nfp: don't hold PF lock while enabling SR-IOV
authorJakub Kicinski <jakub.kicinski@netronome.com>
Wed, 23 Aug 2017 06:22:42 +0000 (23:22 -0700)
committerDavid S. Miller <davem@davemloft.net>
Thu, 24 Aug 2017 03:39:44 +0000 (20:39 -0700)
commitd6e1ab9ea3514840e4f32957c457b094646c2e9d
tree1cd1a376d6902098d629ba0d83bcad2408064b76
parent2f19f50e0f55a8fa067d3705e3f812a4f709567f
nfp: don't hold PF lock while enabling SR-IOV

Enabling SR-IOV VFs will cause the PCI subsystem to schedule a
work and flush its workqueue.  Since the nfp driver schedules its
own work we can't enable VFs while holding driver load.  Commit
6d48ceb27af1 ("nfp: allocate a private workqueue for driver work")
tried to avoid this deadlock by creating a separate workqueue.
Unfortunately, due to the architecture of workqueue subsystem this
does not guarantee a separate thread of execution.  Luckily
we can simply take pci_enable_sriov() from under the driver lock.

Take pci_disable_sriov() from under the lock too for symmetry.

Fixes: 6d48ceb27af1 ("nfp: allocate a private workqueue for driver work")
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/netronome/nfp/nfp_main.c