On Mon, Aug 29, 2016 at 12:53:26PM +0200, Christoph Hellwig wrote: > This series is one third of the earlier "automatic interrupt affinity for > MSI/MSI-X capable devices" series, and make uses of the new irq-level > interrupt / queue mapping code in blk-mq, as well as allowing the driver > to pass in such a mask obtained from the (PCI) interrupt code. To fully > support this feature in drivers the final third in the PCI layer will > be needed as well. > > Note that these patches are on top of Linux 4.8-rc4 and will need several > patches not yet in the block for-4.9 branch. > > A git tree is available at: > > git://git.infradead.org/users/hch/block.git block-queue-mapping > > Gitweb: > > http://git.infradead.org/users/hch/block.git/shortlog/refs/heads/block-queue-mapping > > Changes since V1: > - rebased on top of Linux 4.8-rc4 > > Changes since automatic interrupt affinity for MSI/MSI-X capable devices V3: > - a trivial cleanup in blk_mq_create_mq_map pointed out by Alexander > I really like how this is looking, but the pairing isn't coming out right when I applied this on one of my test machines. I have 32 CPUs and 31 MSI-x vectors, and this is the resulting blk-mq cpu list: # cat /sys/block/nvme0n1/mq/*/cpu_list 0 10 11 12 13 14 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31 1 2 3 4 5 6 7 8 9 Before, each mq hardware queue would have 2 CPUs, now it's terribly unbalanced. I'll look again tomorrow to make sure I didn't mess something up when merging your tree, but just wanted to let you know now. -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
