From mboxrd@z Thu Jan 1 00:00:00 1970 From: martin.petersen@oracle.com (Martin K. Petersen) Date: Tue, 12 Jun 2018 20:35:55 -0400 Subject: [PATCH 08/17] nvme: limit integrity segments to be <= data segments In-Reply-To: <694e0f4e-bf79-6139-e5ba-86c01fcce4b2@mellanox.com> (Max Gurtovoy's message of "Sat, 9 Jun 2018 04:33:58 +0300") References: <1527436222-15494-1-git-send-email-maxg@mellanox.com> <1527436222-15494-9-git-send-email-maxg@mellanox.com> <19c8562a-8478-196c-c9f5-385d2076448e@grimberg.me> <694e0f4e-bf79-6139-e5ba-86c01fcce4b2@mellanox.com> Message-ID: Max, > I agree, but do you think it's reasonable to set max_integrity_segments > > max_segments ? The upper bound for max_integrity_segments is number of sectors in the I/O. That's clearly a pathological corner case. But we used to be able to hit it with the "right" choice of kernel memory allocator and dd with bs=512 (effectively because the allocator always allocated backwards in memory so no two 8-byte PI allocations for a 512-byte buffer_head could be combined into one contiguous segment). -- Martin K. Petersen Oracle Linux Engineering