dmaengine: iota: ioat_alloc_chan_resources should not perform sleeping allocations.
authorKrister Johansen <kjlx@templeofstupid.com>
Wed, 4 Jan 2017 09:22:52 +0000 (01:22 -0800)
committerVinod Koul <vinod.koul@intel.com>
Tue, 10 Jan 2017 05:11:51 +0000 (10:41 +0530)
commit21d25f6a4217e755906cb548b55ddab39d0e88b9
treeb6bc61498679621dc4c6c26e592edc16c308b7fc
parent5c9e6c2b2ba3ec3a442e2fb5b4286498f8b4dcb7
dmaengine: iota: ioat_alloc_chan_resources should not perform sleeping allocations.

On a kernel with DEBUG_LOCKS, ioat_free_chan_resources triggers an
in_interrupt() warning.  With PROVE_LOCKING, it reports detecting a
SOFTIRQ-safe to SOFTIRQ-unsafe lock ordering in the same code path.

This is because dma_generic_alloc_coherent() checks if the GFP flags
permit blocking.  It allocates from different subsystems if blocking is
permitted.  The free path knows how to return the memory to the correct
allocator.  If GFP_KERNEL is specified then the alloc and free end up
going through cma_alloc(), which uses mutexes.

Given that ioat_free_chan_resources() can be called in interrupt
context, ioat_alloc_chan_resources() must specify GFP_NOWAIT so that the
allocations do not block and instead use an allocator that uses
spinlocks.

Signed-off-by: Krister Johansen <kjlx@templeofstupid.com>
Acked-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
drivers/dma/ioat/init.c