Re: [PATCH] PCI/PM: Put devices to low power state on shutdown

[Mika Westerberg]

Hi,

On Fri, Sep 13, 2024 at 02:00:58PM +0800, Kai-Heng Feng wrote:
> On Fri, Sep 13, 2024 at 12:57 AM Bjorn Helgaas <helgaas@xxxxxxxxxx> wrote:
> >
> > [+cc Rafael]
> >
> > On Thu, Sep 12, 2024 at 11:00:43AM +0800, Kai-Heng Feng wrote:
> > > On Thu, Sep 12, 2024 at 3:05 AM Bjorn Helgaas <helgaas@xxxxxxxxxx> wrote:
> > > > On Fri, Jul 12, 2024 at 02:24:11PM +0800, Kai-Heng Feng wrote:
> > > > > Some laptops wake up after poweroff when HP Thunderbolt Dock G4 is
> > > > > connected.
> > > > >
> > > > > The following error message can be found during shutdown:
> > > > > pcieport 0000:00:1d.0: AER: Correctable error message received from 0000:09:04.0
> > > > > pcieport 0000:09:04.0: PCIe Bus Error: severity=Correctable, type=Data Link Layer, (Receiver ID)
> > > > > pcieport 0000:09:04.0:   device [8086:0b26] error status/mask=00000080/00002000
> > > > > pcieport 0000:09:04.0:    [ 7] BadDLLP
> > > > >
> > > > > Calling aer_remove() during shutdown can quiesce the error message,
> > > > > however the spurious wakeup still happens.
> > > > >
> > > > > The issue won't happen if the device is in D3 before system shutdown, so
> > > > > putting device to low power state before shutdown to solve the issue.
> > > > >
> > > > > I don't have a sniffer so this is purely guesswork, however I believe
> > > > > putting device to low power state it's the right thing to do.
> > > >
> > > > My objection here is that we don't have an explanation of why this
> > > > should matter or a pointer to any spec language about this situation,
> > > > so it feels a little bit random.
> > >
> > > I have the same feeling too. The PCIe spec doesn't specify what's the
> > > correct power state for shutdown.
> > > So we can only "logically" think the software should put devices to
> > > low power state during shutdown.
> > >
> > > > I suppose the problem wouldn't happen if AER interrupts were disabled?
> > > > We already do disable them in aer_suspend(), but maybe that's not used
> > > > in the shutdown path?
> > >
> > > That was my first thought, so I modified pcie_port_shutdown_service()
> > > to disable AER interrupt.
> > > That approach didn't work though.
> > >
> > > > My understanding is that .shutdown() should turn off device interrupts
> > > > and stop DMA.  So maybe we need an aer_shutdown() that disables
> > > > interrupts?
> > >
> > > Logically we should do that. However that approach doesn't solve this issue.
> >
> > I'm not completely clear on the semantics of the .shutdown()
> > interface.  The doc at
> > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/include/linux/device/driver.h?id=v6.10#n73
> > says "@shutdown: Called at shut-down time to quiesce the device"
> >
> > Turning off device interrupts and DMA *would* fit within the idea of
> > quiescing the device.  Does that also include changing the device
> > power state?  I dunno.  The power state isn't *mentioned* in the
> > .shutdown() context, while it *is* mentioned for .suspend().
> 
> IMO putting a device to low power also qualifies as "quiesce the device".
> 
> >
> > IIUC, this patch and commit log uses "shutdown" to refer to a
> > system-wide *poweroff*, which is a different concept despite using the
> > same "shutdown" name.
> 
> For ACPI based system, there are .suspend for S3/s2idle, .poweroff for
> S4, and .shutdown for S5.
> Unless we want to introduce a new callback for S5, I think the concept
> is quite similar.
> 
> For DT based system, the OS should also perform the same thing, as
> there's no firmware to cleanup the power state.
> 
> We can also move .shutdown to be part of pm_ops, but I don't think
> it's necessary,
> 
> >
> > So should the system poweroff procedure use .suspend()?  Should it use
> > both .shutdown() and .suspend()?  I think it only uses .shutdown()
> > today:
> >
> >   kernel_power_off
> >     kernel_shutdown_prepare(SYSTEM_POWER_OFF)
> >       device_shutdown
> >         while (!list_empty(&devices_kset->list))
> >           dev->bus->shutdown(dev)
> >             pci_device_shutdown
> >
> > There are several driver .shutdown() methods that do things like this:
> >
> >   e1000_shutdown
> >     if (system_state == SYSTEM_POWER_OFF)
> >       pci_set_power_state(pdev, PCI_D3hot)
> >
> > Maybe that's the right thing and should be done by the PCI core, which
> > is similar to what you propose here.  But I think it muddies the
> > definition of .shutdown() a bit by mixing in power management stuff.
> 
> Do you think adding a new "low power state" callback to be called
> after .shutdown a good idea?
> That would make the concept of .shutdown different to .suspend and
> .poweroff. I personally see .suspend, .poweroff and .shutdown the same
> action but target different power states.

I don't mean to confuse you guys but with this one too, I wonder if you
tried to "disable" the device instead of putting it into D3? On another
thread (Mario at least is aware of this) I mentioned that our PCIe SV
folks identified a couple issues in Linux implementation around power
management and one thing that we are missing is to disable I/O and MMIO
upon entering D3.

I know this is about entering S5 (power off) but I wonder if simply
disabling the device (I/O, MMIO and bus mastering) could stop it from
waking up? To my understanding this can be interpreted as quiesce too :)
Something like the below patch (it also includes the runtime suspend
path which should not matter here. This is the similar patch I shared in
another thread).

diff --git a/drivers/pci/pci-driver.c b/drivers/pci/pci-driver.c
index f412ef73a6e4..79406814699d 100644
--- a/drivers/pci/pci-driver.c
+++ b/drivers/pci/pci-driver.c
@@ -514,11 +514,9 @@ static void pci_device_shutdown(struct device *dev)
 	 * If this is a kexec reboot, turn off Bus Master bit on the
 	 * device to tell it to not continue to do DMA. Don't touch
 	 * devices in D3cold or unknown states.
-	 * If it is not a kexec reboot, firmware will hit the PCI
-	 * devices with big hammer and stop their DMA any way.
 	 */
-	if (kexec_in_progress && (pci_dev->current_state <= PCI_D3hot))
-		pci_clear_master(pci_dev);
+	if (pci_dev->current_state <= PCI_D3hot)
+		pci_disable_device(pci_dev);
 }
 
 #ifdef CONFIG_PM_SLEEP
@@ -1332,6 +1330,7 @@ static int pci_pm_runtime_suspend(struct device *dev)
 
 	if (!pci_dev->state_saved) {
 		pci_save_state(pci_dev);
+		pci_pm_default_suspend(pci_dev);
 		pci_finish_runtime_suspend(pci_dev);
 	}
 
diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
index ffaaca0978cb..91f4e7a03c94 100644
--- a/drivers/pci/pci.c
+++ b/drivers/pci/pci.c
@@ -2218,6 +2218,13 @@ static void do_pci_disable_device(struct pci_dev *dev)
 		pci_command &= ~PCI_COMMAND_MASTER;
 		pci_write_config_word(dev, PCI_COMMAND, pci_command);
 	}
+	/*
+	 * PCI PM 1.2 sec 8.2.2 says that when a function is put into D3
+	 * the OS needs to disable I/O and MMIO space in addition to bus
+	 * mastering so do that here.
+	 */
+	pci_command &= ~(PCI_COMMAND_IO | PCI_COMMAND_MEMORY);
+	pci_write_config_word(dev, PCI_COMMAND, pci_command);
 
 	pcibios_disable_device(dev);
 }