Hi Bjorn,
I never got a reply to this one so resending with updated Subject in
case it got buried in your inbox.
On Mon, Jan 08, 2024 at 09:39:07AM +0100, Johan Hovold wrote:
> On Tue, Jan 02, 2024 at 05:25:50PM -0600, Bjorn Helgaas wrote:
> > From: Bjorn Helgaas <bhelgaas@xxxxxxxxxx>
> >
> > This reverts commit 08d0cc5f34265d1a1e3031f319f594bd1970976c.
> >
> > Michael reported that when attempting to resume from suspend to RAM on ASUS
> > mini PC PN51-BB757MDE1 (DMI model: MINIPC PN51-E1), 08d0cc5f3426
> > ("PCI/ASPM: Remove pcie_aspm_pm_state_change()") caused a 12-second delay
> > with no output, followed by a reboot.
> >
> > Workarounds include:
> >
> > - Reverting 08d0cc5f3426 ("PCI/ASPM: Remove pcie_aspm_pm_state_change()")
> > - Booting with "pcie_aspm=off"
> > - Booting with "pcie_aspm.policy=performance"
> > - "echo 0 | sudo tee /sys/bus/pci/devices/0000:03:00.0/link/l1_aspm"
> > before suspending
> > - Connecting a USB flash drive
> >
> > Fixes: 08d0cc5f3426 ("PCI/ASPM: Remove pcie_aspm_pm_state_change()")
> > Reported-by: Michael Schaller <michael@xxxxxxxxxxx>
> > Link: https://lore.kernel.org/r/76c61361-b8b4-435f-a9f1-32b716763d62@xxxxxxxxxxx
> > Signed-off-by: Bjorn Helgaas <bhelgaas@xxxxxxxxxx>
> > Cc: <stable@xxxxxxxxxxxxxxx>
> > ---
>
> > +/* @pdev: the root port or switch downstream port */
> > +void pcie_aspm_pm_state_change(struct pci_dev *pdev)
> > +{
> > + struct pcie_link_state *link = pdev->link_state;
> > +
> > + if (aspm_disabled || !link)
> > + return;
> > + /*
> > + * Devices changed PM state, we should recheck if latency
> > + * meets all functions' requirement
> > + */
> > + down_read(&pci_bus_sem);
> > + mutex_lock(&aspm_lock);
> > + pcie_update_aspm_capable(link->root);
> > + pcie_config_aspm_path(link);
> > + mutex_unlock(&aspm_lock);
> > + up_read(&pci_bus_sem);
> > +}
>
> This function is now restored in 6.7 final and is called in paths which
> already hold the pci_bus_sem as reported by lockdep (see splat below).
>
> This can potentially lead to a deadlock and specifically prevents using
> lockdep on Qualcomm platforms.
>
> Not sure if you want to propagate whether the bus semaphore is held to
> pcie_aspm_pm_state_change() or if there was some alternative to
> restoring this function which should be explored instead.
So to summarise, this patch, which is now commit
f93e71aea6c6 ("Revert "PCI/ASPM: Remove pcie_aspm_pm_state_change()"")
introduced a regression in 6.7-final for Qualcomm platforms (and some
Intel platforms) similar to the one recently fixed by commit
f352ce999260 ("PCI: qcom: Fix potential deadlock when enabling ASPM").
Johan
#regzbot introduced: f93e71aea6c6
> ============================================
> WARNING: possible recursive locking detected
> 6.7.0 #40 Not tainted
> --------------------------------------------
> kworker/u16:5/90 is trying to acquire lock:
> ffffacfa78ced000 (pci_bus_sem){++++}-{3:3}, at: pcie_aspm_pm_state_change+0x58/0xdc
> pcieport 0002:00:00.0: PME: Signaling with IRQ 197
>
> but task is already holding lock:
> ffffacfa78ced000 (pci_bus_sem){++++}-{3:3}, at: pci_walk_bus+0x34/0xbc
>
> other info that might help us debug this:
> Possible unsafe locking scenario:
>
> CPU0
> ----
> lock(pci_bus_sem);
> lock(pci_bus_sem);
>
> *** DEADLOCK ***
>
> May be due to missing lock nesting notation
>
> 4 locks held by kworker/u16:5/90:
> #0: ffff06c5c0008d38 ((wq_completion)events_unbound){+.+.}-{0:0}, at: process_one_work+0x150/0x53c
> #1: ffff800081c0bdd0 ((work_completion)(&entry->work)){+.+.}-{0:0}, at: process_one_work+0x150/0x53c
> #2: ffff06c5c0b7d0f8 (&dev->mutex){....}-{3:3}, at: __driver_attach_async_helper+0x3c/0xf4
> #3: ffffacfa78ced000 (pci_bus_sem){++++}-{3:3}, at: pci_walk_bus+0x34/0xbc
>
> stack backtrace:
> CPU: 1 PID: 90 Comm: kworker/u16:5 Not tainted 6.7.0 #40
> Hardware name: LENOVO 21BYZ9SRUS/21BYZ9SRUS, BIOS N3HET53W (1.25 ) 10/12/2022
> Workqueue: events_unbound async_run_entry_fn
> Call trace:
> dump_backtrace+0x9c/0x11c
> show_stack+0x18/0x24
> dump_stack_lvl+0x60/0xac
> dump_stack+0x18/0x24
> print_deadlock_bug+0x25c/0x348
> __lock_acquire+0x10a4/0x2064
> lock_acquire+0x1e8/0x318
> down_read+0x60/0x184
> pcie_aspm_pm_state_change+0x58/0xdc
> pci_set_full_power_state+0xa8/0x114
> pci_set_power_state+0xc4/0x120
> qcom_pcie_enable_aspm+0x1c/0x3c [pcie_qcom]
> pci_walk_bus+0x64/0xbc
> qcom_pcie_host_post_init_2_7_0+0x28/0x34 [pcie_qcom]
