perf doesn't work with kernel 3.1.0-1002-linaro-lt-mx5

[Paweł Moll]

Quickstart board

root@linaro-nano:~# uname -a
Linux linaro-nano 3.1.0-1002-linaro-lt-mx5 #7-Ubuntu PREEMPT Tue Nov 15 13:07:56 UTC 2011 armv7l armv7l armv7l GNU/Linux
root@linaro-nano:~# perf stat sleep 1
[ 28.489709] Unable to handle kernel NULL pointer dereference at virtual address 0000008c
[ 28.502323] pgd = d9534000
[ 28.507638] [0000008c] *pgd=c96cd831, *pte=00000000, *ppte=00000000
[ 28.517243] Internal error: Oops: 17 [#1] PREEMPT
[ 28.524366] Modules linked in: mma8450 input_polldev rtc_da9052 leds_da9052 da9052_onkey da9052_wdt
[ 28.536057] CPU: 0 Not tainted (3.1.0-1002-linaro-lt-mx5 #7-Ubuntu)
[ 28.545226] PC is at armpmu_event_init+0xc0/0x328
[ 28.552497] LR is at reserve_pmu+0x1c/0x2c
[ 28.559155] pc : [<80016e14>] lr : [<800164e8>] psr: 600f0013
[ 28.559161] sp : d966de18 ip : 00000000 fp : d966de84
[ 28.575828] r10: 00000000 r9 : 00000000 r8 : d9647d30
[ 28.583623] r7 : 00000000 r6 : 8084fca4 r5 : 8084fc18 r4 : d9647c00
[ 28.592759] r3 : 00000001 r2 : 00000001 r1 : 808aa324 r0 : 00000000
[ 28.601897] Flags: nZCv IRQs on FIQs on Mode SVC_32 ISA ARM Segment user
[ 28.611702] Control: 10c5387d Table: c9534019 DAC: 00000015
[ 28.620138] Process perf_3.1.0-1002 (pid: 1146, stack limit = 0xd966c2f0)
[ 28.629659] Stack: (0xd966de18 to 0xd966e000)
[ 28.636740] de00: 8085e028 00000000
[ 28.647673] de20: 00000000 00000002 8084fca8 8084fca4 d966de84 d966de40 8006aa04 80068a54
[ 28.658614] de40: 00000001 00000000 00000000 800b29ac 00000000 8002de70 00000000 d9647c00
[ 28.669563] de60: 8084fc18 808742e4 00000000 d9647d30 d8152b80 00000000 d966debc d966de88
[ 28.680525] de80: 800b2a84 80016d60 00000001 00000000 800b29ac d966dea0 80057850 8084fc18
[ 28.691468] dea0: d966df28 d9647c00 80e14490 00000000 d966defc d966dec0 800b2d44 800b29b8
[ 28.702435] dec0: 00000005 00000000 00000025 d8152f40 d966defc d8152b80 00000000 00000000
[ 28.713461] dee0: 00000000 ffffffff 00000000 ffffffff d966dfa4 d966df00 800b3658 800b2b08
[ 28.724484] df00: 00000000 00000000 00000000 00000010 d966df2c d966df20 8010a710 8010a5f0
[ 28.735494] df20: 00000008 00000000 00000000 00000048 00000000 00000000 00000000 00000000
[ 28.746513] df40: 00000000 00000000 00000003 00000000 00001003 00000000 00000000 00000000
[ 28.757531] df60: 00000000 00000000 00000000 00000000 001116ac 00000000 d966df94 00000000
[ 28.768568] df80: 001116ac 006cf450 0000016c 8000dc08 d966c000 00000000 00000000 d966dfa8
[ 28.779654] dfa0: 8000d980 800b32e0 00000000 001116ac 006cf458 0000047d ffffffff ffffffff
[ 28.790798] dfc0: 00000000 001116ac 006cf450 0000016c 00000004 00000000 00000000 00084cb8
[ 28.802003] dfe0: 7ee8a588 7ee8a578 00028c65 76ea79c0 200f0010 006cf458 00000000 00000000
[ 28.813299] [<80016e14>] (armpmu_event_init+0xc0/0x328) from [<800b2a84>] (perf_init_event+0xd8/0x150)
[ 28.825806] [<800b2a84>] (perf_init_event+0xd8/0x150) from [<800b2d44>] (perf_event_alloc+0x248/0x470)
[ 28.838382] [<800b2d44>] (perf_event_alloc+0x248/0x470) from [<800b3658>] (sys_perf_event_open+0x384/0x784)
[ 28.851398] [<800b3658>] (sys_perf_event_open+0x384/0x784) from [<8000d980>] (ret_fast_syscall+0x0/0x3c)
[ 28.864158] Code: 0a000002 e59f0254 eb166c05 ea00003f (e597308c)
[ 28.880969] ---[ end trace b9e247a5e3243127 ]---
Segmentation fault

[Will Deacon (will-deacon)]

Hi Pawel,

Do you have a pointer to the kernel sources for this?

Will

[Paweł Moll (pawel-moll)]

The kernel comes from this PPA: https://launchpad.net/~linaro-landing-team-freescale/+archive/public/+packages (the linux-linaro-lt-mx5-3.1 - 3.1.0-1002.8 packages)

I _suppose_ that this comes from that repo: http://git.linaro.org/gitweb?p=landing-teams/leb/freescale/kernel.git;a=summary but that's just a guess. The LT folk will (hopefully ;-) know...

[Scott Bambrough (scottb)]

Yes. There should be a tag for that release there.

Scott

Sent from my iPhone

On 2011-11-22, at 11:17, Paweł Moll <email address hidden> wrote:

> The kernel comes from this PPA: https://launchpad.net/~linaro-landing-
> team-freescale/+archive/public/+packages (the linux-linaro-lt-mx5-3.1 -
> 3.1.0-1002.8 packages)
>
> I _suppose_ that this comes from that repo:
> http://git.linaro.org/gitweb?p=landing-
> teams/leb/freescale/kernel.git;a=summary but that's just a guess. The LT
> folk will (hopefully ;-) know...
>
> --
> You received this bug notification because you are a member of Linaro
> Landing Team: Freescale, which is subscribed to the bug report.
> https://bugs.launchpad.net/bugs/893653
>
> Title:
> perf doesn't work with kernel 3.1.0-1002-linaro-lt-mx5
>
> Status in Linaro Freescale Landing Team Project:
> New
>
> Bug description:
> Quickstart board
>
> root@linaro-nano:~# uname -a
> Linux linaro-nano 3.1.0-1002-linaro-lt-mx5 #7-Ubuntu PREEMPT Tue Nov 15 13:07:56 UTC 2011 armv7l armv7l armv7l GNU/Linux
> root@linaro-nano:~# perf stat sleep 1
> [ 28.489709] Unable to handle kernel NULL pointer dereference at virtual address 0000008c
> [ 28.502323] pgd = d9534000
> [ 28.507638] [0000008c] *pgd=c96cd831, *pte=00000000, *ppte=00000000
> [ 28.517243] Internal error: Oops: 17 [#1] PREEMPT
> [ 28.524366] Modules linked in: mma8450 input_polldev rtc_da9052 leds_da9052 da9052_onkey da9052_wdt
> [ 28.536057] CPU: 0 Not tainted (3.1.0-1002-linaro-lt-mx5 #7-Ubuntu)
> [ 28.545226] PC is at armpmu_event_init+0xc0/0x328
> [ 28.552497] LR is at reserve_pmu+0x1c/0x2c
> [ 28.559155] pc : [<80016e14>] lr : [<800164e8>] psr: 600f0013
> [ 28.559161] sp : d966de18 ip : 00000000 fp : d966de84
> [ 28.575828] r10: 00000000 r9 : 00000000 r8 : d9647d30
> [ 28.583623] r7 : 00000000 r6 : 8084fca4 r5 : 8084fc18 r4 : d9647c00
> [ 28.592759] r3 : 00000001 r2 : 00000001 r1 : 808aa324 r0 : 00000000
> [ 28.601897] Flags: nZCv IRQs on FIQs on Mode SVC_32 ISA ARM Segment user
> [ 28.611702] Control: 10c5387d Table: c9534019 DAC: 00000015
> [ 28.620138] Process perf_3.1.0-1002 (pid: 1146, stack limit = 0xd966c2f0)
> [ 28.629659] Stack: (0xd966de18 to 0xd966e000)
> [ 28.636740] de00: 8085e028 00000000
> [ 28.647673] de20: 00000000 00000002 8084fca8 8084fca4 d966de84 d966de40 8006aa04 80068a54
> [ 28.658614] de40: 00000001 00000000 00000000 800b29ac 00000000 8002de70 00000000 d9647c00
> [ 28.669563] de60: 8084fc18 808742e4 00000000 d9647d30 d8152b80 00000000 d966debc d966de88
> [ 28.680525] de80: 800b2a84 80016d60 00000001 00000000 800b29ac d966dea0 80057850 8084fc18
> [ 28.691468] dea0: d966df28 d9647c00 80e14490 00000000 d966defc d966dec0 800b2d44 800b29b8
> [ 28.702435] dec0: 00000005 00000000 00000025 d8152f40 d966defc d8152b80 00000000 00000000
> [ 28.713461] dee0: 00000000 ffffffff 00000000 ffffffff d966dfa4 d966df00 800b3658 800b2b08
> [ 28.724484] df00: 00000000 00000000 00000000 00000010 d966df2c d966df20 8010a710 8010a5f0
> [ 28.735494] df2...

[Will Deacon (will-deacon)]

Hmm, do you know which of these two tags it corresponds to?:

linux-release-2011-10-1

lt-3.1-2011.11-0

[Will Deacon (will-deacon)]

Also - Pawel, can I have a dmesg please?

[Will Deacon (will-deacon)]

Scratch that - please give this patch a try:

diff --git a/arch/arm/kernel/perf_event.c b/arch/arm/kernel/perf_event.c
index e6e5d7c..9bc3683 100644
--- a/arch/arm/kernel/perf_event.c
+++ b/arch/arm/kernel/perf_event.c
@@ -396,6 +396,9 @@ armpmu_reserve_hardware(struct arm_pmu *armpmu)
        int i, err, irq, irqs;
        struct platform_device *pmu_device = armpmu->plat_device;

+ if (!pmu_device)
+ return -ENODEV;
+
        err = reserve_pmu(armpmu->type);
        if (err) {
                pr_warning("unable to reserve pmu\n");

[Will Deacon (will-deacon)]

Well that patch does fix a real bug, so I've included it in my perf/fixes branch and updated my outstanding pull request for RMK.

If it is causing your problem, then it means that the mx5 BSP isn't registering a platform_device for the PMU. If you want to use hardware perf counters, that's something you'll need to do.

[Scott Bambrough (scottb)]

On 11-11-22 12:44 PM, Will Deacon wrote:
> Hmm, do you know which of these two tags it corresponds to?:
>
> linux-release-2011-10-1

This one I believe.

Scott

[Paweł Moll (pawel-moll)]

Will is right about the platform device not being available:

root@linaro-nano:~# ls /sys/bus/platform/devices/ | grep pmu
root@linaro-nano:~#

[Will Deacon (will-deacon)]

grepping around in mach-mx5 didn't reveal what the IRQ numbers might be, so this needs to be added by somebody who knows the platform. The PMU has a DT binding, so you can use that if you like.

[Eric Miao (eric.y.miao)]

Checked, arm-pmu device not added. The attached patch will be a quick fix for this.

[Al Grant (al-grant)]

No sign of pmu platform device on PandaBoard (Linaro 11.10) either:

root@sdperf-board1:/hd/linux# ls /sys/bus/platform/devices | grep pmu
root@sdperf-board1:/hd/linux#

[Will Deacon (will-deacon)]

Eric - can we get a variant of your patch into mainline please?

[David Long (dave-long)]

We are once again missing the patch for Panda that adds support to route the PMU interrupts through the Cross Trigger Interface. That code performs the missing platform_device registration. I am in the process of up-leveling that code.

Presumably this interrupt is routed differently in the MX5 SoC.

[Tom Gall (tom-gall)]

When you run perf with the 11.12 RC1 hwpack/ubuntu-desktop on imx53, it will complain that perf_3.1.0 isn't installed. This isn't fatal. If you force it to run perf_3.0.0-14 it will run successfully. If you install linux-linaro-lt-xm5-tools-3.1.0 perf will run successfully.