| Message ID | linux-patch-20160313-1@vittgam.net (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Sun, 13 Mar 2016 22:19:34 +0100, Vittorio Gambaletta (VittGam) wrote: > > The clock measurement on the AC'97 audio card found in the IBM ThinkPad X41 > will often fail, so add a quirk entry to fix it. > > Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=441087 > Cc: <stable@vger.kernel.org> > Cc: Jaroslav Kysela <perex@perex.cz> > Cc: Takashi Iwai <tiwai@suse.de> > Signed-off-by: Vittorio Gambaletta <linuxbugs@vittgam.net> > --- > > When the clock measurement doesn't fail, the `pos` variable after line 2861 > will be around 48367, but when it does fail it will be somewhere between > 45000 and 47500. So the bus clock gets set between 48500 and 52000 (!), > resulting in faster audio for rates with final clocks below 48000, and > distorted audio for rates with final clocks above 48000. > > The problem is always reproducible on kernel 4.4.5, but very rarely on 3.13. Hrm, does this mean that the ktime() isn't accurate enough, or do we sample too short time? i.e. Is it improved if we enlarge the measurement time? > This patch fixes the problem for my computer, and will also speed up module > loading a bit as said on Bugzilla, but I've got some questions regarding > other hardware... > > Do we want to protect against overclock for all hardware, by never setting > the measured bus clock higher than 48000? Do we actually risk burning or > otherwise permanently ruining something here, or not? Does hardware clocked > above 48000 exist? > > Also, is line 2874 actually correct? As `pos` gets higher, the bus clock gets > lower, and vice versa. Note that in this function the bus clock is always > 48000 before it gets updated here; so the code first computes 48000 * 48000, > and then it divides the result by `pos`. It doesn't seem right to me... By the > way, the history of that line goes beyond Git (so before kernel 2.6.12-rc2). I don't remember of any machines with clock > 48k, so I guess the code never hits on any real machines, and I'm not sure whether it's really correct at a quick glance. In anyway, I applied your patch now. Thanks. Takashi
--- a/sound/pci/intel8x0.c +++ b/sound/pci/intel8x0.c @@ -2879,6 +2879,7 @@ static struct snd_pci_quirk intel8x0_clock_list[] = { SND_PCI_QUIRK(0x0e11, 0x008a, "AD1885", 41000), + SND_PCI_QUIRK(0x1014, 0x0581, "AD1981B", 48000), SND_PCI_QUIRK(0x1028, 0x00be, "AD1885", 44100), SND_PCI_QUIRK(0x1028, 0x0177, "AD1980", 48000), SND_PCI_QUIRK(0x1028, 0x01ad, "AD1981B", 48000),
The clock measurement on the AC'97 audio card found in the IBM ThinkPad X41 will often fail, so add a quirk entry to fix it. Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=441087 Cc: <stable@vger.kernel.org> Cc: Jaroslav Kysela <perex@perex.cz> Cc: Takashi Iwai <tiwai@suse.de> Signed-off-by: Vittorio Gambaletta <linuxbugs@vittgam.net> --- When the clock measurement doesn't fail, the `pos` variable after line 2861 will be around 48367, but when it does fail it will be somewhere between 45000 and 47500. So the bus clock gets set between 48500 and 52000 (!), resulting in faster audio for rates with final clocks below 48000, and distorted audio for rates with final clocks above 48000. The problem is always reproducible on kernel 4.4.5, but very rarely on 3.13. This patch fixes the problem for my computer, and will also speed up module loading a bit as said on Bugzilla, but I've got some questions regarding other hardware... Do we want to protect against overclock for all hardware, by never setting the measured bus clock higher than 48000? Do we actually risk burning or otherwise permanently ruining something here, or not? Does hardware clocked above 48000 exist? Also, is line 2874 actually correct? As `pos` gets higher, the bus clock gets lower, and vice versa. Note that in this function the bus clock is always 48000 before it gets updated here; so the code first computes 48000 * 48000, and then it divides the result by `pos`. It doesn't seem right to me... By the way, the history of that line goes beyond Git (so before kernel 2.6.12-rc2).