x86, fpu: always use kernel_fpu_begin/end() for in-kernel FPU usage
use kernel_fpu_begin/end() instead of unconditionally accessing cr0 and
saving/restoring just the few used xmm/ymm registers.
This has some advantages like:
* If the task's FPU state is already active, then kernel_fpu_begin()
will just save the user-state and avoiding the read/write of cr0.
In general, cr0 accesses are much slower.
* Manual save/restore of xmm/ymm registers will affect the 'modified' and
the 'init' optimizations brought in the by xsaveopt/xrstor
infrastructure.
* Foward compatibility with future vector register extensions will be a
problem if the xmm/ymm registers are manually saved and restored
(corrupting the extended state of those vector registers).
With this patch, there was no significant difference in the xor throughput
using AVX, measured during boot.
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Link: http://lkml.kernel.org/r/1345842782-24175-5-git-send-email-suresh.b.siddha@intel.com
Cc: Jim Kukunas <james.t.kukunas@linux.intel.com>
Cc: NeilBrown <neilb@suse.de>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
diff --git a/arch/x86/include/asm/xor_avx.h b/arch/x86/include/asm/xor_avx.h
index 2510d35..7ea79c5 100644
--- a/arch/x86/include/asm/xor_avx.h
+++ b/arch/x86/include/asm/xor_avx.h
@@ -20,32 +20,6 @@
#include <linux/compiler.h>
#include <asm/i387.h>
-#define ALIGN32 __aligned(32)
-
-#define YMM_SAVED_REGS 4
-
-#define YMMS_SAVE \
-do { \
- preempt_disable(); \
- cr0 = read_cr0(); \
- clts(); \
- asm volatile("vmovaps %%ymm0, %0" : "=m" (ymm_save[0]) : : "memory"); \
- asm volatile("vmovaps %%ymm1, %0" : "=m" (ymm_save[32]) : : "memory"); \
- asm volatile("vmovaps %%ymm2, %0" : "=m" (ymm_save[64]) : : "memory"); \
- asm volatile("vmovaps %%ymm3, %0" : "=m" (ymm_save[96]) : : "memory"); \
-} while (0);
-
-#define YMMS_RESTORE \
-do { \
- asm volatile("sfence" : : : "memory"); \
- asm volatile("vmovaps %0, %%ymm3" : : "m" (ymm_save[96])); \
- asm volatile("vmovaps %0, %%ymm2" : : "m" (ymm_save[64])); \
- asm volatile("vmovaps %0, %%ymm1" : : "m" (ymm_save[32])); \
- asm volatile("vmovaps %0, %%ymm0" : : "m" (ymm_save[0])); \
- write_cr0(cr0); \
- preempt_enable(); \
-} while (0);
-
#define BLOCK4(i) \
BLOCK(32 * i, 0) \
BLOCK(32 * (i + 1), 1) \
@@ -60,10 +34,9 @@
static void xor_avx_2(unsigned long bytes, unsigned long *p0, unsigned long *p1)
{
- unsigned long cr0, lines = bytes >> 9;
- char ymm_save[32 * YMM_SAVED_REGS] ALIGN32;
+ unsigned long lines = bytes >> 9;
- YMMS_SAVE
+ kernel_fpu_begin();
while (lines--) {
#undef BLOCK
@@ -82,16 +55,15 @@
p1 = (unsigned long *)((uintptr_t)p1 + 512);
}
- YMMS_RESTORE
+ kernel_fpu_end();
}
static void xor_avx_3(unsigned long bytes, unsigned long *p0, unsigned long *p1,
unsigned long *p2)
{
- unsigned long cr0, lines = bytes >> 9;
- char ymm_save[32 * YMM_SAVED_REGS] ALIGN32;
+ unsigned long lines = bytes >> 9;
- YMMS_SAVE
+ kernel_fpu_begin();
while (lines--) {
#undef BLOCK
@@ -113,16 +85,15 @@
p2 = (unsigned long *)((uintptr_t)p2 + 512);
}
- YMMS_RESTORE
+ kernel_fpu_end();
}
static void xor_avx_4(unsigned long bytes, unsigned long *p0, unsigned long *p1,
unsigned long *p2, unsigned long *p3)
{
- unsigned long cr0, lines = bytes >> 9;
- char ymm_save[32 * YMM_SAVED_REGS] ALIGN32;
+ unsigned long lines = bytes >> 9;
- YMMS_SAVE
+ kernel_fpu_begin();
while (lines--) {
#undef BLOCK
@@ -147,16 +118,15 @@
p3 = (unsigned long *)((uintptr_t)p3 + 512);
}
- YMMS_RESTORE
+ kernel_fpu_end();
}
static void xor_avx_5(unsigned long bytes, unsigned long *p0, unsigned long *p1,
unsigned long *p2, unsigned long *p3, unsigned long *p4)
{
- unsigned long cr0, lines = bytes >> 9;
- char ymm_save[32 * YMM_SAVED_REGS] ALIGN32;
+ unsigned long lines = bytes >> 9;
- YMMS_SAVE
+ kernel_fpu_begin();
while (lines--) {
#undef BLOCK
@@ -184,7 +154,7 @@
p4 = (unsigned long *)((uintptr_t)p4 + 512);
}
- YMMS_RESTORE
+ kernel_fpu_end();
}
static struct xor_block_template xor_block_avx = {