2024-06-04 13:41:31

by Niklas Schnelle

Subject: [PATCH] PCI: Add missing lockdep assertion in pci_cfg_access_trylock()

In commit 7e89efc6e9e4 ("PCI: Lock upstream bridge for
pci_reset_function()") it was missed that pci_cfg_access_trylock() needs
the same lockdep assertion as pci_cfg_access_lock(). This leads to false
positive lockdep splats for users of pci_cfg_access_trylock(). Add the
missing assertion when the lock was successfully acquired.

Fixes: 7e89efc6e9e4 ("PCI: Lock upstream bridge for pci_reset_function()")
Signed-off-by: Niklas Schnelle <[email protected]>
---
drivers/pci/access.c | 2 ++
1 file changed, 2 insertions(+)

diff --git a/drivers/pci/access.c b/drivers/pci/access.c
index 30f031de9cfe..5b6620da30d7 100644
--- a/drivers/pci/access.c
+++ b/drivers/pci/access.c
@@ -319,6 +319,8 @@ bool pci_cfg_access_trylock(struct pci_dev *dev)
dev->block_cfg_access = 1;
raw_spin_unlock_irqrestore(&pci_lock, flags);

+ if (locked)
+ lock_map_acquire(&dev->cfg_access_lock);
return locked;
}
EXPORT_SYMBOL_GPL(pci_cfg_access_trylock);

---
base-commit: c3f38fa61af77b49866b006939479069cd451173
change-id: 20240604-pci_cfg_lockdep-b6914e62d726

Best regards,
--
Niklas Schnelle

2024-06-04 14:24:58

by Niklas Schnelle

Subject: Re: [PATCH] PCI: Add missing lockdep assertion in pci_cfg_access_trylock()

On Tue, 2024-06-04 at 15:38 +0200, Niklas Schnelle wrote:
> In commit 7e89efc6e9e4 ("PCI: Lock upstream bridge for
> pci_reset_function()") it was missed that pci_cfg_access_trylock() needs
> the same lockdep assertion as pci_cfg_access_lock(). This leads to false
> positive lockdep splats for users of pci_cfg_access_trylock(). Add the
> missing assertion when the lock was successfully acquired.
>
> Fixes: 7e89efc6e9e4 ("PCI: Lock upstream bridge for pci_reset_function()")
> Signed-off-by: Niklas Schnelle <[email protected]>
> ---
> drivers/pci/access.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/drivers/pci/access.c b/drivers/pci/access.c
> index 30f031de9cfe..5b6620da30d7 100644
> --- a/drivers/pci/access.c
> +++ b/drivers/pci/access.c
> @@ -319,6 +319,8 @@ bool pci_cfg_access_trylock(struct pci_dev *dev)
> dev->block_cfg_access = 1;
> raw_spin_unlock_irqrestore(&pci_lock, flags);
>
> + if (locked)
> + lock_map_acquire(&dev->cfg_access_lock);
> return locked;
> }
> EXPORT_SYMBOL_GPL(pci_cfg_access_trylock);
>
> ---
> base-commit: c3f38fa61af77b49866b006939479069cd451173
> change-id: 20240604-pci_cfg_lockdep-b6914e62d726
>
> Best regards,

Matt just pointed out to me that the same issue is also discussed with
a proposed fix in the discussion of the fixed commit[0]. That
additionally moves the lockdep asserts into the pci_lock critical
section.

[0]
https://lore.kernel.org/all/[email protected]/