Skip to contentSkip to conversation

perf/x86/intel/uncore: Add Haswell-EP uncore support

DISCUSSION Dave Jones 3 participants 5 messages Nov 25, 2014 – Nov 27, 2014 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

Dave Jones

author
#1 lore raw
Quoted 7 lines
Commit:     e735b9db12d76d45f74aee78bd63bbd2f8f480e1
Author:     Yan, Zheng <[email protected]>
AuthorDate: Thu Sep 4 16:08:26 2014 -0700
Committer:  Ingo Molnar <[email protected]>
CommitDate: Wed Sep 24 14:48:21 2014 +0200

perf/x86/intel/uncore: Add Haswell-EP uncore support

This commit added some code which looks a bit fishy, and got flagged by coverity in yesterdays scan.

There are a few cases like this :

Quoted 4 lines
(quoted diff) +4 -0
+static void hswep_cbox_enable_event(struct intel_uncore_box *box,
+				  struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;

Review

Quoted 10 lines
(quoted diff) +10 -0
+	struct hw_perf_event_extra *reg1 = &hwc->extra_reg;
+
+	if (reg1->idx != EXTRA_REG_NONE) {
+		u64 filter = uncore_shared_reg_config(box, 0);
+		wrmsrl(reg1->reg, filter & 0xffffffff);
+		wrmsrl(reg1->reg + 1, filter >> 32);
+	}
+
+	wrmsrl(hwc->config_base, hwc->config | SNBEP_PMON_CTL_EN);
+}

given the definition of wrmsrl ..

#define wrmsrl(msr, val)	\
       native_write_msr((msr), (u32)((u64)(val)), (u32)((u64)(val) >> 32))

We can see that coverity got upset because it realised that the code is turned into constant expressions.

result_independent_of_operands: (u64)(filter & 4294967295U) >> 32 is 0 regardless of the values of its operands.
a function call.

result_independent_of_operands: (u64)(filter >> 32) >> 32 is 0 regardless of the values of its operands.

I couldn't quickly decipher which MSRs we're writing to here in the SDM, but are these really 64-bit wide registers that need 32 bits always set to zero ? I'm wondering if these should be just wrmsr's instead of wrmsrl's.

If someone more familiar with perf hw can let me know if this is safe, I'll dismiss the reports in coverity.

thanks,

Dave

#2 lore raw
Quoted 2 lines

I couldn't quickly decipher which MSRs we're writing to here in the SDM, but are these really 64-bit wide registers that need 32 bits always set to zero ?

Yes the filter MSRs are only 32bit.

Quoted 1 line

I'm wondering if these should be just wrmsr's instead of wrmsrl's.

Can we just shut up coverity please?

wrmsr() is stupid and error prone and should have never been added as an interface.

-Andi

Signature · 1 line
[email protected] -- Speaking for myself only
author
#3 lore raw
Quoted 4 lines · Andi Kleen
Quoted 2 lines

I couldn't quickly decipher which MSRs we're writing to here in the SDM, but are these really 64-bit wide registers that need 32 bits always set to zero ?

Yes the filter MSRs are only 32bit.

thanks for checking.

Quoted 3 lines
Quoted 1 line

I'm wondering if these should be just wrmsr's instead of wrmsrl's.

Can we just shut up coverity please?

I can manually dismiss them, I don't think the checker has the information to determine this for itself.

Dave

#4 lore raw
Quoted 6 lines
Quoted 3 lines
Quoted 1 line

I'm wondering if these should be just wrmsr's instead of wrmsrl's.

Can we just shut up coverity please?

I can manually dismiss them, I don't think the checker has the information to determine this for itself.

I suspect my out of line MSR patchkit would fix the warning as a side effect, because there would be a function call inbetween and the function always gets full u64.

https://lkml.org/lkml/2014/10/10/376

So far it hasn't been reviewed. I'll keep reposting.

-Andi

#5 lore raw
Quoted 7 lines · Andi Kleen
Quoted 6 lines
Quoted 3 lines
Quoted 1 line

I'm wondering if these should be just wrmsr's instead of wrmsrl's.

Can we just shut up coverity please?

I can manually dismiss them, I don't think the checker has the information to determine this for itself.

Quoted 7 lines

I suspect my out of line MSR patchkit would fix the warning as a side effect, because there would be a function call inbetween and the function always gets full u64.

https://lkml.org/lkml/2014/10/10/376

So far it hasn't been reviewed. I'll keep reposting.

And if you do so, can you please add a proper 0/n cover letter which explains the scope of the series. I'm tired of wading through N patches to figure out whether this is something I should care of or not.

Thanks,

tglx