On Wed, Jan 19, 2022 at 12:32:51PM +0300, Alexander Sergeyev wrote:
No, I mean "IO_PAGE_FAULT" and "out of range cmd" don't appear on every unbind & bind. Sometimes it works cleanly.
Unbind & bind generally works without error messages on the first attempt. The second unbind & bind tends to generate io page faults like "AMD-Vi: Event logged [IO_PAGE_FAULT ...]", but might work fine as well.
In some cases "snd_hda_codec_realtek: out of range cmd" errors are triggered in addition to io page faults.
This seems to be a bogus COEF. But I have no idea from where this comes. The values look completely wrong.
In such cases unexpected COEF values are coming from COEF reads during read/write pairs that implement update operations.
For example (these traces are from added printk statements):
alc_update_coef_led(codec, led {.idx=0x19, .mask=0x2000, .on=0x2000, .off=0x0}, polarity=0, on=1) alc_update_coefex_idx(codec, nid, coef_idx=0x19, mask=0x2000, bits_set=0x2000) alc_update_coef_led(codec, led {.idx=0xb, .mask=0x8, .on=0x8, .off=0x0}, polarity=0, on=1) alc_update_coefex_idx(codec, nid, coef_idx=0xb, mask=0x8, bits_set=0x8) snd_hdac_codec_write(hdac, nid, flags=0, verb=0x500, parm=0x19) = 0x0 snd_hdac_codec_write(hdac, nid, flags=0, verb=0x500, parm=0xb) = 0x0 snd_hdac_codec_read(hdac, nid, flags=0, verb=0xc00, parm=0x0) = 0x90170110 alc_update_coefex_idx: alc_read_coefex_idx(codec, nid, coef_idx=0xb) returned 0x90170110 alc_update_coefex_idx: calling alc_write_coefex_idx(codec, nid, coef_idx=0xb, coef_val=0x90170118) snd_hdac_codec_read(hdac, nid, flags=0, verb=0xc00, parm=0x0) = 0x0 alc_update_coefex_idx: alc_read_coefex_idx(codec, nid, coef_idx=0x19) returned 0x0 alc_update_coefex_idx: calling alc_write_coefex_idx(codec, nid, coef_idx=0x19, coef_val=0x2000) snd_hdac_codec_write(hdac, nid, flags=0, verb=0x500, parm=0xb) = 0x0 snd_hda_codec_realtek hdaudioC1D0: out of range cmd 0:20:400:90170118 snd_hdac_codec_write(hdac, nid, flags=0, verb=0x400, parm=0x90170118) = 0xffffffff
Then I've specifically added printk on alc_update_coefex_idx entry and exit:
[4.036211] alc_update_coefex_idx(codec, nid, coef_idx=0x10, mask=0x200, bits_set=0x0): entering [4.036503] alc_update_coefex_idx(codec, nid, coef_idx=0x10, mask=0x200, bits_set=0x0): exiting [4.036543] alc_update_coefex_idx(codec, nid, coef_idx=0xb, mask=0x8, bits_set=0x0): entering [4.036546] alc_update_coefex_idx(codec, nid, coef_idx=0x19, mask=0x2000, bits_set=0x0): entering [4.037139] alc_update_coefex_idx(codec, nid, coef_idx=0xb, mask=0x8, bits_set=0x0): exiting [4.037609] alc_update_coefex_idx(codec, nid, coef_idx=0x19, mask=0x2000, bits_set=0x0): exiting
I'm not sure about kernel log buffering or maybe the device support for pipelining, but is it okay that alc_update_coefex_idx() seem to overlap?