[v3] target/ppc: Assorted ppc target fixes

[PATCH v3 2/9] target/ppc: Fix PMU MMCR0[PMCjCE] bit in hflags calculation

Posted by Nicholas Piggin 2 years, 8 months ago

A store to MMCR0 with PMCjCE=1 fails to update hflags correctly and
results in hflags mismatch:

  qemu: fatal: TCG hflags mismatch (current:0x2408003d rebuilt:0x240a003d)

This can be reproduced by running perf on a recent machine.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
---
Since v2: new patch.

 target/ppc/power8-pmu.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/target/ppc/power8-pmu.c b/target/ppc/power8-pmu.c
index 64a64865d7..29e0012ed6 100644
--- a/target/ppc/power8-pmu.c
+++ b/target/ppc/power8-pmu.c
@@ -236,14 +236,16 @@ void helper_store_mmcr0(CPUPPCState *env, target_ulong value)
 {
     bool hflags_pmcc0 = (value & MMCR0_PMCC0) != 0;
     bool hflags_pmcc1 = (value & MMCR0_PMCC1) != 0;
+    bool hflags_pmcjce = (value & MMCR0_PMCjCE) != 0;
 
     pmu_update_cycles(env);
 
     env->spr[SPR_POWER_MMCR0] = value;
 
-    /* MMCR0 writes can change HFLAGS_PMCC[01] and HFLAGS_INSN_CNT */
+    /* MMCR0 writes can change HFLAGS_PMCC[01], PMCjCE, and HFLAGS_INSN_CNT */
     env->hflags = deposit32(env->hflags, HFLAGS_PMCC0, 1, hflags_pmcc0);
     env->hflags = deposit32(env->hflags, HFLAGS_PMCC1, 1, hflags_pmcc1);
+    env->hflags = deposit32(env->hflags, HFLAGS_PMCJCE, 1, hflags_pmcjce);
 
     pmu_update_summaries(env);
 
-- 
2.40.1

Re: [PATCH v3 2/9] target/ppc: Fix PMU MMCR0[PMCjCE] bit in hflags calculation

Posted by Daniel Henrique Barboza 2 years, 8 months ago


On 5/15/23 06:26, Nicholas Piggin wrote:
> A store to MMCR0 with PMCjCE=1 fails to update hflags correctly and
> results in hflags mismatch:
> 
>    qemu: fatal: TCG hflags mismatch (current:0x2408003d rebuilt:0x240a003d)
> 
> This can be reproduced by running perf on a recent machine.
> 
> Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
> ---

Fixes: c2eff582a32f ("target/ppc: PMU basic cycle count for pseries TCG")

(not sure why I didn't hit this back in 2021)


Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>



> Since v2: new patch.
> 
>   target/ppc/power8-pmu.c | 4 +++-
>   1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/target/ppc/power8-pmu.c b/target/ppc/power8-pmu.c
> index 64a64865d7..29e0012ed6 100644
> --- a/target/ppc/power8-pmu.c
> +++ b/target/ppc/power8-pmu.c
> @@ -236,14 +236,16 @@ void helper_store_mmcr0(CPUPPCState *env, target_ulong value)
>   {
>       bool hflags_pmcc0 = (value & MMCR0_PMCC0) != 0;
>       bool hflags_pmcc1 = (value & MMCR0_PMCC1) != 0;
> +    bool hflags_pmcjce = (value & MMCR0_PMCjCE) != 0;
>   
>       pmu_update_cycles(env);
>   
>       env->spr[SPR_POWER_MMCR0] = value;
>   
> -    /* MMCR0 writes can change HFLAGS_PMCC[01] and HFLAGS_INSN_CNT */
> +    /* MMCR0 writes can change HFLAGS_PMCC[01], PMCjCE, and HFLAGS_INSN_CNT */
>       env->hflags = deposit32(env->hflags, HFLAGS_PMCC0, 1, hflags_pmcc0);
>       env->hflags = deposit32(env->hflags, HFLAGS_PMCC1, 1, hflags_pmcc1);
> +    env->hflags = deposit32(env->hflags, HFLAGS_PMCJCE, 1, hflags_pmcjce);
>   
>       pmu_update_summaries(env);
>

Re: [PATCH v3 2/9] target/ppc: Fix PMU MMCR0[PMCjCE] bit in hflags calculation

Posted by Nicholas Piggin 2 years, 8 months ago

On Tue May 16, 2023 at 7:32 PM AEST, Daniel Henrique Barboza wrote:
>
>
> On 5/15/23 06:26, Nicholas Piggin wrote:
> > A store to MMCR0 with PMCjCE=1 fails to update hflags correctly and
> > results in hflags mismatch:
> > 
> >    qemu: fatal: TCG hflags mismatch (current:0x2408003d rebuilt:0x240a003d)
> > 
> > This can be reproduced by running perf on a recent machine.
> > 
> > Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
> > ---
>
> Fixes: c2eff582a32f ("target/ppc: PMU basic cycle count for pseries TCG")

Or is it this one? 0625c7760d54 ("target/ppc: do not call
hreg_compute_hflags() in helper_store_mmcr0()")

Ah, neither! It looks like 8b3d1c49a9f0 ("target/ppc: Add new PMC
HFLAGS"). But that shows I have probably missed HFLAGS_PMC_OTHER
here.

Let me do a bit more investigation and send an updated patch if
necessary.

Thanks,
Nick

>
> (not sure why I didn't hit this back in 2021)
>
>
> Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>

Re: [PATCH v3 2/9] target/ppc: Fix PMU MMCR0[PMCjCE] bit in hflags calculation

Posted by Daniel Henrique Barboza 2 years, 8 months ago


On 5/16/23 07:44, Nicholas Piggin wrote:
> On Tue May 16, 2023 at 7:32 PM AEST, Daniel Henrique Barboza wrote:
>>
>>
>> On 5/15/23 06:26, Nicholas Piggin wrote:
>>> A store to MMCR0 with PMCjCE=1 fails to update hflags correctly and
>>> results in hflags mismatch:
>>>
>>>     qemu: fatal: TCG hflags mismatch (current:0x2408003d rebuilt:0x240a003d)
>>>
>>> This can be reproduced by running perf on a recent machine.
>>>
>>> Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
>>> ---
>>
>> Fixes: c2eff582a32f ("target/ppc: PMU basic cycle count for pseries TCG")
> 
> Or is it this one? 0625c7760d54 ("target/ppc: do not call
> hreg_compute_hflags() in helper_store_mmcr0()")
> 
> Ah, neither! It looks like 8b3d1c49a9f0 ("target/ppc: Add new PMC
> HFLAGS"). But that shows I have probably missed HFLAGS_PMC_OTHER
> here.
> 
> Let me do a bit more investigation and send an updated patch if
> necessary.


Sure, let's hold this one for now.


Daniel

> 
> Thanks,
> Nick
> 
>>
>> (not sure why I didn't hit this back in 2021)
>>
>>
>> Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>