Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1738124 > unrolled thread

[char-misc 0/2] mei updates for 4.14-rc2

Back to article view | Back to linux.kernel

Contents

  [char-misc 0/2] mei updates for 4.14-rc2 Tomas Winkler <tomas.winkler@intel.com> - 2017-09-24 10:50 +0200
    [char-misc 1/2] mei: always use domain runtime pm callbacks. Tomas Winkler <tomas.winkler@intel.com> - 2017-09-24 10:50 +0200
      Re: [char-misc 1/2] mei: always use domain runtime pm callbacks. "Rafael J. Wysocki" <rafael@kernel.org> - 2017-09-25 00:40 +0200
        Re: [char-misc 1/2] mei: always use domain runtime pm callbacks. "Winkler, Tomas" <tomas.winkler@intel.com> - 2017-09-25 08:20 +0200

#1738124 — [char-misc 0/2] mei updates for 4.14-rc2

Hello Greg, here are updates for 4.14-rc2 containing:

A regression fix that hit us in 4.13,
unfortunately we haven't spotted it during the 4.13 cycle.

A new device id, for Gemini Lake SoC. This particular patch doesn't
apply cleanly to stable kernels so I will post separated backport patches 
for 4.4 and newer kernel versions.

Alexander Usyskin (1):
  mei: always use domain runtime pm callbacks.

Tomas Winkler (1):
  mei: me: add gemini lake devices id

 drivers/misc/mei/hw-me-regs.h |  2 ++
 drivers/misc/mei/pci-me.c     | 23 +++++++++++++----------
 drivers/misc/mei/pci-txe.c    | 30 +++++++++++-------------------
 3 files changed, 26 insertions(+), 29 deletions(-)

-- 
2.9.5

[toc] | [next] | [standalone]

#1738125 — [char-misc 1/2] mei: always use domain runtime pm callbacks.

From: Alexander Usyskin <alexander.usyskin@intel.com>

This patch fixes a regression caused by the new changes
in the run wake handlers, pci_dev_run_wake() return true
when wake is possible from the software perspective
not whether it is actually enabled.

The mei devices that support D0i3 are no longer receiving an interrupt
after entering runtime suspend state and will stall.

MEI maps runtime suspend/resume to its own defined
power gating (PG) states, (D0i3 or other depending on generation),
hence we need to go around the native PCI runtime service which
eventually brings the device into D3cold/hot state,
but the mei devices cannot wake up from D3 unlike from D0i3/PG state,
which keeps irq running.
To get around PCI device native runtime pm,
MEI uses runtime pm domain handlers which take precedence.

Cc: <stable@vger.kernel.org> #4.13+
Cc: Rafael J. Wysocki <rafael@kernel.org>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
---
 drivers/misc/mei/pci-me.c  | 21 +++++++++++----------
 drivers/misc/mei/pci-txe.c | 30 +++++++++++-------------------
 2 files changed, 22 insertions(+), 29 deletions(-)

diff --git a/drivers/misc/mei/pci-me.c b/drivers/misc/mei/pci-me.c
index 4ff40d319676..630757a4b36a 100644
--- a/drivers/misc/mei/pci-me.c
+++ b/drivers/misc/mei/pci-me.c
@@ -226,12 +226,15 @@ static int mei_me_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 	pdev->dev_flags |= PCI_DEV_FLAGS_NEEDS_RESUME;
 
 	/*
-	* For not wake-able HW runtime pm framework
-	* can't be used on pci device level.
-	* Use domain runtime pm callbacks instead.
-	*/
-	if (!pci_dev_run_wake(pdev))
-		mei_me_set_pm_domain(dev);
+	 * ME maps runtime suspend/resume to D0i states,
+	 * hence we need to go around native PCI runtime service which
+	 * eventually brings the device into D3cold/hot state,
+	 * but the mei device cannot wake up from D3 unlike from D0i3.
+	 * To get around the PCI device native runtime pm,
+	 * ME uses runtime pm domain handlers which take precedence
+	 * over the driver's pm handlers.
+	 */
+	mei_me_set_pm_domain(dev);
 
 	if (mei_pg_is_enabled(dev))
 		pm_runtime_put_noidle(&pdev->dev);
@@ -271,8 +274,7 @@ static void mei_me_shutdown(struct pci_dev *pdev)
 	dev_dbg(&pdev->dev, "shutdown\n");
 	mei_stop(dev);
 
-	if (!pci_dev_run_wake(pdev))
-		mei_me_unset_pm_domain(dev);
+	mei_me_unset_pm_domain(dev);
 
 	mei_disable_interrupts(dev);
 	free_irq(pdev->irq, dev);
@@ -300,8 +302,7 @@ static void mei_me_remove(struct pci_dev *pdev)
 	dev_dbg(&pdev->dev, "stop\n");
 	mei_stop(dev);
 
-	if (!pci_dev_run_wake(pdev))
-		mei_me_unset_pm_domain(dev);
+	mei_me_unset_pm_domain(dev);
 
 	mei_disable_interrupts(dev);
 
diff --git a/drivers/misc/mei/pci-txe.c b/drivers/misc/mei/pci-txe.c
index e38a5f144373..0566f9bfa7de 100644
--- a/drivers/misc/mei/pci-txe.c
+++ b/drivers/misc/mei/pci-txe.c
@@ -144,12 +144,14 @@ static int mei_txe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 	pdev->dev_flags |= PCI_DEV_FLAGS_NEEDS_RESUME;
 
 	/*
-	* For not wake-able HW runtime pm framework
-	* can't be used on pci device level.
-	* Use domain runtime pm callbacks instead.
-	*/
-	if (!pci_dev_run_wake(pdev))
-		mei_txe_set_pm_domain(dev);
+	 * TXE maps runtime suspend/resume to own power gating states,
+	 * hence we need to go around native PCI runtime service which
+	 * eventually brings the device into D3cold/hot state.
+	 * But the TXE device cannot wake up from D3 unlike from own
+	 * power gating. To get around PCI device native runtime pm,
+	 * TXE uses runtime pm domain handlers which take precedence.
+	 */
+	mei_txe_set_pm_domain(dev);
 
 	pm_runtime_put_noidle(&pdev->dev);
 
@@ -186,8 +188,7 @@ static void mei_txe_shutdown(struct pci_dev *pdev)
 	dev_dbg(&pdev->dev, "shutdown\n");
 	mei_stop(dev);
 
-	if (!pci_dev_run_wake(pdev))
-		mei_txe_unset_pm_domain(dev);
+	mei_txe_unset_pm_domain(dev);
 
 	mei_disable_interrupts(dev);
 	free_irq(pdev->irq, dev);
@@ -215,8 +216,7 @@ static void mei_txe_remove(struct pci_dev *pdev)
 
 	mei_stop(dev);
 
-	if (!pci_dev_run_wake(pdev))
-		mei_txe_unset_pm_domain(dev);
+	mei_txe_unset_pm_domain(dev);
 
 	mei_disable_interrupts(dev);
 	free_irq(pdev->irq, dev);
@@ -318,15 +318,7 @@ static int mei_txe_pm_runtime_suspend(struct device *device)
 	else
 		ret = -EAGAIN;
 
-	/*
-	 * If everything is okay we're about to enter PCI low
-	 * power state (D3) therefor we need to disable the
-	 * interrupts towards host.
-	 * However if device is not wakeable we do not enter
-	 * D-low state and we need to keep the interrupt kicking
-	 */
-	if (!ret && pci_dev_run_wake(pdev))
-		mei_disable_interrupts(dev);
+	/* keep irq on we are staying in D0 */
 
 	dev_dbg(&pdev->dev, "rpm: txe: runtime suspend ret=%d\n", ret);
 
-- 
2.9.5

[toc] | [prev] | [next] | [standalone]

#1738582 — Re: [char-misc 1/2] mei: always use domain runtime pm callbacks.

On Sun, Sep 24, 2017 at 10:35 AM, Tomas Winkler <tomas.winkler@intel.com> wrote:
> From: Alexander Usyskin <alexander.usyskin@intel.com>
>
> This patch fixes a regression caused by the new changes
> in the run wake handlers, pci_dev_run_wake() return true
> when wake is possible from the software perspective
> not whether it is actually enabled.

This isn't exactly accurate, because "run wake" used to mean "wakeup
signals can be generated for this device in the working state of the
system", so it could not be enabled or disabled before too.  It's just
that the function returns "true" for (at least) some devices for which
it used to return "false" (arguably incorrectly).

Other than that I agree with the patch.

Thanks,
Rafael

[toc] | [prev] | [next] | [standalone]

#1738760 — Re: [char-misc 1/2] mei: always use domain runtime pm callbacks.

	On Mon, 2017-09-25 at 00:37 +0200, Rafael J. Wysocki wrote:
> On Sun, Sep 24, 2017 at 10:35 AM, Tomas Winkler <tomas.winkler@intel.com> wrote:
> > From: Alexander Usyskin <alexander.usyskin@intel.com>
> > 
> > This patch fixes a regression caused by the new changes
> > in the run wake handlers, pci_dev_run_wake() return true
> > when wake is possible from the software perspective
> > not whether it is actually enabled.
> 
> This isn't exactly accurate, because "run wake" used to mean "wakeup
> signals can be generated for this device in the working state of the
> system", so it could not be enabled or disabled before too.  It's just
> that the function returns "true" for (at least) some devices for which
> it used to return "false" (arguably incorrectly).
> 
> Other than that I agree with the patch.

Thanks for the correction, I will respin the patch with your
explanation.

Thanks
Tomas

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web