]> git.karo-electronics.de Git - karo-tx-linux.git/log
karo-tx-linux.git
12 years agomerge: libsas devel rnc-devel fixes
Dan Williams [Fri, 13 Apr 2012 18:04:17 +0000 (11:04 -0700)]
merge: libsas devel rnc-devel fixes

12 years agoisci: End the RNC resumption wait when the RNC is destroyed.
Jeff Skirvin [Wed, 14 Mar 2012 00:15:11 +0000 (17:15 -0700)]
isci: End the RNC resumption wait when the RNC is destroyed.

While the RNC is suspended for I/O cleanup, the remote device can be
stopped and the RNC setup for destruction.  These changes accomodate that
case in the abort path.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Fixed RNC bug that lost the suspension or resumption during destroy
Jeff Skirvin [Wed, 14 Mar 2012 00:03:00 +0000 (17:03 -0700)]
isci: Fixed RNC bug that lost the suspension or resumption during destroy

This fix corrects the saving of resume parameters when the destruction
of the RNC has already been directed, and makes sure not to overwrite
the RNC destruction callbacks.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Fix RNC AWAIT_SUSPENSION->INVALIDATING transition.
Jeff Skirvin [Tue, 13 Mar 2012 23:36:35 +0000 (16:36 -0700)]
isci: Fix RNC AWAIT_SUSPENSION->INVALIDATING transition.

The RNC state machine would incorrectly transition from
SCI_RNC_AWAIT_SUSPENSION directly to SCI_RNC_INVALIDATING when a destruct
request was made.  This would skip the increment of the suspension count
and the abort of pending TCs (although the invalidating state would at
least cleanup outstanding TCs).

Instead, the RNC will transition to SCI_RNC_SUSPENDED and then start the
destruction process.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Manage the IREQ_NO_AUTO_FREE_TAG under scic_lock.
Jeff Skirvin [Tue, 13 Mar 2012 00:29:51 +0000 (17:29 -0700)]
isci: Manage the IREQ_NO_AUTO_FREE_TAG under scic_lock.

Since there is a possibilty of a timeout waiting for the RNC suspension,
handle the exit case from the task termination under scic_lock, and leave
the tag allocated if the termination timed-out.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Remove obviated host callback list.
Jeff Skirvin [Sun, 4 Mar 2012 12:44:53 +0000 (12:44 +0000)]
isci: Remove obviated host callback list.

Since the callbacks to libsas now occur under scic_lock, there is no
longer any reason to save the completed requests in a separate list
for completion to libsas.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Check IDEV_GONE before performing abort path operations.
Jeff Skirvin [Sat, 10 Mar 2012 05:46:46 +0000 (05:46 +0000)]
isci: Check IDEV_GONE before performing abort path operations.

In the link fail path, set IDEV_GONE for every device on the domain
when the last link in the port fails.

In the abort path functions like isci_reset_device, make sure that
there has not already been a detected domain failure with the device
by checking IDEV_GONE, before performing any kind of hard reset, SMP
phy control, or TMF operation.

The check for IDEV_GONE makes sure that the device in the abort path
really has control of the port with which it is associated.  This
prevents starting hard resets at incorrect times and scheduling
unnecessary LUN resets for SATA devices.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Restore the ATAPI device RNC management code.
Jeff Skirvin [Fri, 9 Mar 2012 06:42:09 +0000 (22:42 -0800)]
isci: Restore the ATAPI device RNC management code.

The ATAPI specific and STP general RNC suspension code had been
incorrectly removed from the remote device code.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Don't wait for an RNC suspend if it's being destroyed.
Jeff Skirvin [Fri, 9 Mar 2012 06:42:08 +0000 (22:42 -0800)]
isci: Don't wait for an RNC suspend if it's being destroyed.

Make sure that the wait for suspend can handle the RNC destruction case.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Change the phy control and link reset interface for HW reasons.
Jeff Skirvin [Fri, 9 Mar 2012 06:42:07 +0000 (22:42 -0800)]
isci: Change the phy control and link reset interface for HW reasons.

There is an apparent HW lockup caused when the PE is disabled while there
is an outstanding TC in progress.  This change puts the link into OOB to
force the TC to end before the PE is disabled.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Added timeouts to RNC suspensions in the abort path.
Jeff Skirvin [Fri, 9 Mar 2012 06:42:06 +0000 (22:42 -0800)]
isci: Added timeouts to RNC suspensions in the abort path.

This change adds timeouts to the RNC suspension wait.  It makes the
suspend and resume timeouts the same.

The previous resume timeout of 5 ms was too short, and timeouts were
seen in resumptions of devices in the abort task/LUN reset path - which
would receive an RNC resumed message within a tenth of a second later.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Add protocol indicator for TMF requests.
Jeff Skirvin [Fri, 9 Mar 2012 06:42:06 +0000 (22:42 -0800)]
isci: Add protocol indicator for TMF requests.

Requests contructed as task management requests need to have the protocol
indicator set so the completion decode can observe any RNC suspension
conditions.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Directly control IREQ_ABORT_PATH_ACTIVE when completing TMFs.
Jeff Skirvin [Fri, 9 Mar 2012 06:42:05 +0000 (22:42 -0800)]
isci: Directly control IREQ_ABORT_PATH_ACTIVE when completing TMFs.

TMF requests, unlike normal I/O requests, need to handle I/O management
conditions in the completion function because TMFs are not handled in the
completion tasklet.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Wait for RNC resumption before leaving the abort path.
Jeff Skirvin [Fri, 9 Mar 2012 06:42:04 +0000 (22:42 -0800)]
isci: Wait for RNC resumption before leaving the abort path.

In the case of TMF execution, or device resets, wait for the RNC to fully
resume before returning to the caller.  This ensures that the remote
device will not fail I/O requests while waiting for the RNC resumption to
complete.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Fix RNC suspend call for SCI_RESUMING state.
Jeff Skirvin [Fri, 9 Mar 2012 06:42:04 +0000 (22:42 -0800)]
isci: Fix RNC suspend call for SCI_RESUMING state.

Instead of immediately transitioning to the SCI_RNC_AWAIT_SUSPENSION
state, handle the SCI_RNC_RESUMING suspend transition from the
SCI_RNC_READY state like the SCI_RNC_INVALIDATING --> SCI_RNC_POSTING
transitions do now, by setting the destination state for the entry
into the READY state.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Manage tag releases differently when aborting tasks.
Jeff Skirvin [Fri, 9 Mar 2012 06:42:03 +0000 (22:42 -0800)]
isci: Manage tag releases differently when aborting tasks.

When an individual request is being terminated, the request's tag
is managed in the terminate function.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Callbacks to libsas occur under scic_lock and are synchronized.
Jeff Skirvin [Fri, 9 Mar 2012 06:42:03 +0000 (22:42 -0800)]
isci: Callbacks to libsas occur under scic_lock and are synchronized.

This patch changes the callback mechanism to libsas to only occur while
the scic_lock is held; the abort path cleanup of I/Os also checks to make
sure IREQ_ABORT_PATH_ACTIVE is clear before proceding.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: When in the abort path, defeat other resume calls until done.
Jeff Skirvin [Fri, 9 Mar 2012 06:42:02 +0000 (22:42 -0800)]
isci: When in the abort path, defeat other resume calls until done.

Completion of I/Os during the one of the abort path interface calls
from libsas can drive remote device state changes and the resumption
of the device RNC.  This is a problem when the abort path is
attempting to cleanup outstanding I/O at the same time - the resumption
can prevent the termination from occuring correctly.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Implement waiting for suspend in the abort path.
Jeff Skirvin [Fri, 9 Mar 2012 06:42:01 +0000 (22:42 -0800)]
isci: Implement waiting for suspend in the abort path.

In order to prevent a device from receiving an I/O request while still
in an RNC suspending or resuming state (and therefore failing that
I/O back to libsas with a reset required status) wait for the RNC state
change before proceding in the abort path.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Make sure all TCs are terminated and cleaned in LUN reset.
Jeff Skirvin [Fri, 9 Mar 2012 06:42:00 +0000 (22:42 -0800)]
isci: Make sure all TCs are terminated and cleaned in LUN reset.

In the libsas error path, SATA disks require extra handling in
libata to recover operation.  However, libsas expects to be able
to immediately recover all outstanding I/O once the error handler
escalation stops.  This patch fixes the condition where the libata
error handler is scheduled for operation but libsas has already
deleted the outstanding sas_tasks.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Manage the LLHANG timer enable/disable per-device.
Jeff Skirvin [Fri, 9 Mar 2012 06:42:00 +0000 (22:42 -0800)]
isci: Manage the LLHANG timer enable/disable per-device.

The LLHANG timer should be enabled once per device.  This patch corrects
both the timer enable and the timer disable for the remote device.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Save the suspension hint for upcoming suspensions.
Jeff Skirvin [Fri, 9 Mar 2012 06:41:59 +0000 (22:41 -0800)]
isci: Save the suspension hint for upcoming suspensions.

In the case of a suspend call while in SCI_RNC_POSTING or INVALIDATING
states, the LLHANG detect needed to be saved so the upcoming suspension
would enable it correctly.  The unused suspend callback parameters were
removed.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Fix the terminated I/O to not call sas_task_abort().
Jeff Skirvin [Fri, 9 Mar 2012 06:41:58 +0000 (22:41 -0800)]
isci: Fix the terminated I/O to not call sas_task_abort().

This addresses a regression from the commit "isci: Redesign
device suspension, abort, cleanup." in which the sas_task end
condition for terminated I/Os was made to call back on
sas_task_abort()".
This commit will be rolled into the original.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Distinguish between remote device suspension cases
Jeff Skirvin [Fri, 9 Mar 2012 06:41:58 +0000 (22:41 -0800)]
isci: Distinguish between remote device suspension cases

For NCQ error conditions among others, there is no need to enable
the link layer hang detect timer.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Remove isci_device reqs_in_process and dev_node from isci_device.
Jeff Skirvin [Fri, 9 Mar 2012 06:41:57 +0000 (22:41 -0800)]
isci: Remove isci_device reqs_in_process and dev_node from isci_device.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Only set IDEV_GONE in the device stop path.
Jeff Skirvin [Fri, 9 Mar 2012 06:41:56 +0000 (22:41 -0800)]
isci: Only set IDEV_GONE in the device stop path.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: All pending TCs are terminated when the RNC is invalidated.
Jeff Skirvin [Fri, 9 Mar 2012 06:41:56 +0000 (22:41 -0800)]
isci: All pending TCs are terminated when the RNC is invalidated.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Device access in the error path does not depend on IDEV_GONE.
Jeff Skirvin [Fri, 9 Mar 2012 06:41:55 +0000 (22:41 -0800)]
isci: Device access in the error path does not depend on IDEV_GONE.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Add suspension cases for RNC INVALIDATING, POSTING states.
Jeff Skirvin [Fri, 9 Mar 2012 06:41:54 +0000 (22:41 -0800)]
isci: Add suspension cases for RNC INVALIDATING, POSTING states.

The RNC can be any of the states in the loop from suspended to
ready when the API "suspend" or "resume" are called.  This change
adds destination states parameters that control the suspension /
resumption action of the RNC statemachine for those transition states.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Redesign device suspension, abort, cleanup.
Jeff Skirvin [Fri, 9 Mar 2012 06:41:54 +0000 (22:41 -0800)]
isci: Redesign device suspension, abort, cleanup.

This commit changes the means by which outstanding I/Os are handled
for cleanup.
The likelihood is that this commit will be broken into smaller pieces,
however that will be a later revision.  Among the changes:

- All completion structures have been removed from the tmf and
abort paths.
- Now using one completed I/O list, with the I/O completed in host bit being
used to select error or normal callback paths.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Escalate to I_T_Nexus_Reset when the device is gone.
Jeff Skirvin [Fri, 9 Mar 2012 06:41:53 +0000 (22:41 -0800)]
isci: Escalate to I_T_Nexus_Reset when the device is gone.

If LUN reset sees that the device is gone, it returns TMF_RESP_FUNC_FAILED
to cause libsas to escalate to an I_T_Nexus_Reset.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Remote device stop also suspends the RNC and terminates I/O.
Jeff Skirvin [Fri, 9 Mar 2012 06:41:52 +0000 (22:41 -0800)]
isci: Remote device stop also suspends the RNC and terminates I/O.

Fixing the remote device state machine to suspend and terminate
all outstanding I/O before the device stopped state is reached.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Remote device must be suspended for NCQ cleanup.
Jeff Skirvin [Fri, 9 Mar 2012 06:41:52 +0000 (22:41 -0800)]
isci: Remote device must be suspended for NCQ cleanup.

When the remote device enters the NCQ error state, the device must
be suspended so that the I/O terminations can take place.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Manage device suspensions during TC terminations.
Jeff Skirvin [Fri, 9 Mar 2012 06:41:51 +0000 (22:41 -0800)]
isci: Manage device suspensions during TC terminations.

TCs must be terminated only while the RNC is suspended.  This commit
adds remote device suspensions and resumptions in the abort, reset and
termination paths.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Terminate outstanding TCs on TX/RX RNC suspensions.
Jeff Skirvin [Fri, 9 Mar 2012 06:41:50 +0000 (22:41 -0800)]
isci: Terminate outstanding TCs on TX/RX RNC suspensions.

TCs must only be terminated when RNCs are suspended.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Handle all suspending TC completions
Jeff Skirvin [Fri, 9 Mar 2012 06:41:50 +0000 (22:41 -0800)]
isci: Handle all suspending TC completions

Add comprehensive decode for all TC completions that generate RNC
suspensions.

Note that this commit also removes unconditional resumptions of ATAPI
devices when in the SCI_STP_DEV_ATAPI_ERROR state, and STP devices
when in the SCI_STP_DEV_IDLE state. This is because the SCI_STP_DEV_IDLE
and SCI_STP_DEV_ATAPI state entry functions manage the RNC resumption.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Fixed bug in resumption from RNC Tx/Rx suspend state.
Jeff Skirvin [Fri, 9 Mar 2012 06:41:49 +0000 (22:41 -0800)]
isci: Fixed bug in resumption from RNC Tx/Rx suspend state.

The resumption from the Tx/Rx suspended state should work the same
as the Tx suspended state.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Manage the link layer hang detect timer for RNC suspensions.
Jeff Skirvin [Fri, 9 Mar 2012 06:41:48 +0000 (22:41 -0800)]
isci: Manage the link layer hang detect timer for RNC suspensions.

For STP devices under certain protocol conditions, an RNC will not
suspend until the current transfer state is broken with a SYNC/ESC
sequence from the SCU.  The SYNC/ESC driven by expiration of the
SCU link layer hang detect timer, which has too small a dynamic
range to support slow SATA devices, so normally it is disabled.

This change enables the timer with the minimum period at the point
when the suspension is requested.

Note that there is potential collateral damage to other open
connections to slow SATA devices on the same port, since there
is no alternative but to enable the LLHANG timer on every phy in
the port for the current suspension request - there is no way to
tell on which phy the RNC in question is currently active.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: enable BCN in sci_port_add_phy()
Maciej Trela [Mon, 12 Mar 2012 23:29:30 +0000 (23:29 +0000)]
isci: enable BCN in sci_port_add_phy()

Ensure we enable receiving BCN's from the
hardware when adding phy to isci_port.
Otherwise if we get BCN before the port is
created we won't see any BCN

Signed-off-by: Maciej Trela <maciej.trela@intel.com>
Reported-by: Richard Boyd <richard.g.boyd@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Changes in COMSAS timings enabling ISCI to detect buggy disc drives.
Andrzej Jakowski [Thu, 8 Mar 2012 19:38:50 +0000 (19:38 +0000)]
isci: Changes in COMSAS timings enabling ISCI to detect buggy disc drives.

This patch extends timings in COMSAS signaling, so ISCI can detect disc
drives having issues to send COMSAS in correct time frame.

Signed-off-by: Andrzej Jakowski <andrzej.jakowski@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: implement suspend/resume support
Artur Wojcik [Fri, 24 Feb 2012 23:42:45 +0000 (23:42 +0000)]
isci: implement suspend/resume support

Provide a "simple-dev-pm-ops" implementation that shuts down the domain
and the device on suspend, and resumes the device and the domain on
resume.  All of the mechanics of restoring domain connectivity are
handled by libsas once isci has notified libsas that all links should be
back up.  libsas is in charge of handling links that did not resume, or
resumed out of order.

Signed-off-by: Artur Wojcik <artur.wojcik@intel.com>
Signed-off-by: Jacek Danecki <jacek.danecki@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: kill isci_host.shost
Dan Williams [Sat, 25 Feb 2012 22:29:49 +0000 (14:29 -0800)]
isci: kill isci_host.shost

We can retrieve the shost from the sas_ha like the rest of libsas and
drop this out of our local data structure.

Acked-by: Jacek Danecki <jacek.danecki@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: fix interrupt disable
Dan Williams [Fri, 2 Mar 2012 01:06:24 +0000 (17:06 -0800)]
isci: fix interrupt disable

There is a (dubious?) lost irq workaround in sci_controller_isr() that
effectively nullifies attempts to disable interrupts.  Until the
workaround can be re-evaluated add some infrastructure to prevent the
interrupt handler from inadvertantly re-enabling interrupts.

The failure mode was interrupts continuing to run after the driver had
been removed and its iomappings torn down.

Reported-by: Jacek Danecki <jacek.danecki@intel.com>
Tested-by: Jacek Danecki <jacek.danecki@intel.com>
[richard: clear remaining interrupts at the end of reset]
Acked-by: Richard Boyd <richard.g.boyd@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: fix controller stop
Dan Williams [Thu, 23 Feb 2012 09:12:10 +0000 (01:12 -0800)]
isci: fix controller stop

1/ notify waiters when controller stop completes (fixes 10 second stall
   unloading the driver)
2/ make sure phy stop is after port and device stop

Cc: Richard Boyd <richard.g.boyd@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: fix 'link-up' events occur after 'start-complete'
Dan Williams [Wed, 29 Feb 2012 09:07:56 +0000 (01:07 -0800)]
isci: fix 'link-up' events occur after 'start-complete'

The call to wait_for_start() is meant to ensure that all links have been
given a chance to come up before letting the kernel proceed with
probing.  However, the implementation is not correctly syncing with the
port configuration agent.  In the MPC case the ports are hard-coded, in
the APC case we need to wait for the port-configuration to form ports
from the started phys.

Towards that end increase the timeout for the APC agent to form ports,
and delay start complete until all phys are out of link-training.

Cc: <stable@vger.kernel.org>
Cc: Richard Boyd <richard.g.boyd@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: refactor initialization for S3/S4
Dan Williams [Wed, 15 Feb 2012 21:58:42 +0000 (13:58 -0800)]
isci: refactor initialization for S3/S4

Based on an original implementation by Ed Nadolski and Artur Wojcik

In preparation for S3/S4 support refactor initialization so that
driver-load and resume-from-suspend can share the common init path of
isci_host_init().  Organize the initialization into objects that are
self-contained to the driver (initialized by isci_host_init) versus
those that have some upward registration (initialized at allocation time
asd_sas_phy, asd_sas_port, dma allocations).  The largest change is
moving the the validation of the oem and module parameters from
isci_host_init() to isci_host_alloc().

The S3/S4 approach being taken is that libsas will be tasked with
remembering the state of the domain and the lldd is free to be
forgetful.  In the case of isci we'll just re-init using a subset of the
normal driver load path.

[clean up some unused / mis-indented function definitions in host.h]

Signed-off-by: Ed Nadolski <edmund.nadolski@intel.com>
Signed-off-by: Artur Wojcik <artur.wojcik@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: kill isci_port.domain_dev_list
Dan Williams [Sat, 18 Feb 2012 00:30:47 +0000 (16:30 -0800)]
isci: kill isci_port.domain_dev_list

Another unused field, and isci_port_init is overkill.

Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: kill ->status, and ->state_lock in isci_host
Dan Williams [Wed, 15 Feb 2012 21:20:31 +0000 (13:20 -0800)]
isci: kill ->status, and ->state_lock in isci_host

They serve no incremental purpose over the existing sas_ha state.

Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Don't filter BROADCAST CHANGE primitives
Tom Jackson [Fri, 24 Feb 2012 09:38:49 +0000 (09:38 +0000)]
isci: Don't filter BROADCAST CHANGE primitives

Per the SAS spec, several types of BROADCAST CHANGE primitives
must cause re-discovery of the originating expander.
Only the standard BROADCAST CHANGE primitive was being
sent to the LIBSAS layer.  The other BC primitives have been
added to the sci_phy_event_handler()

Signed-off-by: Tom Jackson <thomas.p.jackson@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: kill sci_phy_protocol and sci_request_protocol
Dan Williams [Wed, 1 Feb 2012 08:44:14 +0000 (00:44 -0800)]
isci: kill sci_phy_protocol and sci_request_protocol

Holdovers from the initial driver cleanup, replace with enum sas_protocol.

Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: kill ->is_direct_attached
Dan Williams [Wed, 1 Feb 2012 08:23:10 +0000 (00:23 -0800)]
isci: kill ->is_direct_attached

domain_device ->parent conveys the same information.

Occurrences of ->is_direct_attached appear next to incomplete open-coded
versions of dev_is_sata(), clean those up as well.

Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: improve 'invalid state' warnings
Dan Williams [Fri, 10 Feb 2012 09:05:43 +0000 (01:05 -0800)]
isci: improve 'invalid state' warnings

Convert controller state machine warnings to emit the state number (it
missed the number to string conversion, but since these error rarely
happen not much motivation to go further).

Fix up the rnc warnings to use the state name.

Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: Just #include "host.h" once in host.c
Jesper Juhl [Sun, 5 Feb 2012 00:37:56 +0000 (01:37 +0100)]
isci: Just #include "host.h" once in host.c

There's no need to include the header twice.

Signed-off-by: Jesper Juhl <jj@chaosbits.net>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoscsi: cleanup setting task state in scsi_error_handler()
Dan Williams [Fri, 6 Apr 2012 00:29:12 +0000 (17:29 -0700)]
scsi: cleanup setting task state in scsi_error_handler()

Reading scsi_error_handler() one could easily 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: Dan Williams <dan.j.williams@intel.com>
12 years agolibsas: suspend / resume support
Dan Williams [Thu, 22 Mar 2012 04:09:12 +0000 (21:09 -0700)]
libsas: suspend / resume support

libsas power management routines to suspend and recover the sas domain
based on a model where the lldd is allowed and expected to be
"forgetful".

sas_suspend_ha - disable event processing allowing the lldd to take down
                 links without concern for causing hotplug events.
                 Regardless of whether the lldd actually posts link down
                 messages libsas notifies the lldd that all
                 domain_devices are gone.

sas_prep_resume_ha - on the way back up before the lldd starts link
                     training clean out any spurious events that were
                     generated on the way down, and re-enable event
                     processing

sas_resume_ha - after the lldd has started and decided that all phys
have posted link-up events this routine is called to let
libsas start it's own timeout of any phys that did not
resume.  After the timeout an lldd can cancel the
                phy teardown by posting a link-up event.

Storage for ex_change_count (u16) and phy_change_count (u8) are changed
to int so they can be set to -1 to indicate 'invalidated'.

Cc: Alan Stern <stern@rowland.harvard.edu>
Reviewed-by: Jacek Danecki <jacek.danecki@intel.com>
Tested-by: Maciej Patelczyk <maciej.patelczyk@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoscsi, sd: limit the scope of the async probe domain
Dan Williams [Thu, 22 Mar 2012 04:09:11 +0000 (21:09 -0700)]
scsi, sd: limit the scope of the async probe domain

sd injects and synchronizes probe work on the global kernel-wide domain.
This runs into conflict with PM that wants to perform resume actions in
async context:

[  494.237079] INFO: task kworker/u:3:554 blocked for more than 120 seconds.
[  494.294396] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[  494.360809] kworker/u:3     D 0000000000000000     0   554      2 0x00000000
[  494.420739]  ffff88012e4d3af0 0000000000000046 ffff88013200c160 ffff88012e4d3fd8
[  494.484392]  ffff88012e4d3fd8 0000000000012500 ffff8801394ea0b0 ffff88013200c160
[  494.548038]  ffff88012e4d3ae0 00000000000001e3 ffffffff81a249e0 ffff8801321c5398
[  494.611685] Call Trace:
[  494.632649]  [<ffffffff8149dd25>] schedule+0x5a/0x5c
[  494.674687]  [<ffffffff8104b968>] async_synchronize_cookie_domain+0xb6/0x112
[  494.734177]  [<ffffffff810461ff>] ? __init_waitqueue_head+0x50/0x50
[  494.787134]  [<ffffffff8131a224>] ? scsi_remove_target+0x48/0x48
[  494.837900]  [<ffffffff8104b9d9>] async_synchronize_cookie+0x15/0x17
[  494.891567]  [<ffffffff8104ba49>] async_synchronize_full+0x54/0x70  <-- here we wait for async contexts to complete
[  494.943783]  [<ffffffff8104b9f5>] ? async_synchronize_full_domain+0x1a/0x1a
[  495.002547]  [<ffffffffa00114b1>] sd_remove+0x2c/0xa2 [sd_mod]
[  495.051861]  [<ffffffff812fe94f>] __device_release_driver+0x86/0xcf
[  495.104807]  [<ffffffff812fe9bd>] device_release_driver+0x25/0x32  <-- here we take device_lock()

[  853.511341] INFO: task kworker/u:4:549 blocked for more than 120 seconds.
[  853.568693] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[  853.635119] kworker/u:4     D ffff88013097b5d0     0   549      2 0x00000000
[  853.695129]  ffff880132773c40 0000000000000046 ffff880130790000 ffff880132773fd8
[  853.758990]  ffff880132773fd8 0000000000012500 ffff88013288a0b0 ffff880130790000
[  853.822796]  0000000000000246 0000000000000040 ffff88013097b5c8 ffff880130790000
[  853.886633] Call Trace:
[  853.907631]  [<ffffffff8149dd25>] schedule+0x5a/0x5c
[  853.949670]  [<ffffffff8149cc44>] __mutex_lock_common+0x220/0x351
[  854.001225]  [<ffffffff81304bd7>] ? device_resume+0x58/0x1c4
[  854.049082]  [<ffffffff81304bd7>] ? device_resume+0x58/0x1c4
[  854.097011]  [<ffffffff8149ce48>] mutex_lock_nested+0x2f/0x36   <-- here we wait for device_lock()
[  854.145591]  [<ffffffff81304bd7>] device_resume+0x58/0x1c4
[  854.192066]  [<ffffffff81304d61>] async_resume+0x1e/0x45
[  854.237019]  [<ffffffff8104bc93>] async_run_entry_fn+0xc6/0x173  <-- ...while running in async context

Provide a 'scsi_sd_probe_domain' so that async probe actions actions can
be flushed without regard for the state of PM, and allow for the resume
path to handle devices that have transitioned from SDEV_QUIESCE to
SDEV_DEL prior to resume.

Acked-by: Alan Stern <stern@rowland.harvard.edu>
[alan: uplevel scsi_sd_probe_domain, clarify scsi_device_resume]
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agolibsas: drop sata port multiplier infrastructure
Dan Williams [Thu, 22 Mar 2012 04:09:10 +0000 (21:09 -0700)]
libsas: drop sata port multiplier infrastructure

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: Dan Williams <dan.j.williams@intel.com>
12 years agolibata: export ata_port suspend/resume infrastructure for sas
Dan Williams [Thu, 22 Mar 2012 04:09:08 +0000 (21:09 -0700)]
libata: export ata_port suspend/resume infrastructure for sas

Reuse ata_port_{suspend|resume}_common for sas.  This path is chosen
over adding coordination between ata-tranport and sas-transport because
libsas wants to revalidate the domain at resume-time at the host level.
It can not validate links have resumed properly until libata has had a
chance to perform its revalidation, and any sane placing of an ata_port
in the sas-transport model would delay it's resumption until after the
host.

Export the common portion of port suspend/resume (bypass pm_runtime),
and allow sas to perform these operations asynchronously (similar to the
libsas async-ata probe implmentation).  Async operation is determined by
having an external, rather than stack based, location for storing the
result of the operation.

Reviewed-by: Jacek Danecki <jacek.danecki@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agolibsas: continue revalidation
Dan Williams [Thu, 22 Mar 2012 04:09:06 +0000 (21:09 -0700)]
libsas: continue revalidation

Continue running revalidation until no more broadcast devices are
discovered.

Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agolibata: reset once
Dan Williams [Tue, 20 Mar 2012 17:58:35 +0000 (10:58 -0700)]
libata: reset once

Hotplug testing with libsas currently encounters a 55 second wait for
link recovery to give up.  In the case where the user trusts the
response time of their devices permit the recovery attempts to be
limited to one.

Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agomvsas: remove unused variable in mvs_task_exec()
Dan Carpenter [Thu, 8 Mar 2012 07:26:33 +0000 (10:26 +0300)]
mvsas: remove unused variable in mvs_task_exec()

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: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agolibsas: sas_rediscover_dev did not look at the SMP exec status.
Jeff Skirvin [Sat, 10 Mar 2012 05:55:50 +0000 (05:55 +0000)]
libsas: sas_rediscover_dev did not look at the SMP exec status.

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: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
[djbw: todo sanitize smp_execute_task return values (see: residue)]
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agolibsas: trim sas_task of slow path infrastructure
Dan Williams [Thu, 9 Feb 2012 04:52:40 +0000 (20:52 -0800)]
libsas: trim sas_task of slow path infrastructure

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: Christoph Hellwig <hch@lst.de>
Acked-by: Jack Wang <jack_wang@usish.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: use sas eh strategy handlers
Dan Williams [Wed, 1 Feb 2012 09:42:02 +0000 (01:42 -0800)]
isci: use sas eh strategy handlers

...now that the strategy handlers guarantee eh context and notify
the driver of bus reset.

Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agolibsas: use ->lldd_I_T_nexus_reset for ->eh_bus_reset_handler
Dan Williams [Wed, 1 Feb 2012 09:12:23 +0000 (01:12 -0800)]
libsas: use ->lldd_I_T_nexus_reset for ->eh_bus_reset_handler

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: Jacek Danecki <jacek.danecki@intel.com>
[jacek: modify pm8001_I_T_nexus_reset to return -ENODEV]
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agolibsas: add sas_eh_abort_handler
Dan Williams [Tue, 31 Jan 2012 08:07:27 +0000 (00:07 -0800)]
libsas: add sas_eh_abort_handler

When recovering failed eh-cmnds let the lldd attempt an abort via
scsi_abort_eh_cmnd before escalating.

Acked-by: Jacek Danecki <jacek.danecki@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agolibsas: enforce eh strategy handlers only in eh context
Dan Williams [Tue, 31 Jan 2012 07:39:11 +0000 (23:39 -0800)]
libsas: enforce eh strategy handlers only in eh context

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: Dan Williams <dan.j.williams@intel.com>
12 years agoisci: fix oem parameter validation on single controller skus
Dan Williams [Fri, 13 Apr 2012 04:18:38 +0000 (21:18 -0700)]
isci: fix oem parameter validation on single controller skus

OEM parameters [1] are parsed from the platform option-rom / efi
driver.  By default the driver was validating the parameters for the
dual-controller case, but in single-controller case only the first set
of parameters may be valid.

Limit the validation to the number of actual controllers detected
otherwise the driver may fail to parse the valid parameters leading to
driver-load or runtime failures.

[1] the platform specific set of phy address, configuration,and analog
    tuning values

Reported-by: Dave Jiang <dave.jiang@intel.com>
Tested-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoscsi_transport_sas: fix delete vs scan race
Dan Williams [Tue, 20 Mar 2012 17:58:38 +0000 (10:58 -0700)]
scsi_transport_sas: fix delete vs scan race

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 sas_rphy_remove to wait for async scanning to quiesce before
removing the end_device.  It seems this is a more general problem [1],
but this patch only addresses sas transport.

[1]: 23edb6e [SCSI] mpt2sas: Do not set sas_device->starget to NULL from
the slave_destroy callback when all the LUNS have been deleted

Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agolibsas: fix false positive 'device attached' conditions
Dan Williams [Tue, 20 Mar 2012 17:50:27 +0000 (10:50 -0700)]
libsas: fix false positive 'device attached' conditions

Normalize phy->attached_sas_addr to return a zero-address in the case
when device-type == NO_DEVICE or the linkrate is invalid to handle
expanders that put non-zero sas addresses in the discovery response:

 sas: ex 5001b4da000f903f phy02:U:0 attached: 0100000000000000 (no device)
 sas: ex 5001b4da000f903f phy01:U:0 attached: 0100000000000000 (no device)
 sas: ex 5001b4da000f903f phy03:U:0 attached: 0100000000000000 (no device)
 sas: ex 5001b4da000f903f phy00:U:0 attached: 0100000000000000 (no device)

Reported-by: Andrzej Jakowski <andrzej.jakowski@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoscsi: fix eh wakeup (scsi_schedule_eh vs scsi_restart_operations)
Dan Williams [Fri, 6 Apr 2012 23:35:36 +0000 (16:35 -0700)]
scsi: fix eh wakeup (scsi_schedule_eh vs scsi_restart_operations)

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.

Cc: Tejun Heo <tj@kernel.org>
Reported-by: Tom Jackson <thomas.p.jackson@intel.com>
Tested-by: Tom Jackson <thomas.p.jackson@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agolibsas, libata: fix start of life for a sas ata_port
Dan Williams [Thu, 22 Mar 2012 04:09:07 +0000 (21:09 -0700)]
libsas, libata: fix start of life for a sas ata_port

This changes the ordering of initialization and probing events from:
  1/ allocate rphy in PORTE_BYTES_DMAED, DISCE_REVALIDATE_DOMAIN
  2/ allocate ata_port and schedule port probe in DISCE_PROBE
...to:
  1/ allocate ata_port in PORTE_BYTES_DMAED, DISCE_REVALIDATE_DOMAIN
  2/ allocate rphy in PORTE_BYTES_DMAED, DISCE_REVALIDATE_DOMAIN
  3/ schedule port probe in DISCE_PROBE

This ordering prevents PHYE_SIGNAL_LOSS_EVENTS from sneaking in to
destrory ata devices before they have been fully initialized:

  BUG: unable to handle kernel paging request at 0000000000003b10
  IP: [<ffffffffa0053d7e>] sas_ata_end_eh+0x12/0x5e [libsas]
  ...
  [<ffffffffa004d1af>] sas_unregister_common_dev+0x78/0xc9 [libsas]
  [<ffffffffa004d4d4>] sas_unregister_dev+0x4f/0xad [libsas]
  [<ffffffffa004d5b1>] sas_unregister_domain_devices+0x7f/0xbf [libsas]
  [<ffffffffa004c487>] sas_deform_port+0x61/0x1b8 [libsas]
  [<ffffffffa004bed0>] sas_phye_loss_of_signal+0x29/0x2b [libsas]

...and kills the awkward "sata domain_device briefly existing in the
domain without an ata_port" state.

Reported-by: Michal Kosciowski <michal.kosciowski@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agolibata: make ata_print_id atomic
Dan Williams [Thu, 22 Mar 2012 04:09:05 +0000 (21:09 -0700)]
libata: make ata_print_id atomic

This variable is incremented from multiple contexts (module_init via
libata-lldds and the libsas discovery thread).  Make it atomic to head
off any chance of libsas and libata creating duplicate ids.

Acked-by: Jacek Danecki <jacek.danecki@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agolibsas: fix ata_eh clobbering ex_phys via smp_ata_check_ready
Dan Williams [Tue, 20 Mar 2012 20:24:29 +0000 (13:24 -0700)]
libsas: fix ata_eh clobbering ex_phys via smp_ata_check_ready

The check_ready implementation in the expander-attached ata device case
polls on sas_ex_phy_discover().  The effect is that the ex_phy fields
(critically ->attached_sas_addr) can change.  When ata_eh ends and
libsas comes along to revalidate the domain
sas_unregister_devs_sas_addr() can fail to lookup devices to remove, or
fail to re-add an ata device that ata_eh marked as disabled.  So change
the code to skip the sas_address and change count updates when ata_eh is
active.

Cc: Jack Wang <jack_wang@usish.com>
Tested-by: Maciej Patelczyk <maciej.patelczyk@intel.com>
Tested-by: Bartek Nowakowski <bartek.nowakowski@intel.com>
Tested-by: Jacek Danecki <jacek.danecki@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agolibsas: unify domain_device sas_rphy lifetimes
Dan Williams [Tue, 20 Mar 2012 17:53:24 +0000 (10:53 -0700)]
libsas: unify domain_device sas_rphy lifetimes

Since the domain_device can out live the scsi_target we need the rphy to
follow suit otherwise we run into issues like:

  BUG: unable to handle kernel NULL pointer dereference at 0000000000000050
  IP: [<ffffffffa011561b>] sas_ata_printk+0x43/0x6f [libsas]
  PGD 0
  Oops: 0000 [#1] SMP
  CPU 1
  Modules linked in: ses enclosure isci libsas scsi_transport_sas fuse sunrpc cpufreq_ondemand acpi_cpufreq freq_table mperf microcode pcspkr igb joydev iTCO_wdt ioatdma iTCO_vendor_support i2c_i801 i2c_core dca wmi hed ipv6 pata_acpi ata_generic [last unloaded: scsi_wait_scan]

  Pid: 129, comm: kworker/u:3 Not tainted 3.3.0-rc5-isci+ #1 Intel Corporation SandyBridge Platform/To be filled by O.E.M.
  RIP: 0010:[<ffffffffa011561b>] [<ffffffffa011561b>] sas_ata_printk+0x43/0x6f [libsas]
  RSP: 0018:ffff88042232dd70 EFLAGS: 00010282
  RAX: 0000000000000000 RBX: ffff8804283165b8 RCX: ffff88042232dda0
  RDX: ffff88042232dd78 RSI: ffff8804283165b8 RDI: ffffffffa01188d7
  RBP: ffff88042232ddd0 R08: ffff880388454000 R09: ffff8803edfde1f8
  R10: ffff8803edfde1f8 R11: ffff8803edfde1f8 R12: ffff880428316750
  R13: ffff880388454000 R14: ffff8803f88b31d0 R15: ffff8803f8b21d50
  FS: 0000000000000000(0000) GS:ffff88042ee20000(0000) knlGS:0000000000000000
  CS: 0010 DS: 0000 ES: 0000 CR0: 000000008005003b
  CR2: 0000000000000050 CR3: 0000000001a05000 CR4: 00000000000406e0
  DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
  DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
  Process kworker/u:3 (pid: 129, threadinfo ffff88042232c000, task ffff88042230c920)
  Stack:
  0000000000000000 ffff880400000018 ffff88042232dde0 ffff88042232dda0
  ffffffffa01188c4 ffff88042ee93af0 ffff88042232ddb0 ffffffff8100e047
  ffff88042232de10 ffff880420e5a2c8 ffff8803f8b21d50 ffff8803edfde1f8
  Call Trace:
  [<ffffffff8100e047>] ? load_TLS+0xb/0xf
  [<ffffffffa01156ad>] async_sas_ata_eh+0x66/0x95 [libsas]
  [<ffffffff810655e1>] async_run_entry_fn+0x9e/0x131

Reported-by: Tom Jackson <thomas.p.jackson@intel.com>
Tested-by: Tom Jackson <thomas.p.jackson@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agolibsas: fix sas_get_port_device regression
Dan Williams [Mon, 12 Mar 2012 18:38:26 +0000 (11:38 -0700)]
libsas: fix sas_get_port_device regression

Commit 899fcf4 "[SCSI] libsas: set attached device type and target
protocols for local phys" setup 'phy' to be dereferenced after
list_for_each_entry(phy, &port->phy_list, port_phy_el) (i.e. phy ==
&port->phy_list) resulting in reports like:

  BUG: unable to handle kernel NULL pointer dereference at 00000000000002b0
  IP: [<ffffffffa00ce948>] sas_discover_domain+0x29e/0x4fb [libsas]

...fix by deferring sas_phy_set_target() to the end of
sas_get_port_device().

Reported-by: Tom Jackson <thomas.p.jackson@intel.com>
Tested-by: Tom Jackson <thomas.p.jackson@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agolibsas: fix sas_find_bcast_phy() in the presence of 'vacant' phys
Thomas Jackson [Sat, 18 Feb 2012 02:33:10 +0000 (18:33 -0800)]
libsas: fix sas_find_bcast_phy() in the presence of 'vacant' phys

If an expander reports 'PHY VACANT' for a phy index prior to the one
that generated a BCN libsas fails rediscovery.  Since a vacant phy is
defined as a valid phy index that will never have an attached device
just continue the search.

Cc: <stable@vger.kernel.org>
Signed-off-by: Thomas Jackson <thomas.p.jackson@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
12 years agoLinux 3.4-rc2 v3.4-rc2
Linus Torvalds [Sun, 8 Apr 2012 01:30:41 +0000 (18:30 -0700)]
Linux 3.4-rc2

12 years agoMerge tag 'regmap-3.4-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie...
Linus Torvalds [Sat, 7 Apr 2012 16:56:00 +0000 (09:56 -0700)]
Merge tag 'regmap-3.4-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/regmap

Pull two more small regmap fixes from Mark Brown:
 - Now we have users for it that aren't running Android it turns out
   that regcache_sync_region() is much more useful to drivers if it's
   exported for use by modules.  Who knew?
 - Make sure we don't divide by zero when doing debugfs dumps of
   rbtrees, not visible up until now because everything was providing at
   least some cache on startup.

* tag 'regmap-3.4-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/regmap:
  regmap: prevent division by zero in rbtree_show
  regmap: Export regcache_sync_region()

12 years agoMerge branch 'kvm-updates/3.4' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Linus Torvalds [Sat, 7 Apr 2012 16:53:33 +0000 (09:53 -0700)]
Merge branch 'kvm-updates/3.4' of git://git.kernel.org/pub/scm/virt/kvm/kvm

Pull a few KVM fixes from Avi Kivity:
 "A bunch of powerpc KVM fixes, a guest and a host RCU fix (unrelated),
  and a small build fix."

* 'kvm-updates/3.4' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
  KVM: Resolve RCU vs. async page fault problem
  KVM: VMX: vmx_set_cr0 expects kvm->srcu locked
  KVM: PMU: Fix integer constant is too large warning in kvm_pmu_set_msr()
  KVM: PPC: Book3S: PR: Fix preemption
  KVM: PPC: Save/Restore CR over vcpu_run
  KVM: PPC: Book3S HV: Save and restore CR in __kvmppc_vcore_entry
  KVM: PPC: Book3S HV: Fix kvm_alloc_linear in case where no linears exist
  KVM: PPC: Book3S: Compile fix for ppc32 in HIOR access code

12 years agoMerge tag 'sh-for-linus' of git://github.com/pmundt/linux-sh
Linus Torvalds [Sat, 7 Apr 2012 16:52:46 +0000 (09:52 -0700)]
Merge tag 'sh-for-linus' of git://github.com/pmundt/linux-sh

Pull SuperH fixes from Paul Mundt.

* tag 'sh-for-linus' of git://github.com/pmundt/linux-sh:
  sh: fix clock-sh7757 for the latest sh_mobile_sdhi driver
  serial: sh-sci: use serial_port_in/out vs sci_in/out.
  sh: vsyscall: Fix up .eh_frame generation.
  sh: dma: Fix up device attribute mismatch from sysdev fallout.
  sh: dwarf unwinder depends on SHcompact.
  sh: fix up fallout from system.h disintegration.

12 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris...
Linus Torvalds [Sat, 7 Apr 2012 16:51:36 +0000 (09:51 -0700)]
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris/linux-security

Pull security layer fixlet from James Morris.

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris/linux-security:
  sysctl: fix write access to dmesg_restrict/kptr_restrict

12 years agoMerge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux
Linus Torvalds [Sat, 7 Apr 2012 02:56:04 +0000 (19:56 -0700)]
Merge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux

Pull ACPI & Power Management patches from Len Brown:
 "Two fixes for cpuidle merge-window changes, plus a URL fix in
  MAINTAINERS"

* 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux:
  MAINTAINERS: Update git url for ACPI
  cpuidle: Fix panic in CPU off-lining with no idle driver
  ACPI processor: Use safe_halt() rather than halt() in acpi_idle_play_dead()

12 years agoMerge branch '3.4-rc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/nab...
Linus Torvalds [Sat, 7 Apr 2012 02:54:26 +0000 (19:54 -0700)]
Merge branch '3.4-rc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending

Pull target fixes from Nicholas Bellinger:
 "Pull two tcm_fc fabric related fixes for -rc2:

  Note that both have been CC'ed to stable, and patch #1 is the
  important one that addresses a memory corruption bug related to FC
  exchange timeouts + command abort.

  Thanks again to MDR for tracking down this issue!"

* '3.4-rc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending:
  tcm_fc: Do not free tpg structure during wq allocation failure
  tcm_fc: Add abort flag for gracefully handling exchange timeout

12 years agotcm_fc: Do not free tpg structure during wq allocation failure
Mark Rustad [Tue, 3 Apr 2012 17:24:52 +0000 (10:24 -0700)]
tcm_fc: Do not free tpg structure during wq allocation failure

Avoid freeing a registered tpg structure if an alloc_workqueue call
fails.  This fixes a bug where the failure was leaking memory associated
with se_portal_group setup during the original core_tpg_register() call.

Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Acked-by: Kiran Patil <Kiran.patil@intel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
12 years agotcm_fc: Add abort flag for gracefully handling exchange timeout
Mark Rustad [Tue, 3 Apr 2012 17:24:41 +0000 (10:24 -0700)]
tcm_fc: Add abort flag for gracefully handling exchange timeout

Add abort flag and use it to terminate processing when an exchange
is timed out or is reset. The abort flag is used in place of the
transport_generic_free_cmd function call in the reset and timeout
cases, because calling that function in that context would free
memory that was in use. The aborted flag allows the lifetime to
be managed in a more normal way, while truncating the processing.

This change eliminates a source of memory corruption which
manifested in a variety of ugly ways.

(nab: Drop unused struct fc_exch *ep in ft_recv_seq)

Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Acked-by: Kiran Patil <Kiran.patil@intel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
12 years agoMerge branches 'idle-fix' and 'misc' into release
Len Brown [Sat, 7 Apr 2012 01:48:59 +0000 (21:48 -0400)]
Merge branches 'idle-fix' and 'misc' into release

12 years agoMAINTAINERS: Update git url for ACPI
Igor Murzov [Fri, 30 Mar 2012 18:40:12 +0000 (22:40 +0400)]
MAINTAINERS: Update git url for ACPI

Signed-off-by: Igor Murzov <e-mail@date.by>
Signed-off-by: Len Brown <len.brown@intel.com>
12 years agoMerge branch 'stable' of git://git.kernel.org/pub/scm/linux/kernel/git/cmetcalf/linux...
Linus Torvalds [Sat, 7 Apr 2012 00:56:20 +0000 (17:56 -0700)]
Merge branch 'stable' of git://git.kernel.org/pub/scm/linux/kernel/git/cmetcalf/linux-tile

Pull arch/tile bug fixes from Chris Metcalf:
 "This includes Paul Gortmaker's change to fix the <asm/system.h>
  disintegration issues on tile, a fix to unbreak the tilepro ethernet
  driver, and a backlog of bugfix-only changes from internal Tilera
  development over the last few months.

  They have all been to LKML and on linux-next for the last few days.
  The EDAC change to MAINTAINERS is an oddity but discussion on the
  linux-edac list suggested I ask you to pull that change through my
  tree since they don't have a tree to pull edac changes from at the
  moment."

* 'stable' of git://git.kernel.org/pub/scm/linux/kernel/git/cmetcalf/linux-tile: (39 commits)
  drivers/net/ethernet/tile: fix netdev_alloc_skb() bombing
  MAINTAINERS: update EDAC information
  tilepro ethernet driver: fix a few minor issues
  tile-srom.c driver: minor code cleanup
  edac: say "TILEGx" not "TILEPro" for the tilegx edac driver
  arch/tile: avoid accidentally unmasking NMI-type interrupt accidentally
  arch/tile: remove bogus performance optimization
  arch/tile: return SIGBUS for addresses that are unaligned AND invalid
  arch/tile: fix finv_buffer_remote() for tilegx
  arch/tile: use atomic exchange in arch_write_unlock()
  arch/tile: stop mentioning the "kvm" subdirectory
  arch/tile: export the page_home() function.
  arch/tile: fix pointer cast in cacheflush.c
  arch/tile: fix single-stepping over swint1 instructions on tilegx
  arch/tile: implement panic_smp_self_stop()
  arch/tile: add "nop" after "nap" to help GX idle power draw
  arch/tile: use proper memparse() for "maxmem" options
  arch/tile: fix up locking in pgtable.c slightly
  arch/tile: don't leak kernel memory when we unload modules
  arch/tile: fix bug in delay_backoff()
  ...

12 years agoMerge tag 'stable/for-linus-3.4-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 7 Apr 2012 00:54:53 +0000 (17:54 -0700)]
Merge tag 'stable/for-linus-3.4-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen

Pull xen fixes from Konrad Rzeszutek Wilk:
 "Two fixes for regressions:
   * one is a workaround that will be removed in v3.5 with proper fix in
     the tip/x86 tree,
   * the other is to fix drivers to load on PV (a previous patch made
     them only load in PVonHVM mode).

  The rest are just minor fixes in the various drivers and some cleanup
  in the core code."

* tag 'stable/for-linus-3.4-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen:
  xen/pcifront: avoid pci_frontend_enable_msix() falsely returning success
  xen/pciback: fix XEN_PCI_OP_enable_msix result
  xen/smp: Remove unnecessary call to smp_processor_id()
  xen/x86: Workaround 'x86/ioapic: Add register level checks to detect bogus io-apic entries'
  xen: only check xen_platform_pci_unplug if hvm

12 years agoMerge tag 'mmc-fixes-for-3.4-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sat, 7 Apr 2012 00:22:23 +0000 (17:22 -0700)]
Merge tag 'mmc-fixes-for-3.4-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/cjb/mmc

Pull MMC fixes from Chris Ball:
 - Disable use of MSI in sdhci-pci, which caused multiple chipsets to
   stop working in 3.4-rc1.  I'll wait to turn this on again until we
   have a chipset whitelist for it.
 - Fix a libertas SDIO powered-resume regression introduced in 3.3;
   thanks to Neil Brown and Rafael Wysocki for this fix.
 - Fix module reloading on omap_hsmmc.
 - Stop trusting the spec/card's specified maximum data timeout length,
   and use three seconds instead.  Previously we used 300ms.

Also cleanups and fixes for s3c, atmel, sh_mmcif and omap_hsmmc.

* tag 'mmc-fixes-for-3.4-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/cjb/mmc: (28 commits)
  mmc: use really long write timeout to deal with crappy cards
  mmc: sdhci-dove: Fix compile error by including module.h
  mmc: Prevent 1.8V switch for SD hosts that don't support UHS modes.
  Revert "mmc: sdhci-pci: Add MSI support"
  Revert "mmc: sdhci-pci: add quirks for broken MSI on O2Micro controllers"
  mmc: core: fix power class selection
  mmc: omap_hsmmc: fix module re-insertion
  mmc: omap_hsmmc: convert to module_platform_driver
  mmc: omap_hsmmc: make it behave well as a module
  mmc: omap_hsmmc: trivial cleanups
  mmc: omap_hsmmc: context save after enabling runtime pm
  mmc: omap_hsmmc: use runtime put sync in probe error patch
  mmc: sdio: Use empty system suspend/resume callbacks at the bus level
  mmc: bus: print bus speed mode of UHS-I card
  mmc: sdhci-pci: add quirks for broken MSI on O2Micro controllers
  mmc: sh_mmcif: Simplify calculation of mmc->f_min
  mmc: sh_mmcif: mmc->f_max should be half of the bus clock
  mmc: sh_mmcif: double clock speed
  mmc: block: Remove use of mmc_blk_set_blksize
  mmc: atmel-mci: add support for odd clock dividers
  ...

12 years agoMake the "word-at-a-time" helper functions more commonly usable
Linus Torvalds [Fri, 6 Apr 2012 20:54:56 +0000 (13:54 -0700)]
Make the "word-at-a-time" helper functions more commonly usable

I have a new optimized x86 "strncpy_from_user()" that will use these
same helper functions for all the same reasons the name lookup code uses
them.  This is preparation for that.

This moves them into an architecture-specific header file.  It's
architecture-specific for two reasons:

 - some of the functions are likely to want architecture-specific
   implementations.  Even if the current code happens to be "generic" in
   the sense that it should work on any little-endian machine, it's
   likely that the "multiply by a big constant and shift" implementation
   is less than optimal for an architecture that has a guaranteed fast
   bit count instruction, for example.

 - I expect that if architectures like sparc want to start playing
   around with this, we'll need to abstract out a few more details (in
   particular the actual unaligned accesses).  So we're likely to have
   more architecture-specific stuff if non-x86 architectures start using
   this.

   (and if it turns out that non-x86 architectures don't start using
   this, then having it in an architecture-specific header is still the
   right thing to do, of course)

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agocpuidle: Fix panic in CPU off-lining with no idle driver
Toshi Kani [Sun, 1 Apr 2012 03:37:02 +0000 (21:37 -0600)]
cpuidle: Fix panic in CPU off-lining with no idle driver

Fix a NULL pointer dereference panic in cpuidle_play_dead() during
CPU off-lining when no cpuidle driver is registered.  A cpuidle
driver may be registered at boot-time based on CPU type.  This patch
allows an off-lined CPU to enter HLT-based idle in this condition.

Signed-off-by: Toshi Kani <toshi.kani@hp.com>
Cc: Boris Ostrovsky <boris.ostrovsky@amd.com>
Reviewed-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Tested-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Len Brown <len.brown@intel.com>
12 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Fri, 6 Apr 2012 17:37:38 +0000 (10:37 -0700)]
Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net

Pull networking updates from David Miller:

 1) Fix inaccuracies in network driver interface documentation, from Ben
    Hutchings.

 2) Fix handling of negative offsets in BPF JITs, from Jan Seiffert.

 3) Compile warning, locking, and refcounting fixes in netfilter's
    xt_CT, from Pablo Neira Ayuso.

 4) phonet sendmsg needs to validate user length just like any other
    datagram protocol, fix from Sasha Levin.

 5) Ipv6 multicast code uses wrong loop index, from RongQing Li.

 6) Link handling and firmware fixes in bnx2x driver from Yaniv Rosner
    and Yuval Mintz.

 7) mlx4 erroneously allocates 4 pages at a time, regardless of page
    size, fix from Thadeu Lima de Souza Cascardo.

 8) SCTP socket option wasn't extended in a backwards compatible way,
    fix from Thomas Graf.

 9) Add missing address change event emissions to bonding, from Shlomo
    Pongratz.

10) /proc/net/dev regressed because it uses a private offset to track
    where we are in the hash table, but this doesn't track the offset
    pullback that the seq_file code does resulting in some entries being
    missed in large dumps.

    Fix from Eric Dumazet.

11) do_tcp_sendpage() unloads the send queue way too fast, because it
    invokes tcp_push() when it shouldn't.  Let the natural sequence
    generated by the splice paths, and the assosciated MSG_MORE
    settings, guide the tcp_push() calls.

    Otherwise what goes out of TCP is spaghetti and doesn't batch
    effectively into GSO/TSO clusters.

    From Eric Dumazet.

12) Once we put a SKB into either the netlink receiver's queue or a
    socket error queue, it can be consumed and freed up, therefore we
    cannot touch it after queueing it like that.

    Fixes from Eric Dumazet.

13) PPP has this annoying behavior in that for every transmit call it
    immediately stops the TX queue, then calls down into the next layer
    to transmit the PPP frame.

    But if that next layer can take it immediately, it just un-stops the
    TX queue right before returning from the transmit method.

    Besides being useless work, it makes several facilities unusable, in
    particular things like the equalizers.  Well behaved devices should
    only stop the TX queue when they really are full, and in PPP's case
    when it gets backlogged to the downstream device.

    David Woodhouse therefore fixed PPP to not stop the TX queue until
    it's downstream can't take data any more.

14) IFF_UNICAST_FLT got accidently lost in some recent stmmac driver
    changes, re-add.  From Marc Kleine-Budde.

15) Fix link flaps in ixgbe, from Eric W. Multanen.

16) Descriptor writeback fixes in e1000e from Matthew Vick.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (47 commits)
  net: fix a race in sock_queue_err_skb()
  netlink: fix races after skb queueing
  doc, net: Update ndo_start_xmit return type and values
  doc, net: Remove instruction to set net_device::trans_start
  doc, net: Update netdev operation names
  doc, net: Update documentation of synchronisation for TX multiqueue
  doc, net: Remove obsolete reference to dev->poll
  ethtool: Remove exception to the requirement of holding RTNL lock
  MAINTAINERS: update for Marvell Ethernet drivers
  bonding: properly unset current_arp_slave on slave link up
  phonet: Check input from user before allocating
  tcp: tcp_sendpages() should call tcp_push() once
  ipv6: fix array index in ip6_mc_add_src()
  mlx4: allocate just enough pages instead of always 4 pages
  stmmac: re-add IFF_UNICAST_FLT for dwmac1000
  bnx2x: Clear MDC/MDIO warning message
  bnx2x: Fix BCM57711+BCM84823 link issue
  bnx2x: Clear BCM84833 LED after fan failure
  bnx2x: Fix BCM84833 PHY FW version presentation
  bnx2x: Fix link issue for BCM8727 boards.
  ...

12 years agoxen/pcifront: avoid pci_frontend_enable_msix() falsely returning success
Jan Beulich [Mon, 2 Apr 2012 14:22:39 +0000 (15:22 +0100)]
xen/pcifront: avoid pci_frontend_enable_msix() falsely returning success

The original XenoLinux code has always had things this way, and for
compatibility reasons (in particular with a subsequent pciback
adjustment) upstream Linux should behave the same way (allowing for two
distinct error indications to be returned by the backend).

Signed-off-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
12 years agoxen/pciback: fix XEN_PCI_OP_enable_msix result
Jan Beulich [Mon, 2 Apr 2012 14:32:22 +0000 (15:32 +0100)]
xen/pciback: fix XEN_PCI_OP_enable_msix result

Prior to 2.6.19 and as of 2.6.31, pci_enable_msix() can return a
positive value to indicate the number of vectors (less than the amount
requested) that can be set up for a given device. Returning this as an
operation value (secondary result) is fine, but (primary) operation
results are expected to be negative (error) or zero (success) according
to the protocol. With the frontend fixed to match the XenoLinux
behavior, the backend can now validly return zero (success) here,
passing the upper limit on the number of vectors in op->value.

Signed-off-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
12 years agoxen/smp: Remove unnecessary call to smp_processor_id()
Srivatsa S. Bhat [Thu, 22 Mar 2012 12:59:24 +0000 (18:29 +0530)]
xen/smp: Remove unnecessary call to smp_processor_id()

There is an extra and unnecessary call to smp_processor_id()
in cpu_bringup(). Remove it.

Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
12 years agoxen/x86: Workaround 'x86/ioapic: Add register level checks to detect bogus io-apic...
Konrad Rzeszutek Wilk [Tue, 20 Mar 2012 19:04:18 +0000 (15:04 -0400)]
xen/x86: Workaround 'x86/ioapic: Add register level checks to detect bogus io-apic entries'

The above mentioned patch checks the IOAPIC and if it contains
-1, then it unmaps said IOAPIC. But under Xen we get this:

BUG: unable to handle kernel NULL pointer dereference at 0000000000000040
IP: [<ffffffff8134e51f>] xen_irq_init+0x1f/0xb0
PGD 0
Oops: 0002 [#1] SMP
CPU 0
Modules linked in:

Pid: 1, comm: swapper/0 Not tainted 3.2.10-3.fc16.x86_64 #1 Dell Inc. Inspiron
1525                  /0U990C
RIP: e030:[<ffffffff8134e51f>]  [<ffffffff8134e51f>] xen_irq_init+0x1f/0xb0
RSP: e02b: ffff8800d42cbb70  EFLAGS: 00010202
RAX: 0000000000000000 RBX: 00000000ffffffef RCX: 0000000000000001
RDX: 0000000000000040 RSI: 00000000ffffffef RDI: 0000000000000001
RBP: ffff8800d42cbb80 R08: ffff8800d6400000 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000000 R12: 00000000ffffffef
R13: 0000000000000001 R14: 0000000000000001 R15: 0000000000000010
FS:  0000000000000000(0000) GS:ffff8800df5fe000(0000) knlGS:0000000000000000
CS:  e033 DS: 0000 ES: 0000 CR0:000000008005003b
CR2: 0000000000000040 CR3: 0000000001a05000 CR4: 0000000000002660
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
Process swapper/0 (pid: 1, threadinfo ffff8800d42ca000, task ffff8800d42d0000)
Stack:
 00000000ffffffef 0000000000000010 ffff8800d42cbbe0 ffffffff8134f157
 ffffffff8100a9b2 ffffffff8182ffd1 00000000000000a0 00000000829e7384
 0000000000000002 0000000000000010 00000000ffffffff 0000000000000000
Call Trace:
 [<ffffffff8134f157>] xen_bind_pirq_gsi_to_irq+0x87/0x230
 [<ffffffff8100a9b2>] ? check_events+0x12+0x20
 [<ffffffff814bab42>] xen_register_pirq+0x82/0xe0
 [<ffffffff814bac1a>] xen_register_gsi.part.2+0x4a/0xd0
 [<ffffffff814bacc0>] acpi_register_gsi_xen+0x20/0x30
 [<ffffffff8103036f>] acpi_register_gsi+0xf/0x20
 [<ffffffff8131abdb>] acpi_pci_irq_enable+0x12e/0x202
 [<ffffffff814bc849>] pcibios_enable_device+0x39/0x40
 [<ffffffff812dc7ab>] do_pci_enable_device+0x4b/0x70
 [<ffffffff812dc878>] __pci_enable_device_flags+0xa8/0xf0
 [<ffffffff812dc8d3>] pci_enable_device+0x13/0x20

The reason we are dying is b/c the call acpi_get_override_irq() is used,
which returns the polarity and trigger for the IRQs. That function calls
mp_find_ioapics to get the 'struct ioapic' structure - which along with the
mp_irq[x] is used to figure out the default values and the polarity/trigger
overrides. Since the mp_find_ioapics now returns -1 [b/c the IOAPIC is filled
with 0xffffffff], the acpi_get_override_irq() stops trying to lookup in the
mp_irq[x] the proper INT_SRV_OVR and we can't install the SCI interrupt.

The proper fix for this is going in v3.5 and adds an x86_io_apic_ops
struct so that platforms can override it. But for v3.4 lets carry this
work-around. This patch does that by providing a slightly different variant
of the fake IOAPIC entries.

Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
12 years agoxen: only check xen_platform_pci_unplug if hvm
Igor Mammedov [Tue, 27 Mar 2012 17:31:08 +0000 (19:31 +0200)]
xen: only check xen_platform_pci_unplug if hvm

commit b9136d207f08
  xen: initialize platform-pci even if xen_emul_unplug=never

breaks blkfront/netfront by not loading them because of
xen_platform_pci_unplug=0 and it is never set for PV guest.

Signed-off-by: Andrew Jones <drjones@redhat.com>
Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
12 years agonet: fix a race in sock_queue_err_skb()
Eric Dumazet [Fri, 6 Apr 2012 08:49:10 +0000 (10:49 +0200)]
net: fix a race in sock_queue_err_skb()

As soon as an skb is queued into socket error queue, another thread
can consume it, so we are not allowed to reference skb anymore, or risk
use after free.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>