This patch implements the hotplug support for virtio-scsi.
When there is a device attached/detached, the virtio-scsi driver will be
signaled via event virtual queue and it will add/remove the scsi device
in question automatically.
Signed-off-by: NSen Wang <senwang@linux.vnet.ibm.com>
Signed-off-by: NCong Meng <mc@linux.vnet.ibm.com>
Acked-by: NPaolo Bonzini <pbonzini@redhat.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

To improve performance for I/O to different targets, add a separate
scatterlist for each of them.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

We do not need the sglist after calling virtqueue_add_buf.  Hence we
can "pipeline" the locked operations and start preparing the sglist
for the next request while we kick the virtqueue.

Together with the previous two patches, this improves performance as
follows.  For a simple "if=/dev/sda of=/dev/null bs=128M iflag=direct"
(the source being a 10G disk, residing entirely in the host buffer cache),
the additional locking does not cause any penalty with only one dd
process, but 2 simultaneous I/O operations improve their times by 3%:

               number of simultaneous dd
                   1               2
 ----------------------------------------
 current        5.9958s        10.2640s
 patched        5.9531s         9.8663s

(Times are best of 10).
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

Keep a separate lock for each virtqueue.  While not particularly
important now, it prepares the code for when we will add support
for multiple request queues.  It is also more tidy as soon as
we introduce a separate lock for the sglist.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

Separate virtqueue_kick_prepare from virtqueue_notify, so that the
expensive vmexit is done without holding the lock.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

This is exposed in the case the FCP_DATA frames somehow got lost and fc_fcp got
the FCP_RSP, in fc_fcp_recv_resp(), since xfer_len is less than the expected_len
it resets the the timer to wait to 2 more jiffies in case the data frames are
already queued locally. However, for target does not support REC, it would just
send RJT w/ ELS_RJT_UNSUP. The rec response handler thus only clears the rport
flag for not doing REC later, but does not do fcp_io_complete() on the
associated fsp.

The fix is just check status of FCP_RSP being received already, i.e. using the
FC_SRB_RCV_STATUS flag, in fc_fcp_timeout before start sending REC. We should
have waited long enough if there is truely data frames queued locally.
Signed-off-by: NYi Zou <yi.zou@intel.com>
Tested-by: NRoss Brattain <ross.b.brattain@intel.com>
Signed-off-by: NRobert Love <robert.w.love@intel.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

The FC-GS-3 sepc requires to wait for least 3 times R_A_TOV per
sec 4.6.1 "If the Requesting_CT does not receive a Response
CT_IU from the Responding_CT within three times R_A_TOV,
it shall consider this to be a protocol error."

This means added four new states with management server
could add significant delay with multiple retries
on default 12 second timeout(3 * R_A_TOV), so instead
just skip these states on very first timeout on any of
these states to not stuck with states for such longer
period.
Signed-off-by: NVasu Dev <vasu.dev@intel.com>
Tested-by: NMarcus Dennis <marcusx.e.dennis@intel.com>
Signed-off-by: NRobert Love <robert.w.love@intel.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

The lport_recv(), i.e., fc_lport_recv_req() may get called w/o the sequence ptr
being set in fr_seq(), particularly in the case of vn2vn mode, this may happen
if the passive fcp provider, e.g., tcm_fc, has not been registered yet.
Signed-off-by: NYi Zou <yi.zou@intel.com>
Tested-by: NRoss Brattain <ross.b.brattain@intel.com>
Signed-off-by: NRobert Love <robert.w.love@intel.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

Noticed that we can shuffle the code around in fcoe_percpu_receive_thread a bit
and avoid taking the fcoe_rx_list lock twice per iteration.  This should improve
throughput somewhat.  With this change we take the lock, and check for new
frames in a single critical section.  Only if the list is empty do we drop the
lock and re-acquire it after being signaled to wake up.

Change Notes:
v2) did some further cleanup on the patch by replacing the 2nd call of
spin_lock/splice_init with a goto to the top of the outer loop.  This allows me
to change the inner while loop to an if conditional and remove the sencond check
of kthread_should_stop.  Based on suggestion from Vasu Dev.
Signed-off-by: NNeil Horman <nhorman@tuxdriver.com>
Acked-by: NVasu Dev <vasu.dev@intel.com>
Signed-off-by: NRobert Love <robert.w.love@intel.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

strtoul returns an 'unsigned long' so there is no
reason to check if the value is less than zero.

strtoul already checks for the '-' character deep
in its bowels. It will return an error if the user
has provided a negative value and fcoe_str_to_dev_loss
will return that error to its caller.

This patch fixes the following Coverity reported warning:

CID 703581 -  NO_EFFECT Unsigned compared against 0 - This
less-than-zero comparison of an unsigned value is never true. "*val < 0UL".
drivers/scsi/fcoe/fcoe_sysfs.c:105
Signed-off-by: NRobert Love <robert.w.love@intel.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

Add exch timeout info to have debug log with exch timeout
value to match with retries, also add debug info
on exch timer cancel.

Added common fc_exch_timer_cancel() func and grouped this
along with fc_exch_timer_set() function, so that
added debug code is not repeated.
Signed-off-by: NVasu Dev <vasu.dev@intel.com>
Signed-off-by: NRobert Love <robert.w.love@intel.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

There is a race in scsi_bus_resume_common when set device's runtime
state to active after pm_runtime_put_sync(dev->parent).

Parent device may have been suspended so pm_runtime_set_active(dev) will
fail with -EBUSY.
Signed-off-by: NLin Ming <ming.m.lin@intel.com>
Acked-by: NAlan Stern <stern@rowland.harvard.edu>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

Commit d38bd3aef ("Add -Werror compilation flag") is causing build breakage
with random gcc incarnations.  These look like gcc problems, but we shouldn't
break the build because of a bad gcc.  Fix this by adding a make flag

WARNINGS_BECOME_ERRORS=1

which is the same as aic7xxx uses so ordinarily the build doesn't use -Werror
Reported-by: NFengguang Wu <fengguang.wu@intel.com>
Cc: Alex Iannicelli <alex.iannicelli@emulex.com>
Cc: James Smart <james.smart@emulex.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

We don't use "dev" any more after 07ec747a5f ("libsas: remove
ata_port.lock management duties from lldds") and it causes a compile
warning.
Signed-off-by: NDan Carpenter <dan.carpenter@oracle.com>
Cc: Xiangliang Yu <yuxiangl@marvell.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

The timer and the completion are only used for slow path tasks (smp, and
lldd tmfs), yet we incur the allocation space and cpu setup time for
every fast path task.

Cc: Xiangliang Yu <yuxiangl@marvell.com>
Acked-by: NJack Wang <jack_wang@usish.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

On the way to add a new sata_device field, noticed that libsas is
carrying port multiplier infrastructure that is explicitly disabled by
sas_discover_sata().  The aic94xx touches the unused port_no, so leave
that field in case there was some use for it.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

commit 198439e4 [SCSI] libsas: do not set res = 0 in sas_ex_discover_dev()
commit 19252de6 [SCSI] libsas: fix wide port hotplug issues

The above commits seem to have confused the return value of
sas_ex_discover_dev which is non-zero on failure and
sas_ex_join_wide_port which just indicates short circuiting discovery on
already established ports.  The result is random discovery failures
depending on configuration.

Calls to sas_ex_join_wide_port are the source of the trouble as its
return value is errantly assigned to 'res'.  Convert it to bool and stop
returning its result up the stack.

Cc: <stable@vger.kernel.org>
Tested-by: NDan Melnic <dan.melnic@amd.com>
Reported-by: NDan Melnic <dan.melnic@amd.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Reviewed-by: NJack Wang <jack_wang@usish.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

Continue running revalidation until no more broadcast devices are
discovered.  Fixes cases where re-discovery completes too early in a
domain with multiple expanders with pending re-discovery events.
Servicing BCNs can get backed up behind error recovery.

Cc: <stable@vger.kernel.org>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

The discovery function "sas_rediscover_dev" had two bugs: 1) it did
not pay attention to the return status from the SMP task execution;
2) the stack variable used for the returned SAS address was compared
against 0 without being initialized.
Signed-off-by: NJeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

...now that the strategy handlers guarantee eh context and notify
the driver of bus reset.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

sas_eh_bus_reset_handler() amounts to sas_phy_reset() without
notification of the reset to the lldd.  If this is triggered from
eh-cmnd recovery there may be sas_tasks for the lldd to terminate, so
->lldd_I_T_nexus_reset is warranted.

Cc: Xiangliang Yu <yuxiangl@marvell.com>
Cc: Luben Tuikov <ltuikov@yahoo.com>
Cc: Jack Wang <jack_wang@usish.com>
Reviewed-by: NJacek Danecki <jacek.danecki@intel.com>
[jacek: modify pm8001_I_T_nexus_reset to return -ENODEV]
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

When recovering failed eh-cmnds let the lldd attempt an abort via
scsi_abort_eh_cmnd before escalating.
Reviewed-by: NJacek Danecki <jacek.danecki@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

The strategy handlers may be called in places that are problematic for
libsas (i.e. sata resets outside of domain revalidation filtering /
libata link recovery), or problematic for userspace (non-blocking ioctl
to sleeping reset functions).  However, these routines are also called
for eh escalations and recovery of scsi_eh_prep_cmnd(), so permit them
as long as we are running in the host's error handler, otherwise arrange
for them to be triggered in eh_context.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

A quick reading of scsi_error_handler() one could come away with the
impression that it does its wakeup event check while the task state is
TASK_RUNNING.  In fact it sets TASK_INTERRUPTIBLE at the bottom of the
loop, but that is ~50 lines down.

Just set TASK_INTERRUPTIBLE at the top of loop and be done.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

eh is woken up automatically by the presence of failed commands,
scsi_schedule_eh is reserved for cases where there are no failed
commands.  This guarantees that host_eh_sceduled is only incremented
when an explicit eh request is made.
Reviewed-by: NJacek Danecki <jacek.danecki@intel.com>
Signed-off-by: NMaciej Trela <maciej.trela@intel.com>
[fixed spurious delete of sas_ata_task_abort]
Signed-off-by: NArtur Wojcik <artur.wojcik@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

Rapid ata hotplug on a libsas controller results in cases where libsas
is waiting indefinitely on eh to perform an ata probe.

A race exists between scsi_schedule_eh() and scsi_restart_operations()
in the case when scsi_restart_operations() issues i/o to other devices
in the sas domain.  When this happens the host state transitions from
SHOST_RECOVERY (set by scsi_schedule_eh) back to SHOST_RUNNING and
->host_busy is non-zero so we put the eh thread to sleep even though
->host_eh_scheduled is active.

Before putting the error handler to sleep we need to check if the
host_state needs to return to SHOST_RECOVERY for another trip through
eh.  Since i/o that is released by scsi_restart_operations has been
blocked for at least one eh cycle, this implementation allows those
i/o's to run before another eh cycle starts to discourage hung task
timeouts.

Cc: <stable@vger.kernel.org>
Reported-by: NTom Jackson <thomas.p.jackson@intel.com>
Tested-by: NTom Jackson <thomas.p.jackson@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

When managing shost->host_eh_scheduled libata assumes that there is a
1:1 shost-to-ata_port relationship.  libsas creates a 1:N relationship
so it needs to manage host_eh_scheduled cumulatively at the host level.
The sched_eh and end_eh port port ops allow libsas to track when domain
devices enter/leave the "eh-pending" state under ha->lock (previously
named ha->state_lock, but it is no longer just a lock for ha->state
changes).

Since host_eh_scheduled indicates eh without backing commands pinning
the device it can be deallocated at any time.  Move the taking of the
domain_device reference under the port_lock to guarantee that the
ata_port stays around for the duration of eh.
Reviewed-by: NJacek Danecki <jacek.danecki@intel.com>
Acked-by: NJeff Garzik <jgarzik@redhat.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

The following crash results from cases where the end_device has been
removed before scsi_sysfs_add_sdev has had a chance to run.

 BUG: unable to handle kernel NULL pointer dereference at 0000000000000098
 IP: [<ffffffff8115e100>] sysfs_create_dir+0x32/0xb6
 ...
 Call Trace:
  [<ffffffff8125e4a8>] kobject_add_internal+0x120/0x1e3
  [<ffffffff81075149>] ? trace_hardirqs_on+0xd/0xf
  [<ffffffff8125e641>] kobject_add_varg+0x41/0x50
  [<ffffffff8125e70b>] kobject_add+0x64/0x66
  [<ffffffff8131122b>] device_add+0x12d/0x63a
  [<ffffffff814b65ea>] ? _raw_spin_unlock_irqrestore+0x47/0x56
  [<ffffffff8107de15>] ? module_refcount+0x89/0xa0
  [<ffffffff8132f348>] scsi_sysfs_add_sdev+0x4e/0x28a
  [<ffffffff8132dcbb>] do_scan_async+0x9c/0x145

...teach scsi_sysfs_add_devices() to check for deleted devices() before
trying to add them, and teach scsi_remove_target() how to remove targets
that have not been added via device_add().

Cc: <stable@vger.kernel.org>
Reported-by: NDariusz Majchrzak <dariusz.majchrzak@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

This may not fix all endian issues in this driver, but it does get the
driver working on PowerPC for a PMC SRC card. So it should at least fix
all the problems in the core and in the SRC support.

[jejb: fix >> 32 breakage reported by Fengguang Wu]
Signed-off-by: NBen Collins <bcollins@ubuntu.com>
Acked-by: NAchim Leubner <Achim_Leubner@pmc-sierra.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

The loop that waited for syncronous fib commands was causing a CPU stall
when a timeout actually occured.

1) Switch to using a more accurate timeout mechanism.
2) Do not pace the loop with udelay(). Use cpu_relax() to allow for
   scheduling to occur.
Signed-off-by: NBen Collins <bcollins@ubuntu.com>
Acked-by: NAchim Leubner <Achim_Leubner@pmc-sierra.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

When an error occured that would shut down the driver, some in-flight
events were getting caught up, deadlocking a CPU or two.
Signed-off-by: NBen Collins <bcollins@ubuntu.com>
Acked-by: NAchim Leubner <Achim_Leubner@pmc-sierra.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

This also stops using the "legacy crap" in Scsi_Host (shost->base is an
unsigned long).

This affected 32-bit systems that have 64-bit resource sizes, causing the
IO address to be truncated.
Signed-off-by: NBen Collins <bcollins@ubuntu.com>
Acked-by: NAchim Leubner <Achim_Leubner@pmc-sierra.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

Introduce scsi_dh_attached_handler_name() to retrieve the name of the
scsi_dh that is attached to the scsi_device associated with the provided
request queue.  Returns NULL if a scsi_dh is not attached.

Also, fix scsi_dh_{attach,detach} function header comments to document
@q rather than @sdev.
Signed-off-by: NMike Snitzer <snitzer@redhat.com>
Tested-by: NBabu Moger <babu.moger@netapp.com>
Reviewed-by: NChandra Seetharaman <sekharan@us.ibm.com>
Acked-by: NHannes Reinecke <hare@suse.de>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

Fixed the parentheses so the tcp push bit would be sent properly.
Signed-off-by: NKaren Xie <kxie@chelsio.com>
Reviewed-by: NMike Christie <michaelc@cs.wisc.edu>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

Avoid that the code for requeueing SCSI requests triggers a
crash by making sure that that code isn't scheduled anymore
after a device has been removed.

Also, source code inspection of __scsi_remove_device() revealed
a race condition in this function: no new SCSI requests must be
accepted for a SCSI device after device removal started.
Signed-off-by: NBart Van Assche <bvanassche@acm.org>
Reviewed-by: NMike Christie <michaelc@cs.wisc.edu>
Acked-by: NTejun Heo <tj@kernel.org>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

The return value of scsi_queue_insert() is ignored by all its
callers, hence change the return type of this function into
void.
Signed-off-by: NBart Van Assche <bvanassche@acm.org>
Reviewed-by: NMike Christie <michaelc@cs.wisc.edu>
Reviewed-by: NTejun Heo <tj@kernel.org>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

When we call scsi_unprep_request() the command associated with the request
gets destroyed and therefore drops its reference on the device.  If this was
the only reference, the device may get released and we end up with a NULL
pointer deref when we call blk_requeue_request.
Reported-by: NMike Christie <michaelc@cs.wisc.edu>
Signed-off-by: NBart Van Assche <bvanassche@acm.org>
Reviewed-by: NMike Christie <michaelc@cs.wisc.edu>
Reviewed-by: NTejun Heo <tj@kernel.org>
Cc: <stable@kernel.org>
[jejb: enhance commend and add commit log for stable]
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

Use blk_queue_dead() to test whether the queue is dead instead
of !sdev. Since scsi_prep_fn() may be invoked concurrently with
__scsi_remove_device(), keep the queuedata (sdev) pointer in
__scsi_remove_device(). This patch fixes a kernel oops that
can be triggered by USB device removal. See also
http://www.spinics.net/lists/linux-scsi/msg56254.html.

Other changes included in this patch:
- Swap the blk_cleanup_queue() and kfree() calls in
  scsi_host_dev_release() to make that code easier to grasp.
- Remove the queue dead check from scsi_run_queue() since the
  queue state can change anyway at any point in that function
  where the queue lock is not held.
- Remove the queue dead check from the start of scsi_request_fn()
  since it is redundant with the scsi_device_online() check.
Reported-by: NJun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: NBart Van Assche <bvanassche@acm.org>
Reviewed-by: NMike Christie <michaelc@cs.wisc.edu>
Reviewed-by: NTejun Heo <tj@kernel.org>
Cc: <stable@kernel.org>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

If the queue is dead blk_execute_rq_nowait() doesn't invoke the done()
callback function. That will result in blk_execute_rq() being stuck
in wait_for_completion(). Avoid this by initializing rq->end_io to the
done() callback before we check the queue state. Also, make sure the
queue lock is held around the invocation of the done() callback. Found
this through source code review.
Signed-off-by: NMuthukumar Ratty <muthur@gmail.com>
Signed-off-by: NBart Van Assche <bvanassche@acm.org>
Reviewed-by: NTejun Heo <tj@kernel.org>
Acked-by: NJens Axboe <axboe@kernel.dk>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

We took this lock with spin_lock() so we should unlock it with
spin_unlock() instead of spin_unlock_irq().  This was introduced in
f2c8dc40 "[SCSI] megaraid_mbox: remove scsi_assign_lock usage".
Signed-off-by: NDan Carpenter <dan.carpenter@oracle.com>
Acked-by: NAdam Radford <aradford@gmail.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>