nvme-pci: use appropriate initial chunk size for HMB allocation
authorAkinobu Mita <akinobu.mita@gmail.com>
Wed, 6 Sep 2017 10:15:31 +0000 (12:15 +0200)
committerChristoph Hellwig <hch@lst.de>
Mon, 11 Sep 2017 16:29:38 +0000 (12:29 -0400)
The initial chunk size for host memory buffer allocation is currently
PAGE_SIZE << MAX_ORDER.  MAX_ORDER order allocation is usually failed
without CONFIG_DMA_CMA.  So the HMB allocation is retried with chunk size
PAGE_SIZE << (MAX_ORDER - 1) in general, but there is no problem if the
retry allocation works correctly.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
[hch: rebased]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <keith.busch@intel.com>
Cc: stable@vger.kernel.org
drivers/nvme/host/pci.c

index 82e23c8..a370f70 100644 (file)
@@ -1680,7 +1680,7 @@ static int nvme_alloc_host_mem(struct nvme_dev *dev, u64 min, u64 preferred)
        u32 chunk_size;
 
        /* start big and work our way down */
-       for (chunk_size = min_t(u64, preferred, PAGE_SIZE << MAX_ORDER);
+       for (chunk_size = min_t(u64, preferred, PAGE_SIZE * MAX_ORDER_NR_PAGES);
             chunk_size >= PAGE_SIZE * 2;
             chunk_size /= 2) {
                if (!__nvme_alloc_host_mem(dev, preferred, chunk_size)) {