Skip to contentSkip to conversation

x86/AMD: extend Zenbleed check to models "good" ucode isn't known for

PATCH In review Jan Beulich 2 participants 3 messages Aug 22, 2023

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86/AMD: extend Zenbleed check to models "good" ucode isn't known for

Reportedly the AMD Custom APU 0405 found on SteamDeck, models 0x90 and 0x91, (quoting the respective Linux commit) is similarly affected. Put another instance of our Zen1 vs Zen2 distinction checks in amd_check_zenbleed(), forcing use of the chickenbit irrespective of ucode version (building upon real hardware never surfacing a version of 0xffffffff).

xen/arch/x86/cpu/amd.c +7 -3
@@ -936,10 +936,14 @@ void amd_check_zenbleed(void)
 	case 0xa0 ... 0xaf: good_rev = 0x08a00008; break;
 	default:
 		/*
-		 * With the Fam17h check above, parts getting here are Zen1.
-		 * They're not affected.
+		 * With the Fam17h check above, most parts getting here are
+		 * Zen1.  They're not affected.  Assume Zen2 ones making it
+		 * here are affected regardless of microcode version.
 		 */
-		return;
+		if (!boot_cpu_has(X86_FEATURE_AMD_STIBP))
+			return;
+		good_rev = ~0u;
+		break;
 	}
 
 	rdmsrl(MSR_AMD64_DE_CFG, val);
#1 lore raw
Quoted 12 lines · Jan Beulich

Reportedly the AMD Custom APU 0405 found on SteamDeck, models 0x90 and 0x91, (quoting the respective Linux commit) is similarly affected. Put another instance of our Zen1 vs Zen2 distinction checks in amd_check_zenbleed(), forcing use of the chickenbit irrespective of ucode version (building upon real hardware never surfacing a version of 0xffffffff).

xen/arch/x86/cpu/amd.c void amd_check_zenbleed(void) · line 941

Quoted 8 lines
xen/arch/x86/cpu/amd.c +3 -2
@@ -936,10 +936,14 @@ void amd_check_zenbleed(void)
 	case 0xa0 ... 0xaf: good_rev = 0x08a00008; break;
 	default:
 		/*
-		 * With the Fam17h check above, parts getting here are Zen1.
-		 * They're not affected.
+		 * With the Fam17h check above, most parts getting here are
+		 * Zen1.  They're not affected.  Assume Zen2 ones making it
+		 * here are affected regardless of microcode version.

It's not really "assume Zen2 are vulnerable".  All Zen2 *are* vulnerable, but we keep on finding new CPUs that AMD did for special circumstances and haven't documented in their model lists.

Furthermore, there needs to be another sentence:

"Because we still don't have an correct authoritative list of Zen1 vs Zen2 by model number, use STIBP as a heuristic to distinguish."

Or something like this.  It is important to state that STIBP is our model-heuristic here.

With some kind of note explaining what's going on, Reviewed-by: Andrew Cooper <[email protected]>

xen/arch/x86/cpu/amd.c void amd_check_zenbleed(void) · line 945

Quoted 5 lines
xen/arch/x86/cpu/amd.c +3 -1
 		 */
-		return;
+		if (!boot_cpu_has(X86_FEATURE_AMD_STIBP))
+			return;
+		good_rev = ~0u;

While I hate to review like this, someone is going to come along and swap this u for U for MISRA reasons.  Probably best to adjust it now.

~Andrew

Quoted 4 lines
xen/arch/x86/cpu/amd.c +1 -0
+		break;
 	}
 
 	rdmsrl(MSR_AMD64_DE_CFG, val);
#2 lore raw
Quoted 10 lines · Andrew Cooper
Quoted 9 lines · Jan Beulich

Reportedly the AMD Custom APU 0405 found on SteamDeck, models 0x90 and 0x91, (quoting the respective Linux commit) is similarly affected. Put another instance of our Zen1 vs Zen2 distinction checks in amd_check_zenbleed(), forcing use of the chickenbit irrespective of ucode version (building upon real hardware never surfacing a version of 0xffffffff).

Quoted 13 lines

xen/arch/x86/cpu/amd.c void amd_check_zenbleed(void) · line 941

Quoted 11 lines
xen/arch/x86/cpu/amd.c +3 -2
@@ -936,10 +936,14 @@ void amd_check_zenbleed(void)
 	case 0xa0 ... 0xaf: good_rev = 0x08a00008; break;
 	default:
 		/*
-		 * With the Fam17h check above, parts getting here are Zen1.
-		 * They're not affected.
+		 * With the Fam17h check above, most parts getting here are
+		 * Zen1.  They're not affected.  Assume Zen2 ones making it
+		 * here are affected regardless of microcode version.

It's not really "assume Zen2 are vulnerable".

But that's also not what the comment says. It says "regardless of microcode version".

Quoted 11 lines

  All Zen2 *are*
vulnerable, but we keep on finding new CPUs that AMD did for special circumstances and haven't documented in their model lists.

Furthermore, there needs to be another sentence:

"Because we still don't have an correct authoritative list of Zen1 vs Zen2 by model number, use STIBP as a heuristic to distinguish."

Or something like this.  It is important to state that STIBP is our model-heuristic here.

Will add.

Quoted 2 lines

With some kind of note explaining what's going on, Reviewed-by: Andrew Cooper <[email protected]>

Thanks.

Quoted 8 lines

xen/arch/x86/cpu/amd.c void amd_check_zenbleed(void) · line 945

Quoted 5 lines
xen/arch/x86/cpu/amd.c +3 -1
 		 */
-		return;
+		if (!boot_cpu_has(X86_FEATURE_AMD_STIBP))
+			return;
+		good_rev = ~0u;

While I hate to review like this, someone is going to come along and swap this u for U for MISRA reasons.  Probably best to adjust it now.

Oh, right, will do.

Jan