Patch is predecessor to changes that will be made to storsvc driver in 7.7

Hello

How to test it? just add the  scsi_mod.use_blk_mq to host?

We may not need this patch for RHEL7, which is only for addressing virito-scsi's issue wrt. queue mapping vs. CPU hotplug.

However, there isn't such issue on RHEL7 because of the queue mapping/CPU hotplug handling difference.

We will need it for the storvsc driver update in RHEL7.

(In reply to guazhang from comment #4)
> Hello
> 
> How to test it? just add the  scsi_mod.use_blk_mq to host?

Yes.  With the exception of drivers that individually set
use_blk_mq, verify the following:

Boot with scsi_mod.use_blk_mq=N (the default case in RHEL7)
  find /sys | fgrep use_blk_mq (to find the sysfs nodes for the individual
                                scsi_host object settings for instantiated
                                hosts, they should be "0" unless a driver
                                has enabled it explicitly)

Boot with scsi_mod.use_blk_mq=Y
  find /sys | fgrep use_blk_mq -- sysfs nodes for scsi_host objects should be "1"

(i.e. verify that this change does not force SCSI-MQ "on" or "off" unintentionally
for all drivers.)

The actual logic in question that is being changed is in scsi_host_alloc()

-       shost->use_blk_mq = scsi_use_blk_mq && !shost->hostt->disable_blk_mq;
+       shost->use_blk_mq = (scsi_use_blk_mq || shost->hostt->force_blk_mq)
+                           && !shost->hostt->disable_blk_mq;

Drivers can change shost->use_blk_mq after the alloc but before they add the object,
these are the drivers that set use_blk_mq individual by their own module parameters.
(We did this with some drivers earlier in RHEL7 that wished to use SCSI-MQ but we
did not want to turn on SCSI-MQ for all SCSI drivers that might be in use in the kernel.)

Drivers that set use_blk_mq individually are:
  - Infiniband IB/srp -- default module parameter "use_blk_mq" is "Y"
  - lpfc driver - default module parameter "use_blk_mq" is "0"
  - qla2xxx driver - default module parameter "ql2xmqsupport" is "0"

---

The reason for this change now is that storvsc will not be running in non-MQ mode,
and therefore does not need a module option, it can just set the bit in the host
template.

Once the storvsc driver has been updated in RHEL7 to make use of this
functionality to set .force_blk_mq = 1 in the scsi_host_template, it
should always report "use_blk_mq" in the scsi_host object in sysfs
regardless of the scsi_mod.use_blk_mq setting.

Patch(es) committed on kernel-3.10.0-1017.el7

Hello

RHEL7.7 mq regression test passed. so move to verified.

thanks
guazhang

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

https://access.redhat.com/errata/RHSA-2019:2029