In the first place, the loop 'for' in the macro 'for_each_isci_host'
(drivers/scsi/isci/host.h:314) is incorrect, because it accesses
the 3rd element of 2 element array. After the 2nd iteration it executes
the instruction:
ihost = to_pci_info(pdev)->hosts[2]
(while the size of the 'hosts' array equals 2) and reads an
out of range element.
In the second place, this loop is incorrectly optimized by GCC v4.8
(see http://marc.info/?l=linux-kernel&m=138998871911336&w=2).
As a result, on platforms with two SCU controllers,
the loop is executed more times than it can be (for i=0,1 and 2).
It causes kernel panic during entering the S3 state
and the following oops after 'rmmod isci':
BUG: unable to handle kernel NULL pointer dereference at (null)
IP: [<ffffffff8131360b>] __list_add+0x1b/0xc0
Oops: 0000 [#1] SMP
RIP: 0010:[<ffffffff8131360b>] [<ffffffff8131360b>] __list_add+0x1b/0xc0
Call Trace:
[<ffffffff81661b84>] __mutex_lock_slowpath+0x114/0x1b0
[<ffffffff81661c3f>] mutex_lock+0x1f/0x30
[<ffffffffa03e97cb>] sas_disable_events+0x1b/0x50 [libsas]
[<ffffffffa03e9818>] sas_unregister_ha+0x18/0x60 [libsas]
[<ffffffffa040316e>] isci_unregister+0x1e/0x40 [isci]
[<ffffffffa0403efd>] isci_pci_remove+0x5d/0x100 [isci]
[<ffffffff813391cb>] pci_device_remove+0x3b/0xb0
[<ffffffff813fbf7f>] __device_release_driver+0x7f/0xf0
[<ffffffff813fc8f8>] driver_detach+0xa8/0xb0
[<ffffffff813fbb8b>] bus_remove_driver+0x9b/0x120
[<ffffffff813fcf2c>] driver_unregister+0x2c/0x50
[<ffffffff813381f3>] pci_unregister_driver+0x23/0x80
[<ffffffffa04152f8>] isci_exit+0x10/0x1e [isci]
[<ffffffff810d199b>] SyS_delete_module+0x16b/0x2d0
[<ffffffff81012a21>] ? do_notify_resume+0x61/0xa0
[<ffffffff8166ce29>] system_call_fastpath+0x16/0x1b
The loop has been corrected.
This patch fixes kernel panic during entering the S3 state
and the above oops.
Signed-off-by: Lukasz Dorau <lukasz.dorau@intel.com>
Reviewed-by: Maciej Patelczyk <maciej.patelczyk@intel.com>
Tested-by: Lukasz Dorau <lukasz.dorau@intel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
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>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
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>
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>
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>
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>
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>
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>
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>
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>
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>
This field is a holdover from the OS abstraction conversion. The stable
phy to port lookups are done via iphy->ownining_port under scic_lock.
After this conversion to use port->lldd_port the only volatile lookup is
the initial lookup in isci_port_formed(). After that point any lookup
via a successfully notified domain_device is guaranteed to be valid
until the domain_device is destroyed.
Delete ->start_complete as it is only set once and is set as a
consequence of the port going link up, by definition of getting a port
formed event the port is "ready".
While we are correcting port lookups also move the asd_sas_port table
out from under the isci_port. This is to preclude any temptation to use
container_of() to convert an asd_sas_port to an isci_port, the
association is dynamic and under libsas control.
Tested-by: Maciej Trela <maciej.trela@intel.com>
[dmilburn@redhat.com: fix i686 compile error]
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
v1.3 allows the attenuation of the attached cables to be specified to
the driver in terms of 'short', 'medium', and 'long' (see probe_roms.h).
These settings (per phy) are retrieved from the platform oem-parameters
(BIOS rom) or via a module parameter override.
Reviewed-by: Jiangbi Liu <jiangbi.liu@intel.com>
Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
C1 silicon requires updates to the phy tuning recipe and also support
for user provided cable selects (per-phy) for short, medium, and long
cables. Default to 'short' awaiting support for selecting the cable via
oem parameters.
Reviewed-by: Jiangbi Liu <jiangbi.liu@intel.com>
Signed-off-by: Marcin Tomczak <marcin.tomczak@intel.com>
Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
Basic support to initialize the gpio unit, accept an incomming
SAS_GPIO_REG_TX_GP bitstream, and translate it to the ODx.n fields in
the hardware registers. If register indexes outside the supported range
are specified in the SMP frame we simply accept the write and return how
many registers (SFF-8485) were written (libsas reports this as residue
in the request).
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
Hardware allows both an outstanding number commands and a timeout value
(whichever occurs first) as a gate to the next interrupt generation. This
scheme at completion time looks at the remaining number of outstanding tasks
and sets the timeout to maximize small transaction operation. If transactions
are large (take more than a few 10s of microseconds to complete) then
performance is not interrupt processing bound, so the small timeouts this
scheme generates are overridden by the time it takes for a completion to
arrive.
Tested-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
Perform checking per-pci device (even though all systems will only have
1 pci device in this generation), and delete support for silicon that
does not report a proper revision (i.e. A0).
Reported-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Most of these simple dereference macros are longer than their open coded
equivalent. Deleting enum sci_controller_mode is thrown in for good
measure.
Reported-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
The distinction between scic_sds_ scic_ and sci_ are no longer relevant
so just unify the prefixes on sci_. The distinction between isci_ and
sci_ is historically significant, and useful for comparing the old
'core' to the current Linux driver. 'sci_' represents the former core as
well as the routines that are closer to the hardware and protocol than
their 'isci_' brethren. sci == sas controller interface.
Also unwind the 'sds1' out of the parameter structs.
Reported-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Remove the distinction between these two implementations and unify on
isci_host (local instances named ihost). Hmmm, we had two
'oem_parameters' instances, one was unused... nice.
Reported-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Remove the distinction between these two implementations and unify on
isci_remote_device (local instances named idev).
Reported-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Remove the distinction between these two implementations and unify on
isci_port (local instances named iport). The duplicate '->owning_port' and
'->isci_port' in both isci_phy and isci_remote_device will be fixed in a later
patch... this is just the straightforward rename/unification.
Reported-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
They are one in the same object so remove the distinction. The near
duplicate fields (owning_port, and isci_port) will be cleaned up
after the scic_sds_port isci_port unification.
Reported-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
They are one in the same object so remove the distinction. The near
duplicate fields (owning_controller, and isci_host) will be cleaned up
after the scic_sds_contoller isci_host unification.
Reported-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
the dma_pool interface is optimized for object_size << page_size which
is not the case with isci_request objects and the dma_pool routines show
up in the top of the profile.
The old io_request_table which tracked whether tci slots were in-flight
or not is replaced with an IREQ_ACTIVE flag per request.
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
The tci_pool tracks our outstanding command slots which are also the 'index'
portion of our tags. Grabbing the tag early in ->lldd_execute_task let's us
drop the isci_host_can_queue() and ->was_tag_assigned_by_user infrastructure.
->was_tag_assigned_by_user required the task context to be duplicated in
request-local buffer. With the tci established early we can build the
task_context directly into its final location and skip a memcpy.
With the task context buffer at a known address at request construction we
have the opportunity/obligation to also fix sgl handling. This rework feels
like it belongs in another patch but the sgl handling and task_context are too
intertwined.
1/ fix the 'ab' pair embedded in the task context to point to the 'cd' pair in
the task context (previously we were prematurely linking to the staging
buffer).
2/ fix the broken iteration of pio sgls that assumes all sgls are relative to
the request, and does a dangerous looking reverse lookup of physical
address to virtual address.
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Now that we have upleveled device reassignment protection to the
isci_remote_device reference count we no longer need this level of
self-defense.
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Every single i/o or event completion incurs a test and branch to see if
the cycle bit changed. For power-of-2 queue sizes the cycle bit can be
read directly from the rollover of the queue pointer.
Likely premature optimization, but the hidden if() and hidden
assignments / side-effects in the macros were already asking to be
cleaned up.
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
A tag is a 16 bit number where the upper four bits is a sequence number
and the remainder is the task context index (tci). Sanitize the macro
names and shave 256-bytes out of scic_sds_controller by reducing the size of
io_request_sequence.
scic_sds_io_tag_construct --> ISCI_TAG
scic_sds_io_tag_get_sequence --> ISCI_TAG_SEQ
scic_sds_io_tag_get_index() --> ISCI_TAG_TCI
scic_sds_io_sequence_increment() [delete / open code]
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
The circ_buf macros are ~6% faster, as measured by perf, because they take
advantage of power-of-two math assumptions i.e. no test and branch for
rollover. Their semantics are clearer than the hidden side effects in pool.h
(like sci_pool_get() which hides an assignment).
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
The old 'core' had aspirations of running in severely memory constrained
environments like bios option-rom, it's not needed for Linux and gets in
the way of other cleanups (like unifying/reducing the number of structure
members in scic_sds_controller/isci_host).
This also fixes a theoretical bug in that the driver would blindly override
the silicon advertised limits for number of ports, task contexts, and remote
node contexts.
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
C0 silicon updates the pci revision id and requires new AFE parameters
for phy signal integrity. Support for previous silicon revisions is
deprecated (it's also broken for the theoretical case of multiple
controllers at different silicon revisions, all the more reason to get
it removed as soon as possible)
Signed-off-by: Adam Gruchala <adam.gruchala@intel.com>
[fixed up deprecated silicon support]
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Additional state machine cleanups:
o Remove static functions sci_state_machine_exit_state() and
sci_state_machine_enter_state()
o Combines sci_base_state_machine_construct() and
sci_base_state_machine_start() into a single function,
sci_init_sm()
o Remove sci_base_state_machine_stop() which is unused.
o Kill state_machine.[ch]
Signed-off-by: Edmund Nadolski <edmund.nadolski@intel.com>
[fixed too large to inline functions]
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
This cleans up several areas of the state machine mechanism:
o Rename sci_base_state_machine_change_state to sci_change_state
o Remove sci_base_state_machine_get_state function
o Rename 'state_machine' struct member to 'sm' in client structs
o Shorten the name of request states
o Shorten state machine state names as follows:
SCI_BASE_CONTROLLER_STATE_xxx to SCIC_xxx
SCI_BASE_PHY_STATE_xxx to SCI_PHY_xxx
SCIC_SDS_PHY_STARTING_SUBSTATE_xxx to SCI_PHY_SUB_xxx
SCI_BASE_PORT_STATE_xxx to SCI_PORT_xxx and
SCIC_SDS_PORT_READY_SUBSTATE_xxx to SCI_PORT_SUB_xxx
SCI_BASE_REMOTE_DEVICE_STATE_xxx to SCI_DEV_xxx
SCIC_SDS_STP_REMOTE_DEVICE_READY_SUBSTATE_xxx to SCI_STP_DEV_xxx
SCIC_SDS_SMP_REMOTE_DEVICE_READY_SUBSTATE_xxx to SCI_SMP_DEV_xxx
SCIC_SDS_REMOTE_NODE_CONTEXT_xxx_STATE to SCI_RNC_xxx
Signed-off-by: Edmund Nadolski <edmund.nadolski@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
* Move port configuration agent implementation
* Merge core/scic_sds_port.[ch] into port.[ch]
Reported-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
unify core/sci_base_state.h and core/sci_base_state_machine.[ch] into
state_machine.[ch]
Reported-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Now that the data structures are unified unify the implementation in
host.[ch] and cleanup namespace pollution.
Reported-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
cross driver constants are spread out over multiple header files, consolidate
them into isci.h, and push some includes out to the source files that need
them.
TODO: remove SCI_MODE_SIZE infrastructure.
TODO: task.h is full of inlines that are too large
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Make scic_sds_request a proper member of isci_request. Also let's us
get rid of the dma pool object size tracking since we now know that all
requests are sizeof(isci_request). While cleaning up the construct
routine incidentally replaced SCI_FIELD_OFFSET with offsetof.
Reported-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Make scic_sds_port a member of isci_port and merge their lifetimes which
means removing the port table from scic_sds_controller in favor of the
one at the isci_host level. Merge ihost->sas_ports into ihost->ports.
_
Reported-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Make it explicit that isci_host and scic_sds_controller are one in the same
object.
Signed-off-by: Artur Wojcik <artur.wojcik@intel.com>
[removed ->ihost back pointer]
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Now that the core/lldd remote_device data structures are nominally unified
merge the corresponding sources into the top-level directory. Also move the
remote_node_context infrastructure which has no analog at the lldd level.
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
