PCI: pciehp: fix circular lock dependency b/w pci_rescan_remove_lock and reset_lock
PATCH 1/1 PCI: pciehp: fix circular lock dependency b/w pci_rescan_remove_lock and reset_lock
Resolves a circular locking dependency issue between pci_rescan_remove_lock() and ctrl->reset_lock() in the PCI hotplug subsystem, specifically in the pciehp_unconfigure_device() function.
Commit f5eff5591b8f ("PCI: pciehp: Fix AB-BA deadlock between reset_lock
and device_lock") introduced a change in the locking order withinpciehp_unconfigure_device() to avoid an AB-BA deadlock between ctrl->reset_lock and device_lock. However, this change inadvertently introduced a new circular locking dependency between pci_rescan_remove_lock and ctrl->reset_lock, as detected by lockdep.
The problematic sequence is as follows:
1. pciehp_unconfigure_device() acquires ctrl->reset_lock (read lock).
2. It then acquires pci_rescan_remove_lock.
3. Within the device removal loop, it attempts to reacquire ctrl->reset_lock after releasing it for driver unbinding, while still holding pci_rescan_remove_lock.
This creates a potential for deadlock if another thread acquires the locks in the opposite order, as illustrated by lockdep's report.
To resolve this, change the locking order in pciehp_unconfigure_device() so that ctrl->reset_lock is released before acquiring pci_rescan_remove_lock and before driver unbinding. This avoids holding both locks at the same time and breaks the circular dependency. After the critical section, ctrl->reset_lock is reacquired as needed.
This ensures that the locking order is consistent and prevents the circular dependency, addressing the lockdep warning and potential deadlock.
[ 120.615285] ======================================================
[ 120.615289] WARNING: possible circular locking dependency detected
[ 120.615293] 6.14.5-300.fc42.x86_64+debug #1 Not tainted
[ 120.615297] ------------------------------------------------------
[ 120.615300] irq/36-pciehp/136 is trying to acquire lock:
[ 120.615303] ffff88810d247340 (&ctrl->reset_lock){.+.+}-{4:4}, at: pciehp_unconfigure_device+0x1d0/0x390
[ 120.615323]
but task is already holding lock:
[ 120.615326] ffffffff9c13ce90 (pci_rescan_remove_lock){+.+.}-{4:4}, at: pciehp_unconfigure_device+0xe5/0x390
[ 120.615337]
which lock already depends on the new lock.[ 120.615340]
the existing dependency chain (in reverse order) is:
[ 120.615343]
-> #1 (pci_rescan_remove_lock){+.+.}-{4:4}:
[ 120.615351] lock_acquire.part.0+0x133/0x390
[ 120.615359] __mutex_lock+0x1b3/0x1490
[ 120.615366] pciehp_unconfigure_device+0xe5/0x390
[ 120.615370] pciehp_disable_slot+0xfa/0x2f0
[ 120.615376] pciehp_handle_presence_or_link_change+0xc8/0x310
[ 120.615381] pciehp_ist+0x2d5/0x3e0
[ 120.615386] irq_thread_fn+0x88/0x160
[ 120.615393] irq_thread+0x21e/0x490
[ 120.615399] kthread+0x39d/0x760
[ 120.615406] ret_from_fork+0x31/0x70
[ 120.615412] ret_from_fork_asm+0x1a/0x30
[ 120.615418]
-> #0 (&ctrl->reset_lock){.+.+}-{4:4}:
[ 120.615426] check_prev_add+0x1ab/0x23b0
[ 120.615431] __lock_acquire+0x2311/0x2e10
[ 120.615436] lock_acquire.part.0+0x133/0x390
[ 120.615441] down_read_nested+0xa4/0x490
[ 120.615445] pciehp_unconfigure_device+0x1d0/0x390
[ 120.615448] pciehp_disable_slot+0xfa/0x2f0
[ 120.615453] pciehp_handle_presence_or_link_change+0xc8/0x310
[ 120.615458] pciehp_ist+0x2d5/0x3e0
[ 120.615462] irq_thread_fn+0x88/0x160
[ 120.615465] irq_thread+0x21e/0x490
[ 120.615469] kthread+0x39d/0x760
[ 120.615474] ret_from_fork+0x31/0x70
[ 120.615478] ret_from_fork_asm+0x1a/0x30
[ 120.615483]
other info that might help us debug this:[ 120.615485] Possible unsafe locking scenario:
[ 120.615488] CPU0 CPU1
[ 120.615490] ---- ----
[ 120.615492] lock(pci_rescan_remove_lock);
[ 120.615497] lock(&ctrl->reset_lock);
[ 120.615502] lock(pci_rescan_remove_lock);
[ 120.615506] rlock(&ctrl->reset_lock);
[ 120.615511]
*** DEADLOCK ***--- drivers/pci/hotplug/pciehp_pci.c | 16 +++++++++++----- 1 file changed, 11 insertions(+), 5 deletions(-)
drivers/pci/hotplug/pciehp_pci.c +11 -5
@@ -104,6 +104,11 @@ void pciehp_unconfigure_device(struct controller *ctrl, bool presence)if (!presence)pci_walk_bus (parent, pci_dev_set_disconnected,NULL ); +/* +* Release reset_lock before driver unbinding +* to avoid AB-BA deadlock with device_lock. +*/ +up_read (&ctrl->reset_lock);pci_lock_rescan_remove ();/*
@@ -116,11 +121,6 @@ void pciehp_unconfigure_device(struct controller *ctrl, bool presence) bus_list) {pci_dev_get (dev); -/* -* Release reset_lock during driver unbinding -* to avoid AB-BA deadlock with device_lock. -*/ -up_read (&ctrl->reset_lock);pci_stop_and_remove_bus_device (dev);down_read_nested (&ctrl->reset_lock, ctrl->depth);
@@ -134,8 +134,14 @@ void pciehp_unconfigure_device(struct controller *ctrl, bool presence) command |=PCI_COMMAND_INTX_DISABLE ;pci_write_config_word (dev,PCI_COMMAND , command); } +/* +* Release reset_lock before driver unbinding +* to avoid AB-BA deadlock with device_lock. +*/ +up_read (&ctrl->reset_lock);pci_dev_put (dev); } +down_read_nested (&ctrl->reset_lock, ctrl->depth);pci_unlock_rescan_remove (); }
Signature · 1 line
2.34.1
Quoted 3 lines · Raju Rangoju
Resolves a circular locking dependency issue between
pci_rescan_remove_lock()andctrl->reset_lock()in the PCI hotplug subsystem, specifically in thepciehp_unconfigure_device()function.
This seems to be a false-positive because the two stacktraces are identical but pciehp_ist() is single-threaded. There is only ever a single instance of pciehp_ist() running per hotplug port, so two instances running on separate CPUs can never happen.
Previous reports of the same false-positive lockdep splat, only they refer to pciehp_configure_device() instead of pciehp_unconfigure_device():
https://lore.kernel.org/r/[email protected]/
https://lore.kernel.org/r/[email protected]/
Thanks,
Lukas