Skip to contentSkip to conversation

x86/microcode/intel: replace sync_core() with native_cpuid_reg(eax)

PATCH Superseded Evalds Iodzevics 4 participants 9 messages Apr 20, 2020 – Apr 21, 2020 x86

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86/microcode/intel: replace sync_core() with native_cpuid_reg(eax)

sync_core() always jums past cpuid instruction on 32 bit machines because data structure boot_cpu_data are not populated so early in boot.

It depends on commit 5dedade6dfa243c130b85d1e4daba6f027805033 for native_cpuid_reg(eax) definitions

This patch is for 4.4 but also should apply to 4.9

---
 arch/x86/include/asm/microcode_intel.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/include/asm/microcode_intel.h +1 -1
index 90343ba50485..92ce9c8a508b 100644
@@ -60,7 +60,7 @@ static inline u32 intel_get_microcode_revision(void)
 	native_wrmsrl(MSR_IA32_UCODE_REV, 0);
 
 	/* As documented in the SDM: Do a CPUID 1 here */
-	sync_core();
+	native_cpuid_eax(1);
 
 	/* get the current revision from MSR 0x8B */
 	native_rdmsr(MSR_IA32_UCODE_REV, dummy, rev);
Signature · 1 line
2.17.4
maintainer
#1 lore raw
Quoted 2 lines · Evalds Iodzevics

sync_core() always jums past cpuid instruction on 32 bit machines because data structure boot_cpu_data are not populated so early in boot.

I'm guessing because boot_cpu_data.cpuid_level is not properly set and very early code in head_32.S sets it to -1 temporarily until the highest CPUID level has been detected (or not).

But the microcode loading happens *before* that.

I'd probably be interested how you trigger that but since the backport got changed (see below) and yours is correcting it to the upstream variant then perhaps maybe I don't care that much. :)

Quoted 16 lines

It depends on commit 5dedade6dfa243c130b85d1e4daba6f027805033 for native_cpuid_reg(eax) definitions

This patch is for 4.4 but also should apply to 4.9

---
 arch/x86/include/asm/microcode_intel.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/include/asm/microcode_intel.h +0 -0
index 90343ba50485..92ce9c8a508b 100644

arch/x86/include/asm/microcode_intel.h static inline u32 intel_get_microcode_revision(void) · line 66

Quoted 8 lines
arch/x86/include/asm/microcode_intel.h +1 -1
@@ -60,7 +60,7 @@ static inline u32 intel_get_microcode_revision(void)
 	native_wrmsrl(MSR_IA32_UCODE_REV, 0);
 
 	/* As documented in the SDM: Do a CPUID 1 here */
-	sync_core();
+	native_cpuid_eax(1);
 
 	/* get the current revision from MSR 0x8B */
 	native_rdmsr(MSR_IA32_UCODE_REV, dummy, rev);

Hrm, the original patch of mine did use native_cpuid_eax():

4167709bbf82 ("x86/microcode/intel: Add a helper which gives the microcode revision")

but the backport:

commit 98cc1464cfd6edf9dc7fa96aaaf596aae952029b

Author: Borislav Petkov <[email protected]>
Date:   Mon Jan 9 12:41:45 2017 +0100

x86/microcode/intel: Add a helper which gives the microcode revision

commit 4167709bbf826512a52ebd6aafda2be104adaec9 upstream.

Since on Intel we're required to do CPUID(1) first, before reading the microcode revision MSR, let's add a special helper which does the required steps so that we don't forget to do them next time, when we want to read the microcode revision.

[bwh: Backported to 4.4:
- Don't touch prev_rev variable in apply_microcode()
- Keep using sync_core(), which will alway includes the necessary CPUID ^^^^^^^^^^^^^^^^^^^

decided to use sync_core() for whatever reason. Perhaps because the native_cpuid* things weren't there. Adding Ben to Cc.

I believe this is the background info Greg needed to figure out *why* you're doing this.

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
#2 lore raw
Quoted 9 lines · Borislav Petkov
Quoted 2 lines · Evalds Iodzevics

sync_core() always jums past cpuid instruction on 32 bit machines because data structure boot_cpu_data are not populated so early in boot.

I'm guessing because boot_cpu_data.cpuid_level is not properly set and very early code in head_32.S sets it to -1 temporarily until the highest CPUID level has been detected (or not).

But the microcode loading happens *before* that.

[...]

Quoted 22 lines

Hrm, the original patch of mine did use native_cpuid_eax():

4167709bbf82 ("x86/microcode/intel: Add a helper which gives the microcode revision")

but the backport:

commit 98cc1464cfd6edf9dc7fa96aaaf596aae952029b

Author: Borislav Petkov <[email protected]>
Date:   Mon Jan 9 12:41:45 2017 +0100

x86/microcode/intel: Add a helper which gives the microcode revision

commit 4167709bbf826512a52ebd6aafda2be104adaec9 upstream.

Since on Intel we're required to do CPUID(1) first, before reading the microcode revision MSR, let's add a special helper which does the required steps so that we don't forget to do them next time, when we want to read the microcode revision.

Quoted 7 lines

[bwh: Backported to 4.4:
- Don't touch prev_rev variable in apply_microcode()
- Keep using sync_core(), which will alway includes the necessary CPUID ^^^^^^^^^^^^^^^^^^^

decided to use sync_core() for whatever reason. Perhaps because the native_cpuid* things weren't there. Adding Ben to Cc.

This commit didn't introduce the call to native_cpuid_eax(), but only moved it. So I didn't think it made sense for the backport to change from sync_core().

If it's important to use native_cpuid_eax() then these older branches should presumably get backports of:

484d0e5c7943 x86/microcode/intel: Replace sync_core() with native_cpuid() f3e2a51f568d x86/microcode: Use native CPUID to tickle out microcode revision

Ben.

Quoted 3 lines

I believe this is the background info Greg needed to figure out *why* you're doing this.

Signature · 3 lines
Ben Hutchings
Klipstein's 4th Law of Prototyping and Production:
                               A fail-safe circuit will destroy others.
#3 lore raw
Quoted 54 lines · Ben Hutchings
Quoted 9 lines · Borislav Petkov
Quoted 2 lines · Evalds Iodzevics

sync_core() always jums past cpuid instruction on 32 bit machines because data structure boot_cpu_data are not populated so early in boot.

I'm guessing because boot_cpu_data.cpuid_level is not properly set and very early code in head_32.S sets it to -1 temporarily until the highest CPUID level has been detected (or not).

But the microcode loading happens *before* that.

[...]

Quoted 22 lines

Hrm, the original patch of mine did use native_cpuid_eax():

4167709bbf82 ("x86/microcode/intel: Add a helper which gives the microcode revision")

but the backport:

commit 98cc1464cfd6edf9dc7fa96aaaf596aae952029b

Author: Borislav Petkov <[email protected]>
Date:   Mon Jan 9 12:41:45 2017 +0100

x86/microcode/intel: Add a helper which gives the microcode revision

commit 4167709bbf826512a52ebd6aafda2be104adaec9 upstream.

Since on Intel we're required to do CPUID(1) first, before reading the microcode revision MSR, let's add a special helper which does the required steps so that we don't forget to do them next time, when we want to read the microcode revision.

Quoted 7 lines

[bwh: Backported to 4.4:
- Don't touch prev_rev variable in apply_microcode()
- Keep using sync_core(), which will alway includes the necessary CPUID ^^^^^^^^^^^^^^^^^^^

decided to use sync_core() for whatever reason. Perhaps because the native_cpuid* things weren't there. Adding Ben to Cc.

This commit didn't introduce the call to native_cpuid_eax(), but only moved it. So I didn't think it made sense for the backport to change from sync_core().

If it's important to use native_cpuid_eax() then these older branches should presumably get backports of:

484d0e5c7943 x86/microcode/intel: Replace sync_core() with native_cpuid() f3e2a51f568d x86/microcode: Use native CPUID to tickle out microcode revision

Ben.

Quoted 8 lines
Quoted 3 lines

I believe this is the background info Greg needed to figure out *why* you're doing this.

--
Ben Hutchings
Klipstein's 4th Law of Prototyping and Production:
A fail-safe circuit will destroy others.

As far as i can tell, these two:

484d0e5c7943 x86/microcode/intel: Replace sync_core() with native_cpuid() f3e2a51f568d x86/microcode: Use native CPUID to tickle out microcode revision

are already made obsolete by:

98cc1464cfd6 x86/microcode/intel: Add a helper which gives the microcode revision

except that it still relied on sync_core() witch did not behave correctly on 32 bit machines

#4 lore raw
In reply to Evalds Iodzevics
Quoted 6 lines · Evalds Iodzevics

sync_core() always jums past cpuid instruction on 32 bit machines because data structure boot_cpu_data are not populated so early in boot.

It depends on commit 5dedade6dfa243c130b85d1e4daba6f027805033 for native_cpuid_reg(eax) definitions

Quoted 6 lines

This patch is for 4.4 but also should apply to 4.9

---
 arch/x86/include/asm/microcode_intel.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

<formletter>

This is not the correct way to submit patches for inclusion in the stable kernel tree. Please read:
https://www.kernel.org/doc/html/latest/process/stable-kernel-rules.html
for how to do this properly.

</formletter>

#5 lore raw

PATCH v2 1/1 x86/microcode/intel: replace sync_core() with native_cpuid_reg(eax)

In reply to Evalds Iodzevics

On Intel it is required to do CPUID(1) before reading the microcode revision MSR. Current code in 4.4 an 4.9 relies on sync_core() to call CPUID, unfortunately on 32 bit machines code inside sync_core() always jumps past CPUID instruction as it depends on data structure boot_cpu_data witch are not populated correctly so early in boot sequence.

It depends on:
commit 5dedade6dfa2 ("x86/CPU: Add native CPUID variants returning a single datum")

This patch is for 4.4 but also should apply to 4.9

---
 arch/x86/include/asm/microcode_intel.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/include/asm/microcode_intel.h +1 -1
index 90343ba50485..92ce9c8a508b 100644
@@ -60,7 +60,7 @@ static inline u32 intel_get_microcode_revision(void)
 	native_wrmsrl(MSR_IA32_UCODE_REV, 0);
 
 	/* As documented in the SDM: Do a CPUID 1 here */
-	sync_core();
+	native_cpuid_eax(1);
 
 	/* get the current revision from MSR 0x8B */
 	native_rdmsr(MSR_IA32_UCODE_REV, dummy, rev);
Signature · 1 line
2.17.4
#6 lore raw
Quoted 10 lines · Evalds Iodzevics

On Intel it is required to do CPUID(1) before reading the microcode revision MSR. Current code in 4.4 an 4.9 relies on sync_core() to call CPUID, unfortunately on 32 bit machines code inside sync_core() always jumps past CPUID instruction as it depends on data structure boot_cpu_data witch are not populated correctly so early in boot sequence.

It depends on:
commit 5dedade6dfa2 ("x86/CPU: Add native CPUID variants returning a single datum")

Quoted 6 lines

This patch is for 4.4 but also should apply to 4.9

---
 arch/x86/include/asm/microcode_intel.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Why are you not sending this to the stable mailing list like I have pointed out numerous times by sending you a link to _how_ to get a patch into the stable kernel trees?

Again, here it is:
https://www.kernel.org/doc/html/latest/process/stable-kernel-rules.html

Please follow that so that we can do this correctly.

thanks,

greg k-h

#7 lore raw
Quoted 23 lines · Greg KH
Quoted 10 lines · Evalds Iodzevics

On Intel it is required to do CPUID(1) before reading the microcode revision MSR. Current code in 4.4 an 4.9 relies on sync_core() to call CPUID, unfortunately on 32 bit machines code inside sync_core() always jumps past CPUID instruction as it depends on data structure boot_cpu_data witch are not populated correctly so early in boot sequence.

It depends on:
commit 5dedade6dfa2 ("x86/CPU: Add native CPUID variants returning a single datum")

Quoted 6 lines

This patch is for 4.4 but also should apply to 4.9

---
 arch/x86/include/asm/microcode_intel.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Why are you not sending this to the stable mailing list like I have pointed out numerous times by sending you a link to _how_ to get a patch into the stable kernel trees?

Quoted 8 lines

Again, here it is:
https://www.kernel.org/doc/html/latest/process/stable-kernel-rules.html

Please follow that so that we can do this correctly.

thanks,

greg k-h

Sorry, I might sound dumb here but should i just send it to [email protected] or try to tag it Cc: stable... in sign-off area, its quite confusing for newcomer.
Thanks for patience!

#8 lore raw
Quoted 27 lines · Evalds Iodzevics
Quoted 26 lines · Greg KH
Quoted 10 lines · Evalds Iodzevics

On Intel it is required to do CPUID(1) before reading the microcode revision MSR. Current code in 4.4 an 4.9 relies on sync_core() to call CPUID, unfortunately on 32 bit machines code inside sync_core() always jumps past CPUID instruction as it depends on data structure boot_cpu_data witch are not populated correctly so early in boot sequence.

It depends on:
commit 5dedade6dfa2 ("x86/CPU: Add native CPUID variants returning a single datum")

Quoted 6 lines

This patch is for 4.4 but also should apply to 4.9

---
 arch/x86/include/asm/microcode_intel.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Why are you not sending this to the stable mailing list like I have pointed out numerous times by sending you a link to _how_ to get a patch into the stable kernel trees?

Again, here it is:
https://www.kernel.org/doc/html/latest/process/stable-kernel-rules.html

Quoted 8 lines
Quoted 5 lines

Please follow that so that we can do this correctly.

thanks,

greg k-h

Sorry, I might sound dumb here but should i just send it to [email protected] or try to tag it Cc: stable... in sign-off area, its quite confusing for newcomer.

Yes, be sure to cc: the [email protected] list when you send patches that you want to have applied to a stable kernel tree.

thanks,

greg k-h