Re: [PATCH v2 14/14] perf arm-spe: Add support for ARMv8.3-SPE

From: André Przywara
Date: Wed Oct 21 2020 - 05:27:01 EST


On 21/10/2020 06:10, Leo Yan wrote:

Hi,

> On Tue, Oct 20, 2020 at 10:54:44PM +0100, Andr� Przywara wrote:
>> On 29/09/2020 14:39, Leo Yan wrote:
>>
>> Hi,
>>
>>> From: Wei Li <liwei391@xxxxxxxxxx>
>>>
>>> This patch is to support Armv8.3 extension for SPE, it adds alignment
>>> field in the Events packet and it supports the Scalable Vector Extension
>>> (SVE) for Operation packet and Events packet with two additions:
>>>
>>> - The vector length for SVE operations in the Operation Type packet;
>>> - The incomplete predicate and empty predicate fields in the Events
>>> packet.
>>>
>>> Signed-off-by: Wei Li <liwei391@xxxxxxxxxx>
>>> Signed-off-by: Leo Yan <leo.yan@xxxxxxxxxx>
>>> ---
>>> .../arm-spe-decoder/arm-spe-pkt-decoder.c | 84 ++++++++++++++++++-
>>> .../arm-spe-decoder/arm-spe-pkt-decoder.h | 6 ++
>>> 2 files changed, 87 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/tools/perf/util/arm-spe-decoder/arm-spe-pkt-decoder.c b/tools/perf/util/arm-spe-decoder/arm-spe-pkt-decoder.c
>>> index 05a4c74399d7..3ec381fddfcb 100644
>>> --- a/tools/perf/util/arm-spe-decoder/arm-spe-pkt-decoder.c
>>> +++ b/tools/perf/util/arm-spe-decoder/arm-spe-pkt-decoder.c
>>> @@ -342,14 +342,73 @@ int arm_spe_pkt_desc(const struct arm_spe_pkt *packet, char *buf,
>>> return ret;
>>> }
>>> }
>>> + if (idx > 2) {
>>
>> As I mentioned in the other patch, I doubt this extra comparison is
>> useful. Does that protect us from anything?
>
> It's the same reason with Event packet which have explained for replying
> patch 10, the condition is to respect the SPE specifiction:
>
> E[11], byte 1, bit [11], when SZ == 0b10 , or SZ == 0b11
> Alignment.
> ...
> Otherwise this bit reads-as-zero.
>
> So we gives higher priority for checking payload size than the Event
> bit setting; if you have other thinking for this, please let me know.

Ah, thanks for pointing this out. It looks like a bug in the manual
then, because I don't see why bit 11 should be any different from bits
[10:8] and bits [15:12] in this respect. And in the diagrams above you
clearly see bit 11 being shown even when SZ == 0b01.

I will try to follow this up here.

>>> + if (payload & SPE_EVT_PKT_ALIGNMENT) {
>>
>> Mmh, but this is bit 11, right?
>
> Yes.
>
>> So would need to go into the (idx > 1)
>> section (covering bits 8-15)? Another reason to ditch this comparison above.
>
> As has explained in patch 10, idx is not the same thing with "sz"
> field; "idx" stands for payload length in bytes, so:
>
> idx = 1 << sz
>
> The spec defines the sz is 2 or 3, thus idx is 4 or 8; so this is why
> here use the condition "(idx > 2)".
>
> I think here need to refine code for more explict expression so can
> avoid confusion. So I think it's better to condition such like:
>
> if (payload_len >= 4) {

Yes, that would be (or have been) more helpful, but as mentioned in the
other patch, I'd rather see those comparisons go entirely.

Cheers,
Andre

>
> ...
>
> }
>
>>> + ret = snprintf(buf, buf_len, " ALIGNMENT");
>>> + if (ret < 0)
>>> + return ret;
>>> + buf += ret;
>>> + blen -= ret;
>>
>> Shouldn't we use the new arm_spe_pkt_snprintf() function here as well?
>> Or is there a reason that this doesn't work?
>
> Goot point. Will change to use arm_spe_pkt_snprintf().
>
>>> + }
>>> + if (payload & SPE_EVT_PKT_SVE_PARTIAL_PREDICATE) {
>>> + ret = snprintf(buf, buf_len, " SVE-PARTIAL-PRED");
>>> + if (ret < 0)
>>> + return ret;
>>> + buf += ret;
>>> + blen -= ret;
>>> + }
>>> + if (payload & SPE_EVT_PKT_SVE_EMPTY_PREDICATE) {
>>> + ret = snprintf(buf, buf_len, " SVE-EMPTY-PRED");
>>> + if (ret < 0)
>>> + return ret;
>>> + buf += ret;
>>> + blen -= ret;
>>> + }
>>> + }
>>> +
>>> return buf_len - blen;
>>>
>>> case ARM_SPE_OP_TYPE:
>>> switch (idx) {
>>> case SPE_OP_PKT_HDR_CLASS_OTHER:
>>> - return arm_spe_pkt_snprintf(&buf, &blen,
>>> - payload & SPE_OP_PKT_OTHER_SUBCLASS_COND ?
>>> - "COND-SELECT" : "INSN-OTHER");
>>> + if ((payload & SPE_OP_PKT_OTHER_SVE_SUBCLASS_MASK) ==
>>> + SPE_OP_PKT_OTHER_SUBCLASS_SVG_OP) {
>>> +
>>> + ret = arm_spe_pkt_snprintf(&buf, &blen, "SVE-OTHER");
>>> + if (ret < 0)
>>> + return ret;
>>> +
>>> + /* Effective vector length: step is 32 bits */
>>> + ret = arm_spe_pkt_snprintf(&buf, &blen, " EVLEN %d",
>>> + 32 << ((payload & SPE_OP_PKT_SVE_EVL_MASK) >>
>>> + SPE_OP_PKT_SVE_EVL_SHIFT));
>>
>> Can you move this into a macro, and add a comment about how this works?
>> People might get confused over the "32 << something".
>
> Yeah, will refine for it.
>
> Thanks,
> Leo
>