Re: [PATCH] mpt3sas: Do scsi_remove_host() before deleting SAS PHY objects

[Calvin Owens <calvinowens@fb.com>]

On Monday 05/16 at 15:51 -0600, Sathya Prakash Veerichetty wrote:
> Our understanding is the relationship between the SCSI host and SAS end
> devices is a parent-child and before ripping of SCSI host we need to rip of
> all the children.

I know at least HPSA does the opposite, Elliott observes that other
drivers do as well.

> Why the enclosure ends up trying to re-parent the SCSI device from the
> enclosure to the SAS PHY even after we remove the SAS Phy?

Take this path in /sys:

/sys/devices/pci0000:00/0000:00:01.0/0000:01:00.0/host6/port-6:0/expander-6:0/port-6:0:0/end_device-6:0:0/target6:0:0/6:0:0:0/scsi_device/6:0:0:0

The problem is "port-6:0:0" in the hierarchy: since mpt3sas explicitly
calls sas_port_delete() before scsi_remove_host(), that node is removed
from the device hierarchy and seems to orphan what lies beneath it:

[  955.119328] kobject: '6:0:0:0' (ffff88074e2e1008): kobject_uevent_env
[  955.133890] kobject: '6:0:0:0' (ffff88074e2e1008): fill_kobj_path: path = '/end_device-6:0:0/target6:0:0/6:0:0:0/bsg/6:0:0:0'
<WARN/OOPS sadness follows>

As the above from CONFIG_DEBUG_KOBJECT suggests, "end_device-6:0:0" no longer
has a parent. The path to the enclosured device is:

/sys/devices/pci0000:00/0000:00:01.0/0000:01:00.0/host6/port-6:0/expander-6:0/port-6:0:15/end_device-6:0:15/target6:0:15/6:0:15:0/enclosure/6:0:15:0/ArrayDevice09

Enclosure wants to re-add the device outside the enclosure:

https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/drivers/misc/enclosure.c#n404

We melt when trying to re-add "ArrayDevice09" to the "6:0:15:0" directory
above the "enclosure" directory best I can tell. This is *incredibly*
confusing because kobj_add_internal() actually clobbers the parent pointer
if it's NULL:

https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/lib/kobject.c#n216

So even though it appears that the parent was "6:0:15:0" here, it was
actually NULL when passed into kobj_add_internal(). Thus, we try to do
the add with the kset as our parent, and we get -ENOENT because that
node has been marked as inactive in kernfs as per here:

https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/fs/kernfs/dir.c#n743

Why is it inactive? I think, because __kernfs_remove() deleted an ancestor:

https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/fs/kernfs/dir.c#n1243

...which brings us full circle.

> Doesn't this need to be taken care in enclosure_removal?

Even with enclosure compiled out, I still get a WARN like the following
for every block device being removed at rmmod (-dirty here is just from me
applying a fix to SES I sent to the list a few days ago):

[  224.057286] ------------[ cut here ]------------
[  224.067696] WARNING: CPU: 1 PID: 5412 at fs/sysfs/group.c:237 sysfs_remove_group+0x10f/0x150
[  224.088007] sysfs group ffffffff83109ba0 not found for kobject 'sdc'
[  224.102283] Modules linked in: mpt3sas(-) <snip>
[  224.200483] CPU: 0 PID: 5412 Comm: rmmod Not tainted 4.6.0-dirty #4
[  224.219797] Hardware name: Wiwynn   HoneyBadger/PantherPlus, BIOS HBM6.71 02/03/2016
[  224.237001]  ffffffff82af6280 ffff8806e826f8b0 ffffffff81e13e54 ffff8806e826f928
[  224.253811]  0000000000000000 ffff8806e826f8f8 ffffffff81127822 ffffffff831a5968
[  224.270647]  ffff8806000000ed ffffed00dd04df21 0000000000000000 ffff880754ae3078
[  224.287462] Call Trace:
[  224.292953]  [<ffffffff81e13e54>] dump_stack+0x68/0x94
[  224.304433]  [<ffffffff81127822>] __warn+0x172/0x1b0
[  224.315514]  [<ffffffff811278f7>] warn_slowpath_fmt+0x97/0xb0
[  224.351845]  [<ffffffff816bbc1f>] sysfs_remove_group+0x10f/0x150
[  224.365252]  [<ffffffff81377814>] blk_trace_remove_sysfs+0x14/0x20
[  224.379021]  [<ffffffff81d966bf>] blk_unregister_queue+0xcf/0x120
[  224.392589]  [<ffffffff81dc8c7e>] del_gendisk+0x26e/0x600
[  224.432345]  [<ffffffff8215ea75>] sd_remove+0xf5/0x1b0
[  224.443808]  [<ffffffff820dd210>] __device_release_driver+0x160/0x3a0
[  224.458151]  [<ffffffff820dd475>] device_release_driver+0x25/0x40
[  224.471734]  [<ffffffff820dbd81>] bus_remove_device+0x2e1/0x4b0
[  224.484926]  [<ffffffff820d25ad>] device_del+0x37d/0x680
[  224.555423]  [<ffffffff82148225>] __scsi_remove_device+0x1e5/0x250
[  224.569196]  [<ffffffff821443bc>] scsi_forget_host+0x12c/0x1e0
[  224.582197]  [<ffffffff82120b6c>] scsi_remove_host+0x10c/0x300
[  224.595227]  [<ffffffffa0066cc9>] scsih_remove+0x339/0x670 [mpt3sas]
[  224.609383]  [<ffffffff81edd400>] pci_device_remove+0x70/0x110
[  224.622379]  [<ffffffff820dd210>] __device_release_driver+0x160/0x3a0
[  224.636724]  [<ffffffff820de7b3>] driver_detach+0x183/0x200
[  224.649166]  [<ffffffff820dc61f>] bus_remove_driver+0xdf/0x200
[  224.662185]  [<ffffffff820df2e7>] driver_unregister+0x67/0xa0
[  224.675009]  [<ffffffff81edb2ee>] pci_unregister_driver+0x1e/0xe0
[  224.688613]  [<ffffffffa0093a3a>] _mpt3sas_exit+0x23/0x5e9 [mpt3sas]
[  224.702782]  [<ffffffff812a2d1b>] SyS_delete_module+0x2eb/0x390
[  224.743359]  [<ffffffff829529e5>] entry_SYSCALL_64_fastpath+0x18/0xa8
[  224.759036] ---[ end trace 13f1919bf2ec7bb0 ]---

My patch fixes that too :)

The only driver besides mpt*sas that calls sas_delete_port() explicitly is
HPSA, and it does it in the opposite order mpt3sas does: scsi_remove_host()
first.

Thanks,
Calvin

> -----Original Message-----
> From: Calvin Owens [mailto:calvinowens@fb.com]
> Sent: Monday, May 16, 2016 2:25 PM
> To: Elliott, Robert (Persistent Memory)
> Cc: Sathya Prakash; Chaitra P B; Suganath Prabu Subramani; James E.J.
> Bottomley; Martin K. Petersen; MPT-FusionLinux.pdl@broadcom.com;
> linux-scsi@vger.kernel.org; linux-kernel@vger.kernel.org;
> kernel-team@fb.com; calvinowens@fb.com
> Subject: Re: [PATCH] mpt3sas: Do scsi_remove_host() before deleting SAS PHY
> objects
> 
> On Friday 05/13 at 21:17 +0000, Elliott, Robert (Persistent Memory) wrote:
> >
> >
> > > -----Original Message-----
> > > From: linux-kernel-owner@vger.kernel.org [mailto:linux-kernel-
> > > owner@vger.kernel.org] On Behalf Of Calvin Owens
> > > Sent: Friday, May 13, 2016 3:28 PM
> > ...
> > > Subject: [PATCH] mpt3sas: Do scsi_remove_host() before deleting SAS
> > > PHY objects
> > >
> > ...
> > > The issue is that enclosure_remove_device() expects to be able to
> > > re-add the device that was previously enclosured: so with SES
> > > running, the order we unwind things matters in a way it otherwise
> > > wouldn't.
> > >
> > > Since mpt3sas deletes the SAS objects before the SCSI hosts,
> > > enclosure ends up trying to re-parent the SCSI device from the
> > > enclosure to the SAS PHY which has already been deleted. This obviously
> > > ends in sadness.
> > >
> > > The fix appears to be simple: just call scsi_remove_host() before we
> > > call
> > > sas_port_delete() and/or sas_remove_host().
> > >
> > ...
> > > diff --git a/drivers/scsi/mpt3sas/mpt3sas_scsih.c
> > > @@ -8149,6 +8149,8 @@ void scsih_remove(struct pci_dev *pdev)
> > >  		_scsih_raid_device_remove(ioc, raid_device);
> > >  	}
> > >
> > > +	scsi_remove_host(shost);
> > > +
> > >  	/* free ports attached to the sas_host */
> > >  	list_for_each_entry_safe(mpt3sas_port, next_port,
> > >  	   &ioc->sas_hba.sas_port_list, port_list) { @@ -8172,7 +8174,6 @@
> > > void scsih_remove(struct pci_dev *pdev)
> > >  	}
> > >
> > >  	sas_remove_host(shost);
> > > -	scsi_remove_host(shost);
> > >  	mpt3sas_base_detach(ioc);
> > >  	spin_lock(&gioc_lock);
> > >  	list_del(&ioc->list);
> >
> > That change matches the pattern of all other drivers calling
> > sas_remove_host, except for one:
> > 	drivers/message/fusion/mptsas.c
> >
> > That consensus means this comment in drivers/scsi/scsi_transport_sas.c
> > is wrong:
> >
> > /**
> >  * sas_remove_host  -  tear down a Scsi_Host's SAS data structures
> >  * @shost:      Scsi Host that is torn down
> >  *
> >  * Removes all SAS PHYs and remote PHYs for a given Scsi_Host.
> >  * Must be called just before scsi_remove_host for SAS HBAs.
> >  */
> 
> Yes, exactly: that comment appears to be backwards, and as you say most
> drivers appear to do the opposite (I looked at HPSA at least, which calls
> sas_port_delete() before scsi_remove_host()).
> 
> I suppose I should send a patch to fix the comment as well? It'd be nice to
> get some testing to be certain I'm not breaking somebody else's setup by
> fixing mine though...
> 
> Thanks,
> Calvin