Re: RDMA/rpma + fsdax(ext4) was broken since 36f30e486d

[Jason Gunthorpe]

On Wed, Aug 04, 2021 at 04:06:53PM +0800, Li, Zhijian/李 智坚 wrote:
> convert to text and send again
> 
> 2021/8/4 15:55, Li, Zhijian wrote:
> > 
> > Hey all:
> > 
> > Recently, i reported a issue to rpmahttps://github.com/pmem/rpma/issues/1142
> > where we found that the native rpma + fsdax example failed in recent kernel.
> > 
> > Below is the bisect log
> > 
> > [lizhijian@yl linux]$ git bisect log
> > git bisect start
> > # good: [bbf5c979011a099af5dc76498918ed7df445635b] Linux 5.9
> > git bisect good bbf5c979011a099af5dc76498918ed7df445635b
> > # bad: [2c85ebc57b3e1817b6ce1a6b703928e113a90442] Linux 5.10
> > git bisect bad 2c85ebc57b3e1817b6ce1a6b703928e113a90442
> > # good: [4d0e9df5e43dba52d38b251e3b909df8fa1110be] lib, uaccess: add failure injection to usercopy functions
> > git bisect good 4d0e9df5e43dba52d38b251e3b909df8fa1110be
> > # bad: [6694875ef8045cdb1e6712ee9b68fe08763507d8] ext4: indicate that fast_commit is available via /sys/fs/ext4/feature/...
> > git bisect bad 6694875ef8045cdb1e6712ee9b68fe08763507d8
> > # good: [14c914fcb515c424177bb6848cc2858ebfe717a8] Merge tag 'wireless-drivers-next-2020-10-02' of git://git.kernel.org/pub/scm/linux/kernel/git/kvalo/wireless-drivers-next
> > git bisect good 14c914fcb515c424177bb6848cc2858ebfe717a8
> > # good: [6f78b9acf04fbf9ede7f4265e7282f9fb39d2c8c] Merge tag 'mtd/for-5.10' of git://git.kernel.org/pub/scm/linux/kernel/git/mtd/linux
> > git bisect good 6f78b9acf04fbf9ede7f4265e7282f9fb39d2c8c
> > # bad: [bbe85027ce8019c73ab99ad1c2603e2dcd1afa49] Merge tag 'xfs-5.10-merge-5' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux
> > git bisect bad bbe85027ce8019c73ab99ad1c2603e2dcd1afa49
> > # bad: [9d9af1007bc08971953ae915d88dc9bb21344b53] Merge tag 'perf-tools-for-v5.10-2020-10-15' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux
> > git bisect bad 9d9af1007bc08971953ae915d88dc9bb21344b53
> > # good: [21c2fe94abb2abe894e6aabe6b4e84a255c8d339] RDMA/mthca: Combine special QP struct with mthca QP
> > git bisect good 21c2fe94abb2abe894e6aabe6b4e84a255c8d339
> > # good: [dbaa1b3d9afba3c050d365245a36616ae3f425a7] Merge branch 'perf/urgent' into perf/core
> > git bisect good dbaa1b3d9afba3c050d365245a36616ae3f425a7
> > # bad: [c7a198c700763ac89abbb166378f546aeb9afb33] RDMA/ucma: Fix use after free in destroy id flow
> > git bisect bad c7a198c700763ac89abbb166378f546aeb9afb33
> > # bad: [5ce2dced8e95e76ff7439863a118a053a7fc6f91] RDMA/ipoib: Set rtnl_link_ops for ipoib interfaces
> > git bisect bad 5ce2dced8e95e76ff7439863a118a053a7fc6f91
> > # bad: [a03bfc37d59de316436c46f5691c5a972ed57c82] RDMA/mlx5: Sync device with CPU pages upon ODP MR registration
> > git bisect bad a03bfc37d59de316436c46f5691c5a972ed57c82
> > # good: [a6f0b08dbaf289c3c57284e16ac8043140f2139b] RDMA/core: Remove ucontext->closing
> > git bisect good a6f0b08dbaf289c3c57284e16ac8043140f2139b
> > # bad: [36f30e486dce22345c2dd3a3ba439c12cd67f6ba] IB/core: Improve ODP to use hmm_range_fault()
> > git bisect bad 36f30e486dce22345c2dd3a3ba439c12cd67f6ba
> > # good: [2ee9bf346fbfd1dad0933b9eb3a4c2c0979b633e] RDMA/addr: Fix race with netevent_callback()/rdma_addr_cancel()
> > git bisect good 2ee9bf346fbfd1dad0933b9eb3a4c2c0979b633e
> > # first bad commit: [36f30e486dce22345c2dd3a3ba439c12cd67f6ba] IB/core: Improve ODP to use hmm_range_fault()

This is perhaps not so surprising, but I think you should report it to
the dax people that hmm_range_fault and dax don't work together..

Though I think it is supposed to, and I'm surprised it doesn't.

Jason