lib/test_bitops.ko
rmmod lib/test_bitops.ko
<check dmesg>, there should be no compiler/sparse warnings and no error messages in log.

Seems this didn't make the kernel. Perhaps you need to send it to Andrew Morton.

Signature · 2 lines
With Best Regards,
Andy Shevchenko
Acked-by on v6 1/2
#3 lore raw
In reply to Jesse Brandeburg
Quoted 6 lines · Jesse Brandeburg

Fix many sparse warnings when building with C=1. These are useless noise from the bitops.h file and getting rid of them helps devs make more use of the tools and possibly find real bugs.

When the kernel is compiled with C=1, there are lots of messages like:
arch/x86/include/asm/bitops.h:77:37: warning: cast truncates bits from constant value (ffffff7f becomes 7f)

Acked-by on v6 1/2
#4 lore raw
In reply to Jesse Brandeburg
Quoted 20 lines · Jesse Brandeburg

Fix many sparse warnings when building with C=1. These are useless noise from the bitops.h file and getting rid of them helps devs make more use of the tools and possibly find real bugs.

When the kernel is compiled with C=1, there are lots of messages like:
arch/x86/include/asm/bitops.h:77:37: warning: cast truncates bits from constant value (ffffff7f becomes 7f)

CONST_MASK() is using a signed integer "1" to create the mask which is later cast to (u8), in order to yield an 8-bit value for the assembly instructions to use. Simplify the expressions used to clearly indicate they are working on 8-bit values only, which still keeps sparse happy without an accidental promotion to a 32 bit integer.

The warning was occurring because certain bitmasks that end with a bit set next to a natural boundary like 7, 15, 23, 31, end up with a mask like 0x7f, which then results in sign extension due to the integer type promotion rules[1]. It was really only "clear_bit" that was having problems, and it was only on some bit checks that resulted in a mask like 0xffffff7f being generated after the inversion.

Quoted 7 lines

Verified with a test module (see next patch) and assembly inspection that the patch doesn't introduce any change in generated code.

[1] https://stackoverflow.com/questions/46073295/implicit-type-promotion-rules

Boris, can you make this happen?

Quoted 37 lines

---
v6: reworded commit message, enhanced explanation
v5: changed code to use simple AND and XOR, updated commit message
v4: reverse argument order as suggested by David Laight, added reviewed-by
v3: Clean up the header file changes as per peterz.
v2: use correct CC: list

---
 arch/x86/include/asm/bitops.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/x86/include/asm/bitops.h +2 -2
index 062cdecb2f24..53f246e9df5a 100644
@@ -54,7 +54,7 @@ arch_set_bit(long nr, volatile unsigned long *addr)
 	if (__builtin_constant_p(nr)) {
 		asm volatile(LOCK_PREFIX "orb %1,%0"
 			: CONST_MASK_ADDR(nr, addr)
-			: "iq" ((u8)CONST_MASK(nr))
+			: "iq" (CONST_MASK(nr) & 0xff)
 			: "memory");
 	} else {
 		asm volatile(LOCK_PREFIX __ASM_SIZE(bts) " %1,%0"
@@ -74,7 +74,7 @@ arch_clear_bit(long nr, volatile unsigned long *addr)
 	if (__builtin_constant_p(nr)) {
 		asm volatile(LOCK_PREFIX "andb %1,%0"
 			: CONST_MASK_ADDR(nr, addr)
-			: "iq" ((u8)~CONST_MASK(nr)));
+			: "iq" (CONST_MASK(nr) ^ 0xff));
 	} else {
 		asm volatile(LOCK_PREFIX __ASM_SIZE(btr) " %1,%0"
 			: : RLONG_ADDR(addr), "Ir" (nr) : "memory");

base-commit: 8b614cb8f1dcac8ca77cf4dd85f46ef3055f8238

Signature · 1 line
2.24.1
#5 lore raw

[tip: x86/asm] x86: Fix bitops.h warning with a moved cast

In reply to Jesse Brandeburg
The following commit has been merged into the x86/asm branch of tip:

The following commit has been merged into the x86/asm branch of tip:

Commit-ID:     1651e700664b4597ddf4f8adfe435252a0d11277
Gitweb:        https://git.kernel.org/tip/1651e700664b4597ddf4f8adfe435252a0d11277
Author:        Jesse Brandeburg <[email protected]>
AuthorDate:    Tue, 10 Mar 2020 15:17:46 -07:00
Committer:     Borislav Petkov <[email protected]>
CommitterDate: Wed, 18 Mar 2020 12:30:19 +01:00

x86: Fix bitops.h warning with a moved cast

Fix many sparse warnings when building with C=1. These are useless noise from the bitops.h file and getting rid of them helps developers make more use of the tools and possibly find real bugs.

When the kernel is compiled with C=1, there are lots of messages like:

arch/x86/include/asm/bitops.h:77:37: warning: cast truncates bits from constant value (ffffff7f becomes 7f)

CONST_MASK() is using a signed integer "1" to create the mask which is later cast to (u8), in order to yield an 8-bit value for the assembly instructions to use. Simplify the expressions used to clearly indicate they are working on 8-bit values only, which still keeps sparse happy without an accidental promotion to a 32 bit integer.

The warning was occurring because certain bitmasks that end with a bit set next to a natural boundary like 7, 15, 23, 31, end up with a mask like 0x7f, which then results in sign extension due to the integer type promotion rules[1]. It was really only clear_bit() that was having problems, and it was only on some bit checks that resulted in a mask like 0xffffff7f being generated after the inversion.

Verify with a test module (see next patch) and assembly inspection that the fix doesn't introduce any change in generated code.

[ bp: Massage. ]

---
 arch/x86/include/asm/bitops.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/x86/include/asm/bitops.h +2 -2
index 062cdec..53f246e 100644
@@ -54,7 +54,7 @@ arch_set_bit(long nr, volatile unsigned long *addr)
 	if (__builtin_constant_p(nr)) {
 		asm volatile(LOCK_PREFIX "orb %1,%0"
 			: CONST_MASK_ADDR(nr, addr)
-			: "iq" ((u8)CONST_MASK(nr))
+			: "iq" (CONST_MASK(nr) & 0xff)
 			: "memory");
 	} else {
 		asm volatile(LOCK_PREFIX __ASM_SIZE(bts) " %1,%0"
@@ -74,7 +74,7 @@ arch_clear_bit(long nr, volatile unsigned long *addr)
 	if (__builtin_constant_p(nr)) {
 		asm volatile(LOCK_PREFIX "andb %1,%0"
 			: CONST_MASK_ADDR(nr, addr)
-			: "iq" ((u8)~CONST_MASK(nr)));
+			: "iq" (CONST_MASK(nr) ^ 0xff));
 	} else {
 		asm volatile(LOCK_PREFIX __ASM_SIZE(btr) " %1,%0"
 			: : RLONG_ADDR(addr), "Ir" (nr) : "memory");
#6 lore raw
In reply to Jesse Brandeburg
Quoted 42 lines · Jesse Brandeburg

Fix many sparse warnings when building with C=1. These are useless noise from the bitops.h file and getting rid of them helps devs make more use of the tools and possibly find real bugs.

When the kernel is compiled with C=1, there are lots of messages like:
arch/x86/include/asm/bitops.h:77:37: warning: cast truncates bits from constant value (ffffff7f becomes 7f)

CONST_MASK() is using a signed integer "1" to create the mask which is later cast to (u8), in order to yield an 8-bit value for the assembly instructions to use. Simplify the expressions used to clearly indicate they are working on 8-bit values only, which still keeps sparse happy without an accidental promotion to a 32 bit integer.

The warning was occurring because certain bitmasks that end with a bit set next to a natural boundary like 7, 15, 23, 31, end up with a mask like 0x7f, which then results in sign extension due to the integer type promotion rules[1]. It was really only "clear_bit" that was having problems, and it was only on some bit checks that resulted in a mask like 0xffffff7f being generated after the inversion.

Verified with a test module (see next patch) and assembly inspection that the patch doesn't introduce any change in generated code.

[1] https://stackoverflow.com/questions/46073295/implicit-type-promotion-rules

---
v6: reworded commit message, enhanced explanation
v5: changed code to use simple AND and XOR, updated commit message
v4: reverse argument order as suggested by David Laight, added reviewed-by
v3: Clean up the header file changes as per peterz.
v2: use correct CC: list

---
 arch/x86/include/asm/bitops.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/x86/include/asm/bitops.h +0 -0
index 062cdecb2f24..53f246e9df5a 100644

arch/x86/include/asm/bitops.h arch_set_bit(long nr, volatile unsigned long *addr) · line 57

Quoted 5 lines
arch/x86/include/asm/bitops.h +1 -1
@@ -54,7 +54,7 @@ arch_set_bit(long nr, volatile unsigned long *addr)
 	if (__builtin_constant_p(nr)) {
 		asm volatile(LOCK_PREFIX "orb %1,%0"
 			: CONST_MASK_ADDR(nr, addr)
-			: "iq" ((u8)CONST_MASK(nr))
+			: "iq" (CONST_MASK(nr) & 0xff)

Sorry for the very late report. It turns out that if your config tickles __builtin_constant_p just right, this now produces invalid assembly:

$ cat foo.c
long a(long b, long c) {
  asm("orb\t%1, %0" : "+q"(c): "r"(b));
  return c;
}
$ gcc foo.c
foo.c: Assembler messages:
foo.c:2: Error: `%rax' not allowed with `orb'

The "q" constraint only has meanting on -m32 otherwise is treated as "r".

Since we have the mask (& 0xff), can we drop the b suffix from the instruction? Or is a revert more appropriate? Or maybe another way to skin this cat?

This is blowing up our KernelCI reports.

Quoted 17 lines
arch/x86/include/asm/bitops.h +1 -1
 			: "memory");
 	} else {
 		asm volatile(LOCK_PREFIX __ASM_SIZE(bts) " %1,%0"
@@ -74,7 +74,7 @@ arch_clear_bit(long nr, volatile unsigned long *addr)
 	if (__builtin_constant_p(nr)) {
 		asm volatile(LOCK_PREFIX "andb %1,%0"
 			: CONST_MASK_ADDR(nr, addr)
-			: "iq" ((u8)~CONST_MASK(nr)));
+			: "iq" (CONST_MASK(nr) ^ 0xff));
 	} else {
 		asm volatile(LOCK_PREFIX __ASM_SIZE(btr) " %1,%0"
 			: : RLONG_ADDR(addr), "Ir" (nr) : "memory");

base-commit: 8b614cb8f1dcac8ca77cf4dd85f46ef3055f8238

Signature · 1 line
2.24.1
#7 lore raw
Quoted 13 lines · On Mon, 4 May 2020 12:51:12 -0700 Nick Desaulniers <[email protected]>…

Sorry for the very late report. It turns out that if your config tickles __builtin_constant_p just right, this now produces invalid assembly:

$ cat foo.c
long a(long b, long c) {
  asm("orb\t%1, %0" : "+q"(c): "r"(b));
  return c;
}
$ gcc foo.c
foo.c: Assembler messages:
foo.c:2: Error: `%rax' not allowed with `orb'
Quoted 6 lines

The "q" constraint only has meanting on -m32 otherwise is treated as "r".

Since we have the mask (& 0xff), can we drop the b suffix from the instruction? Or is a revert more appropriate? Or maybe another way to skin this cat?

Figures that such a small change can create problems :-( Sorry for the thrash!

The patches in the link below basically add back the cast, but I'm interested to see if any others can come up with a better fix that
a) passes the above code generation test
b) still keeps sparse happy
c) passes the test module and the code inspection

If need be I'm OK with a revert of the original patch to fix the issue in the short term, but it seems to me there must be a way to satisfy both tools. We went through several iterations on the way to the final patch that we might be able to pluck something useful from.

Quoted 2 lines

This is blowing up our KernelCI reports.

ASIDE:
Bummer, how come none of those KernelCI reports are part of zero-day testing at https://01.org/lkp/documentation/0-day-test-service
I'm interested in your answer but don't want to pollute this thread, feel free to contact me directly for this one or start a new thread?

#8 lore raw
Quoted 32 lines · Jesse Brandeburg
Quoted 13 lines · On Mon, 4 May 2020 12:51:12 -0700 Nick Desaulniers <[email protected]>…

Sorry for the very late report. It turns out that if your config tickles __builtin_constant_p just right, this now produces invalid assembly:

$ cat foo.c
long a(long b, long c) {
  asm("orb\t%1, %0" : "+q"(c): "r"(b));
  return c;
}
$ gcc foo.c
foo.c: Assembler messages:
foo.c:2: Error: `%rax' not allowed with `orb'
Quoted 6 lines

The "q" constraint only has meanting on -m32 otherwise is treated as "r".

Since we have the mask (& 0xff), can we drop the b suffix from the instruction? Or is a revert more appropriate? Or maybe another way to skin this cat?

Figures that such a small change can create problems :-( Sorry for the thrash!

The patches in the link below basically add back the cast, but I'm interested to see if any others can come up with a better fix that
a) passes the above code generation test
b) still keeps sparse happy
c) passes the test module and the code inspection

Quoted 4 lines

If need be I'm OK with a revert of the original patch to fix the issue in the short term, but it seems to me there must be a way to satisfy both tools. We went through several iterations on the way to the final patch that we might be able to pluck something useful from.

For me the below seems to work:

arch/x86/include/asm/bitops.h +2 -2
index b392571c1f1d1..139122e5b25b1 100644
@@ -54,7 +54,7 @@ arch_set_bit(long nr, volatile unsigned long *addr)
 	if (__builtin_constant_p(nr)) {
 		asm volatile(LOCK_PREFIX "orb %1,%0"
 			: CONST_MASK_ADDR(nr, addr)
-			: "iq" (CONST_MASK(nr) & 0xff)
+			: "iq" ((u8)(CONST_MASK(nr) & 0xff))
 			: "memory");
 	} else {
 		asm volatile(LOCK_PREFIX __ASM_SIZE(bts) " %1,%0"
@@ -74,7 +74,7 @@ arch_clear_bit(long nr, volatile unsigned long *addr)
 	if (__builtin_constant_p(nr)) {
 		asm volatile(LOCK_PREFIX "andb %1,%0"
 			: CONST_MASK_ADDR(nr, addr)
-			: "iq" (CONST_MASK(nr) ^ 0xff));
+			: "iq" ((u8)(CONST_MASK(nr) ^ 0xff)));
 	} else {
 		asm volatile(LOCK_PREFIX __ASM_SIZE(btr) " %1,%0"
 			: : RLONG_ADDR(addr), "Ir" (nr) : "memory");
Signature · 2 lines
With Best Regards,
Andy Shevchenko
#9 lore raw
Quoted 34 lines · Andy Shevchenko
Quoted 32 lines · Jesse Brandeburg
Quoted 13 lines · On Mon, 4 May 2020 12:51:12 -0700 Nick Desaulniers <[email protected]>…

Sorry for the very late report. It turns out that if your config tickles __builtin_constant_p just right, this now produces invalid assembly:

$ cat foo.c
long a(long b, long c) {
  asm("orb\t%1, %0" : "+q"(c): "r"(b));
  return c;
}
$ gcc foo.c
foo.c: Assembler messages:
foo.c:2: Error: `%rax' not allowed with `orb'
Quoted 6 lines

The "q" constraint only has meanting on -m32 otherwise is treated as "r".

Since we have the mask (& 0xff), can we drop the b suffix from the instruction? Or is a revert more appropriate? Or maybe another way to skin this cat?

Figures that such a small change can create problems :-( Sorry for the thrash!

The patches in the link below basically add back the cast, but I'm interested to see if any others can come up with a better fix that
a) passes the above code generation test
b) still keeps sparse happy
c) passes the test module and the code inspection

Quoted 6 lines
Quoted 4 lines

If need be I'm OK with a revert of the original patch to fix the issue in the short term, but it seems to me there must be a way to satisfy both tools. We went through several iterations on the way to the final patch that we might be able to pluck something useful from.

For me the below seems to work:

Yep:
https://github.com/ClangBuiltLinux/linux/issues/961#issuecomment-623785987
https://github.com/ClangBuiltLinux/linux/issues/961#issuecomment-624162497
Sedat wrote the same patch 22 days ago; I didn't notice before starting this thread. I will sign off on his patch, add your Suggested by tag, and send shortly.

Quoted 31 lines
arch/x86/include/asm/bitops.h +2 -3
index b392571c1f1d1..139122e5b25b1 100644
@@ -54,7 +54,7 @@ arch_set_bit(long nr, volatile unsigned long *addr)
        if (__builtin_constant_p(nr)) {
                asm volatile(LOCK_PREFIX "orb %1,%0"
                        : CONST_MASK_ADDR(nr, addr)
-                       : "iq" (CONST_MASK(nr) & 0xff)
+                       : "iq" ((u8)(CONST_MASK(nr) & 0xff))
                        : "memory");
        } else {
                asm volatile(LOCK_PREFIX __ASM_SIZE(bts) " %1,%0"
@@ -74,7 +74,7 @@ arch_clear_bit(long nr, volatile unsigned long *addr)
        if (__builtin_constant_p(nr)) {
                asm volatile(LOCK_PREFIX "andb %1,%0"
                        : CONST_MASK_ADDR(nr, addr)
-                       : "iq" (CONST_MASK(nr) ^ 0xff));
+                       : "iq" ((u8)(CONST_MASK(nr) ^ 0xff)));
        } else {
                asm volatile(LOCK_PREFIX __ASM_SIZE(btr) " %1,%0"
                        : : RLONG_ADDR(addr), "Ir" (nr) : "memory");
 
 
--

With Best Regards,
Andy Shevchenko

Signature · 2 lines
Thanks,
~Nick Desaulniers
#10 lore raw
Quoted 42 lines · Nick Desaulniers
Quoted 39 lines · Andy Shevchenko
Quoted 36 lines · Jesse Brandeburg
Quoted 13 lines · On Mon, 4 May 2020 12:51:12 -0700 Nick Desaulniers <[email protected]>…

Sorry for the very late report. It turns out that if your config tickles __builtin_constant_p just right, this now produces invalid assembly:

$ cat foo.c
long a(long b, long c) {
  asm("orb\t%1, %0" : "+q"(c): "r"(b));
  return c;
}
$ gcc foo.c
foo.c: Assembler messages:
foo.c:2: Error: `%rax' not allowed with `orb'
Quoted 6 lines

The "q" constraint only has meanting on -m32 otherwise is treated as "r".

Since we have the mask (& 0xff), can we drop the b suffix from the instruction? Or is a revert more appropriate? Or maybe another way to skin this cat?

Figures that such a small change can create problems :-( Sorry for the thrash!

The patches in the link below basically add back the cast, but I'm interested to see if any others can come up with a better fix that
a) passes the above code generation test
b) still keeps sparse happy
c) passes the test module and the code inspection

If need be I'm OK with a revert of the original patch to fix the issue in the short term, but it seems to me there must be a way to satisfy both tools. We went through several iterations on the way to the final patch that we might be able to pluck something useful from.

Quoted 8 lines
Quoted 1 line

For me the below seems to work:

Yep:
https://github.com/ClangBuiltLinux/linux/issues/961#issuecomment-623785987
https://github.com/ClangBuiltLinux/linux/issues/961#issuecomment-624162497
Sedat wrote the same patch 22 days ago; I didn't notice before starting this thread. I will sign off on his patch, add your Suggested by tag, and send shortly.

Started a new thread:
https://lore.kernel.org/lkml/[email protected]/T/#u

Quoted 37 lines
Quoted 28 lines
arch/x86/include/asm/bitops.h +2 -3
index b392571c1f1d1..139122e5b25b1 100644
@@ -54,7 +54,7 @@ arch_set_bit(long nr, volatile unsigned long *addr)
        if (__builtin_constant_p(nr)) {
                asm volatile(LOCK_PREFIX "orb %1,%0"
                        : CONST_MASK_ADDR(nr, addr)
-                       : "iq" (CONST_MASK(nr) & 0xff)
+                       : "iq" ((u8)(CONST_MASK(nr) & 0xff))
                        : "memory");
        } else {
                asm volatile(LOCK_PREFIX __ASM_SIZE(bts) " %1,%0"
@@ -74,7 +74,7 @@ arch_clear_bit(long nr, volatile unsigned long *addr)
        if (__builtin_constant_p(nr)) {
                asm volatile(LOCK_PREFIX "andb %1,%0"
                        : CONST_MASK_ADDR(nr, addr)
-                       : "iq" (CONST_MASK(nr) ^ 0xff));
+                       : "iq" ((u8)(CONST_MASK(nr) ^ 0xff)));
        } else {
                asm volatile(LOCK_PREFIX __ASM_SIZE(btr) " %1,%0"
                        : : RLONG_ADDR(addr), "Ir" (nr) : "memory");
 
 
--
Quoted 4 lines

With Best Regards,
Andy Shevchenko

--
Thanks,
~Nick Desaulniers

Signature · 2 lines
Thanks,
~Nick Desaulniers