Skip to contentSkip to conversation

PCI / PM: Really allow runtime PM without callback functions

PATCH Superseded Raag Jadav 6 participants 12 messages Feb 12, 2024 – Feb 26, 2024 PCI

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Raag Jadav

author
#0 lore raw

PATCH 1/1 PCI / PM: Really allow runtime PM without callback functions

Commit c5eb1190074c ("PCI / PM: Allow runtime PM without callback functions") tried to eliminate the need for runtime PM callbacks by modifying pci_pm_runtime_suspend() and pci_pm_runtime_resume(), but didn't modify pci_pm_runtime_idle() with relevant changes, which still returns -ENOSYS if the driver supplies no runtime PM callbacks.

Fix this by modifying pci_pm_runtime_idle() such that it allows PCI device power state transitions without runtime PM callbacks.

 0)               |  pm_runtime_work() {
 0)               |    rpm_idle() {
 0)               |      rpm_check_suspend_allowed() {
 0)   1.500 us    |        __dev_pm_qos_resume_latency(); /* = 0x7fffffff */
 0)   4.840 us    |      } /* rpm_check_suspend_allowed = 0x0 */
 0)   1.550 us    |      __rpm_get_callback(); /* = 0xffffffffb4bc84f0 */
 0)   1.800 us    |      pci_pm_runtime_idle(); /* = -38 */
 0) + 17.070 us   |    } /* rpm_idle = -38 */
 0) + 22.450 us   |  } /* pm_runtime_work = -38 */

---

This is not marked for linux-stable for the need of extensive testing and can be backported after a few releases if no issues are reported.

 drivers/pci/pci-driver.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)
drivers/pci/pci-driver.c +1 -4
index 51ec9e7e784f..bb7f6775b350 100644
@@ -1382,10 +1382,7 @@ static int pci_pm_runtime_idle(struct device *dev)
 	if (!pci_dev->driver)
 		return 0;
 
-	if (!pm)
-		return -ENOSYS;
-
-	if (pm->runtime_idle)
+	if (pm && pm->runtime_idle)
 		return pm->runtime_idle(dev);
 
 	return 0;
Signature · 1 line
2.35.3
Acked-by on v1 1/1
#1 lore raw
Quoted 20 lines · Raag Jadav

Commit c5eb1190074c ("PCI / PM: Allow runtime PM without callback functions") tried to eliminate the need for runtime PM callbacks by modifying pci_pm_runtime_suspend() and pci_pm_runtime_resume(), but didn't modify pci_pm_runtime_idle() with relevant changes, which still returns -ENOSYS if the driver supplies no runtime PM callbacks.

Fix this by modifying pci_pm_runtime_idle() such that it allows PCI device power state transitions without runtime PM callbacks.

 0)               |  pm_runtime_work() {
 0)               |    rpm_idle() {
 0)               |      rpm_check_suspend_allowed() {
 0)   1.500 us    |        __dev_pm_qos_resume_latency(); /* = 0x7fffffff */
 0)   4.840 us    |      } /* rpm_check_suspend_allowed = 0x0 */
 0)   1.550 us    |      __rpm_get_callback(); /* = 0xffffffffb4bc84f0 */
 0)   1.800 us    |      pci_pm_runtime_idle(); /* = -38 */
 0) + 17.070 us   |    } /* rpm_idle = -38 */
 0) + 22.450 us   |  } /* pm_runtime_work = -38 */
Quoted 5 lines
Quoted 28 lines

---

This is not marked for linux-stable for the need of extensive testing and can be backported after a few releases if no issues are reported.

 drivers/pci/pci-driver.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)
drivers/pci/pci-driver.c +1 -5
index 51ec9e7e784f..bb7f6775b350 100644
@@ -1382,10 +1382,7 @@ static int pci_pm_runtime_idle(struct device *dev)
        if (!pci_dev->driver)
                return 0;
 
-       if (!pm)
-               return -ENOSYS;
-
-       if (pm->runtime_idle)
+       if (pm && pm->runtime_idle)
                return pm->runtime_idle(dev);
 
        return 0;
--

2.35.3

#2 lore raw
In reply to Raag Jadav
Quoted 9 lines · Raag Jadav

Commit c5eb1190074c ("PCI / PM: Allow runtime PM without callback functions") tried to eliminate the need for runtime PM callbacks by modifying pci_pm_runtime_suspend() and pci_pm_runtime_resume(), but didn't modify pci_pm_runtime_idle() with relevant changes, which still returns -ENOSYS if the driver supplies no runtime PM callbacks.

Fix this by modifying pci_pm_runtime_idle() such that it allows PCI device power state transitions without runtime PM callbacks.

Quoted 9 lines
 0)               |  pm_runtime_work() {
 0)               |    rpm_idle() {
 0)               |      rpm_check_suspend_allowed() {
 0)   1.500 us    |        __dev_pm_qos_resume_latency(); /* = 0x7fffffff */
 0)   4.840 us    |      } /* rpm_check_suspend_allowed = 0x0 */
 0)   1.550 us    |      __rpm_get_callback(); /* = 0xffffffffb4bc84f0 */
 0)   1.800 us    |      pci_pm_runtime_idle(); /* = -38 */
 0) + 17.070 us   |    } /* rpm_idle = -38 */
 0) + 22.450 us   |  } /* pm_runtime_work = -38 */

What is this timing information telling me?

Quoted 1 line

Sounds like this resolves a problem report? Is there a URL we can cite? If not, at least a mention of what the user-visible problem is?

From the c5eb1190074c commit log, it sounds like maybe this allows devices to be autosuspended when they previously could not be?

Possibly this should have "Fixes: c5eb1190074c ("PCI / PM: Allow runtime PM without callback functions")" since it sounds like it goes with it?

Quoted 8 lines

---

This is not marked for linux-stable for the need of extensive testing and can be backported after a few releases if no issues are reported.

If you think this should not get backported to stable, you'll have to watch the backports to prevent it. Lots of stuff gets auto-backported even though not explicitly marked for stable. This comment won't prevent it (and won't even appear in the commit log).

Quoted 22 lines
 drivers/pci/pci-driver.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)
drivers/pci/pci-driver.c +1 -4
index 51ec9e7e784f..bb7f6775b350 100644
@@ -1382,10 +1382,7 @@ static int pci_pm_runtime_idle(struct device *dev)
 	if (!pci_dev->driver)
 		return 0;
 
-	if (!pm)
-		return -ENOSYS;
-
-	if (pm->runtime_idle)
+	if (pm && pm->runtime_idle)
 		return pm->runtime_idle(dev);
 
 	return 0;
Signature · 1 line
2.35.3
#3 lore raw

Hi

Quoted 5 lines · Bjorn Helgaas
Quoted 1 line

Sounds like this resolves a problem report? Is there a URL we can cite? If not, at least a mention of what the user-visible problem is?

Quoted 7 lines

From the c5eb1190074c commit log, it sounds like maybe this allows
devices to be autosuspended when they previously could not be?

Possibly this should have "Fixes: c5eb1190074c ("PCI / PM: Allow runtime PM without callback functions")" since it sounds like it goes with it?

I don't think there's known regression but my above commit wasn't complete. Autosuspending works without runtime PM callback as long as the driver has the PM callbacks structure set.

For example the drivers/i2c/busses/i2c-i801.c has system suspend/resume callbacks. I tested this patch by hack-removing them and yes, autosuspend doesn't work without this patch.

Raag and Mika noticed the issue when cleaning up empty runtime PM callbacks from an another driver which doesn't have any other PM callbacks.

#4 lore raw
Quoted 9 lines · Jarkko Nikula
Quoted 8 lines · Bjorn Helgaas
Quoted 1 line

Sounds like this resolves a problem report? Is there a URL we can cite? If not, at least a mention of what the user-visible problem is?

From the c5eb1190074c commit log, it sounds like maybe this allows
devices to be autosuspended when they previously could not be?

Quoted 7 lines
Quoted 4 lines

Possibly this should have "Fixes: c5eb1190074c ("PCI / PM: Allow runtime PM without callback functions")" since it sounds like it goes with it?

I don't think there's known regression but my above commit wasn't complete. Autosuspending works without runtime PM callback as long as the driver has the PM callbacks structure set.

I didn't suggest there was a regression, but if we mention that Mika debugged something, I want to know what the something was.

I'm guessing runtime PM doesn't work for some subset of drivers, and this patch fixes that. So let's say exactly how to find that subset of drivers, e.g., "drivers that implement X but not Y" or whatever.

Quoted 6 lines

For example the drivers/i2c/busses/i2c-i801.c has system suspend/resume callbacks. I tested this patch by hack-removing them and yes, autosuspend doesn't work without this patch.

Raag and Mika noticed the issue when cleaning up empty runtime PM callbacks from an another driver which doesn't have any other PM callbacks.

Bjorn

author
#5 lore raw
Quoted 14 lines · Bjorn Helgaas
Quoted 13 lines · Jarkko Nikula
Quoted 12 lines · Bjorn Helgaas
Quoted 1 line

Sounds like this resolves a problem report? Is there a URL we can cite? If not, at least a mention of what the user-visible problem is?

From the c5eb1190074c commit log, it sounds like maybe this allows
devices to be autosuspended when they previously could not be?

Possibly this should have "Fixes: c5eb1190074c ("PCI / PM: Allow runtime PM without callback functions")" since it sounds like it goes with it?

Quoted 6 lines
Quoted 3 lines

I don't think there's known regression but my above commit wasn't complete. Autosuspending works without runtime PM callback as long as the driver has the PM callbacks structure set.

I didn't suggest there was a regression, but if we mention that Mika debugged something, I want to know what the something was.

Considering it's not a bug to begin with, perhaps we can change it to Suggested-by or Co-developed-by?

Raag

author
#6 lore raw
Quoted 19 lines · Raag Jadav
Quoted 18 lines · Bjorn Helgaas
Quoted 16 lines · Jarkko Nikula
Quoted 5 lines · Bjorn Helgaas
Quoted 1 line

Sounds like this resolves a problem report? Is there a URL we can cite? If not, at least a mention of what the user-visible problem is?

Quoted 7 lines

From the c5eb1190074c commit log, it sounds like maybe this allows
devices to be autosuspended when they previously could not be?

Possibly this should have "Fixes: c5eb1190074c ("PCI / PM: Allow runtime PM without callback functions")" since it sounds like it goes with it?

I don't think there's known regression but my above commit wasn't complete. Autosuspending works without runtime PM callback as long as the driver has the PM callbacks structure set.

Quoted 5 lines
Quoted 2 lines

I didn't suggest there was a regression, but if we mention that Mika debugged something, I want to know what the something was.

Considering it's not a bug to begin with, perhaps we can change it to Suggested-by or Co-developed-by?

Hi Mika,

If you are okay with this, please let me know and perhaps suggest a better fit for the scenario.

Raag

#7 lore raw
Quoted 23 lines · Raag Jadav
Quoted 22 lines · Raag Jadav
Quoted 20 lines · Bjorn Helgaas
Quoted 9 lines · Jarkko Nikula
Quoted 8 lines · Bjorn Helgaas
Quoted 1 line

Sounds like this resolves a problem report? Is there a URL we can cite? If not, at least a mention of what the user-visible problem is?

From the c5eb1190074c commit log, it sounds like maybe this allows
devices to be autosuspended when they previously could not be?

Quoted 7 lines
Quoted 4 lines

Possibly this should have "Fixes: c5eb1190074c ("PCI / PM: Allow runtime PM without callback functions")" since it sounds like it goes with it?

I don't think there's known regression but my above commit wasn't complete. Autosuspending works without runtime PM callback as long as the driver has the PM callbacks structure set.

I didn't suggest there was a regression, but if we mention that Mika debugged something, I want to know what the something was.

Quoted 7 lines
Quoted 2 lines

Considering it's not a bug to begin with, perhaps we can change it to Suggested-by or Co-developed-by?

Hi Mika,

If you are okay with this, please let me know and perhaps suggest a better fit for the scenario.

You can just drop my name from it completely.

author
#8 lore raw
In reply to Bjorn Helgaas
Quoted 20 lines · Bjorn Helgaas
Quoted 9 lines · Raag Jadav

Commit c5eb1190074c ("PCI / PM: Allow runtime PM without callback functions") tried to eliminate the need for runtime PM callbacks by modifying pci_pm_runtime_suspend() and pci_pm_runtime_resume(), but didn't modify pci_pm_runtime_idle() with relevant changes, which still returns -ENOSYS if the driver supplies no runtime PM callbacks.

Fix this by modifying pci_pm_runtime_idle() such that it allows PCI device power state transitions without runtime PM callbacks.

Quoted 9 lines
 0)               |  pm_runtime_work() {
 0)               |    rpm_idle() {
 0)               |      rpm_check_suspend_allowed() {
 0)   1.500 us    |        __dev_pm_qos_resume_latency(); /* = 0x7fffffff */
 0)   4.840 us    |      } /* rpm_check_suspend_allowed = 0x0 */
 0)   1.550 us    |      __rpm_get_callback(); /* = 0xffffffffb4bc84f0 */
 0)   1.800 us    |      pci_pm_runtime_idle(); /* = -38 */
 0) + 17.070 us   |    } /* rpm_idle = -38 */
 0) + 22.450 us   |  } /* pm_runtime_work = -38 */
Quoted 1 line

What is this timing information telling me?

It's a raw ftrace dump.

Quoted 11 lines
Quoted 1 line

Sounds like this resolves a problem report? Is there a URL we can cite? If not, at least a mention of what the user-visible problem is?

From the c5eb1190074c commit log, it sounds like maybe this allows devices to be autosuspended when they previously could not be?

Possibly this should have "Fixes: c5eb1190074c ("PCI / PM: Allow runtime PM without callback functions")" since it sounds like it goes with it?

As pointed out by Jarkko, it's not a regression. The implementation in original commit is incomplete. We discovered it while cleaning up another PCI based driver.

Quoted 6 lines
Quoted 6 lines

---

Quoted 7 lines
Quoted 2 lines

This is not marked for linux-stable for the need of extensive testing and can be backported after a few releases if no issues are reported.

If you think this should not get backported to stable, you'll have to watch the backports to prevent it. Lots of stuff gets auto-backported even though not explicitly marked for stable. This comment won't prevent it (and won't even appear in the commit log).

This is why I've added Greg and Sasha here.

Raag

#9 lore raw
Quoted 2 lines · Raag Jadav
Quoted 1 line · Bjorn Helgaas

On Mon, Feb 12, 2024 at 12:02:33PM +0530, Raag Jadav wrote:

...

Quoted 10 lines
Quoted 10 lines
Quoted 9 lines
 0)               |  pm_runtime_work() {
 0)               |    rpm_idle() {
 0)               |      rpm_check_suspend_allowed() {
 0)   1.500 us    |        __dev_pm_qos_resume_latency(); /* = 0x7fffffff */
 0)   4.840 us    |      } /* rpm_check_suspend_allowed = 0x0 */
 0)   1.550 us    |      __rpm_get_callback(); /* = 0xffffffffb4bc84f0 */
 0)   1.800 us    |      pci_pm_runtime_idle(); /* = -38 */
 0) + 17.070 us   |    } /* rpm_idle = -38 */
 0) + 22.450 us   |  } /* pm_runtime_work = -38 */
Quoted 3 lines
Quoted 1 line

What is this timing information telling me?

It's a raw ftrace dump.

(Told ya that people would be surprised with this without seeing how you get
this and what fields mean)

Signature · 2 lines
With Best Regards,
Andy Shevchenko
author
#10 lore raw
Quoted 16 lines · Andy Shevchenko
Quoted 2 lines · Raag Jadav
Quoted 1 line · Bjorn Helgaas

On Mon, Feb 12, 2024 at 12:02:33PM +0530, Raag Jadav wrote:

...

Quoted 10 lines
Quoted 10 lines
Quoted 9 lines
 0)               |  pm_runtime_work() {
 0)               |    rpm_idle() {
 0)               |      rpm_check_suspend_allowed() {
 0)   1.500 us    |        __dev_pm_qos_resume_latency(); /* = 0x7fffffff */
 0)   4.840 us    |      } /* rpm_check_suspend_allowed = 0x0 */
 0)   1.550 us    |      __rpm_get_callback(); /* = 0xffffffffb4bc84f0 */
 0)   1.800 us    |      pci_pm_runtime_idle(); /* = -38 */
 0) + 17.070 us   |    } /* rpm_idle = -38 */
 0) + 22.450 us   |  } /* pm_runtime_work = -38 */
Quoted 6 lines
Quoted 3 lines
Quoted 1 line

What is this timing information telling me?

It's a raw ftrace dump.

(Told ya that people would be surprised with this without seeing how you get
this and what fields mean)

I can add stat headers in v2 which I think will be more helpful.

Raag

#11 lore raw
Quoted 17 lines · Raag Jadav
Quoted 16 lines · Andy Shevchenko
Quoted 2 lines · Raag Jadav
Quoted 1 line · Bjorn Helgaas

On Mon, Feb 12, 2024 at 12:02:33PM +0530, Raag Jadav wrote:

...

Quoted 10 lines
Quoted 10 lines
Quoted 9 lines
 0)               |  pm_runtime_work() {
 0)               |    rpm_idle() {
 0)               |      rpm_check_suspend_allowed() {
 0)   1.500 us    |        __dev_pm_qos_resume_latency(); /* = 0x7fffffff */
 0)   4.840 us    |      } /* rpm_check_suspend_allowed = 0x0 */
 0)   1.550 us    |      __rpm_get_callback(); /* = 0xffffffffb4bc84f0 */
 0)   1.800 us    |      pci_pm_runtime_idle(); /* = -38 */
 0) + 17.070 us   |    } /* rpm_idle = -38 */
 0) + 22.450 us   |  } /* pm_runtime_work = -38 */
Quoted 8 lines
Quoted 6 lines
Quoted 3 lines
Quoted 1 line

What is this timing information telling me?

It's a raw ftrace dump.

(Told ya that people would be surprised with this without seeing how you get
this and what fields mean)

I can add stat headers in v2 which I think will be more helpful.

That's not what I was asking. *Why* is the ftrace dump here? Is the point that we're calling a function we shouldn't? That this patch improves performance? Without some interpretation of what the dump shows, it's just noise.

Bjorn