Re: [linus:master] [perf vendor events intel] 31efbec216: perf-stat-tests.UNC_IIO_NUM_TGT_MATCHED_REQ_OF_CPU.fail

[Ian Rogers]

On Wed, Nov 4, 2026 at 5:47 PM kernel test robot <yi1.lai@xxxxxxxxx> wrote:
>
>
>
> Hello,
>
> kernel test robot noticed "perf-stat-tests.UNC_IIO_NUM_TGT_MATCHED_REQ_OF_CPU.fail" on:
>
> commit: 31efbec2161d2d31b020b086958e41f35f070a38 ("perf vendor events intel: Update sapphirerapids events from 1.39 to 1.40") https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git master
>
> [test failed on linux-next/master 7649ba73f9e7e1e04d0b46213e39c79ae09dd55c]
>
> in testcase: perf-stat-tests
> with following parameters:
>
>
> config: x86_64-rhel-9.4-bpf
> compiler: gcc-14
> test machine: 224 threads 2 sockets Intel(R) Xeon(R) Platinum 8480CTDX (Sapphire Rapids) with 256G memory
>
> (please refer to attached dmesg/kmsg for entire log/backtrace)
>
>
> If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
> | Reported-by: kernel test robot <yi1.lai@xxxxxxxxx>
> | Closes: https://lore.kernel.org/oe-lkp/202611050912.2cdf1ff5-lkp@xxxxxxxxx
>
>
> EventName: UNC_IIO_NUM_TGT_MATCHED_REQ_OF_CPU
> 2026-10-05 17:15:24 perf stat -e UNC_IIO_NUM_TGT_MATCHED_REQ_OF_CPU -a sleep 0.1
> event syntax error: 'UNC_IIO_NUM_TGT_MATCHED_REQ_OF_CPU'
>                      \___ Bad event name
>
> Unable to find event on a PMU of 'UNC_IIO_NUM_TGT_MATCHED_REQ_OF_CPU'

Ok, this is an ugly issue. When we fixed the issue so that umasks of 0
weren't dropped, we added new umasks:
https://web.git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit?id=31efbec2161d2d31b020b086958e41f35f070a38
```
...
@@ -1923,6 +1974,7 @@
         "PerPkg": "1",
         "PortMask": "0x0000",
         "PublicDescription": "UNC_IIO_NUM_TGT_MATCHED_REQ_OF_CPU",
+        "UMask": "0x7000000",
...
```
The umask calculation is odd:
https://github.com/intel/perfmon/blob/main/scripts/create_perf_json.py#L361
```
        # Combine the umask with the umask_ext. We don't know the
        # width of the umask as we concatenate the strings, removing
        # "0x"
        if umask_ext:
            self.umask = umask_ext + self.umask[2:] if self.umask else
umask_ext + "00"
```
For Sapphire Rapids the UMaskExt is:
https://github.com/intel/perfmon/blob/main/SPR/events/sapphirerapids_uncore_experimental.json#L17422
```
...
    {
      "Unit": "IIO",
      "EventCode": "0x8f",
      "UMask": "0x00",
      "PortMask": "0x0000",
      "FCMask": "0x00",
      "UMaskExt": "0x00070000",
      "EventName": "UNC_IIO_NUM_TGT_MATCHED_REQ_OF_CPU",
      "BriefDescription": "ITC address map 1",
      "PublicDescription": "UNC_IIO_NUM_TGT_MATCHED_REQ_OF_CPU",
      "Counter": "0,1,2,3",
      "ELLC": "0",
      "Filter": "na",
      "ExtSel": "0",
      "Deprecated": "0",
      "FILTER_VALUE": "0",
      "CounterType": "PGMABLE"
    },
...
```
So that's hex 7 followed by 4 zeros. In the generated perf json it is
hex 7 followed by 6 zeros as we concatenated the UMask's 00 to the
front.

> event syntax error: '..O_NUM_TGT_MATCHED_REQ_OF_CPU'
>                                   \___ value too big for format (umask), maximum is 255

This error indicates that the UMask for the IIO PMU only expects
values from 0 to 255.

So something is wrong in how we interpret the UMask and UMaskExt for
Sapphire Rapids' IIO. Could someone from Intel take a look? Is the
problem in the original perfmon event json, is there a different way
to encode the UMaskExt for the IIO on SPR or is the IIO PMU on SPR
constraining the umask too much?

Thanks,
Ian


> event syntax error: '..O_NUM_TGT_MATCHED_REQ_OF_CPU'
>                                   \___ value too big for format (umask), maximum is 255
>
> event syntax error: '..O_NUM_TGT_MATCHED_REQ_OF_CPU'
>                                   \___ value too big for format (umask), maximum is 255
>
> event syntax error: '..O_NUM_TGT_MATCHED_REQ_OF_CPU'
>                                   \___ value too big for format (umask), maximum is 255
>
> event syntax error: '..O_NUM_TGT_MATCHED_REQ_OF_CPU'
>                                   \___ value too big for format (umask), maximum is 255
>
> event syntax error: '..O_NUM_TGT_MATCHED_REQ_OF_CPU'
>                                   \___ value too big for format (umask), maximum is 255
>
> event syntax error: '..O_NUM_TGT_MATCHED_REQ_OF_CPU'
>                                   \___ value too big for format (umask), maximum is 255
>
> event syntax error: '..O_NUM_TGT_MATCHED_REQ_OF_CPU'
>                                   \___ value too big for format (umask), maximum is 255
>
> event syntax error: '..O_NUM_TGT_MATCHED_REQ_OF_CPU'
>                                   \___ value too big for format (umask), maximum is 255
>
> event syntax error: '..O_NUM_TGT_MATCHED_REQ_OF_CPU'
>                                   \___ value too big for format (umask), maximum is 255
> Run 'perf list' for a list of valid events
>
>  Usage: perf stat [<options>] [<command>]
>
>     -e, --event <event>   event selector. use 'perf list' to list available events
>
>
> The kernel config and materials to reproduce are available at:
> https://download.01.org/0day-ci/archive/20261105/202611050912.2cdf1ff5-lkp@xxxxxxxxx
>
>
> --
> 0-DAY CI Kernel Test Service
> https://github.com/intel/lkp-tests/wiki
>
>