Skip to contentSkip to conversation

use __asm__ and __volatile__ in asm-x86_64/msr.h

PATCH NAK Mike Frysinger 4 participants 8 messages Jun 18, 2007 – Jul 23, 2007

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 use __asm__ and __volatile__ in asm-x86_64/msr.h

Most of the macros in msr.h already use __asm__ and __volatile__, this patch brings the rest into line. This is needed since these get exported into userspace (need to use strict __asm__ vs asm when gcc compiles with GNU extensions turned off).

---

include/asm-x86_64/msr.h +5 -5
index d5c55b8..ed15d2b 100644
@@ -33,7 +33,7 @@
 
 /* wrmsr with exception handling */
 #define wrmsr_safe(msr,a,b) ({ int ret__;			\
-	asm volatile("2: wrmsr ; xorl %0,%0\n"			\
+	__asm__ __volatile__("2: wrmsr ; xorl %0,%0\n"			\
 		     "1:\n\t"					\
 		     ".section .fixup,\"ax\"\n\t"		\
 		     "3:  movl %4,%0 ; jmp 1b\n\t"		\
@@ -50,7 +50,7 @@
 
 #define rdmsr_safe(msr,a,b) \
 	({ int ret__;						\
-	  asm volatile ("1:       rdmsr\n"			\
+	  __asm__ __volatile__ ("1:       rdmsr\n"			\
                       "2:\n"					\
                       ".section .fixup,\"ax\"\n"		\
                       "3:       movl %4,%0\n"			\
@@ -70,17 +70,17 @@
      __asm__ __volatile__ ("rdtsc" : "=a" (low) : : "edx")
 
 #define rdtscp(low,high,aux) \
-     asm volatile (".byte 0x0f,0x01,0xf9" : "=a" (low), "=d" (high), "=c" (aux))
+     __asm__ __volatile__ (".byte 0x0f,0x01,0xf9" : "=a" (low), "=d" (high), "=c" (aux))
 
 #define rdtscll(val) do { \
      unsigned int __a,__d; \
-     asm volatile("rdtsc" : "=a" (__a), "=d" (__d)); \
+     __asm__ __volatile__("rdtsc" : "=a" (__a), "=d" (__d)); \
      (val) = ((unsigned long)__a) | (((unsigned long)__d)<<32); \
 } while(0)
 
 #define rdtscpll(val, aux) do { \
      unsigned long __a, __d; \
-     asm volatile (".byte 0x0f,0x01,0xf9" : "=a" (__a), "=d" (__d), "=c" (aux)); \
+     __asm__ __volatile__ (".byte 0x0f,0x01,0xf9" : "=a" (__a), "=d" (__d), "=c" (aux)); \
      (val) = (__d << 32) | __a; \
 } while (0)
 
#1 lore raw
Quoted 19 lines · Mike Frysinger

Most of the macros in msr.h already use __asm__ and __volatile__, this patch brings the rest into line. This is needed since these get exported into userspace (need to use strict __asm__ vs asm when gcc compiles with GNU extensions turned off).
---

include/asm-x86_64/msr.h +1 -1
index d5c55b8..ed15d2b 100644
@@ -33,7 +33,7 @@
 
 /* wrmsr with exception handling */
 #define wrmsr_safe(msr,a,b) ({ int ret__;			\
-	asm volatile("2: wrmsr ; xorl %0,%0\n"			\
+	__asm__ __volatile__("2: wrmsr ; xorl %0,%0\n"			\
 		     "1:\n\t"					\
 		     ".section .fixup,\"ax\"\n\t"		\
 		     "3:  movl %4,%0 ; jmp 1b\n\t"		\

include/asm-x86_64/msr.h line 56

Quoted 8 lines
include/asm-x86_64/msr.h +1 -1
@@ -50,7 +50,7 @@
 
 #define rdmsr_safe(msr,a,b) \
 	({ int ret__;						\
-	  asm volatile ("1:       rdmsr\n"			\
+	  __asm__ __volatile__ ("1:       rdmsr\n"			\
                       "2:\n"					\
                       ".section .fixup,\"ax\"\n"		\
                       "3:       movl %4,%0\n"			\

NAK.

This is wrong for two reasons:

- MSRs are not accessible in user space.
- .section .fixup is not applicable to user space.

Anything MSR-related should be protected by __KERNEL__ instead.

This NAK doesn't apply to rdtsc* and rdpmc* changes, however, as those instructions are legitimate in userspace.

-hpa

#2 lore raw
Quoted 33 lines · H. Peter Anvin
Quoted 19 lines · Mike Frysinger

Most of the macros in msr.h already use __asm__ and __volatile__, this patch brings the rest into line. This is needed since these get exported into userspace (need to use strict __asm__ vs asm when gcc compiles with GNU extensions turned off).
---

include/asm-x86_64/msr.h +1 -1
index d5c55b8..ed15d2b 100644
@@ -33,7 +33,7 @@
 
 /* wrmsr with exception handling */
 #define wrmsr_safe(msr,a,b) ({ int ret__;			\
-	asm volatile("2: wrmsr ; xorl %0,%0\n"			\
+	__asm__ __volatile__("2: wrmsr ; xorl %0,%0\n"			\
 		     "1:\n\t"					\
 		     ".section .fixup,\"ax\"\n\t"		\
 		     "3:  movl %4,%0 ; jmp 1b\n\t"		\

include/asm-x86_64/msr.h line 56

Quoted 8 lines
include/asm-x86_64/msr.h +1 -1
@@ -50,7 +50,7 @@
 
 #define rdmsr_safe(msr,a,b) \
 	({ int ret__;						\
-	  asm volatile ("1:       rdmsr\n"			\
+	  __asm__ __volatile__ ("1:       rdmsr\n"			\
                       "2:\n"					\
                       ".section .fixup,\"ax\"\n"		\
                       "3:       movl %4,%0\n"			\

NAK.

This is wrong for two reasons:

Quoted 7 lines

- MSRs are not accessible in user space.
- .section .fixup is not applicable to user space.

Anything MSR-related should be protected by __KERNEL__ instead.

This NAK doesn't apply to rdtsc* and rdpmc* changes, however, as those instructions are legitimate in userspace.

then someone needs to audit msr.h for this because currently it lacks __KERNEL__ of any kind yet is exported completely to userspace ... i dont know anything about this instructions so that someone != me

this is similar to what i heard last time i tried to fixup msr.h for userspace, but the file still hasnt been updated ... -mike

#3 lore raw
Quoted 10 lines · Mike Frysinger
Quoted 9 lines

This is wrong for two reasons:

- MSRs are not accessible in user space.
- .section .fixup is not applicable to user space.

Anything MSR-related should be protected by __KERNEL__ instead.

This NAK doesn't apply to rdtsc* and rdpmc* changes, however, as those instructions are legitimate in userspace.

Quoted 6 lines

then someone needs to audit msr.h for this because currently it lacks __KERNEL__ of any kind yet is exported completely to userspace ... i dont know anything about this instructions so that someone != me

this is similar to what i heard last time i tried to fixup msr.h for userspace, but the file still hasnt been updated ...

asm-i386/msr.h should not be exported to userspace at all, it contains nothing but kernel-internal helpers.

#4 lore raw
Quoted 3 lines

asm-i386/msr.h should not be exported to userspace at all, it contains nothing but kernel-internal helpers.

Actually rdtsc and rdtscll and potentially rdpmc which is in there can be very useful in user space if you know what you're doing. Unfortunately a lot of its users don't, but not having the include probably won't stop them either.

-Andi

#5 lore raw
Quoted 6 lines · Andi Kleen
Quoted 2 lines

asm-i386/msr.h should not be exported to userspace at all, it contains nothing but kernel-internal helpers.

Actually rdtsc and rdtscll and potentially rdpmc which is in there can be very useful in user space if you know what you're doing. Unfortunately a lot of its users don't, but not having the include probably won't stop them either.

More likely, people will just re-implement them incorrectly.

However, the rdtsc() definition in the kernel is weird (and removable, I think there are no more users -- I have it removed in my MSR driver rewrite tree which I need to get off my arse and push.) Most users would expect the rdtscll() functionality with the rdtsc() name.

-hpa

#6 lore raw
Quoted 9 lines · H. Peter Anvin
Quoted 7 lines · Andi Kleen
Quoted 2 lines

asm-i386/msr.h should not be exported to userspace at all, it contains nothing but kernel-internal helpers.

Actually rdtsc and rdtscll and potentially rdpmc which is in there can be very useful in user space if you know what you're doing. Unfortunately a lot of its users don't, but not having the include probably won't stop them either.

Quoted 6 lines

More likely, people will just re-implement them incorrectly.

However, the rdtsc() definition in the kernel is weird (and removable, I think there are no more users -- I have it removed in my MSR driver rewrite tree which I need to get off my arse and push.) Most users would expect the rdtscll() functionality with the rdtsc() name.

does that mean you'll also take care of cleaning up msr.h ? or do i need to post another patch ?
-mike

#7 lore raw
Quoted 10 lines · Mike Frysinger
Quoted 9 lines · H. Peter Anvin
Quoted 6 lines · Andi Kleen
Quoted 2 lines

asm-i386/msr.h should not be exported to userspace at all, it contains nothing but kernel-internal helpers.

Actually rdtsc and rdtscll and potentially rdpmc which is in there can be very useful in user space if you know what you're doing. Unfortunately a lot of its users don't, but not having the include probably won't stop them either.

More likely, people will just re-implement them incorrectly.

Quoted 7 lines
Quoted 4 lines

However, the rdtsc() definition in the kernel is weird (and removable, I think there are no more users -- I have it removed in my MSR driver rewrite tree which I need to get off my arse and push.) Most users would expect the rdtscll() functionality with the rdtsc() name.

does that mean you'll also take care of cleaning up msr.h ? or do i need to post another patch ?

I'll put that on my list.

-hpa