x86: Remove the prefetch() specific implementation on x86_64
Subsystems (3)
PATCH 1/2 x86: Remove the prefetch() specific implementation on x86_64
From: Youling Tang <[email protected]>
After commit ab483570a13b ("x86 & generic: change to __builtin_prefetch()"), x86_64 directly uses __builtin_prefetch() without the specific implementation of prefetch(). Also, x86_64 use a generic definition until commit ae2e15eb3b6c ("x86: unify prefetch operations"). So remove it.
--- arch/x86/include/asm/processor.h | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-)
arch/x86/include/asm/processor.h +4 -3
@@ -599,9 +599,6 @@ extern char ignore_fpu_irq;#ifdef CONFIG_X86_32 # define BASE_PREFETCH ""# define ARCH_HAS_PREFETCH -#else -# define BASE_PREFETCH "prefetcht0 %1" -#endif /* * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
@@ -616,6 +613,10 @@ static inline void prefetch(const void *x) "m" (*(const char *)x)); } +#else +# define BASE_PREFETCH "prefetcht0 %1" +#endif +/* * 3dnow prefetch to get an exclusive cache line. * Useful for spinlocks to avoid one state transition in the
Signature · 1 line
2.34.1
PATCH 2/2 prefetch: Add ARCH_HAS_PREFETCH definition when the architecture is not defined
From: Youling Tang <[email protected]>
After commit ab483570a13b ("x86 & generic: change to __builtin_prefetch()"), the __builtin_prefetch implementation will be used by default, so add its definition to return to a state similar to before commit[1].
Currently prefetch_range() will be empty implemented under the x86_64 architecture, there was a concrete implementation before "x86 & generic: change to __builtin_prefetch()", so fix it.
No similar changes have been made to ARCH_HAS_PREFETCHW at this time.
--- include/linux/prefetch.h | 1 + 1 file changed, 1 insertion(+)
include/linux/prefetch.h +1 -0
@@ -36,6 +36,7 @@ struct page;*/ #ifndef ARCH_HAS_PREFETCH +#define ARCH_HAS_PREFETCH #define prefetch (x) __builtin_prefetch(x)#endif
Signature · 1 line
2.34.1
kernel test robot
build botkernel test robot noticed the following build errors:
Hi Youling,
kernel test robot noticed the following build errors:
[auto build test ERROR on tip/master]
[also build test ERROR on linus/master v6.10-rc1 next-20240529] [cannot apply to tip/auto-latest tip/x86/core bp/for-next] [If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]
url: https://github.com/intel-lab-lkp/linux/commits/Youling-Tang/prefetch-Add-ARCH_HAS_PREFETCH-definition-when-the-architecture-is-not-defined/20240529-112345 base: tip/master patch link: https://lore.kernel.org/r/20240529032059.899347-1-youling.tang%40linux.dev patch subject: [PATCH 1/2] x86: Remove the prefetch() specific implementation on x86_64 config: x86_64-buildonly-randconfig-006-20240530 (https://download.01.org/0day-ci/archive/20240530/[email protected]/config) compiler: gcc-13 (Ubuntu 13.2.0-4ubuntu3) 13.2.0 reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20240530/[email protected]/reproduce)
If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/
All errors (new ones prefixed by >>):
drivers/soc/fsl/dpio/dpio-service.c: In function 'dpaa2_io_store_next':
>> drivers/soc/fsl/dpio/dpio-service.c:745:17: error: implicit declaration of function 'prefetch'; did you mean 'prefetchw'? [-Werror=implicit-function-declaration]
745 | prefetch(&s->vaddr[s->idx]);
| ^~~~~~~~
| prefetchw
cc1: some warnings being treated as errors
--
drivers/soc/fsl/dpio/qbman-portal.c: In function 'qbman_swp_dqrr_next_direct':
>> drivers/soc/fsl/dpio/qbman-portal.c:1213:17: error: implicit declaration of function 'prefetch'; did you mean 'prefetchw'? [-Werror=implicit-function-declaration]
1213 | prefetch(qbman_get_cmd(s,
| ^~~~~~~~
| prefetchwcc1: some warnings being treated as errors
vim +745 drivers/soc/fsl/dpio/dpio-service.c
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 703
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 704 /**
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 705 * dpaa2_io_store_next() - Determine when the next dequeue result is available.
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 706 * @s: the dpaa2_io_store object.
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 707 * @is_last: indicate whether this is the last frame in the pull command.
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 708 *
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 709 * When an object driver performs dequeues to a dpaa2_io_store, this function
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 710 * can be used to determine when the next frame result is available. Once
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 711 * this function returns non-NULL, a subsequent call to it will try to find
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 712 * the next dequeue result.
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 713 *
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 714 * Note that if a pull-dequeue has a NULL result because the target FQ/channel
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 715 * was empty, then this function will also return NULL (rather than expecting
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 716 * the caller to always check for this. As such, "is_last" can be used to
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 717 * differentiate between "end-of-empty-dequeue" and "still-waiting".
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 718 *
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 719 * Return dequeue result for a valid dequeue result, or NULL for empty dequeue.
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 720 */
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 721 struct dpaa2_dq *dpaa2_io_store_next(struct dpaa2_io_store *s, int *is_last)
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 722 {
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 723 int match;
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 724 struct dpaa2_dq *ret = &s->vaddr[s->idx];
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 725
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 726 match = qbman_result_has_new_result(s->swp, ret);
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 727 if (!match) {
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 728 *is_last = 0;
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 729 return NULL;
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 730 }
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 731
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 732 s->idx++;
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 733
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 734 if (dpaa2_dq_is_pull_complete(ret)) {
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 735 *is_last = 1;
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 736 s->idx = 0;
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 737 /*
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 738 * If we get an empty dequeue result to terminate a zero-results
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 739 * vdqcr, return NULL to the caller rather than expecting him to
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 740 * check non-NULL results every time.
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 741 */
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 742 if (!(dpaa2_dq_flags(ret) & DPAA2_DQ_STAT_VALIDFRAME))
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 743 ret = NULL;
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 744 } else {
f1e250bf365962 drivers/soc/fsl/dpio/dpio-service.c Ioana Ciocoi Radulescu 2018-12-14 @745 prefetch(&s->vaddr[s->idx]);
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 746 *is_last = 0;
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 747 }
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 748
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 749 return ret;
780b626323d721 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Roy Pledge 2017-03-13 750 }
53639c64c686f0 drivers/staging/fsl-mc/bus/dpio/dpio-service.c Laurentiu Tudor 2017-11-17 751 EXPORT_SYMBOL_GPL(dpaa2_io_store_next);
e80081c34b0358 drivers/soc/fsl/dpio/dpio-service.c Roy Pledge 2018-12-18 752 Signature · 2 lines
0-DAY CI Kernel Test Service https://github.com/intel/lkp-tests/wiki
Quoted 31 lines · kernel test robot
Hi Youling,
kernel test robot noticed the following build errors:
[auto build test ERROR on tip/master]
[also build test ERROR on linus/master v6.10-rc1 next-20240529] [cannot apply to tip/auto-latest tip/x86/core bp/for-next] [If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]url: https://github.com/intel-lab-lkp/linux/commits/Youling-Tang/prefetch-Add-ARCH_HAS_PREFETCH-definition-when-the-architecture-is-not-defined/20240529-112345 base: tip/master patch link: https://lore.kernel.org/r/20240529032059.899347-1-youling.tang%40linux.dev patch subject: [PATCH 1/2] x86: Remove the prefetch() specific implementation on x86_64 config: x86_64-buildonly-randconfig-006-20240530 (https://download.01.org/0day-ci/archive/20240530/[email protected]/config) compiler: gcc-13 (Ubuntu 13.2.0-4ubuntu3) 13.2.0 reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20240530/[email protected]/reproduce)If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/All errors (new ones prefixed by >>):
drivers/soc/fsl/dpio/dpio-service.c: In function 'dpaa2_io_store_next':Quoted 1 line
Quoted 1 line
drivers/soc/fsl/dpio/dpio-service.c:745:17: error: implicit declaration of function 'prefetch'; did you mean 'prefetchw'? [-Werror=implicit-function-declaration]745 | prefetch(&s->vaddr[s->idx]); | ^~~~~~~~ | prefetchw
Quoted 8 lines
cc1: some warnings being treated as errors
--
drivers/soc/fsl/dpio/qbman-portal.c: In function 'qbman_swp_dqrr_next_direct':Quoted 1 line
Quoted 1 line
drivers/soc/fsl/dpio/qbman-portal.c:1213:17: error: implicit declaration of function 'prefetch'; did you mean 'prefetchw'? [-Werror=implicit-function-declaration]1213 | prefetch(qbman_get_cmd(s, | ^~~~~~~~ | prefetchwcc1: some warnings being treated as errors
This problem is caused by not including the linux/prefetch.h file. There were no build errors earlier because the definitions in processor.h were used indirectly. (For architectures that do not implement prefetch, this build error can occur without the patch).
We can fix it in the following way:
drivers/soc/fsl/dpio/dpio-service.c +0 -0
@@ -9,6 +9,7 @@
#include <soc/fsl/dpaa2-io.h>
#include <linux/init.h>
#include <linux/module.h>
+#include <linux/prefetch.h>
#include <linux/platform_device.h>
#include <linux/interrupt.h>
#include <linux/dma-mapping.h>
drivers/soc/fsl/dpio/qbman-portal.c +0 -0
@@ -7,6 +7,7 @@
#include <asm/cacheflush.h>
#include <linux/io.h>
+#include <linux/prefetch.h>
#include <linux/slab.h>
Thanks,
Youling.
Quoted 6 lines · Youling Tang
From: Youling Tang <[email protected]>
After commit
ab483570a13b("x86 & generic: change to__builtin_prefetch()"),x86_64directly uses__builtin_prefetch()without the specific implementation ofprefetch(). Also,x86_64use a generic definition until commitae2e15eb3b6c("x86: unify prefetch operations"). So remove it.
So this patch just ensures the x86-specific prefetch() implementation is defined only for 32bit case, otherwise we have it defined for the 64bit case as well but effectively it's not used since ARCH_HAS_PREFETCH is not defined for 64bit, meaning in the 64bit case prefetch() is still defined to __builtint_prefetch in include/linux/prefetch.h.
In essence this is a purely cosmetic cleanup , am I right?
I compiled a file that utilizes prefetch with and without your patch and the generated assembly is identical.
Quoted 31 lines
Signed-off-by Youling Tang <[email protected]>--- arch/x86/include/asm/processor.h | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-)arch/x86/include/asm/processor.h +4
-3@@ -599,9 +599,6 @@ extern char ignore_fpu_irq;#ifdef CONFIG_X86_32 # define BASE_PREFETCH ""# define ARCH_HAS_PREFETCH -#else -# define BASE_PREFETCH "prefetcht0 %1" -#endif /* * Prefetch instructions for Pentium III (+) and AMD Athlon (+) @@ -616,6 +613,10 @@ static inline void prefetch(const void *x) "m" (*(const char *)x)); } +#else +# define BASE_PREFETCH "prefetcht0 %1" +#endif +/* * 3dnow prefetch to get an exclusive cache line. * Useful for spinlocks to avoid one state transition in the
Re: [PATCH 2/2] prefetch: Add ARCH_HAS_PREFETCH definition when the architecture is not defined
In reply to Youling TangQuoted 5 lines · Youling Tang
From: Youling Tang <[email protected]>
After commit
ab483570a13b("x86 & generic: change to__builtin_prefetch()"), the__builtin_prefetchimplementation will be used by default, so add its definition to return to a state similar to before commit[1].
I find those references to the past state somewhat confusing and not really adding sustenance to the explanation.
Quoted 4 lines
Currently
prefetch_range()will be empty implemented under thex86_64architecture, there was a concrete implementation before "x86 & generic: change to__builtin_prefetch()", so fix it.
Actually prefetch_range will be empty for every architecture which doesn't defined ARCH_HAS_PREFETCH and since we have a working generic fallback this indeed seems backwards. So defining ARCH_HAS_PREFETCH makes sense.
Quoted 23 lines
No similar changes have been made to
ARCH_HAS_PREFETCHWat this time.Fixes ab483570a13b ("x86 & generic: change to __builtin_prefetch()")Signed-off-by Youling Tang <[email protected]>--- include/linux/prefetch.h | 1 + 1 file changed, 1 insertion(+)include/linux/prefetch.h +1
-0@@ -36,6 +36,7 @@ struct page;*/ #ifndef ARCH_HAS_PREFETCH +#define ARCH_HAS_PREFETCH #define prefetch (x) __builtin_prefetch(x)#endif
Hi, Nikolay
Quoted 11 lines · Nikolay Borisov
Quoted 6 lines · Youling Tang
From: Youling Tang <[email protected]>
After commit
ab483570a13b("x86 & generic: change to__builtin_prefetch()"),x86_64directly uses__builtin_prefetch()without the specific implementation ofprefetch(). Also,x86_64use a generic definition until commitae2e15eb3b6c("x86: unify prefetch operations"). So remove it.
Quoted 4 lines
So this patch just ensures the x86-specific
prefetch()implementation is defined only for 32bit case, otherwise we have it defined for the 64bit case as well but effectively it's not used sinceARCH_HAS_PREFETCHis not defined for 64bit, meaning in the 64bit caseprefetch()is still defined to__builtint_prefetchininclude/linux/prefetch.h.In essence this is a purely cosmetic cleanup , am I right?
Yes, when arch customization and __builtint_prefetch are implemented with the same instructions, it looks like pure cleaning (without changing the generated assembly).
Thanks,
Youling.
Quoted 37 lines
I compiled a file that utilizes prefetch with and without your patch and the generated assembly is identical.
Reviewed-by Nikolay Borisov <[email protected]>Quoted 29 lines
Signed-off-by Youling Tang <[email protected]>---
arch/x86/include/asm/processor.h| 7 ++++---1 file changed, 4 insertions(+), 3 deletions(-)arch/x86/include/asm/processor.h +0
-3@@ -599,9 +599,6 @@ extern char ignore_fpu_irq; #ifdef CONFIG_X86_32 # define BASE_PREFETCH "" # define ARCH_HAS_PREFETCH -#else -# define BASE_PREFETCH "prefetcht0 %1" -#endif /** Prefetch instructions for Pentium III (+) and AMD Athlon (+) @@ -616,6 +613,10 @@ static inline void prefetch(const void *x) "m" (*(const char *)x)); }+#else
arch/x86/include/asm/processor.h +3
-0+# define BASE_PREFETCH "prefetcht0 %1" +#endif +/*
* 3dnow prefetch to get an exclusive cache line.
* Useful for spinlocks to avoid one state transition in the
Hi,
Quoted 44 lines · Nikolay Borisov
Quoted 6 lines · Youling Tang
From: Youling Tang <[email protected]>
After commit
ab483570a13b("x86 & generic: change to__builtin_prefetch()"),x86_64directly uses__builtin_prefetch()without the specific implementation ofprefetch(). Also,x86_64use a generic definition until commitae2e15eb3b6c("x86: unify prefetch operations"). So remove it.So this patch just ensures the x86-specific
prefetch()implementation is defined only for 32bit case, otherwise we have it defined for the 64bit case as well but effectively it's not used sinceARCH_HAS_PREFETCHis not defined for 64bit, meaning in the 64bit caseprefetch()is still defined to__builtint_prefetchininclude/linux/prefetch.h.In essence this is a purely cosmetic cleanup , am I right?
I compiled a file that utilizes prefetch with and without your patch and the generated assembly is identical.
Reviewed-by Nikolay Borisov <[email protected]>Quoted 21 lines
Signed-off-by Youling Tang <[email protected]>---
arch/x86/include/asm/processor.h| 7 ++++---1 file changed, 4 insertions(+), 3 deletions(-)arch/x86/include/asm/processor.h +0
-3@@ -599,9 +599,6 @@ extern char ignore_fpu_irq; #ifdef CONFIG_X86_32 # define BASE_PREFETCH "" # define ARCH_HAS_PREFETCH -#else -# define BASE_PREFETCH "prefetcht0 %1" -#endif /** Prefetch instructions for Pentium III (+) and AMD Athlon (+) @@ -616,6 +613,10 @@ static inline void prefetch(const void *x) "m" (*(const char *)x));
Quoted 8 lines
Quoted 8 lines
arch/x86/include/asm/processor.h +0
-0}+#else
arch/x86/include/asm/processor.h +3
-0+# define BASE_PREFETCH "prefetcht0 %1" +#endif +/*
* 3dnow prefetch to get an exclusive cache line.
* Useful for spinlocks to avoid one state transition in the
Sorry to bother you, but do we still need this patchset? (Do I need to modify the commit message and send v2 if necessary?)
Thanks,
Youling.