Re: WARNING: CPU: 1 PID: 1 at kernel/events/core.c:7825 perf_pmu_register+0x385/0x390

[Peter Zijlstra <peterz@infradead.org>]

On Sat, Apr 23, 2016 at 03:03:22PM +0200, Borislav Petkov wrote:
> Yo,
> 
> did the fix for this go anywhere? I'm still seeing it on rc4+tip/master:
> 
> [    0.760493] AMD Power PMU detected
> [    0.760689] LVT offset 0 assigned for vector 0x400
> [    0.761072] perf: AMD IBS detected (0x000007ff)
> [    0.761340] ------------[ cut here ]------------
> [    0.761571] WARNING: CPU: 1 PID: 1 at kernel/events/core.c:7825 perf_pmu_register+0x385/0x390
> [    0.761909] Modules linked in:
> [    0.762093] CPU: 1 PID: 1 Comm: swapper/0 Not tainted 4.6.0-rc4+ #1
> [    0.762331] Hardware name: HP HP EliteBook 745 G3/807E, BIOS N73 Ver. 01.08 01/28/2016
> [    0.762674]  0000000000000000 ffffffff812dafa9 0000000000000000 0000000000000000
> [    0.763185]  ffffffff810589ac ffffffff81a0fbe0 000000000000000a ffffffff8172ee85
> [    0.763672]  0000000000000000 ffff88042bdf9130 000000000008b000 ffffffff8112ff55
> [    0.764101] Call Trace:
> [    0.764240]  [<ffffffff812dafa9>] ? dump_stack+0x5c/0x83
> [    0.764454]  [<ffffffff810589ac>] ? __warn+0xec/0x110
> [    0.764680]  [<ffffffff8112ff55>] ? perf_pmu_register+0x385/0x390
> [    0.764956]  [<ffffffff81b1b51d>] ? msr_init+0xbe/0xbe
> [    0.765202]  [<ffffffff81b1b5f1>] ? amd_iommu_pc_init+0xd4/0x141
> [    0.765475]  [<ffffffff810003ef>] ? do_one_initcall+0xaf/0x200
> [    0.765718]  [<ffffffff810768eb>] ? parse_args+0x2ab/0x4c0
> [    0.765935]  [<ffffffff81b18fe1>] ? kernel_init_freeable+0x111/0x190
> [    0.766175]  [<ffffffff815df02a>] ? kernel_init+0xa/0x100
> [    0.766389]  [<ffffffff815eaa22>] ? ret_from_fork+0x22/0x40
> [    0.766620]  [<ffffffff815df020>] ? rest_init+0x90/0x90
> [    0.766885] ---[ end trace 9285cdb6cf96a9b2 ]---
> [    0.767120] perf: amd_iommu: Detected. (0 banks, 0 counters/bank)
> 

Oh, cute there's two different ones.

31d50c551e30 ("perf/x86/amd/uncore: Do not register a task ctx for uncore PMUs")

Doth the below fixeth thingies?

---
Subject: perf/amd/iommu: Do not register a task ctx for uncore like PMUs

The new sanity check introduced by:

  26657848502b ("perf/core: Verify we have a single perf_hw_context PMU")

... triggered on the AMD IOMMU driver.

IOMMUs are not per logical CPU, they cannot have per-task counters. Fix it.

Cc: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Cc: Joerg Roedel <jroedel@suse.de>
Reported-by: Borislav Petkov <bp@alien8.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
 arch/x86/events/amd/iommu.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/arch/x86/events/amd/iommu.c b/arch/x86/events/amd/iommu.c
index 40625ca7a190..6011a573dd64 100644
--- a/arch/x86/events/amd/iommu.c
+++ b/arch/x86/events/amd/iommu.c
@@ -474,6 +474,7 @@ static __init int _init_perf_amd_iommu(
 
 static struct perf_amd_iommu __perf_iommu = {
 	.pmu = {
+		.task_ctx_nr    = perf_invalid_context,
 		.event_init	= perf_iommu_event_init,
 		.add		= perf_iommu_add,
 		.del		= perf_iommu_del,