[dm-devel] [PATCH 2/7] dm: Simplify dm_table_determine_type()

[Bart Van Assche]

On 11/16/2016 06:54 AM, Mike Snitzer wrote:
> On Tue, Nov 15 2016 at  6:33pm -0500,
> Bart Van Assche <bart.vanassche at sandisk.com> wrote:
>> +/*
>> + * Returns true if all paths are blk-mq devices. Returns false if all paths
>> + * are single queue block devices or if there are no paths.
>> + */
>
> This code isn't specific to multipath.  So "paths" is misplaced.
> "devices" is more appropriate.  Not seeing the need for the comment to
> be honest.  The function name is pretty descriptive.
>
>>  bool dm_table_all_blk_mq_devices(struct dm_table *t)
>>  {
>>  	return t->all_blk_mq;

Hello Mike,

After having added WARN_ON_ONCE(sq_count == 0 && mq_count == 0) close to 
the end of dm_table_determine_type() the following output appeared:

WARNING: CPU: 0 PID: 2458 at drivers/md/dm-table.c:966 
dm_table_complete+0x60e/0x6b0 [dm_mod]
[ ... ]
Call Trace:
  [<ffffffff81329b45>] dump_stack+0x68/0x93
  [<ffffffff810650e6>] __warn+0xc6/0xe0
  [<ffffffff810651b8>] warn_slowpath_null+0x18/0x20
  [<ffffffffa0021c8e>] dm_table_complete+0x60e/0x6b0 [dm_mod]
  [<ffffffffa00245cd>] table_load+0x13d/0x330 [dm_mod]
  [<ffffffffa0024490>] ? retrieve_status+0x1b0/0x1b0 [dm_mod]
  [<ffffffffa0025145>] ctl_ioctl+0x1f5/0x520 [dm_mod]
  [<ffffffffa002547e>] dm_ctl_ioctl+0xe/0x20 [dm_mod]
  [<ffffffff811f26cf>] do_vfs_ioctl+0x8f/0x6b0
  [<ffffffff811ff41c>] ? __fget+0x10c/0x200
  [<ffffffff811ff310>] ? expand_files+0x2a0/0x2a0
  [<ffffffff811f2d2c>] SyS_ioctl+0x3c/0x70
  [<ffffffff8163f9aa>] entry_SYSCALL_64_fastpath+0x18/0xad

Is it intentional that dm_table_determine_type() can get called for a dm 
device with an empty table?

Thanks,

Bart.