On Wed, Apr 27, 2016 at 10:07 AM, Haggai Abramovsky <hag...@mellanox.com> wrote: > The dma_alloc_coherent() function returns a virtual address which can > be used for coherent access to the underlying memory. On some > architectures, like arm64, undefined behavior results if this memory is > also accessed via virtual mappings that are not coherent. Because of > their undefined nature, operations like virt_to_page() return garbage > when passed virtual addresses obtained from dma_alloc_coherent(). Any > subsequent mappings via vmap() of the garbage page values are unusable > and result in bad things like bus errors (synchronous aborts in ARM64 > speak). > > The mlx4 driver contains code that does the equivalent of: > vmap(virt_to_page(dma_alloc_coherent)), this results in an OOPs when the > device is opened. > > Prevent Ethernet driver to run this problematic code by forcing it to > allocate contiguous memory. As for the Infiniband driver, at first we > are trying to allocate contiguous memory, but in case of failure roll > back to work with fragmented memory.
Dave, The patch changes the driver to do single allocation for potentially very large HW WQE descriptor buffers such as those used by the RDMA (mlx5_ib) driver. The IB driver does have the means to cope with fragmented allocations, and under RDMA use cases, QPs are being frequently set not only on system startup, but rather throughout all the lifecycles (e.g every now and then in production systems). As of all the above, we prefer the patch to go to net-next and not net. This will make the code (1) correct, and (2) we have the chance to do the whatever investigations needed and if required add a follow up fix for 4.7-rc Or. > Signed-off-by: Haggai Abramovsky <hag...@mellanox.com> > Signed-off-by: Yishai Hadas <yish...@mellanox.com> > Reported-by: David Daney <david.da...@cavium.com> > Tested-by: Sinan Kaya <ok...@codeaurora.org>
