Re: [kvm-unit-tests PATCH v10 08/15] powerpc: add pmu tests

["Nicholas Piggin" <npiggin@gmail.com>]

On Wed Jun 19, 2024 at 4:39 AM AEST, Thomas Huth wrote:
> On 12/06/2024 07.23, Nicholas Piggin wrote:
> > Add some initial PMU testing.
> > 
> > - PMC5/6 tests
> > - PMAE / PMI test
> > - BHRB basic tests
> > 
> > Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
> > ---
> ...
> > diff --git a/powerpc/pmu.c b/powerpc/pmu.c
> > new file mode 100644
> > index 000000000..bdc45e167
> > --- /dev/null
> > +++ b/powerpc/pmu.c
> > @@ -0,0 +1,562 @@
> ...
> > +static void test_pmc5_with_ldat(void)
> > +{
> > +	unsigned long pmc5_1, pmc5_2;
> > +	register unsigned long r4 asm("r4");
> > +	register unsigned long r5 asm("r5");
> > +	register unsigned long r6 asm("r6");
> > +	uint64_t val;
> > +
> > +	reset_mmcr0();
> > +	mtspr(SPR_PMC5, 0);
> > +	mtspr(SPR_MMCR0, mfspr(SPR_MMCR0) & ~(MMCR0_FC | MMCR0_FC56));
> > +	asm volatile(".rep 20 ; nop ; .endr" ::: "memory");
> > +	mtspr(SPR_MMCR0, mfspr(SPR_MMCR0) | (MMCR0_FC | MMCR0_FC56));
> > +	pmc5_1 = mfspr(SPR_PMC5);
> > +
> > +	val = 0xdeadbeef;
> > +	r4 = 0;
> > +	r5 = 0xdeadbeef;
> > +	r6 = 100;
> > +	reset_mmcr0();
> > +	mtspr(SPR_PMC5, 0);
> > +	mtspr(SPR_MMCR0, mfspr(SPR_MMCR0) & ~(MMCR0_FC | MMCR0_FC56));
> > +	asm volatile(".rep 10 ; nop ; .endr ; ldat %0,%3,0x10 ; .rep 10 ; nop ; .endr" : "=r"(r4), "+r"(r5), "+r"(r6) : "r"(&val) :"memory");
>
> Looks like older versions of Clang do not like this instruction:
>
>   /tmp/pmu-4fda98.s: Assembler messages:
>   /tmp/pmu-4fda98.s:1685: Error: unrecognized opcode: `ldat'
>   clang-13: error: assembler command failed with exit code 1 (use -v to see 
> invocation)
>
> Could you please work-around that issue?

Ah, just catching up with this -- thanks for working it out.

I will fix.

>
> Also, please break the very long line here. Thanks!

Sure.

Thanks,
Nick

>
> > +	mtspr(SPR_MMCR0, mfspr(SPR_MMCR0) | (MMCR0_FC | MMCR0_FC56));
> > +	pmc5_2 = mfspr(SPR_PMC5);
> > +	assert(r4 == 0xdeadbeef);
> > +	assert(val == 0xdeadbeef);
> > +
> > +	/* TCG does not count instructions around syscalls correctly */
> > +	report_kfail(host_is_tcg, pmc5_1 != pmc5_2 + 1,
> > +		     "PMC5 counts instructions with ldat");
> > +}
>
>   Thomas