scsi: libsas: direct call probe and destruct
authorJason Yan <yanaijie@huawei.com>
Fri, 8 Dec 2017 09:42:09 +0000 (17:42 +0800)
committerMartin K. Petersen <martin.petersen@oracle.com>
Thu, 11 Jan 2018 04:24:02 +0000 (23:24 -0500)
commit0558f33c06bb910e2879e355192227a8e8f0219d
treebbcddcb5078ec6f66d40e9a9855535a762f01b1c
parent517e5153d242cb2dd0a1150d2a7bd6788d501ca9
scsi: libsas: direct call probe and destruct

In commit 87c8331fcf72 ("[SCSI] libsas: prevent domain rediscovery
competing with ata error handling") introduced disco mutex to prevent
rediscovery competing with ata error handling and put the whole
revalidation in the mutex. But the rphy add/remove needs to wait for the
error handling which also grabs the disco mutex. This may leads to dead
lock.So the probe and destruct event were introduce to do the rphy
add/remove asynchronously and out of the lock.

The asynchronously processed workers makes the whole discovery process
not atomic, the other events may interrupt the process. For example,
if a loss of signal event inserted before the probe event, the
sas_deform_port() is called and the port will be deleted.

And sas_port_delete() may run before the destruct event, but the
port-x:x is the top parent of end device or expander. This leads to
a kernel WARNING such as:

[   82.042979] sysfs group 'power' not found for kobject 'phy-1:0:22'
[   82.042983] ------------[ cut here ]------------
[   82.042986] WARNING: CPU: 54 PID: 1714 at fs/sysfs/group.c:237
sysfs_remove_group+0x94/0xa0
[   82.043059] Call trace:
[   82.043082] [<ffff0000082e7624>] sysfs_remove_group+0x94/0xa0
[   82.043085] [<ffff00000864e320>] dpm_sysfs_remove+0x60/0x70
[   82.043086] [<ffff00000863ee10>] device_del+0x138/0x308
[   82.043089] [<ffff00000869a2d0>] sas_phy_delete+0x38/0x60
[   82.043091] [<ffff00000869a86c>] do_sas_phy_delete+0x6c/0x80
[   82.043093] [<ffff00000863dc20>] device_for_each_child+0x58/0xa0
[   82.043095] [<ffff000008696f80>] sas_remove_children+0x40/0x50
[   82.043100] [<ffff00000869d1bc>] sas_destruct_devices+0x64/0xa0
[   82.043102] [<ffff0000080e93bc>] process_one_work+0x1fc/0x4b0
[   82.043104] [<ffff0000080e96c0>] worker_thread+0x50/0x490
[   82.043105] [<ffff0000080f0364>] kthread+0xfc/0x128
[   82.043107] [<ffff0000080836c0>] ret_from_fork+0x10/0x50

Make probe and destruct a direct call in the disco and revalidate function,
but put them outside the lock. The whole discovery or revalidate won't
be interrupted by other events. And the DISCE_PROBE and DISCE_DESTRUCT
event are deleted as a result of the direct call.

Introduce a new list to destruct the sas_port and put the port delete after
the destruct. This makes sure the right order of destroying the sysfs
kobject and fix the warning above.

In sas_ex_revalidate_domain() have a loop to find all broadcasted
device, and sometimes we have a chance to find the same expander twice.
Because the sas_port will be deleted at the end of the whole revalidate
process, sas_port with the same name cannot be added before this.
Otherwise the sysfs will complain of creating duplicate filename. Since
the LLDD will send broadcast for every device change, we can only
process one expander's revalidation.

[mkp: kbuild test robot warning]

Signed-off-by: Jason Yan <yanaijie@huawei.com>
CC: John Garry <john.garry@huawei.com>
CC: Johannes Thumshirn <jthumshirn@suse.de>
CC: Ewan Milne <emilne@redhat.com>
CC: Christoph Hellwig <hch@lst.de>
CC: Tomas Henzl <thenzl@redhat.com>
CC: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
drivers/scsi/libsas/sas_ata.c
drivers/scsi/libsas/sas_discover.c
drivers/scsi/libsas/sas_expander.c
drivers/scsi/libsas/sas_internal.h
drivers/scsi/libsas/sas_port.c
include/scsi/libsas.h
include/scsi/scsi_transport_sas.h