Corrected use of the request state_lock in the completion callback.

In the case where an abort (or reset) thread is trying to terminate an
I/O request, it sets the request state to "aborting" (or "terminating")
if the state is still "starting".  One of the bugs was to never set the
state to "completed".  Another was to not correctly recognize the
situation where the I/O had completed but the sas_task was still pending
callback to task_done - this was typically a problem in the LUN and
device reset cases.

It is now possible that we leave isci_task_abort_task() with
request->io_request_completion pointing to localy allocated
aborted_io_completion struct. It may result in a system crash.
Signed-off-by: NJeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: NMaciej Trela <Maciej.Trela@intel.com>
Signed-off-by: NJacek Danecki <Jacek.Danecki@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Changes to move management of the reqs_in_process entry for the request here.
Made changes to note when the task is already in the abort path and
cannot be completed through callbacks.
Signed-off-by: NJeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: NJacek Danecki <Jacek.Danecki@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

In the condition where outstanding I/Os are being cleaned from the device
requests in process list, the cleanup function needs to check that the
request is actually a sas-task and not a task management function.
Signed-off-by: NJeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Reported-by: NJames Bottomley <James.Bottomley@suse.de>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

The remote_device_lock is currently used to protect a controller global
resource (RNCs), but the remote_device_lock is per-port.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Until we synchronize against device removal this limits the damage of
use after free bugs to the driver's own objects.  Unless we implement
reference counting we need to ensure at least a subset of a remote
device is valid at all times.  We follow the lead of other libsas
drivers that also preallocate devices.

This also enforces maximum remote device accounting at the lldd layer,
but the core may still run out of RNC's before we hit this limit.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Replace the device completion infrastructure with the controller wide
event queue.  There was a potential for the stop and ready notifications
to corrupt each other, now that cannot happen.

The stop pending flag cannot be used until devices are statically
allocated.  We temporarily need to maintain a completion to handle
waiting for an object that has disappeared, but we can at least stop
scribbling on freed memory.

A future change will also get rid of the "stopping" state as it should
not be exposed to the rest of the driver.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

The midlayer is already throttling i/o in the places where host_quiesce
was trying to prevent further i/o to the device.  It's also problematic
in that it holds a lock over GFP_KERNEL allocations.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

It belies the fact that isci_remote_device and scic_sds_remote_device
are one in same object with the same lifetime rules.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

isci_host_by_id() should have been a clue that an array would have been
a simpler approach.
Reported-by: NJames Bottomley <James.Bottomley@suse.de>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Now that phys_to_virt() and virt_to_phys() have been removed we are no
longer violating the dma mapping (or kmap apis).
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Ross says:
 "The memory allocation for these requests doesn’t take into account the
  additional memory needed when the code in
  scic_sds_s[mst]p_request_assign_buffers() shifts the struct
  scu_task_context so that it is cache line aligned:

  In an example from my machine, total buffer that I’ve given to SCIC goes
  from 0x410024566f84 to 0x410024567308.  From this same example, this
  call shifts my task_context_buffer from 0x410024567208 to
  0x410024567240.

  This means that the task_context_buffer that used to range from
  0x410024567208 to 0x410024567308 instead now goes from 0x410024567240 to
  0x410024567340.

  When the memset() call at the end of scic_task_request_construct()
  clears out this task_context_buffer, it does so from 0x410024567240 to
  0x410024567340, effectively killing whatever buffer follows this
  allocation in memory."

djbw:
Use the kernel's PTR_ALIGN instead of
scic_sds_request_align_task_context_buffer() and SMP_CACHE_BYTES instead of
the local CACHE_LINE_SIZE definition.

TODO: These allocations really want to be better defined in a union rather
than opaque buffers carved up by macros.
Reported-by: NRoss Zwisler <ross.zwisler@intel.com>
Signed-off-by: NJacek Danecki <Jacek.Danecki@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

When aborting a task context we need to be sure that the hardware has acted on
this request (retrieved the task context) before invalidating the remote node
context.  In the case of the "dummy" task context and remote node we do not
have the full state machine that goes through the complete tc abort and rnc
invalidate states.  Instead we ensure the hardware has seen and acted on
Signed-off-by: NJacek Danecki <Jacek.Danecki@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Moving some of the chattiness of warning messages to debug so only the Linux
system messages are shown.
Signed-off-by: NDave Jiang <dave.jiang@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Adding support for PHY_FUNC_LINK_RESET and PHY_FUNC_DISABLE. This allow the
sysfs knob enable (both 0 and 1) and link_reset to work properly.
Signed-off-by: NDave Jiang <dave.jiang@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Core reworks to support stopping and re-starting the controller, lays the
groundwork for phy disable / re-enable and fixes other bugs around port/phy
setup/teardown.
Signed-off-by: NPawel Marek <pawel.marek@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Observed that some devices return a d2h fis, treat like an sdb error fis.
Signed-off-by: NPiotr Sawicki <piotr.sawicki@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

There is a condition whereby TCs (task contexts) can jump to the head of
the round robin queue causing indefinite starvation of pending tasks.
Posting a TC to a suspended RNC (remote node context) causes the
hardware to select that task first, but since the RNC is suspended the
scheduler proceeds to the next task in the expected round robin fashion,
restoring TC arbitration fairness.
Signed-off-by: NTomasz Chudy <tomasz.chudy@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Prepare the timer api for the arrival of dynamic creation and
destruction events from the core.  It pretended to do this previously
but the core to date only used it in a static init-time only fashion.
This is an interim fix until a cleaner event queue can be developed.

1/ make all locking external to the api (add WARN_ONCE to verify)
2/ add a timer_destroy interface (to be used by the core)
3/ use del_timer_sync() prior to deallocating timer data
4/ delete the "timer_list" indirection, we only have timers allocated
   for the isci_host
5/ fix detection of timer list allocation errors
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Undo the open coded and incorrect translation of the oem parameter sas
address to its libsas expected format.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Removed all callbacks in the deprecated.c. Core will call the appropriate
functions directly.
Signed-off-by: NDave Jiang <dave.jiang@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Renaming the callbacks to apparopriate event notify calls for the LLDD.
Signed-off-by: NDave Jiang <dave.jiang@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Remove abstraction for SG building and get rid of callbacks for getting
DMA memory mapping.
Signed-off-by: NDave Jiang <dave.jiang@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

We can copy the data directly to and from sg for SATA PIO read operations.
There is no reason to involve the hardware SGL. In the process we also need
to kmap the sg because we don't know where that can come from.

We also do to not call phys_to_virt(). The driver already has the information.
We can just calculcate the appropriate offets.
Signed-off-by: NDave Jiang <dave.jiang@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

These macros are not necessary. We can do 64bit math directly.
Signed-off-by: NDave Jiang <dave.jiang@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Sending aborts/resets to SAS/SATA targets in APC mode eventually causes
an assert in scic_sds_apc_agent_link_up().  We need to handle the hard reset
case for apc mode ports.
Signed-off-by: NPiotr Sawicki <piotr.sawicki@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Update the SCI Core to comprehend the changes in the TC completion
codes from A0 to B0.  Specifically, there isnew R_ER code
differences for command and data FISes.

Changes are as follows:

1) 0x16 now additionally indicates an R_ERR received for a COMMAND
FIS being sent to a SATA target.  0x16 for SSP still indicates a
NAK received for a COMMAND frame.  Fix is to retry TC to be compliant
with SATA spec or ensure proper error handling of return value
(not spec compliant I don't believe).
2) 0x1B was previously called DONE_BREAK_RCVD for STP and
DONE_LL_ABORT_ERR for SSP.  Now it is universally called
DONE_LL_ABORT_ERR.  This is purely a superficial change.
3) 0x32 is no longer a reserved code.  Now it indicates
DONE_CMD_SDMA_ERR for STP/SSP.  There was a fatal error on the
SDMA for a command IU (includes Raw frames).  Consider retry,
but at a minimum gracefully fail the request.
4) 0x33 is no longer a reserved code.  Now it indicates
DONE_CMD_LL_ABORT_ERR for SSP.  There was a break receivd
during transmission of a command IU.  Consider retry, but
at a minimum gracefully fail the request.
Signed-off-by: NTomasz Chudy <Tomasz.Chudy@intel.com>
Signed-off-by: NJacek Danecki <Jacek.Danecki@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Use the dynamic revision detection code in
scic_sds_phy_link_layer_initialization() and apply some coding style
fixups (long deref chains).  The compile time max link rate setting is
removed in favor of honoring the user-parameter max.
Reported-by: NKrzysztof Wierzbicki <Krzysztof.Wierzbicki@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Add support for the following parameters in SCIC:

     /**
       * This field specifies the NOTIFY (ENABLE SPIN UP) primitive
       * insertion frequency for this phy index.
       */
      u32  notify_enable_spin_up_insertion_frequency;

      /**
       * This method specifies the number of transmitted DWORDs within which
       * to transmit a single ALIGN primitive.  This value applies regardless
       * of what type of device is attached or connection state.  A value of
       * 0 indicates that no ALIGN primitives will be inserted.
       */
      u16  align_insertion_frequency;

      /**
       * This method specifies the number of transmitted DWORDs within which
       * to transmit 2 ALIGN primitives.  This applies for SAS connections
       * only.  A minimum value of 3 is required for this field.
       */
      u16  in_connection_align_insertion_frequency;
Signed-off-by: NKrzysztof Wierzbicki <Krzysztof.Wierzbicki@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

At init and RNC resume we need to touch every phy in a port to be sure
we have initialized STP properties in the case where port_index !=
phy_index.  Also add some missing __iomem annotations.
Signed-off-by: NHenryk Dembkowski <henryk.dembkowski@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

The default should be 5us.  The hardware encodes it in 256ns increments,
so the value should be 20 to approximate a 5us timeout.
Signed-off-by: NTomasz Chudy <Tomasz.Chudy@intel.com>
Signed-off-by: NJacek Danecki <Jacek.Danecki@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

 c99 the struct initializers:
	1/ allows grep to consistently show method name associations.  The
	   naming is mostly consistent (except when it isn't) so this guarantees
	   coverage of present and future exception cases.
	2/ let's the compiler guarantee that the state table array entry
	   correlates with an actual state name and detect accidental reordering or
	   deletion of states.
	/ allows default handler's to be identified easily
Signed-off-by: NJacek Danecki <Jacek.Danecki@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Moved the firmware loading from per adapter to per PCI device. This should
prevent firmware from being loaded twice becuase of 2 SCU controller per
PCI device. We do have to do it per PCI device because request_firmware()
requires a struct device passed in.
Signed-off-by: NDave Jiang <dave.jiang@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

The proc_name field in struct scsi_host_template is exported through sysfs and
allows userspace tools to identify the driver behind a particular SCSI host
controller.

Initialize this field so that userspace tools can easily identify isci host
controllers through sysfs.
Signed-off-by: NHavard Skinnemoen <hskinnemoen@google.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

This removes scic_controller_get_handler_methods and its
associated unused code.
Signed-off-by: NEdmund Nadolski <edmund.nadolski@intel.com>
[djbw: kill off the legacy handler, now that we have basic error isr support]
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Some of the chain walks to get back to our dev are invalid.

isci_remote_device_change_state: delete rather than adding conditional deref
chain walking
isci_request_change_state: fix, it was being called too early
isci_request_ssp_io_request_get_lun: fix compile breakage hidden by ifdef DEBUG
Signed-off-by: NMaciej Trela <maciej.trela@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Inform libsas of the linkrate of direct attached links.
Reported-by: NHaavard Skinnemoen <hskinnemoen@gmail.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Add basic support for handling/reporting error interrupts.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Polling the event queue during scan is an unneeded holdover from the
original driver.
Signed-off-by: NEdmund Nadolski <edmund.nadolski@intel.com>
[djbw: ensure we flush all port events and domain discovery]
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

The lldd actively disallows requests in the "starting" state.  Retrying
or holding off commands in this state is sub-optimal:
1/ it adds another state check to the fast path
2/ retrying can cause libsas to give up

However, isci's ->lldd_dev_found() routine already waits for controller
start to complete before allowing further progress.  Checking the
"starting" state in isci_task_execute_task and the isr is redundant and
misleading.  Clean this up and introduce a controller-wide event queue
to start reeling in "completion" proliferation in the driver.

The "stopping" state cleanups are in a similar vein, rely on the the isr
and other paths being precluded from occurring rather than implementing
state checking logic.
Reported-by: NChristoph Hellwig <hch@infradead.org>
Cc: Jeff Garzik <jeff@garzik.org>
Signed-off-by: NEdmund Nadolski <edmund.nadolski@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>