Skip to contentSkip to conversation

x86/speculation/srbds: do not try to turn mitigation off when not supported

PATCH In review Thadeu Lima de Souza Cascardo 2 participants 4 messages Jun 9, 2020 – Jun 15, 2020

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86/speculation/srbds: do not try to turn mitigation off when not supported

When SRBDS is mitigated by TSX OFF, update_srbds_msr will still read and write to MSR_IA32_MCU_OPT_CTRL even when that is not supported by the microcode.

Checking for X86_FEATURE_SRBDS_CTRL as a CPU feature available makes more sense than checking for SRBDS_MITIGATION_UCODE_NEEDED as the found "mitigation".

---
 arch/x86/kernel/cpu/bugs.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/kernel/cpu/bugs.c +1 -1
index b6f887be440c..ee5bdca7fd30 100644
@@ -432,7 +432,7 @@ void update_srbds_msr(void)
 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
 		return;
 
-	if (srbds_mitigation == SRBDS_MITIGATION_UCODE_NEEDED)
+	if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
 		return;
 
 	rdmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
Signature · 1 line
2.25.1
maintainer
#1 lore raw
Quoted 1 line · Thadeu Lima de Souza Cascardo

When SRBDS is mitigated by TSX OFF, update_srbds_msr will still read and

Are you talking about this case in srbds_select_mitigation():

        if ((ia32_cap & ARCH_CAP_MDS_NO) && !boot_cpu_has(X86_FEATURE_RTM))
                srbds_mitigation = SRBDS_MITIGATION_TSX_OFF;

?

and you have a system which:

* Check to see if this is one of the MDS_NO systems supporting
* TSX that are only exposed to SRBDS when TSX is enabled.

i.e., no SRBDS microcode for it and the fix is to disable TSX?

If so, I think the right fix should be to do:

	if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
		return;

in update_srbds_msr() with a comment above it explaining why that check is being done.

Hmmm.

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
#2 lore raw
Quoted 10 lines · Borislav Petkov
Quoted 1 line · Thadeu Lima de Souza Cascardo

When SRBDS is mitigated by TSX OFF, update_srbds_msr will still read and

Are you talking about this case in srbds_select_mitigation():

        if ((ia32_cap & ARCH_CAP_MDS_NO) && !boot_cpu_has(X86_FEATURE_RTM))
                srbds_mitigation = SRBDS_MITIGATION_TSX_OFF;

?

That's the case that it hits, correct.

Quoted 6 lines

and you have a system which:

* Check to see if this is one of the MDS_NO systems supporting
* TSX that are only exposed to SRBDS when TSX is enabled.

i.e., no SRBDS microcode for it and the fix is to disable TSX?

It was booted without the microcode update. There was microcode available, but systems may be booted without it, thus causing warnings due to the MSR read/write.

Quoted 8 lines

If so, I think the right fix should be to do:

	if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
		return;

in update_srbds_msr() with a comment above it explaining why that check is being done.

That's exactly the fix in the patch I sent, right? Do you want me to resend with a comment, then?

Thanks.
Cascardo.

Quoted 8 lines

Hmmm.

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
maintainer
#3 lore raw
Quoted 3 lines · Thadeu Lima de Souza Cascardo

It was booted without the microcode update. There was microcode available, but systems may be booted without it, thus causing warnings due to the MSR read/write.

Right.

Quoted 2 lines

That's exactly the fix in the patch I sent, right? Do you want me to resend with a comment, then?

Your patch replaced

- if (srbds_mitigation == SRBDS_MITIGATION_UCODE_NEEDED)

Thinking about this more, I think the proper thing to do is this:

---

arch/x86/kernel/cpu/bugs.c +2 -2
index 0b71970d2d3d..ce2931563f8f 100644
@@ -432,14 +432,14 @@ void update_srbds_msr(void)
 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
 		return;
 
-	if (srbds_mitigation == SRBDS_MITIGATION_UCODE_NEEDED)
+	if (srbds_mitigation == SRBDS_MITIGATION_UCODE_NEEDED ||
+	    srbds_mitigation == SRBDS_MITIGATION_TSX_OFF)
 		return;
 
 	rdmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
 
 	switch (srbds_mitigation) {
 	case SRBDS_MITIGATION_OFF:
-	case SRBDS_MITIGATION_TSX_OFF:
 		mcu_ctrl |= RNGDS_MITG_DIS;
 		break;
 	case SRBDS_MITIGATION_FULL:

---

because looking at:

7e5b3c267d25 ("x86/speculation: Add Special Register Buffer Data Sampling (SRBDS) mitigation")

it says:

While it is present on all affected CPU models, the microcode mitigation is not needed on models that enumerate ARCH_CAPABILITIES[MDS_NO] in the cases where TSX is not supported or has been disabled with TSX_CTRL.

which could be also understood as "there won't be microcode for those CPUs and thus MSR_IA32_MCU_OPT_CTRL won't be there."

Because if that is the case, then SRBDS_MITIGATION_TSX_OFF means the MSR is not there and therefore we should not touch it.

And you've actually shown that without the microcode loaded, you can have a system which is MDS_NO and which hasn't generated MSR_IA32_MCU_OPT_CTRL (because the microcode is not loaded) and thus can't touch said MSR.

Mark?

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette