Tag completion writes for direct cache access to reduce the latency of
checking for descriptor completions.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Export driver attributes for diagnostic purposes:
'ring_size': total number of descriptors available to the engine
'ring_active': number of descriptors in-flight
'capabilities': supported operation types for this channel
'version': Intel(R) QuickData specfication revision

This also allows some chattiness to be removed from the driver startup
as this information is now available via sysfs.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Up until this point the driver for Intel(R) QuickData Technology
engines, specification versions 2 and 3, were mostly identical save for
a few quirks.  Version 3.2 hardware adds many new capabilities (like
raid offload support) requiring some infrastructure that is not relevant
for v2.  For better code organization of the new funcionality move v3
and v3.2 support to its own file dma_v3.c, and export some routines from
the base files (dma.c and dma_v2.c) that can be reused directly.

The first new capability included in this code reorganization is support
for v3.2 memset operations.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

ioat3.2 adds raid5 and raid6 offload capabilities.
Signed-off-by: NTom Picard <tom.s.picard@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

In preparation for adding more operation types to the ioat3 path the
driver needs to honor the DMA_PREP_FENCE flag.  For example the async_tx api
will hand xor->memcpy->xor chains to the driver with the 'fence' flag set on
the first xor and the memcpy operation.  This flag in turn sets the 'fence'
flag in the descriptor control field telling the hardware that future
descriptors in the chain depend on the result of the current descriptor, so
wait for all writes to complete before starting the next operation.

Note that ioat1 does not prefetch the descriptor chain, so does not
require/support fenced operations.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Some engines have transfer size and address alignment restrictions.  Add
a per-operation alignment property to struct dma_device that the async
routines and dmatest can use to check alignment capabilities.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

No drivers currently implement these operation types, so they can be
deleted.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Channel switching is problematic for some dmaengine drivers as the
architecture precludes separating the ->prep from ->submit.  In these
cases the driver can select ASYNC_TX_DISABLE_CHANNEL_SWITCH to modify
the async_tx allocator to only return channels that support all of the
required asynchronous operations.

For example MD_RAID456=y selects support for asynchronous xor, xor
validate, pq, pq validate, and memcpy.  When
ASYNC_TX_DISABLE_CHANNEL_SWITCH=y any channel with all these
capabilities is marked DMA_ASYNC_TX allowing async_tx_find_channel() to
quickly locate compatible channels with the guarantee that dependency
chains will remain on one channel.  When
ASYNC_TX_DISABLE_CHANNEL_SWITCH=n async_tx_find_channel() may select
channels that lead to operation chains that need to cross channel
boundaries using the async_tx channel switch capability.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Some engines optimize operation by reading ahead in the descriptor chain
such that descriptor2 may start execution before descriptor1 completes.
If descriptor2 depends on the result from descriptor1 then a fence is
required (on descriptor2) to disable this optimization.  The async_tx
api could implicitly identify dependencies via the 'depend_tx'
parameter, but that would constrain cases where the dependency chain
only specifies a completion order rather than a data dependency.  So,
provide an ASYNC_TX_FENCE to explicitly identify data dependencies.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Conflicts:
	include/linux/dmaengine.h

Handle descriptor allocation failures by polling for a descriptor.  The
driver will force forward progress when polled.  In the best case this
polling interval will be the time it takes for one dma memcpy
transaction to complete.  In the worst case, channel hang, we will need
to wait 100ms for the cleanup watchdog to fire (ioatdma driver).
Signed-off-by: NMaciej Sosnowski <maciej.sosnowski@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Increment the allocation order of the descriptor ring every time we run
out of descriptors up to a maximum of allocation order specified by the
module parameter 'ioat_max_alloc_order'.  After each idle period
decrement the allocation order to a minimum order of
'ioat_ring_alloc_order' (i.e. the default ring size, tunable as a module
parameter).
Signed-off-by: NMaciej Sosnowski <maciej.sosnowski@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

In order to support dynamic resizing of the descriptor ring or polling
for a descriptor in the presence of a hung channel the reset handler
needs to make progress while in a non-preemptible context.  The current
workqueue implementation precludes polling channel reset completion
under spin_lock().

This conversion also allows us to return to opportunistic cleanup in the
ioat2 case as the timer implementation guarantees at least one cleanup
after every descriptor is submitted.  This means the worst case
completion latency becomes the timer frequency (for exceptional
circumstances), but with the benefit of avoiding busy waiting when the
lock is contended.
Signed-off-by: NMaciej Sosnowski <maciej.sosnowski@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Save 4 bytes per software descriptor by transmitting tx_cnt in an unused
portion of the hardware descriptor.
Signed-off-by: NMaciej Sosnowski <maciej.sosnowski@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Mark all single use initialization routines with __devinit.
Signed-off-by: NMaciej Sosnowski <maciej.sosnowski@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

The register write in ioat_dma_cleanup_tasklet is unfortunate in two
ways:
1/ It clears the extra 'enable' bits that we set at alloc_chan_resources time
2/ It gives the impression that it disables interrupts when it is in
   fact re-arming interrupts

[ Impact: fix, persist the value of the chanctrl register when re-arming ]
Signed-off-by: NMaciej Sosnowski <maciej.sosnowski@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Don't trust that the reserved bits are always zero, also sanity check
the returned value.
Signed-off-by: NMaciej Sosnowski <maciej.sosnowski@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

The cleanup path makes an effort to only perform an atomic read of the
64-bit completion address.  However in the 32-bit case it does not
matter if we read the upper-32 and lower-32 non-atomically because the
upper-32 will always be zero.
Signed-off-by: NMaciej Sosnowski <maciej.sosnowski@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Provide some output for debugging the driver.
Signed-off-by: NMaciej Sosnowski <maciej.sosnowski@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

The unified ioat1/ioat2 ioat_dma_unmap() implementation derives the
source and dest addresses from the unmap descriptor.  There is no longer
a need to track this information in struct ioat_desc_sw.
Signed-off-by: NMaciej Sosnowski <maciej.sosnowski@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Replace the current linked list munged into a ring with a native ring
buffer implementation.  The benefit of this approach is reduced overhead
as many parameters can be derived from ring position with simple pointer
comparisons and descriptor allocation/freeing becomes just a
manipulation of head/tail pointers.

It requires a contiguous allocation for the software descriptor
information.

Since this arrangement is significantly different from the ioat1 chain,
move ioat2,3 support into its own file and header.  Common routines are
exported from driver/dma/ioat/dma.[ch].
Signed-off-by: NMaciej Sosnowski <maciej.sosnowski@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Prepare the code for the conversion of the ioat2 linked-list-ring into a
native ring buffer.  After this conversion ioat2 channels will share
less of the ioat1 infrastructure, but there will still be places where
sharing is possible.  struct ioat_chan_common is created to house the
channel attributes that will remain common between ioat1 and ioat2
channels.

For every routine that accesses both common and hardware specific fields
the old unified 'ioat_chan' pointer is split into an 'ioat' and  'chan'
pointer.  Where 'chan' references common fields and 'ioat' the
hardware/version specific.

[ Impact: pure structure member movement/variable renames, no logic changes ]
Signed-off-by: NMaciej Sosnowski <maciej.sosnowski@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

If a callback is to be attached to a descriptor the channel needs to
know at ->prep time so it can set the interrupt enable bit.  This is in
preparation for moving descriptor ioat2 descriptor preparation from
->submit to ->prep.
Signed-off-by: NMaciej Sosnowski <maciej.sosnowski@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

The async_tx api assumes that after a successful ->prep a subsequent
->submit will not fail due to a lack of resources.

This also fixes a bug in the allocation failure case.  Previously the
descriptors allocated prior to the allocation failure would not be
returned to the free list.
Signed-off-by: NMaciej Sosnowski <maciej.sosnowski@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

This cleans up a mess of and'ing and or'ing bit definitions, and allows
simple assignments from the specified dma_ctrl_flags parameter.
Signed-off-by: NMaciej Sosnowski <maciej.sosnowski@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

->dmacount tracks the sequence number of active descriptors.  It is
written to the DMACOUNT register to update the channel's view of pending
descriptors in the chain.  The register is 16-bits so ->dmacount should
be unsigned and 16-bit as well.  Also modify ->desccount to maintain
alignment.

This was never a problem in practice because we never compared dmacount
values, but this is a bug waiting to happen.
Signed-off-by: NMaciej Sosnowski <maciej.sosnowski@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Towards the removal of ioatdma_device.version split the initialization
path into distinct versions.  This conversion:
1/ moves version specific probe code to version specific routines
2/ removes the need for ioat_device
3/ turns off the ioat1 msi quirk if the device is reinitialized for intx
Signed-off-by: NMaciej Sosnowski <maciej.sosnowski@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

The only .c files that utilize these protected prototypes depend on
CONFIG_INTEL_IOATDMA=y, so there is no value gained in providing empty
prototypes.

[ Impact: pure cleanup ]
Signed-off-by: NMaciej Sosnowski <maciej.sosnowski@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

* reduce device->common. to dma-> in ioat_dma_{probe,remove,selftest}
* ioat_lookup_chan_by_index to ioat_chan_by_index
* multi-line function definitions
* ioat_desc_sw.async_tx to ioat_desc_sw.txd
* desc->txd. to tx-> in cleanup routine
Signed-off-by: NMaciej Sosnowski <maciej.sosnowski@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

The driver currently duplicates much of what these routines offer, so
just use the common code.  For example ->irq_mode tracks what interrupt
mode was initialized, which duplicates the ->msix_enabled and
->msi_enabled handling in pcim_release.

This also adds a check to the return value of dma_async_device_register,
which can fail.
Signed-off-by: NMaciej Sosnowski <maciej.sosnowski@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Some of these defines may be useful outside of dma.c and the header is
private so there are no namespace pollution concerns.
Signed-off-by: NMaciej Sosnowski <maciej.sosnowski@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Now that the resources to handle stripe_head operations are allocated
percpu it is possible for raid5d to distribute stripe handling over
multiple cores.  This conversion also adds a call to cond_resched() in
the non-multicore case to prevent one core from getting monopolized for
raid operations.

Cc: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

These routines have been replaced by there asynchronous counterparts.
Signed-off-by: NYuri Tikhonov <yur@emcraft.com>
Signed-off-by: NIlya Yanok <yanok@emcraft.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

1/ Use STRIPE_OP_BIOFILL to offload completion of read requests to
   raid_run_ops
2/ Implement a handler for sh->reconstruct_state similar to the raid5 case
   (adds handling of Q parity)
3/ Prevent handle_parity_checks6 from running concurrently with 'compute'
   operations
4/ Hook up raid_run_ops
Signed-off-by: NYuri Tikhonov <yur@emcraft.com>
Signed-off-by: NIlya Yanok <yanok@emcraft.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

[ Based on an original patch by Yuri Tikhonov ]

Implement the state machine for handling the RAID-6 parities check and
repair functionality.  Note that the raid6 case does not need to check
for new failures, like raid5, as it will always writeback the correct
disks.  The raid5 case can be updated to check zero_sum_result to avoid
getting confused by new failures rather than retrying the entire check
operation.
Signed-off-by: NYuri Tikhonov <yur@emcraft.com>
Signed-off-by: NIlya Yanok <yanok@emcraft.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

In the synchronous implementation of stripe dirtying we processed a
degraded stripe with one call to handle_stripe_dirtying6().  I.e.
compute the missing blocks from the other drives, then copy in the new
data and reconstruct the parities.

In the asynchronous case we do not perform stripe operations directly.
Instead, operations are scheduled with flags to be later serviced by
raid_run_ops.  So, for the degraded case the final reconstruction step
can only be carried out after all blocks have been brought up to date by
being read, or computed.  Like the raid5 case schedule_reconstruction()
sets STRIPE_OP_RECONSTRUCT to request a parity generation pass and
through operation chaining can handle compute and reconstruct in a
single raid_run_ops pass.

[dan.j.williams@intel.com: fixup handle_stripe_dirtying6 gating]
Signed-off-by: NYuri Tikhonov <yur@emcraft.com>
Signed-off-by: NIlya Yanok <yanok@emcraft.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Modify handle_stripe_fill6 to work asynchronously by introducing
fetch_block6 as the raid6 analog of fetch_block5 (schedule compute
operations for missing/out-of-sync disks).

[dan.j.williams@intel.com: compute D+Q in one pass]
Signed-off-by: NYuri Tikhonov <yur@emcraft.com>
Signed-off-by: NIlya Yanok <yanok@emcraft.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Extend schedule_reconstruction5 for reuse by the raid6 path.  Add
support for generating Q and BUG() if a request is made to perform
'prexor'.
Signed-off-by: NYuri Tikhonov <yur@emcraft.com>
Signed-off-by: NIlya Yanok <yanok@emcraft.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

[ Based on an original patch by Yuri Tikhonov ]

The raid_run_ops routine uses the asynchronous offload api and
the stripe_operations member of a stripe_head to carry out xor+pq+copy
operations asynchronously, outside the lock.

The operations performed by RAID-6 are the same as in the RAID-5 case
except for no support of STRIPE_OP_PREXOR operations. All the others
are supported:
STRIPE_OP_BIOFILL
 - copy data into request buffers to satisfy a read request
STRIPE_OP_COMPUTE_BLK
 - generate missing blocks (1 or 2) in the cache from the other blocks
STRIPE_OP_BIODRAIN
 - copy data out of request buffers to satisfy a write request
STRIPE_OP_RECONSTRUCT
 - recalculate parity for new data that has entered the cache
STRIPE_OP_CHECK
 - verify that the parity is correct

The flow is the same as in the RAID-5 case, and reuses some routines, namely:
1/ ops_complete_postxor (renamed to ops_complete_reconstruct)
2/ ops_complete_compute (updated to set up to 2 targets uptodate)
3/ ops_run_check (renamed to ops_run_check_p for xor parity checks)

[neilb@suse.de: fixes to get it to pass mdadm regression suite]
Reviewed-by: NAndre Noll <maan@systemlinux.org>
Signed-off-by: NYuri Tikhonov <yur@emcraft.com>
Signed-off-by: NIlya Yanok <yanok@emcraft.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

ops_complete_compute5 can be reused in the raid6 path if it is updated to
generically handle a second target.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>