2020-11-19 00:19:47

by David E. Box

[permalink] [raw]
Subject: [PATCH 2/2] PCI: Disable Precision Time Measurement during suspend

On Intel client platforms that support suspend-to-idle, like Ice Lake,
root ports that have Precision Time Management (PTM) enabled can prevent
the port from being fully power gated, causing higher power consumption
while suspended. To prevent this, after saving the PTM control register,
disable the feature. The feature will be returned to its previous state
during restore.

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=209361
Reported-by: Len Brown <[email protected]>
Suggested-by: Rafael J. Wysocki <[email protected]>
Signed-off-by: David E. Box <[email protected]>
---
drivers/pci/pci.c | 14 +++++++++++++-
1 file changed, 13 insertions(+), 1 deletion(-)

diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
index 6fd4ae910a88..a2b40497d443 100644
--- a/drivers/pci/pci.c
+++ b/drivers/pci/pci.c
@@ -21,6 +21,7 @@
#include <linux/module.h>
#include <linux/spinlock.h>
#include <linux/string.h>
+#include <linux/suspend.h>
#include <linux/log2.h>
#include <linux/logic_pio.h>
#include <linux/pm_wakeup.h>
@@ -1543,7 +1544,7 @@ static void pci_save_ptm_state(struct pci_dev *dev)
{
int ptm;
struct pci_cap_saved_state *save_state;
- u16 *cap;
+ u16 *cap, ctrl;

if (!pci_is_pcie(dev))
return;
@@ -1560,6 +1561,17 @@ static void pci_save_ptm_state(struct pci_dev *dev)

cap = (u16 *)&save_state->cap.data[0];
pci_read_config_word(dev, ptm + PCI_PTM_CTRL, cap);
+
+ /*
+ * On Intel systems that support suspend-to-idle, additional
+ * power savings can be gained by disabling PTM on root ports,
+ * as this allows the port to enter a deeper pm state.
+ */
+ if (pm_suspend_target_state == PM_SUSPEND_TO_IDLE &&
+ pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT) {
+ ctrl = *cap & ~(PCI_PTM_CTRL_ENABLE | PCI_PTM_CTRL_ROOT);
+ pci_write_config_word(dev, ptm + PCI_PTM_CTRL, ctrl);
+ }
}

static void pci_restore_ptm_state(struct pci_dev *dev)
--
2.20.1


2020-11-19 12:05:57

by Rafael J. Wysocki

[permalink] [raw]
Subject: Re: [PATCH 2/2] PCI: Disable Precision Time Measurement during suspend

On Thu, Nov 19, 2020 at 1:17 AM David E. Box
<[email protected]> wrote:
>
> On Intel client platforms that support suspend-to-idle, like Ice Lake,
> root ports that have Precision Time Management (PTM) enabled can prevent
> the port from being fully power gated, causing higher power consumption
> while suspended. To prevent this, after saving the PTM control register,
> disable the feature. The feature will be returned to its previous state
> during restore.
>
> Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=209361
> Reported-by: Len Brown <[email protected]>
> Suggested-by: Rafael J. Wysocki <[email protected]>
> Signed-off-by: David E. Box <[email protected]>
> ---
> drivers/pci/pci.c | 14 +++++++++++++-
> 1 file changed, 13 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
> index 6fd4ae910a88..a2b40497d443 100644
> --- a/drivers/pci/pci.c
> +++ b/drivers/pci/pci.c
> @@ -21,6 +21,7 @@
> #include <linux/module.h>
> #include <linux/spinlock.h>
> #include <linux/string.h>
> +#include <linux/suspend.h>
> #include <linux/log2.h>
> #include <linux/logic_pio.h>
> #include <linux/pm_wakeup.h>
> @@ -1543,7 +1544,7 @@ static void pci_save_ptm_state(struct pci_dev *dev)
> {
> int ptm;
> struct pci_cap_saved_state *save_state;
> - u16 *cap;
> + u16 *cap, ctrl;
>
> if (!pci_is_pcie(dev))
> return;
> @@ -1560,6 +1561,17 @@ static void pci_save_ptm_state(struct pci_dev *dev)
>
> cap = (u16 *)&save_state->cap.data[0];
> pci_read_config_word(dev, ptm + PCI_PTM_CTRL, cap);
> +
> + /*
> + * On Intel systems that support suspend-to-idle, additional
> + * power savings can be gained by disabling PTM on root ports,
> + * as this allows the port to enter a deeper pm state.

I would say "There are systems (for example, ...) where the power
drawn while suspended can be significantly reduced by disabling PTM on
PCIe root ports, as this allows the port to enter a lower-power PM
state and the SoC to reach a lower-power idle state as a whole".

> + */
> + if (pm_suspend_target_state == PM_SUSPEND_TO_IDLE &&

AFAICS the target sleep state doesn't matter here, so I'd skip the
check above, but otherwise it LGTM.

> + pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT) {
> + ctrl = *cap & ~(PCI_PTM_CTRL_ENABLE | PCI_PTM_CTRL_ROOT);
> + pci_write_config_word(dev, ptm + PCI_PTM_CTRL, ctrl);
> + }
> }
>
> static void pci_restore_ptm_state(struct pci_dev *dev)
> --

2020-11-19 17:47:53

by David E. Box

[permalink] [raw]
Subject: Re: [PATCH 2/2] PCI: Disable Precision Time Measurement during suspend

On Thu, 2020-11-19 at 13:01 +0100, Rafael J. Wysocki wrote:
> On Thu, Nov 19, 2020 at 1:17 AM David E. Box
> <[email protected]> wrote:
> > On Intel client platforms that support suspend-to-idle, like Ice
> > Lake,
> > root ports that have Precision Time Management (PTM) enabled can
> > prevent
> > the port from being fully power gated, causing higher power
> > consumption
> > while suspended. To prevent this, after saving the PTM control
> > register,
> > disable the feature. The feature will be returned to its previous
> > state
> > during restore.
> >
> > Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=209361
> > Reported-by: Len Brown <[email protected]>
> > Suggested-by: Rafael J. Wysocki <[email protected]>
> > Signed-off-by: David E. Box <[email protected]>
> > ---
> > drivers/pci/pci.c | 14 +++++++++++++-
> > 1 file changed, 13 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
> > index 6fd4ae910a88..a2b40497d443 100644
> > --- a/drivers/pci/pci.c
> > +++ b/drivers/pci/pci.c
> > @@ -21,6 +21,7 @@
> > #include <linux/module.h>
> > #include <linux/spinlock.h>
> > #include <linux/string.h>
> > +#include <linux/suspend.h>
> > #include <linux/log2.h>
> > #include <linux/logic_pio.h>
> > #include <linux/pm_wakeup.h>
> > @@ -1543,7 +1544,7 @@ static void pci_save_ptm_state(struct pci_dev
> > *dev)
> > {
> > int ptm;
> > struct pci_cap_saved_state *save_state;
> > - u16 *cap;
> > + u16 *cap, ctrl;
> >
> > if (!pci_is_pcie(dev))
> > return;
> > @@ -1560,6 +1561,17 @@ static void pci_save_ptm_state(struct
> > pci_dev *dev)
> >
> > cap = (u16 *)&save_state->cap.data[0];
> > pci_read_config_word(dev, ptm + PCI_PTM_CTRL, cap);
> > +
> > + /*
> > + * On Intel systems that support suspend-to-idle,
> > additional
> > + * power savings can be gained by disabling PTM on root
> > ports,
> > + * as this allows the port to enter a deeper pm state.
>
> I would say "There are systems (for example, ...) where the power
> drawn while suspended can be significantly reduced by disabling PTM
> on
> PCIe root ports, as this allows the port to enter a lower-power PM
> state and the SoC to reach a lower-power idle state as a whole".

Okay.

>
> > + */
> > + if (pm_suspend_target_state == PM_SUSPEND_TO_IDLE &&
>
> AFAICS the target sleep state doesn't matter here, so I'd skip the
> check above, but otherwise it LGTM.

The target sleep state doesn't matter so much but that it's suspending
does. pci_save_state() is called during probe for the root ports (and
many other pci devices - I'm curious as to why). So without this check
the capability gets disabled on boot.

>
> > + pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT) {
> > + ctrl = *cap & ~(PCI_PTM_CTRL_ENABLE |
> > PCI_PTM_CTRL_ROOT);
> > + pci_write_config_word(dev, ptm + PCI_PTM_CTRL,
> > ctrl);
> > + }
> > }
> >
> > static void pci_restore_ptm_state(struct pci_dev *dev)
> > --

David

2020-11-19 18:16:49

by Rafael J. Wysocki

[permalink] [raw]
Subject: Re: [PATCH 2/2] PCI: Disable Precision Time Measurement during suspend

On Thu, Nov 19, 2020 at 6:45 PM David E. Box
<[email protected]> wrote:
>
> On Thu, 2020-11-19 at 13:01 +0100, Rafael J. Wysocki wrote:
> > On Thu, Nov 19, 2020 at 1:17 AM David E. Box
> > <[email protected]> wrote:
> > > On Intel client platforms that support suspend-to-idle, like Ice
> > > Lake,
> > > root ports that have Precision Time Management (PTM) enabled can
> > > prevent
> > > the port from being fully power gated, causing higher power
> > > consumption
> > > while suspended. To prevent this, after saving the PTM control
> > > register,
> > > disable the feature. The feature will be returned to its previous
> > > state
> > > during restore.
> > >
> > > Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=209361
> > > Reported-by: Len Brown <[email protected]>
> > > Suggested-by: Rafael J. Wysocki <[email protected]>
> > > Signed-off-by: David E. Box <[email protected]>
> > > ---
> > > drivers/pci/pci.c | 14 +++++++++++++-
> > > 1 file changed, 13 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
> > > index 6fd4ae910a88..a2b40497d443 100644
> > > --- a/drivers/pci/pci.c
> > > +++ b/drivers/pci/pci.c
> > > @@ -21,6 +21,7 @@
> > > #include <linux/module.h>
> > > #include <linux/spinlock.h>
> > > #include <linux/string.h>
> > > +#include <linux/suspend.h>
> > > #include <linux/log2.h>
> > > #include <linux/logic_pio.h>
> > > #include <linux/pm_wakeup.h>
> > > @@ -1543,7 +1544,7 @@ static void pci_save_ptm_state(struct pci_dev
> > > *dev)
> > > {
> > > int ptm;
> > > struct pci_cap_saved_state *save_state;
> > > - u16 *cap;
> > > + u16 *cap, ctrl;
> > >
> > > if (!pci_is_pcie(dev))
> > > return;
> > > @@ -1560,6 +1561,17 @@ static void pci_save_ptm_state(struct
> > > pci_dev *dev)
> > >
> > > cap = (u16 *)&save_state->cap.data[0];
> > > pci_read_config_word(dev, ptm + PCI_PTM_CTRL, cap);
> > > +
> > > + /*
> > > + * On Intel systems that support suspend-to-idle,
> > > additional
> > > + * power savings can be gained by disabling PTM on root
> > > ports,
> > > + * as this allows the port to enter a deeper pm state.
> >
> > I would say "There are systems (for example, ...) where the power
> > drawn while suspended can be significantly reduced by disabling PTM
> > on
> > PCIe root ports, as this allows the port to enter a lower-power PM
> > state and the SoC to reach a lower-power idle state as a whole".
>
> Okay.
>
> >
> > > + */
> > > + if (pm_suspend_target_state == PM_SUSPEND_TO_IDLE &&
> >
> > AFAICS the target sleep state doesn't matter here, so I'd skip the
> > check above, but otherwise it LGTM.
>
> The target sleep state doesn't matter so much but that it's suspending
> does. pci_save_state() is called during probe for the root ports (and
> many other pci devices - I'm curious as to why).

I tend to forget about this, sorry.

> So without this check the capability gets disabled on boot.
>

So instead of calling this from here, why don't we invoke the code
below from pci_prepare_to_sleep() and pci_finish_runtime_suspend(),
before enabling wakeup (and it needs to be re-done on failures, eg. by
restoring the cap from the saved copy)?

> > > + pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT) {
> > > + ctrl = *cap & ~(PCI_PTM_CTRL_ENABLE |
> > > PCI_PTM_CTRL_ROOT);
> > > + pci_write_config_word(dev, ptm + PCI_PTM_CTRL,
> > > ctrl);
> > > + }
> > > }
> > >
> > > static void pci_restore_ptm_state(struct pci_dev *dev)
> > > --

2020-11-19 19:41:30

by David E. Box

[permalink] [raw]
Subject: Re: [PATCH 2/2] PCI: Disable Precision Time Measurement during suspend

On Thu, 2020-11-19 at 19:13 +0100, Rafael J. Wysocki wrote:
> On Thu, Nov 19, 2020 at 6:45 PM David E. Box
> <[email protected]> wrote:
> > On Thu, 2020-11-19 at 13:01 +0100, Rafael J. Wysocki wrote:
> > > On Thu, Nov 19, 2020 at 1:17 AM David E. Box
> > > <[email protected]> wrote:

...

> > > >
> > > > + */
> > > > + if (pm_suspend_target_state == PM_SUSPEND_TO_IDLE &&
> > >
> > > AFAICS the target sleep state doesn't matter here, so I'd skip
> > > the
> > > check above, but otherwise it LGTM.
> >
> > The target sleep state doesn't matter so much but that it's
> > suspending
> > does. pci_save_state() is called during probe for the root ports
> > (and
> > many other pci devices - I'm curious as to why).
>
> I tend to forget about this, sorry.
>
> > So without this check the capability gets disabled on boot.
> >
>
> So instead of calling this from here, why don't we invoke the code
> below from pci_prepare_to_sleep() and pci_finish_runtime_suspend(),
> before enabling wakeup (and it needs to be re-done on failures, eg.
> by
> restoring the cap from the saved copy)?

Ok.