From: "Borislav Petkov (AMD)" <[email protected]>
We've had enough fun with the spectral chicken bit - name it what it
really does: it suppresses non-branch predictions.
Rename defines to make it clear that it is Zen2 only.
No functional changes.
Signed-off-by: Borislav Petkov (AMD) <[email protected]>
---
arch/x86/include/asm/msr-index.h | 6 ++++--
arch/x86/kernel/cpu/amd.c | 22 +++++++++++-----------
2 files changed, 15 insertions(+), 13 deletions(-)
diff --git a/arch/x86/include/asm/msr-index.h b/arch/x86/include/asm/msr-index.h
index ad35355ee43e..0094e6a1b553 100644
--- a/arch/x86/include/asm/msr-index.h
+++ b/arch/x86/include/asm/msr-index.h
@@ -625,8 +625,10 @@
/* Fam 17h MSRs */
#define MSR_F17H_IRPERF 0xc00000e9
-#define MSR_ZEN2_SPECTRAL_CHICKEN 0xc00110e3
-#define MSR_ZEN2_SPECTRAL_CHICKEN_BIT BIT_ULL(1)
+/* Fam 17h, Zen2 MSRs */
+#define MSR_ZEN2_DE_CFG2 0xc00110e3
+#define MSR_ZEN2_DE_CFG2_SUPPRESS_NOBR_PRED_BIT 1
+#define MSR_ZEN2_DE_CFG2_SUPPRESS_NOBR_PRED BIT(MSR_ZEN2_DE_CFG2_SUPPRESS_NOBR_PRED_BIT)
/* Fam 16h MSRs */
#define MSR_F16H_L2I_PERF_CTL 0xc0010230
diff --git a/arch/x86/kernel/cpu/amd.c b/arch/x86/kernel/cpu/amd.c
index 571abf808ea3..70a41088e900 100644
--- a/arch/x86/kernel/cpu/amd.c
+++ b/arch/x86/kernel/cpu/amd.c
@@ -860,23 +860,23 @@ static void init_amd_bd(struct cpuinfo_x86 *c)
clear_rdrand_cpuid_bit(c);
}
-void init_spectral_chicken(struct cpuinfo_x86 *c)
+void init_amd_zn2(struct cpuinfo_x86 *c)
{
#ifdef CONFIG_CPU_UNRET_ENTRY
u64 value;
/*
- * On Zen2 we offer this chicken (bit) on the altar of Speculation.
+ * Suppress speculation from the middle of a basic block, i.e.
+ * non-branch predictions.
*
- * This suppresses speculation from the middle of a basic block, i.e. it
- * suppresses non-branch predictions.
- *
- * We use STIBP as a heuristic to filter out Zen2 from the rest of F17H
+ * STIBP is used as a heuristic to filter out Zen2 from the rest
+ * of family 0x17.
*/
- if (!cpu_has(c, X86_FEATURE_HYPERVISOR) && cpu_has(c, X86_FEATURE_AMD_STIBP)) {
- if (!rdmsrl_safe(MSR_ZEN2_SPECTRAL_CHICKEN, &value)) {
- value |= MSR_ZEN2_SPECTRAL_CHICKEN_BIT;
- wrmsrl_safe(MSR_ZEN2_SPECTRAL_CHICKEN, value);
+ if (!cpu_has(c, X86_FEATURE_HYPERVISOR) &&
+ cpu_has(c, X86_FEATURE_AMD_STIBP)) {
+ if (!rdmsrl_safe(MSR_ZEN2_DE_CFG2, &value)) {
+ value |= MSR_ZEN2_DE_CFG2_SUPPRESS_NOBR_PRED;
+ wrmsrl_safe(MSR_ZEN2_DE_CFG2, value);
}
}
#endif
@@ -949,7 +949,7 @@ static void init_amd(struct cpuinfo_x86 *c)
case 0x12: init_amd_ln(c); break;
case 0x15: init_amd_bd(c); break;
case 0x16: init_amd_jg(c); break;
- case 0x17: init_spectral_chicken(c);
+ case 0x17: init_amd_zn2(c);
fallthrough;
case 0x19: init_amd_zn(c); break;
}
--
2.35.1
On Tue, Apr 25, 2023 at 09:50:24PM +0200, Borislav Petkov wrote:
> From: "Borislav Petkov (AMD)" <[email protected]>
>
> We've had enough fun with the spectral chicken bit - name it what it
> really does: it suppresses non-branch predictions.
>
> Rename defines to make it clear that it is Zen2 only.
>
> No functional changes.
>
> Signed-off-by: Borislav Petkov (AMD) <[email protected]>
> ---
> arch/x86/include/asm/msr-index.h | 6 ++++--
> arch/x86/kernel/cpu/amd.c | 22 +++++++++++-----------
> 2 files changed, 15 insertions(+), 13 deletions(-)
>
> diff --git a/arch/x86/include/asm/msr-index.h b/arch/x86/include/asm/msr-index.h
> index ad35355ee43e..0094e6a1b553 100644
> --- a/arch/x86/include/asm/msr-index.h
> +++ b/arch/x86/include/asm/msr-index.h
> @@ -625,8 +625,10 @@
> /* Fam 17h MSRs */
> #define MSR_F17H_IRPERF 0xc00000e9
>
> -#define MSR_ZEN2_SPECTRAL_CHICKEN 0xc00110e3
> -#define MSR_ZEN2_SPECTRAL_CHICKEN_BIT BIT_ULL(1)
> +/* Fam 17h, Zen2 MSRs */
> +#define MSR_ZEN2_DE_CFG2 0xc00110e3
> +#define MSR_ZEN2_DE_CFG2_SUPPRESS_NOBR_PRED_BIT 1
> +#define MSR_ZEN2_DE_CFG2_SUPPRESS_NOBR_PRED BIT(MSR_ZEN2_DE_CFG2_SUPPRESS_NOBR_PRED_BIT)
NAK, I had to write all this without *any* official communication from
AMD, mostly on hear-say.
This gets to be the spectral chicken forever more as punishment. Next
time AMD can try again, and if they manage to get their act together and
publish something before I get to write the code and invent a name for a
magical bit, they get to name it how they like.
On Tue, Apr 25, 2023 at 10:59:44PM +0200, Peter Zijlstra wrote:
> This gets to be the spectral chicken forever more as punishment. Next
> time AMD can try again, and if they manage to get their act together and
> publish something before I get to write the code and invent a name for a
> magical bit, they get to name it how they like.
It is unlikely that you'll have to do that again.
--
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquette
On Wed, Apr 26, 2023 at 11:14:44AM +0200, Thomas Gleixner wrote:
> That's a purely speculative assumption.
No, it's not.
--
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquette
On Tue, Apr 25 2023 at 23:15, Borislav Petkov wrote:
> On Tue, Apr 25, 2023 at 10:59:44PM +0200, Peter Zijlstra wrote:
>> This gets to be the spectral chicken forever more as punishment. Next
>> time AMD can try again, and if they manage to get their act together and
>> publish something before I get to write the code and invent a name for a
>> magical bit, they get to name it how they like.
>
> It is unlikely that you'll have to do that again.
That's a purely speculative assumption.