IB/hfi1: Re-order IRQ cleanup to address driver cleanup race
authorMichael J. Ruhl <michael.j.ruhl@intel.com>
Thu, 1 Feb 2018 18:43:42 +0000 (10:43 -0800)
committerJason Gunthorpe <jgg@mellanox.com>
Thu, 1 Feb 2018 22:24:32 +0000 (15:24 -0700)
commit82a979265638c505e12fbe7ba40980dc0901436d
tree4ea19e1a41f220c466e822fb12ed97215ab90a85
parentf34727a1351362c8f349d29efcd0e72702571e6a
IB/hfi1: Re-order IRQ cleanup to address driver cleanup race

The pci_request_irq() interfaces always adds the IRQF_SHARED bit to
all IRQ requests.

When the kernel is built with CONFIG_DEBUG_SHIRQ config flag, if the
IRQF_SHARED bit is set, a call to the IRQ handler is made from the
__free_irq() function. This is testing a race condition between the
IRQ cleanup and an IRQ racing the cleanup.  The HFI driver should be
able to handle this race, but does not.

This race can cause traces that start with this footprint:

BUG: unable to handle kernel NULL pointer dereference at   (null)
Call Trace:
 <hfi1 irq handler>
 ...
 __free_irq+0x1b3/0x2d0
 free_irq+0x35/0x70
 pci_free_irq+0x1c/0x30
 clean_up_interrupts+0x53/0xf0 [hfi1]
 hfi1_start_cleanup+0x122/0x190 [hfi1]
 postinit_cleanup+0x1d/0x280 [hfi1]
 remove_one+0x233/0x250 [hfi1]
 pci_device_remove+0x39/0xc0

Export IRQ cleanup function so it can be called from other modules.

Using the exported cleanup function:

  Re-order the driver cleanup code to clean up IRQ resources before
  other resources, eliminating the race.

  Re-order error path for init so that the race does not occur.

Reduce severity on spurious error message for SDMA IRQs to info.

Reviewed-by: Alex Estrin <alex.estrin@intel.com>
Reviewed-by: Patel Jay P <jay.p.patel@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
drivers/infiniband/hw/hfi1/chip.c
drivers/infiniband/hw/hfi1/hfi.h
drivers/infiniband/hw/hfi1/init.c