Skip to contentSkip to conversation

fix c/s 18938

PATCH In review Jan Beulich 3 participants 6 messages Mar 27, 2009 – Mar 30, 2009

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 fix c/s 18938

Provide for up to 16/32 on (32/64-bit) extended MCE MSRs, and use actually existing extended MSRs on 64-bits that were inaccessible so far.

xen/arch/x86/cpu/mcheck/mce_intel.c +32 -23
@@ -117,6 +117,16 @@ static void intel_init_thermal(struct cp
 }
 #endif /* CONFIG_X86_MCE_THERMAL */
 
+static inline void intel_get_extended_msr(struct mcinfo_extended *ext, u32 msr)
+{
+    if ( ext->mc_msrs < ARRAY_SIZE(ext->mc_msr)
+         && msr < MSR_IA32_MCG_EAX + nr_intel_ext_msrs ) {
+        ext->mc_msr[ext->mc_msrs].reg = msr;
+        rdmsrl(msr, ext->mc_msr[ext->mc_msrs].value);
+        ++ext->mc_msrs;
+    }
+}
+
 static enum mca_extinfo
 intel_get_extended_msrs(struct mc_info *mci, uint16_t bank, uint64_t status)
 {
@@ -129,30 +139,29 @@ intel_get_extended_msrs(struct mc_info *
     memset(&mc_ext, 0, sizeof(struct mcinfo_extended));
     mc_ext.common.type = MC_TYPE_EXTENDED;
     mc_ext.common.size = sizeof(mc_ext);
-    mc_ext.mc_msrs = 10;
 
-    mc_ext.mc_msr[0].reg = MSR_IA32_MCG_EAX;
-    rdmsrl(MSR_IA32_MCG_EAX, mc_ext.mc_msr[0].value);
-    mc_ext.mc_msr[1].reg = MSR_IA32_MCG_EBX;
-    rdmsrl(MSR_IA32_MCG_EBX, mc_ext.mc_msr[1].value);
-    mc_ext.mc_msr[2].reg = MSR_IA32_MCG_ECX;
-    rdmsrl(MSR_IA32_MCG_ECX, mc_ext.mc_msr[2].value);
-
-    mc_ext.mc_msr[3].reg = MSR_IA32_MCG_EDX;
-    rdmsrl(MSR_IA32_MCG_EDX, mc_ext.mc_msr[3].value);
-    mc_ext.mc_msr[4].reg = MSR_IA32_MCG_ESI;
-    rdmsrl(MSR_IA32_MCG_ESI, mc_ext.mc_msr[4].value);
-    mc_ext.mc_msr[5].reg = MSR_IA32_MCG_EDI;
-    rdmsrl(MSR_IA32_MCG_EDI, mc_ext.mc_msr[5].value);
-
-    mc_ext.mc_msr[6].reg = MSR_IA32_MCG_EBP;
-    rdmsrl(MSR_IA32_MCG_EBP, mc_ext.mc_msr[6].value);
-    mc_ext.mc_msr[7].reg = MSR_IA32_MCG_ESP;
-    rdmsrl(MSR_IA32_MCG_ESP, mc_ext.mc_msr[7].value);
-    mc_ext.mc_msr[8].reg = MSR_IA32_MCG_EFLAGS;
-    rdmsrl(MSR_IA32_MCG_EFLAGS, mc_ext.mc_msr[8].value);
-    mc_ext.mc_msr[9].reg = MSR_IA32_MCG_EIP;
-    rdmsrl(MSR_IA32_MCG_EIP, mc_ext.mc_msr[9].value);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_EAX);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_EBX);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_ECX);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_EDX);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_ESI);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_EDI);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_EBP);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_ESP);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_EFLAGS);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_EIP);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_MISC);
+
+#ifdef __x86_64__
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_R8);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_R9);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_R10);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_R11);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_R12);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_R13);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_R14);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_R15);
+#endif
 
     x86_mcinfo_add(mci, &mc_ext);
 
xen/include/asm-x86/msr-index.h +9 -1
@@ -326,7 +326,15 @@
 #define MSR_IA32_MCG_ESP		0x00000187
 #define MSR_IA32_MCG_EFLAGS		0x00000188
 #define MSR_IA32_MCG_EIP		0x00000189
-#define MSR_IA32_MCG_RESERVED		0x0000018a
+#define MSR_IA32_MCG_MISC		0x0000018a
+#define MSR_IA32_MCG_R8			0x00000190
+#define MSR_IA32_MCG_R9			0x00000191
+#define MSR_IA32_MCG_R10		0x00000192
+#define MSR_IA32_MCG_R11		0x00000193
+#define MSR_IA32_MCG_R12		0x00000194
+#define MSR_IA32_MCG_R13		0x00000195
+#define MSR_IA32_MCG_R14		0x00000196
+#define MSR_IA32_MCG_R15		0x00000197
 
 /* Pentium IV performance counter MSRs */
 #define MSR_P4_BPU_PERFCTR0		0x00000300
xen/include/public/arch-x86/xen-mca.h +5 -5
@@ -166,11 +166,11 @@ struct mcinfo_extended {
 
     uint32_t mc_msrs; /* Number of msr with valid values. */
     /*
-     * Currently Intel extended MSR (32/64) including all gp registers
-     * and E(R)DI, E(R)BP, E(R)SP, E(R)FLAGS, E(R)IP, E(R)MISC, only 10
-     * of them might be useful. So expend this array to 10.
-    */
-    struct mcinfo_msr mc_msr[10];
+     * Currently Intel extended MSR (32/64) include all gp registers
+     * and E(R)FLAGS, E(R)IP, E(R)MISC, up to 11/19 of them might be
+     * useful at present. So expand this array to 16/32 to leave room.
+     */
+    struct mcinfo_msr mc_msr[sizeof(void *) * 4];
 };
 
 /* Recovery Action flags. Giving recovery result information to DOM0 */
#1 lore raw

This patch looks very good to me. Just one comment at the bottom.

Christoph

Quoted 107 lines · On Friday 27 March 2009 17:34:07 Jan Beulich wrote:

Provide for up to 16/32 on (32/64-bit) extended MCE MSRs, and use actually existing extended MSRs on 64-bits that were inaccessible so far.

--- 2009-03-27.orig/xen/arch/x86/cpu/mcheck/mce_intel.c 2009-03-27
16:43:26.000000000 +0100 +++
2009-03-27/xen/arch/x86/cpu/mcheck/mce_intel.c 2009-03-27
12:06:52.000000000 +0100 @@ -117,6 +117,16 @@ static void intel_init_thermal(struct cp
}
#endif /* CONFIG_X86_MCE_THERMAL */

+static inline void intel_get_extended_msr(struct mcinfo_extended *ext, u32 msr) +{

xen/arch/x86/cpu/mcheck/mce_intel.c +8 -0
+    if ( ext->mc_msrs < ARRAY_SIZE(ext->mc_msr)
+         && msr < MSR_IA32_MCG_EAX + nr_intel_ext_msrs ) {
+        ext->mc_msr[ext->mc_msrs].reg = msr;
+        rdmsrl(msr, ext->mc_msr[ext->mc_msrs].value);
+        ++ext->mc_msrs;
+    }
+}
+
 static enum mca_extinfo
 intel_get_extended_msrs(struct mc_info *mci, uint16_t bank, uint64_t

status) {

xen/arch/x86/cpu/mcheck/mce_intel.c +22 -24
@@ -129,30 +139,29 @@ intel_get_extended_msrs(struct mc_info *
     memset(&mc_ext, 0, sizeof(struct mcinfo_extended));
     mc_ext.common.type = MC_TYPE_EXTENDED;
     mc_ext.common.size = sizeof(mc_ext);
-    mc_ext.mc_msrs = 10;
 
-    mc_ext.mc_msr[0].reg = MSR_IA32_MCG_EAX;
-    rdmsrl(MSR_IA32_MCG_EAX, mc_ext.mc_msr[0].value);
-    mc_ext.mc_msr[1].reg = MSR_IA32_MCG_EBX;
-    rdmsrl(MSR_IA32_MCG_EBX, mc_ext.mc_msr[1].value);
-    mc_ext.mc_msr[2].reg = MSR_IA32_MCG_ECX;
-    rdmsrl(MSR_IA32_MCG_ECX, mc_ext.mc_msr[2].value);
-
-    mc_ext.mc_msr[3].reg = MSR_IA32_MCG_EDX;
-    rdmsrl(MSR_IA32_MCG_EDX, mc_ext.mc_msr[3].value);
-    mc_ext.mc_msr[4].reg = MSR_IA32_MCG_ESI;
-    rdmsrl(MSR_IA32_MCG_ESI, mc_ext.mc_msr[4].value);
-    mc_ext.mc_msr[5].reg = MSR_IA32_MCG_EDI;
-    rdmsrl(MSR_IA32_MCG_EDI, mc_ext.mc_msr[5].value);
-
-    mc_ext.mc_msr[6].reg = MSR_IA32_MCG_EBP;
-    rdmsrl(MSR_IA32_MCG_EBP, mc_ext.mc_msr[6].value);
-    mc_ext.mc_msr[7].reg = MSR_IA32_MCG_ESP;
-    rdmsrl(MSR_IA32_MCG_ESP, mc_ext.mc_msr[7].value);
-    mc_ext.mc_msr[8].reg = MSR_IA32_MCG_EFLAGS;
-    rdmsrl(MSR_IA32_MCG_EFLAGS, mc_ext.mc_msr[8].value);
-    mc_ext.mc_msr[9].reg = MSR_IA32_MCG_EIP;
-    rdmsrl(MSR_IA32_MCG_EIP, mc_ext.mc_msr[9].value);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_EAX);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_EBX);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_ECX);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_EDX);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_ESI);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_EDI);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_EBP);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_ESP);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_EFLAGS);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_EIP);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_MISC);
+
+#ifdef __x86_64__
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_R8);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_R9);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_R10);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_R11);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_R12);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_R13);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_R14);
+    intel_get_extended_msr(&mc_ext, MSR_IA32_MCG_R15);
+#endif
 
     x86_mcinfo_add(mci, &mc_ext);
 
--- 2009-03-27.orig/xen/include/asm-x86/msr-index.h	2009-03-27

16:43:26.000000000 +0100 +++
2009-03-27/xen/include/asm-x86/msr-index.h 2009-03-27 13:18:04.000000000
+0100 @@ -326,7 +326,15 @@
#define MSR_IA32_MCG_ESP 0x00000187
#define MSR_IA32_MCG_EFLAGS 0x00000188
#define MSR_IA32_MCG_EIP 0x00000189

xen/include/asm-x86/msr-index.h +9 -2
-#define MSR_IA32_MCG_RESERVED		0x0000018a
+#define MSR_IA32_MCG_MISC		0x0000018a
+#define MSR_IA32_MCG_R8			0x00000190
+#define MSR_IA32_MCG_R9			0x00000191
+#define MSR_IA32_MCG_R10		0x00000192
+#define MSR_IA32_MCG_R11		0x00000193
+#define MSR_IA32_MCG_R12		0x00000194
+#define MSR_IA32_MCG_R13		0x00000195
+#define MSR_IA32_MCG_R14		0x00000196
+#define MSR_IA32_MCG_R15		0x00000197
 
 /* Pentium IV performance counter MSRs */
 #define MSR_P4_BPU_PERFCTR0		0x00000300
--- 2009-03-27.orig/xen/include/public/arch-x86/xen-mca.h	2009-03-27

16:47:30.000000000 +0100 +++
2009-03-27/xen/include/public/arch-x86/xen-mca.h 2009-03-27
16:48:35.000000000 +0100 @@ -166,11 +166,11 @@ struct mcinfo_extended {

uint32_t mc_msrs; /* Number of msr with valid values. */ /*

xen/include/public/arch-x86/xen-mca.h struct mcinfo_extended { · line 173

Quoted 10 lines
xen/include/public/arch-x86/xen-mca.h +5 -5
-     * Currently Intel extended MSR (32/64) including all gp registers
-     * and E(R)DI, E(R)BP, E(R)SP, E(R)FLAGS, E(R)IP, E(R)MISC, only 10
-     * of them might be useful. So expend this array to 10.
-    */
-    struct mcinfo_msr mc_msr[10];
+     * Currently Intel extended MSR (32/64) include all gp registers
+     * and E(R)FLAGS, E(R)IP, E(R)MISC, up to 11/19 of them might be
+     * useful at present. So expand this array to 16/32 to leave room.
+     */
+    struct mcinfo_msr mc_msr[sizeof(void *) * 4];

Please make this a fixed sized array. There are users like Oracle who run a 32bit PAE Dom0 on a 64bit Xen ...

Quoted 3 lines

};

/* Recovery Action flags. Giving recovery result information to DOM0 */

Signature · 6 lines
---to satisfy European Law for business letters:
Advanced Micro Devices GmbH
Karl-Hammerschmidt-Str. 34, 85609 Dornach b. Muenchen
Geschaeftsfuehrer: Jochen Polster, Thomas M. McCoy, Giuliano Meroni
Sitz: Dornach, Gemeinde Aschheim, Landkreis Muenchen
Registergericht Muenchen, HRB Nr. 43632
#2 lore raw
Quoted 13 lines
Quoted 1 line
Quoted 1 line

Christoph Egger <[email protected]> 27.03.09 17:49 >>>

xen/include/public/arch-x86/xen-mca.h struct mcinfo_extended { · line 173

Quoted 10 lines · On Friday 27 March 2009 17:34:07 Jan Beulich wrote:
xen/include/public/arch-x86/xen-mca.h +5 -5
-     * Currently Intel extended MSR (32/64) including all gp registers
-     * and E(R)DI, E(R)BP, E(R)SP, E(R)FLAGS, E(R)IP, E(R)MISC, only 10
-     * of them might be useful. So expend this array to 10.
-    */
-    struct mcinfo_msr mc_msr[10];
+     * Currently Intel extended MSR (32/64) include all gp registers
+     * and E(R)FLAGS, E(R)IP, E(R)MISC, up to 11/19 of them might be
+     * useful at present. So expand this array to 16/32 to leave room.
+     */
+    struct mcinfo_msr mc_msr[sizeof(void *) * 4];
Quoted 2 lines

Please make this a fixed sized array. There are users like Oracle who run a 32bit PAE Dom0 on a 64bit Xen ...

And you expect a 32-bit kernel to be able to make sense of the MSRs corresponding to 64-bit-only registers?

But you remind me that I failed to handle the difference in size of that array for 32-on-64 - I really need to check why the structure layout checking logic didn't catch the difference in size. Oh, right, sizeof(void *) needs special treatment (and I really don#t want to sue sizeof(long) here due to the implied dependency on the OS ABI).

Jan

#3 lore raw
Quoted 17 lines · On Friday 27 March 2009 18:08:00 Jan Beulich wrote:
Quoted 14 lines
Quoted 1 line
Quoted 1 line

Christoph Egger <[email protected]> 27.03.09 17:49 >>>

xen/include/public/arch-x86/xen-mca.h struct mcinfo_extended { · line 173

Quoted 10 lines · On Friday 27 March 2009 17:34:07 Jan Beulich wrote:
xen/include/public/arch-x86/xen-mca.h +5 -5
-     * Currently Intel extended MSR (32/64) including all gp registers
-     * and E(R)DI, E(R)BP, E(R)SP, E(R)FLAGS, E(R)IP, E(R)MISC, only 10
-     * of them might be useful. So expend this array to 10.
-    */
-    struct mcinfo_msr mc_msr[10];
+     * Currently Intel extended MSR (32/64) include all gp registers
+     * and E(R)FLAGS, E(R)IP, E(R)MISC, up to 11/19 of them might be
+     * useful at present. So expand this array to 16/32 to leave room.
+     */
+    struct mcinfo_msr mc_msr[sizeof(void *) * 4];
Quoted 2 lines

Please make this a fixed sized array. There are users like Oracle who run a 32bit PAE Dom0 on a 64bit Xen ...

Quoted 8 lines

And you expect a 32-bit kernel to be able to make sense of the MSRs corresponding to 64-bit-only registers?

But you remind me that I failed to handle the difference in size of that array for 32-on-64 - I really need to check why the structure layout checking logic didn't catch the difference in size. Oh, right, sizeof(void *) needs special treatment (and I really don#t want to sue sizeof(long) here due to the implied dependency on the OS ABI).

Having the same size for both 32bit and 64bit makes the Dom0 code simpler to use it. Taking care about the different sizes would enforce either hackish code or two different implementations. That's not acceptable IMO.

Christoph

Signature · 6 lines
---to satisfy European Law for business letters:
Advanced Micro Devices GmbH
Karl-Hammerschmidt-Str. 34, 85609 Dornach b. Muenchen
Geschaeftsfuehrer: Jochen Polster, Thomas M. McCoy, Giuliano Meroni
Sitz: Dornach, Gemeinde Aschheim, Landkreis Muenchen
Registergericht Muenchen, HRB Nr. 43632
#4 lore raw
Quoted 14 lines
Quoted 1 line
Quoted 1 line

Christoph Egger <[email protected]> 27.03.09 18:21 >>>

Quoted 11 lines · On Friday 27 March 2009 18:08:00 Jan Beulich wrote:
Quoted 2 lines

Please make this a fixed sized array. There are users like Oracle who run a 32bit PAE Dom0 on a 64bit Xen ...

And you expect a 32-bit kernel to be able to make sense of the MSRs corresponding to 64-bit-only registers?

But you remind me that I failed to handle the difference in size of that array for 32-on-64 - I really need to check why the structure layout checking logic didn't catch the difference in size. Oh, right, sizeof(void *) needs special treatment (and I really don#t want to sue sizeof(long) here due to the implied dependency on the OS ABI).

Quoted 4 lines

Having the same size for both 32bit and 64bit makes the Dom0 code simpler to use it. Taking care about the different sizes would enforce either hackish code or two different implementations. That's not acceptable IMO.

After some more thinking about it I actually like the way the patch currently is. This is because the formal structure layout now properly represents the differences between the two bit widths, while it specifically does *not* require different code paths in the guest: The way the structure gets used, regardless of Dom0's bitness all registers will always be put in the structure (since there's no is_pv_32on64_???() check in that code), and Dom0 will also not have any issues reading them (unless it applies very pedantic checks on the structure contents) since the structure is to be parsed using its mcinfo_common header, which correctly states the size of the structure.

If anything, I'd be inclined to either make the structure variable length (but that would imply all compilers possibly used to compile producing and consuming code understand this construct) or add a comment to the effect of the above explanation.

Jan

#5 lore raw
In reply to Jan Beulich
Quoted 3 lines · Jan Beulich

And you expect a 32-bit kernel to be able to make sense of the MSRs corresponding to 64-bit-only registers?

The other fixes that you did for the MC hypercalls all only apply to the case of a 32bit dom0 on 64bit Xen too, so if we care about that case, we should care about this case too, I guess.

I was aware of some of the alignment issues, but basically forgot about them, since we do 64bit dom0 on 64bit Xen only.

- Frank