Hi
Some time ago, a patch was merged that removed pci_save_state() and
pci_restore_state() from various drivers. That patch also added
pci_restore_state() to sound/core/init.c but didn't add pci_save_state()
anywhere.
My laptop doesn't resume (gets what I assume is an ACPI timeout and
hangs solid) without this small obvious patch.
Signed-off-by: Martin Josefsson <[email protected]>
--- linux-2.6.10-rc1-bk21.orig/sound/core/init.c 2004-11-11 18:51:17.000000000 +0100
+++ linux-2.6.10-rc1-bk21/sound/core/init.c 2004-11-11 20:57:52.000000000 +0100
@@ -789,6 +789,8 @@ int snd_card_pci_suspend(struct pci_dev
return 0;
if (card->power_state == SNDRV_CTL_POWER_D3hot)
return 0;
+ /* save the PCI config space */
+ pci_save_state(dev);
/* FIXME: correct state value? */
return card->pm_suspend(card, 0);
}
--
/Martin
At Thu, 11 Nov 2004 21:39:46 +0100,
Martin Josefsson wrote:
>
> Hi
>
> Some time ago, a patch was merged that removed pci_save_state() and
> pci_restore_state() from various drivers. That patch also added
> pci_restore_state() to sound/core/init.c but didn't add pci_save_state()
> anywhere.
pci_save_state() is called internally in
drivers/pci/pci-driver.c:pci_device_suspend(), so it's redundant.
> My laptop doesn't resume (gets what I assume is an ACPI timeout and
> hangs solid) without this small obvious patch.
I'm wondering how this can fix your problem...
Takashi
> Signed-off-by: Martin Josefsson <[email protected]>
>
> --- linux-2.6.10-rc1-bk21.orig/sound/core/init.c 2004-11-11 18:51:17.000000000 +0100
> +++ linux-2.6.10-rc1-bk21/sound/core/init.c 2004-11-11 20:57:52.000000000 +0100
> @@ -789,6 +789,8 @@ int snd_card_pci_suspend(struct pci_dev
> return 0;
> if (card->power_state == SNDRV_CTL_POWER_D3hot)
> return 0;
> + /* save the PCI config space */
> + pci_save_state(dev);
> /* FIXME: correct state value? */
> return card->pm_suspend(card, 0);
> }
>
> --
> /Martin
Takashi Iwai wrote:
> pci_save_state() is called internally in
> drivers/pci/pci-driver.c:pci_device_suspend(), so it's redundant.
>
>> My laptop doesn't resume (gets what I assume is an ACPI timeout and
>> hangs solid) without this small obvious patch.
>
> I'm wondering how this can fix your problem...
For example, some devices call pci_save_state before pci_disable_device
in
->suspend, but don't pci_enable_device in ->resume. This works before,
but
is broken after the pci_save_state() change. We need to find those
drivers out
and change the individual drivers instead of this simple fix.
Martin, which sound driver do you use?
Thanks,
-yi
>> Signed-off-by: Martin Josefsson <[email protected]>
>>
>> --- linux-2.6.10-rc1-bk21.orig/sound/core/init.c
> 2004-11-11 18:51:17.000000000 +0100
>> +++ linux-2.6.10-rc1-bk21/sound/core/init.c 2004-11-11
>> 20:57:52.000000000 +0100 @@ -789,6 +789,8 @@ int
>> snd_card_pci_suspend(struct pci_dev return 0; if
>> (card->power_state == SNDRV_CTL_POWER_D3hot) return 0;
>> + /* save the PCI config space */
>> + pci_save_state(dev);
>> /* FIXME: correct state value? */
>> return card->pm_suspend(card, 0);
>> }
>>
>> --
>> /Martin
> -
> To unsubscribe from this list: send the line "unsubscribe
> linux-kernel" in the body of a message to
> [email protected] More majordomo info at
> http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
At Fri, 12 Nov 2004 21:00:30 +0800,
Zhu, Yi wrote:
>
> Takashi Iwai wrote:
> > pci_save_state() is called internally in
> > drivers/pci/pci-driver.c:pci_device_suspend(), so it's redundant.
> >
> >> My laptop doesn't resume (gets what I assume is an ACPI timeout and
> >> hangs solid) without this small obvious patch.
> >
> > I'm wondering how this can fix your problem...
>
> For example, some devices call pci_save_state before pci_disable_device
> in
> ->suspend, but don't pci_enable_device in ->resume. This works before,
> but
> is broken after the pci_save_state() change. We need to find those
> drivers out
> and change the individual drivers instead of this simple fix.
But pci_save_state() is called again after the driver's suspend
callback is called. So, the final saved state must be anyway same.
> Martin, which sound driver do you use?
Yep that's important to know.
Takashi
On Fri, 2004-11-12 at 14:26 +0100, Takashi Iwai wrote:
> But pci_save_state() is called again after the driver's suspend
> callback is called. So, the final saved state must be anyway same.
no that changed recently in the upstream kernel.
pci_save_state() is now only called if there is no suspend callback in the driver!
At Fri, 12 Nov 2004 14:45:40 +0100,
Arjan van de Ven wrote:
>
> On Fri, 2004-11-12 at 14:26 +0100, Takashi Iwai wrote:
> > But pci_save_state() is called again after the driver's suspend
> > callback is called. So, the final saved state must be anyway same.
>
> no that changed recently in the upstream kernel.
> pci_save_state() is now only called if there is no suspend callback in the driver!
Ah, thanks, that explains why (I referred 2.6.10-rc1).
If so, the patch is almost correct, but pci_save_state() should be put
after the call of callback.
Takashi
--- linux/sound/core/init.c 8 Nov 2004 11:37:08 -0000 1.48
+++ linux/sound/core/init.c 12 Nov 2004 13:56:32 -0000
@@ -782,12 +782,15 @@
int snd_card_pci_suspend(struct pci_dev *dev, u32 state)
{
snd_card_t *card = pci_get_drvdata(dev);
+ int err;
if (! card || ! card->pm_suspend)
return 0;
if (card->power_state == SNDRV_CTL_POWER_D3hot)
return 0;
/* FIXME: correct state value? */
- return card->pm_suspend(card, 0);
+ err = card->pm_suspend(card, 0);
+ pci_save_state(dev);
+ return err;
}
int snd_card_pci_resume(struct pci_dev *dev)
On Fri, 2004-11-12 at 14:58, Takashi Iwai wrote:
> At Fri, 12 Nov 2004 14:45:40 +0100,
> Arjan van de Ven wrote:
> >
> > On Fri, 2004-11-12 at 14:26 +0100, Takashi Iwai wrote:
> > > But pci_save_state() is called again after the driver's suspend
> > > callback is called. So, the final saved state must be anyway same.
> >
> > no that changed recently in the upstream kernel.
> > pci_save_state() is now only called if there is no suspend callback in the driver!
>
> Ah, thanks, that explains why (I referred 2.6.10-rc1).
I realized that I forgot to include som vital information, sorry about
that.
Kernel 2.6.10-rc1-bk21 which has the pci_save_state() change in
pci-driver.c
My driver(s) are intel8x0.c and intel8x0m.c
First I just thought the missing pci_save_state() was a small mistake
and wondered why nobody else had reported the problem, later I found out
that a "generic" pci_save_state() for all pci-devices was removed
recently but I didn't have time to send a mail.
> If so, the patch is almost correct, but pci_save_state() should be put
> after the call of callback.
Great, thanks for fixing the fix.
I havn't had any problems with my broken patch, I'll see if I have the
time to test this one as well this weekend.
--
/Martin