[PATCH V13 - RESEND 02/10] arm64/perf: Add BRBE registers and fields

[James Clark]

On 31/07/2023 03:33, Anshuman Khandual wrote:
> 
> 
> On 7/28/23 22:22, James Clark wrote:
>>
>>
>> On 28/07/2023 17:20, Will Deacon wrote:
>>> On Tue, Jul 11, 2023 at 01:54:47PM +0530, Anshuman Khandual wrote:
>>>> This adds BRBE related register definitions and various other related field
>>>> macros there in. These will be used subsequently in a BRBE driver which is
>>>> being added later on.
>>>>
>>>> Cc: Catalin Marinas <catalin.marinas at arm.com>
>>>> Cc: Will Deacon <will at kernel.org>
>>>> Cc: Marc Zyngier <maz at kernel.org>
>>>> Cc: Mark Rutland <mark.rutland at arm.com>
>>>> Cc: linux-arm-kernel at lists.infradead.org
>>>> Cc: linux-kernel at vger.kernel.org
>>>> Tested-by: James Clark <james.clark at arm.com>
>>>> Reviewed-by: Mark Brown <broonie at kernel.org>
>>>> Signed-off-by: Anshuman Khandual <anshuman.khandual at arm.com>
>>>> ---
>>>>  arch/arm64/include/asm/sysreg.h | 103 +++++++++++++++++++++
>>>>  arch/arm64/tools/sysreg         | 158 ++++++++++++++++++++++++++++++++
>>>>  2 files changed, 261 insertions(+)
>>>>
>>>> diff --git a/arch/arm64/include/asm/sysreg.h b/arch/arm64/include/asm/sysreg.h
>>>> index b481935e9314..f95e30c13c8b 100644
>>>> --- a/arch/arm64/include/asm/sysreg.h
>>>> +++ b/arch/arm64/include/asm/sysreg.h
>>>> @@ -163,6 +163,109 @@
>>>>  #define SYS_DBGDTRTX_EL0		sys_reg(2, 3, 0, 5, 0)
>>>>  #define SYS_DBGVCR32_EL2		sys_reg(2, 4, 0, 7, 0)
>>>>  
>>>> +#define __SYS_BRBINFO(n)		sys_reg(2, 1, 8, ((n) & 0xf), ((((n) & 0x10)) >> 2 + 0))
>>>> +#define __SYS_BRBSRC(n)			sys_reg(2, 1, 8, ((n) & 0xf), ((((n) & 0x10)) >> 2 + 1))
>>>> +#define __SYS_BRBTGT(n)			sys_reg(2, 1, 8, ((n) & 0xf), ((((n) & 0x10)) >> 2 + 2))
>>>
>>> It's that time on a Friday but... aren't these macros busted? I think you
>>> need brackets before adding the offset, otherwise wouldn't, for example,
>>> target registers 0-15 all access info register 0 and __SYS_BRBTGT(16) would
>>> then start accessing source register 0?
>>>
>>> I'm surprised that the compiler doesn't warn about this, but even more
>>> surprised that you managed to test this.
>>>
>>> Please tell me I'm wrong!
>>>
>>> Will
>>
>> No I think you are right, it is wrong. Luckily there is already an
>> extraneous bracket so you you can fix it by moving one a place down:
>>
>>   sys_reg(2, 1, 8, ((n) & 0xf), ((((n) & 0x10) >> 2) + 2))
>>
>> It's interesting because the test [1] is doing quite a bit and looking
>> at the branch info, and that src and targets match up to function names.
>> I also manually looked at the branch buffers and didn't see anything
>> obviously wrong like things that looked like branch infos in the source
>> or target fields. Will have to take another look to see if it would be
>> possible for the test to catch this.
>>
>> James
>>
>> [1]:
>> https://gitlab.arm.com/linux-arm/linux-jc/-/commit/3a7ddce70c2daadb63fcc511de0a89055ca48b32
> 
> ((((n) & 0x10)) >> 2 + 2) ---> ((((n) & 0x10) >> 2) + 2)
> 
> The additional brackets are useful in explicitly telling the compiler but
> what it the compiler is just doing the right thing implicitly i.e computing
> the shifting operation before doing the offset addition. During testing, all
> those captured branch records looked alright. But that is no excuse, for not
> doing the right thing to begin with i.e adding explicit brackets. I will fix
> these in next version.

Are you sure? If you see the return value here, it's 0 until register
16, then it becomes 1:

  https://godbolt.org/z/c7zhbno3n

If you add the bracket it does actually change the return value.