[PATCH] PCI / PM: Use per-device D3 delays

[PATCH] PCI / PM: Use per-device D3 delays

[Rafael J. Wysocki]

From: Rafael J. Wysocki <rjw@sisk.pl>

It turns out that some PCI devices require extra delays when changing
power state from D3 to D0 (and the other way around).  Although this
is against the PCI specification, we can handle it quite easily by
allowing drivers to define arbitrary D3 delays for devices known to
require extra time for switching power states.

Introduce additional field d3_delay in struct pci_dev and use it to
store the value of the device's D0->D3 delay, in miliseconds.  Make
the PCI PM core code use the per-device d3_delay unless
pci_pm_d3_delay is greater (in which case the latter is used).
[This also allows the driver to specify d3_delay shorter than the
 10 ms required by the PCI standard if the device is known to be able
 to handle that.]

Make the sky2 driver set d3_delay to 150 for devices handled by it.

Fixes http://bugzilla.kernel.org/show_bug.cgi?id=14730 which is a
listed regression from 2.6.30.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
---
 drivers/net/sky2.c  |    1 +
 drivers/pci/pci.c   |   19 +++++++++++++++----
 include/linux/pci.h |    1 +
 3 files changed, 17 insertions(+), 4 deletions(-)

Index: linux-2.6/include/linux/pci.h
===================================================================
--- linux-2.6.orig/include/linux/pci.h
+++ linux-2.6/include/linux/pci.h
@@ -244,6 +244,7 @@ struct pci_dev {
 	unsigned int	d2_support:1;	/* Low power state D2 is supported */
 	unsigned int	no_d1d2:1;	/* Only allow D0 and D3 */
 	unsigned int	wakeup_prepared:1;
+	unsigned int	d3_delay;	/* D3->D0 transition time in ms */
 
 #ifdef CONFIG_PCIEASPM
 	struct pcie_link_state	*link_state;	/* ASPM link state. */
Index: linux-2.6/drivers/pci/pci.c
===================================================================
--- linux-2.6.orig/drivers/pci/pci.c
+++ linux-2.6/drivers/pci/pci.c
@@ -30,7 +30,17 @@ const char *pci_power_names[] = {
 };
 EXPORT_SYMBOL_GPL(pci_power_names);
 
-unsigned int pci_pm_d3_delay = PCI_PM_D3_WAIT;
+unsigned int pci_pm_d3_delay;
+
+static void pci_dev_d3_sleep(struct pci_dev *dev)
+{
+	unsigned int delay = dev->d3_delay;
+
+	if (delay < pci_pm_d3_delay)
+		delay = pci_pm_d3_delay;
+
+	msleep(delay);
+}
 
 #ifdef CONFIG_PCI_DOMAINS
 int pci_domains_supported = 1;
@@ -529,7 +539,7 @@ static int pci_raw_set_power_state(struc
 	/* Mandatory power management transition delays */
 	/* see PCI PM 1.1 5.6.1 table 18 */
 	if (state == PCI_D3hot || dev->current_state == PCI_D3hot)
-		msleep(pci_pm_d3_delay);
+		pci_dev_d3_sleep(dev);
 	else if (state == PCI_D2 || dev->current_state == PCI_D2)
 		udelay(PCI_PM_D2_DELAY);
 
@@ -1544,6 +1554,7 @@ void pci_pm_init(struct pci_dev *dev)
 	}
 
 	dev->pm_cap = pm;
+	dev->d3_delay = PCI_PM_D3_WAIT;
 
 	dev->d1_support = false;
 	dev->d2_support = false;
@@ -2382,12 +2393,12 @@ static int pci_pm_reset(struct pci_dev *
 	csr &= ~PCI_PM_CTRL_STATE_MASK;
 	csr |= PCI_D3hot;
 	pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
-	msleep(pci_pm_d3_delay);
+	pci_dev_d3_sleep(dev);
 
 	csr &= ~PCI_PM_CTRL_STATE_MASK;
 	csr |= PCI_D0;
 	pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
-	msleep(pci_pm_d3_delay);
+	pci_dev_d3_sleep(dev);
 
 	return 0;
 }
Index: linux-2.6/drivers/net/sky2.c
===================================================================
--- linux-2.6.orig/drivers/net/sky2.c
+++ linux-2.6/drivers/net/sky2.c
@@ -4684,6 +4684,7 @@ static int __devinit sky2_probe(struct p
 	INIT_WORK(&hw->restart_work, sky2_restart);
 
 	pci_set_drvdata(pdev, hw);
+	pdev->d3_delay = 150;
 
 	return 0;
 

Re: [PATCH] PCI / PM: Use per-device D3 delays

[Jesse Barnes]

On Thu, 31 Dec 2009 12:15:54 +0100
"Rafael J. Wysocki" <rjw@sisk.pl> wrote:

> From: Rafael J. Wysocki <rjw@sisk.pl>
> 
> It turns out that some PCI devices require extra delays when changing
> power state from D3 to D0 (and the other way around).  Although this
> is against the PCI specification, we can handle it quite easily by
> allowing drivers to define arbitrary D3 delays for devices known to
> require extra time for switching power states.
> 
> Introduce additional field d3_delay in struct pci_dev and use it to
> store the value of the device's D0->D3 delay, in miliseconds.  Make
> the PCI PM core code use the per-device d3_delay unless
> pci_pm_d3_delay is greater (in which case the latter is used).
> [This also allows the driver to specify d3_delay shorter than the
>  10 ms required by the PCI standard if the device is known to be able
>  to handle that.]
> 
> Make the sky2 driver set d3_delay to 150 for devices handled by it.
> 
> Fixes http://bugzilla.kernel.org/show_bug.cgi?id=14730 which is a
> listed regression from 2.6.30.
> 
> Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>

Applied to my for-linus branch, thanks.

-- 
Jesse Barnes, Intel Open Source Technology Center

Re: [PATCH] PCI / PM: Use per-device D3 delays

[Andreas Mohr]

Hi,

While the bug report mentions "So it's just quirky hardware.",
the implementation of your patch makes it seem like this delay attribute is
totally "norm"al behaviour - I'm missing some more aggressive wording.

Perhaps rename d3_delay to d3_delay__quirk or add something to the
comment, like "D3->D0 transition time in ms (out-of-spec PCI device quirk)"?
Or maybe something like
"custom D3->D0 transition time in ms (for quirky hardware violating the PCI spec's <= 100ms)".

Thanks for your ongoing great PCI efforts,

Andreas Mohr

Re: [PATCH] PCI / PM: Use per-device D3 delays

[Rafael J. Wysocki]

On Friday 01 January 2010, Andreas Mohr wrote:
> Hi,
> 
> While the bug report mentions "So it's just quirky hardware.",
> the implementation of your patch makes it seem like this delay attribute is
> totally "norm"al behaviour - I'm missing some more aggressive wording.

That's because it works both ways (please look at the changelog).

I know of a few devices that don't need the PCI-prescribed 10 ms wait when
going from D3 to D0 and their drivers may use the d3_delay field to actually
set a _shorter_ delay.

> Perhaps rename d3_delay to d3_delay__quirk or add something to the
> comment, like "D3->D0 transition time in ms (out-of-spec PCI device quirk)"?
> Or maybe something like
> "custom D3->D0 transition time in ms (for quirky hardware violating the PCI spec's <= 100ms)".
> 
> Thanks for your ongoing great PCI efforts,

You're welcome.

Rafael

Re: [PATCH] PCI / PM: Use per-device D3 delays

[Andreas Mohr]

Hi,

On Fri, Jan 01, 2010 at 07:55:27PM +0100, Rafael J. Wysocki wrote:
> On Friday 01 January 2010, Andreas Mohr wrote:
> > Hi,
> > 
> > While the bug report mentions "So it's just quirky hardware.",
> > the implementation of your patch makes it seem like this delay attribute is
> > totally "norm"al behaviour - I'm missing some more aggressive wording.
> 
> That's because it works both ways (please look at the changelog).

Ah, ok.

> I know of a few devices that don't need the PCI-prescribed 10 ms wait when
> going from D3 to D0 and their drivers may use the d3_delay field to actually
> set a _shorter_ delay.

Then why is the value lower-bounded by pci_pm_d3_delay
(which, puzzlingly, was initialized to PCI_PM_D3_WAIT and thus 10
before, which the patch now removes!), in pci_dev_d3_sleep()?
(and pci_pm_d3_delay is being quirked in drivers/pci/quirks.c only,
to 120)

Confused,

Andreas Mohr

Re: [PATCH] PCI / PM: Use per-device D3 delays

[Rafael J. Wysocki]

On Friday 01 January 2010, Andreas Mohr wrote:
> Hi,
> 
> On Fri, Jan 01, 2010 at 07:55:27PM +0100, Rafael J. Wysocki wrote:
> > On Friday 01 January 2010, Andreas Mohr wrote:
> > > Hi,
> > > 
> > > While the bug report mentions "So it's just quirky hardware.",
> > > the implementation of your patch makes it seem like this delay attribute is
> > > totally "norm"al behaviour - I'm missing some more aggressive wording.
> > 
> > That's because it works both ways (please look at the changelog).
> 
> Ah, ok.
> 
> > I know of a few devices that don't need the PCI-prescribed 10 ms wait when
> > going from D3 to D0 and their drivers may use the d3_delay field to actually
> > set a _shorter_ delay.
> 
> Then why is the value lower-bounded by pci_pm_d3_delay
> (which, puzzlingly, was initialized to PCI_PM_D3_WAIT and thus 10
> before, which the patch now removes!),

That's because dev->d3_delay is initialized to PCI_PM_D3_WAIT for all devices.

> in pci_dev_d3_sleep()? (and pci_pm_d3_delay is being quirked in
> drivers/pci/quirks.c only, to 120)

Exactly because pci_pm_d3_delay is only necessary for some quirky chipsets
that require longer delays for _all_ devices (note that this cannot be handled
at the driver level).  So, we use dev->d3_delay (that the driver gave us),
unless the chipset is known to be quirky and requires a longer delay for
all devices (the driver has no chance to know about that).

Rafael

Re: [PATCH] PCI / PM: Use per-device D3 delays

[Andreas Mohr]

On Fri, Jan 01, 2010 at 10:42:28PM +0100, Rafael J. Wysocki wrote:
> On Friday 01 January 2010, Andreas Mohr wrote:
> > Then why is the value lower-bounded by pci_pm_d3_delay
> > (which, puzzlingly, was initialized to PCI_PM_D3_WAIT and thus 10
> > before, which the patch now removes!),
> 
> That's because dev->d3_delay is initialized to PCI_PM_D3_WAIT for all devices.

Doh, I should have viewn the altered mechanism correctly, sorry.
Seems quite correct and useful after all.

Andreas Mohr