PCI: switchtec: Fix stdev_release() crash after surprise hot remove
authorDaniel Stodden <dns@arista.com>
Wed, 22 Nov 2023 04:23:16 +0000 (20:23 -0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Mon, 5 Feb 2024 20:14:31 +0000 (20:14 +0000)
commit0233b836312e39a3c763fb53512b3fa455b473b3
treead3bbb995c82be6a81f9e4f6fe402684f5435b17
parentcc56867d3226c1b5f142c77af8de149c0d5f774c
PCI: switchtec: Fix stdev_release() crash after surprise hot remove

[ Upstream commit df25461119d987b8c81d232cfe4411e91dcabe66 ]

A PCI device hot removal may occur while stdev->cdev is held open. The call
to stdev_release() then happens during close or exit, at a point way past
switchtec_pci_remove(). Otherwise the last ref would vanish with the
trailing put_device(), just before return.

At that later point in time, the devm cleanup has already removed the
stdev->mmio_mrpc mapping. Also, the stdev->pdev reference was not a counted
one. Therefore, in DMA mode, the iowrite32() in stdev_release() will cause
a fatal page fault, and the subsequent dma_free_coherent(), if reached,
would pass a stale &stdev->pdev->dev pointer.

Fix by moving MRPC DMA shutdown into switchtec_pci_remove(), after
stdev_kill(). Counting the stdev->pdev ref is now optional, but may prevent
future accidents.

Reproducible via the script at
https://lore.kernel.org/r/20231113212150.96410-1-dns@arista.com

Link: https://lore.kernel.org/r/20231122042316.91208-2-dns@arista.com
Signed-off-by: Daniel Stodden <dns@arista.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Logan Gunthorpe <logang@deltatee.com>
Reviewed-by: Dmitry Safonov <dima@arista.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
drivers/pci/switch/switchtec.c