x86/mem changes for v2.6.40

GIT PULL Ingo Molnar 1 participant 1 message May 19, 2011 x86

lore

Share & tools

lore

Ingo Molnar

author

2011-05-19 21:09 UTC #0 lore raw

Linus,

Please pull the latest x86-mem-for-linus git tree from:

git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip.git x86-mem-for-linus

Thanks,

Ingo

------------------>
Fenghua Yu (10):
x86, cpufeature: Fix cpuid leaf 7 feature detection
x86, cpufeature: Add CPU feature bit for enhanced REP MOVSB/STOSB x86, mem, intel: Initialize Enhanced REP MOVSB/STOSB
x86, alternative, doc: Add comment for applying alternatives order x86, alternative: Add altinstruction_entry macro
x86, mem: clear_page_64.S: Support clear_page() with enhanced REP MOVSB/STOSB x86, mem: copy_user_64.S: Support copy_to/from_user by enhanced REP MOVSB/STOSB x86, mem: memcpy_64.S: Optimize memcpy by enhanced REP MOVSB/STOSB x86, mem: memmove_64.S: Optimize memmove by enhanced REP MOVSB/STOSB x86, mem: memset_64.S: Optimize memset by enhanced REP MOVSB/STOSB

Jiri Olsa (1):
x86, 64-bit: Fix copy_[to/from]_user() checks for the userspace address limit

 arch/x86/include/asm/alternative-asm.h |    9 ++++
 arch/x86/include/asm/cpufeature.h      |    1 +
 arch/x86/include/asm/uaccess.h         |    2 +-
 arch/x86/kernel/alternative.c          |    9 ++++
 arch/x86/kernel/cpu/common.c           |    3 +-
 arch/x86/kernel/cpu/intel.c            |   19 +++++++--
 arch/x86/lib/clear_page_64.S           |   33 +++++++++++----
 arch/x86/lib/copy_user_64.S            |   69 ++++++++++++++++++++++++++------
 arch/x86/lib/memcpy_64.S               |   45 +++++++++++++++------
 arch/x86/lib/memmove_64.S              |   29 +++++++++++++-
 arch/x86/lib/memset_64.S               |   54 +++++++++++++++++++------
 11 files changed, 219 insertions(+), 54 deletions(-)

arch/x86/include/asm/alternative-asm.h +9 -0

index a63a68b..94d420b 100644
@@ -15,4 +15,13 @@
 	.endm
 #endif
 
+.macro altinstruction_entry orig alt feature orig_len alt_len
+	.align 8
+	.quad \orig
+	.quad \alt
+	.word \feature
+	.byte \orig_len
+	.byte \alt_len
+.endm
+
 #endif  /*  __ASSEMBLY__  */

arch/x86/include/asm/cpufeature.h +1 -0

index 91f3e087..7f2f7b1 100644
@@ -195,6 +195,7 @@
 
 /* Intel-defined CPU features, CPUID level 0x00000007:0 (ebx), word 9 */
 #define X86_FEATURE_FSGSBASE	(9*32+ 0) /* {RD/WR}{FS/GS}BASE instructions*/
+#define X86_FEATURE_ERMS	(9*32+ 9) /* Enhanced REP MOVSB/STOSB */
 
 #if defined(__KERNEL__) && !defined(__ASSEMBLY__)

arch/x86/include/asm/uaccess.h +1 -1

index abd3e0e..99f0ad7 100644
@@ -42,7 +42,7 @@
  * Returns 0 if the range is valid, nonzero otherwise.
  *
  * This is equivalent to the following test:
- * (u33)addr + (u33)size >= (u33)current->addr_limit.seg (u65 for x86_64)
+ * (u33)addr + (u33)size > (u33)current->addr_limit.seg (u65 for x86_64)
  *
  * This needs 33-bit (65-bit for x86_64) arithmetic. We have a carry...
  */

arch/x86/kernel/alternative.c +9 -0

index 4a23467..f4fe15d 100644
@@ -210,6 +210,15 @@ void __init_or_module apply_alternatives(struct alt_instr *start,
 	u8 insnbuf[MAX_PATCH_LEN];
 
 	DPRINTK("%s: alt table %p -> %p\n", __func__, start, end);
+	/*
+	 * The scan order should be from start to end. A later scanned
+	 * alternative code can overwrite a previous scanned alternative code.
+	 * Some kernel functions (e.g. memcpy, memset, etc) use this order to
+	 * patch code.
+	 *
+	 * So be careful if you want to change the scan order to any other
+	 * order.
+	 */
 	for (a = start; a < end; a++) {
 		u8 *instr = a->instr;
 		BUG_ON(a->replacementlen > a->instrlen);

arch/x86/kernel/cpu/common.c +1 -2

index e2ced007..173f3a3 100644
@@ -565,8 +565,7 @@ void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
 
 		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
 
-		if (eax > 0)
-			c->x86_capability[9] = ebx;
+		c->x86_capability[9] = ebx;
 	}
 
 	/* AMD-defined flags: level 0x80000001 */

arch/x86/kernel/cpu/intel.c +15 -4

index df86bc8..fc73a34 100644
@@ -29,10 +29,10 @@
 
 static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
 {
+	u64 misc_enable;
+
 	/* Unmask CPUID levels if masked: */
 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
-		u64 misc_enable;
-
 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
 
 		if (misc_enable & MSR_IA32_MISC_ENABLE_LIMIT_CPUID) {

@@ -118,8 +118,6 @@ static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
 	 * (model 2) with the same problem.
 	 */
 	if (c->x86 == 15) {
-		u64 misc_enable;
-
 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
 
 		if (misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING) {

@@ -130,6 +128,19 @@ static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
 		}
 	}
 #endif
+
+	/*
+	 * If fast string is not enabled in IA32_MISC_ENABLE for any reason,
+	 * clear the fast string and enhanced fast string CPU capabilities.
+	 */
+	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
+		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
+		if (!(misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING)) {
+			printk(KERN_INFO "Disabled fast string operations\n");
+			setup_clear_cpu_cap(X86_FEATURE_REP_GOOD);
+			setup_clear_cpu_cap(X86_FEATURE_ERMS);
+		}
+	}
 }
 
 #ifdef CONFIG_X86_32

arch/x86/lib/clear_page_64.S +24 -9

index aa4326b..f2145cf 100644
@@ -1,5 +1,6 @@
 #include <linux/linkage.h>
 #include <asm/dwarf2.h>
+#include <asm/alternative-asm.h>
 
 /*
  * Zero a page.

@@ -14,6 +15,15 @@ ENTRY(clear_page_c)
 	CFI_ENDPROC
 ENDPROC(clear_page_c)
 
+ENTRY(clear_page_c_e)
+	CFI_STARTPROC
+	movl $4096,%ecx
+	xorl %eax,%eax
+	rep stosb
+	ret
+	CFI_ENDPROC
+ENDPROC(clear_page_c_e)
+
 ENTRY(clear_page)
 	CFI_STARTPROC
 	xorl   %eax,%eax

@@ -38,21 +48,26 @@ ENTRY(clear_page)
 .Lclear_page_end:
 ENDPROC(clear_page)
 
-	/* Some CPUs run faster using the string instructions.
-	   It is also a lot simpler. Use this when possible */
+	/*
+	 * Some CPUs support enhanced REP MOVSB/STOSB instructions.
+	 * It is recommended to use this when possible.
+	 * If enhanced REP MOVSB/STOSB is not available, try to use fast string.
+	 * Otherwise, use original function.
+	 *
+	 */
 
 #include <asm/cpufeature.h>
 
 	.section .altinstr_replacement,"ax"
 1:	.byte 0xeb					/* jmp <disp8> */
 	.byte (clear_page_c - clear_page) - (2f - 1b)	/* offset */
-2:
+2:	.byte 0xeb					/* jmp <disp8> */
+	.byte (clear_page_c_e - clear_page) - (3f - 2b)	/* offset */
+3:
 	.previous
 	.section .altinstructions,"a"
-	.align 8
-	.quad clear_page
-	.quad 1b
-	.word X86_FEATURE_REP_GOOD
-	.byte .Lclear_page_end - clear_page
-	.byte 2b - 1b
+	altinstruction_entry clear_page,1b,X86_FEATURE_REP_GOOD,\
+			     .Lclear_page_end-clear_page, 2b-1b
+	altinstruction_entry clear_page,2b,X86_FEATURE_ERMS,   \
+			     .Lclear_page_end-clear_page,3b-2b
 	.previous

arch/x86/lib/copy_user_64.S +57 ~~-12~~

99 lines not on this page · raw message

arch/x86/lib/memcpy_64.S +32 ~~-13~~

65 lines not on this page · raw message

arch/x86/lib/memmove_64.S +28 -1

index 0ecb843..d0ec9c2 100644
@@ -8,6 +8,7 @@
 #define _STRING_C
 #include <linux/linkage.h>
 #include <asm/dwarf2.h>
+#include <asm/cpufeature.h>
 
 #undef memmove

@@ -24,6 +25,7 @@
  */
 ENTRY(memmove)
 	CFI_STARTPROC
+
 	/* Handle more 32bytes in loop */
 	mov %rdi, %rax
 	cmp $0x20, %rdx

@@ -31,8 +33,13 @@ ENTRY(memmove)
 
 	/* Decide forward/backward copy mode */
 	cmp %rdi, %rsi
-	jb	2f
+	jge .Lmemmove_begin_forward
+	mov %rsi, %r8
+	add %rdx, %r8
+	cmp %rdi, %r8
+	jg 2f
 
+.Lmemmove_begin_forward:
 	/*
 	 * movsq instruction have many startup latency
 	 * so we handle small size by general register.

@@ -78,6 +85,8 @@ ENTRY(memmove)
 	rep movsq
 	movq %r11, (%r10)
 	jmp 13f
+.Lmemmove_end_forward:
+
 	/*
 	 * Handle data backward by movsq.
 	 */

@@ -194,4 +203,22 @@ ENTRY(memmove)
 13:
 	retq
 	CFI_ENDPROC
+
+	.section .altinstr_replacement,"ax"
+.Lmemmove_begin_forward_efs:
+	/* Forward moving data. */
+	movq %rdx, %rcx
+	rep movsb
+	retq
+.Lmemmove_end_forward_efs:
+	.previous
+
+	.section .altinstructions,"a"
+	.align 8
+	.quad .Lmemmove_begin_forward
+	.quad .Lmemmove_begin_forward_efs
+	.word X86_FEATURE_ERMS
+	.byte .Lmemmove_end_forward-.Lmemmove_begin_forward
+	.byte .Lmemmove_end_forward_efs-.Lmemmove_begin_forward_efs
+	.previous
 ENDPROC(memmove)

arch/x86/lib/memset_64.S +42 ~~-12~~

73 lines not on this page · raw message