]> err.no Git - linux-2.6/commitdiff
dmaengine: driver for the iop32x, iop33x, and iop13xx raid engines
authorDan Williams <dan.j.williams@intel.com>
Tue, 2 Jan 2007 20:52:26 +0000 (13:52 -0700)
committerDan Williams <dan.j.williams@intel.com>
Fri, 13 Jul 2007 15:06:18 +0000 (08:06 -0700)
The Intel(R) IOP series of i/o processors integrate an Xscale core with
raid acceleration engines.  The capabilities per platform are:

iop219:
 (2) copy engines
iop321:
 (2) copy engines
 (1) xor and block fill engine
iop33x:
 (2) copy and crc32c engines
 (1) xor, xor zero sum, pq, pq zero sum, and block fill engine
iop34x (iop13xx):
 (2) copy, crc32c, xor, xor zero sum, and block fill engines
 (1) copy, crc32c, xor, xor zero sum, pq, pq zero sum, and block fill engine

The driver supports the features of the async_tx api:
* asynchronous notification of operation completion
* implicit (interupt triggered) handling of inter-channel transaction
  dependencies

The driver adapts to the platform it is running by two methods.
1/ #include <asm/arch/adma.h> which defines the hardware specific
   iop_chan_* and iop_desc_* routines as a series of static inline
   functions
2/ The private platform data attached to the platform_device defines the
   capabilities of the channels

20070626: Callbacks are run in a tasklet.  Given the recent discussion on
LKML about killing tasklets in favor of workqueues I did a quick conversion
of the driver.  Raid5 resync performance dropped from 50MB/s to 30MB/s, so
the tasklet implementation remains until a generic softirq interface is
available.

Changelog:
* fixed a slot allocation bug in do_iop13xx_adma_xor that caused too few
slots to be requested eventually leading to data corruption
* enabled the slot allocation routine to attempt to free slots before
returning -ENOMEM
* switched the cleanup routine to solely use the software chain and the
status register to determine if a descriptor is complete.  This is
necessary to support other IOP engines that do not have status writeback
capability
* make the driver iop generic
* modified the allocation routines to understand allocating a group of
slots for a single operation
* added a null xor initialization operation for the xor only channel on
iop3xx
* support xor operations on buffers larger than the hardware maximum
* split the do_* routines into separate prep, src/dest set, submit stages
* added async_tx support (dependent operations initiation at cleanup time)
* simplified group handling
* added interrupt support (callbacks via tasklets)
* brought the pending depth inline with ioat (i.e. 4 descriptors)
* drop dma mapping methods, suggested by Chris Leech
* don't use inline in C files, Adrian Bunk
* remove static tasklet declarations
* make iop_adma_alloc_slots easier to read and remove chances for a
  corrupted descriptor chain
* fix locking bug in iop_adma_alloc_chan_resources, Benjamin Herrenschmidt
* convert capabilities over to dma_cap_mask_t
* fixup sparse warnings
* add descriptor flush before iop_chan_enable
* checkpatch.pl fixes
* gpl v2 only correction
* move set_src, set_dest, submit to async_tx methods
* move group_list and phys to async_tx

Cc: Russell King <rmk@arm.linux.org.uk>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
drivers/dma/Kconfig
drivers/dma/Makefile
drivers/dma/iop-adma.c [new file with mode: 0644]
include/asm-arm/hardware/iop_adma.h [new file with mode: 0644]

index 492aa080562dca216a4d13cab622bd0345d2f7fc..b31756d599787b6872077f1829c506574a8f6a24 100644 (file)
@@ -31,4 +31,13 @@ config INTEL_IOATDMA
        default m
        ---help---
          Enable support for the Intel(R) I/OAT DMA engine.
+
+config INTEL_IOP_ADMA
+        tristate "Intel IOP ADMA support"
+        depends on DMA_ENGINE && (ARCH_IOP32X || ARCH_IOP33X || ARCH_IOP13XX)
+       select ASYNC_CORE
+        default m
+        ---help---
+          Enable support for the Intel(R) IOP Series RAID engines.
+
 endmenu
index bdcfdbdb1aeceb7a753cf98f12aa7a5f6d4ebeef..b3839b687ae033841f04baa7193ec3c24dd2b05b 100644 (file)
@@ -1,3 +1,4 @@
 obj-$(CONFIG_DMA_ENGINE) += dmaengine.o
 obj-$(CONFIG_NET_DMA) += iovlock.o
 obj-$(CONFIG_INTEL_IOATDMA) += ioatdma.o
+obj-$(CONFIG_INTEL_IOP_ADMA) += iop-adma.o
diff --git a/drivers/dma/iop-adma.c b/drivers/dma/iop-adma.c
new file mode 100644 (file)
index 0000000..5a1d426
--- /dev/null
@@ -0,0 +1,1467 @@
+/*
+ * offload engine driver for the Intel Xscale series of i/o processors
+ * Copyright © 2006, Intel Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ */
+
+/*
+ * This driver supports the asynchrounous DMA copy and RAID engines available
+ * on the Intel Xscale(R) family of I/O Processors (IOP 32x, 33x, 134x)
+ */
+
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/async_tx.h>
+#include <linux/delay.h>
+#include <linux/dma-mapping.h>
+#include <linux/spinlock.h>
+#include <linux/interrupt.h>
+#include <linux/platform_device.h>
+#include <linux/memory.h>
+#include <linux/ioport.h>
+
+#include <asm/arch/adma.h>
+
+#define to_iop_adma_chan(chan) container_of(chan, struct iop_adma_chan, common)
+#define to_iop_adma_device(dev) \
+       container_of(dev, struct iop_adma_device, common)
+#define tx_to_iop_adma_slot(tx) \
+       container_of(tx, struct iop_adma_desc_slot, async_tx)
+
+/**
+ * iop_adma_free_slots - flags descriptor slots for reuse
+ * @slot: Slot to free
+ * Caller must hold &iop_chan->lock while calling this function
+ */
+static void iop_adma_free_slots(struct iop_adma_desc_slot *slot)
+{
+       int stride = slot->slots_per_op;
+
+       while (stride--) {
+               slot->slots_per_op = 0;
+               slot = list_entry(slot->slot_node.next,
+                               struct iop_adma_desc_slot,
+                               slot_node);
+       }
+}
+
+static dma_cookie_t
+iop_adma_run_tx_complete_actions(struct iop_adma_desc_slot *desc,
+       struct iop_adma_chan *iop_chan, dma_cookie_t cookie)
+{
+       BUG_ON(desc->async_tx.cookie < 0);
+       spin_lock_bh(&desc->async_tx.lock);
+       if (desc->async_tx.cookie > 0) {
+               cookie = desc->async_tx.cookie;
+               desc->async_tx.cookie = 0;
+
+               /* call the callback (must not sleep or submit new
+                * operations to this channel)
+                */
+               if (desc->async_tx.callback)
+                       desc->async_tx.callback(
+                               desc->async_tx.callback_param);
+
+               /* unmap dma addresses
+                * (unmap_single vs unmap_page?)
+                */
+               if (desc->group_head && desc->unmap_len) {
+                       struct iop_adma_desc_slot *unmap = desc->group_head;
+                       struct device *dev =
+                               &iop_chan->device->pdev->dev;
+                       u32 len = unmap->unmap_len;
+                       u32 src_cnt = unmap->unmap_src_cnt;
+                       dma_addr_t addr = iop_desc_get_dest_addr(unmap,
+                               iop_chan);
+
+                       dma_unmap_page(dev, addr, len, DMA_FROM_DEVICE);
+                       while (src_cnt--) {
+                               addr = iop_desc_get_src_addr(unmap,
+                                                       iop_chan,
+                                                       src_cnt);
+                               dma_unmap_page(dev, addr, len,
+                                       DMA_TO_DEVICE);
+                       }
+                       desc->group_head = NULL;
+               }
+       }
+
+       /* run dependent operations */
+       async_tx_run_dependencies(&desc->async_tx);
+       spin_unlock_bh(&desc->async_tx.lock);
+
+       return cookie;
+}
+
+static int
+iop_adma_clean_slot(struct iop_adma_desc_slot *desc,
+       struct iop_adma_chan *iop_chan)
+{
+       /* the client is allowed to attach dependent operations
+        * until 'ack' is set
+        */
+       if (!desc->async_tx.ack)
+               return 0;
+
+       /* leave the last descriptor in the chain
+        * so we can append to it
+        */
+       if (desc->chain_node.next == &iop_chan->chain)
+               return 1;
+
+       dev_dbg(iop_chan->device->common.dev,
+               "\tfree slot: %d slots_per_op: %d\n",
+               desc->idx, desc->slots_per_op);
+
+       list_del(&desc->chain_node);
+       iop_adma_free_slots(desc);
+
+       return 0;
+}
+
+static void __iop_adma_slot_cleanup(struct iop_adma_chan *iop_chan)
+{
+       struct iop_adma_desc_slot *iter, *_iter, *grp_start = NULL;
+       dma_cookie_t cookie = 0;
+       u32 current_desc = iop_chan_get_current_descriptor(iop_chan);
+       int busy = iop_chan_is_busy(iop_chan);
+       int seen_current = 0, slot_cnt = 0, slots_per_op = 0;
+
+       dev_dbg(iop_chan->device->common.dev, "%s\n", __FUNCTION__);
+       /* free completed slots from the chain starting with
+        * the oldest descriptor
+        */
+       list_for_each_entry_safe(iter, _iter, &iop_chan->chain,
+                                       chain_node) {
+               pr_debug("\tcookie: %d slot: %d busy: %d "
+                       "this_desc: %#x next_desc: %#x ack: %d\n",
+                       iter->async_tx.cookie, iter->idx, busy,
+                       iter->async_tx.phys, iop_desc_get_next_desc(iter),
+                       iter->async_tx.ack);
+               prefetch(_iter);
+               prefetch(&_iter->async_tx);
+
+               /* do not advance past the current descriptor loaded into the
+                * hardware channel, subsequent descriptors are either in
+                * process or have not been submitted
+                */
+               if (seen_current)
+                       break;
+
+               /* stop the search if we reach the current descriptor and the
+                * channel is busy, or if it appears that the current descriptor
+                * needs to be re-read (i.e. has been appended to)
+                */
+               if (iter->async_tx.phys == current_desc) {
+                       BUG_ON(seen_current++);
+                       if (busy || iop_desc_get_next_desc(iter))
+                               break;
+               }
+
+               /* detect the start of a group transaction */
+               if (!slot_cnt && !slots_per_op) {
+                       slot_cnt = iter->slot_cnt;
+                       slots_per_op = iter->slots_per_op;
+                       if (slot_cnt <= slots_per_op) {
+                               slot_cnt = 0;
+                               slots_per_op = 0;
+                       }
+               }
+
+               if (slot_cnt) {
+                       pr_debug("\tgroup++\n");
+                       if (!grp_start)
+                               grp_start = iter;
+                       slot_cnt -= slots_per_op;
+               }
+
+               /* all the members of a group are complete */
+               if (slots_per_op != 0 && slot_cnt == 0) {
+                       struct iop_adma_desc_slot *grp_iter, *_grp_iter;
+                       int end_of_chain = 0;
+                       pr_debug("\tgroup end\n");
+
+                       /* collect the total results */
+                       if (grp_start->xor_check_result) {
+                               u32 zero_sum_result = 0;
+                               slot_cnt = grp_start->slot_cnt;
+                               grp_iter = grp_start;
+
+                               list_for_each_entry_from(grp_iter,
+                                       &iop_chan->chain, chain_node) {
+                                       zero_sum_result |=
+                                           iop_desc_get_zero_result(grp_iter);
+                                           pr_debug("\titer%d result: %d\n",
+                                           grp_iter->idx, zero_sum_result);
+                                       slot_cnt -= slots_per_op;
+                                       if (slot_cnt == 0)
+                                               break;
+                               }
+                               pr_debug("\tgrp_start->xor_check_result: %p\n",
+                                       grp_start->xor_check_result);
+                               *grp_start->xor_check_result = zero_sum_result;
+                       }
+
+                       /* clean up the group */
+                       slot_cnt = grp_start->slot_cnt;
+                       grp_iter = grp_start;
+                       list_for_each_entry_safe_from(grp_iter, _grp_iter,
+                               &iop_chan->chain, chain_node) {
+                               cookie = iop_adma_run_tx_complete_actions(
+                                       grp_iter, iop_chan, cookie);
+
+                               slot_cnt -= slots_per_op;
+                               end_of_chain = iop_adma_clean_slot(grp_iter,
+                                       iop_chan);
+
+                               if (slot_cnt == 0 || end_of_chain)
+                                       break;
+                       }
+
+                       /* the group should be complete at this point */
+                       BUG_ON(slot_cnt);
+
+                       slots_per_op = 0;
+                       grp_start = NULL;
+                       if (end_of_chain)
+                               break;
+                       else
+                               continue;
+               } else if (slots_per_op) /* wait for group completion */
+                       continue;
+
+               /* write back zero sum results (single descriptor case) */
+               if (iter->xor_check_result && iter->async_tx.cookie)
+                       *iter->xor_check_result =
+                               iop_desc_get_zero_result(iter);
+
+               cookie = iop_adma_run_tx_complete_actions(
+                                       iter, iop_chan, cookie);
+
+               if (iop_adma_clean_slot(iter, iop_chan))
+                       break;
+       }
+
+       BUG_ON(!seen_current);
+
+       iop_chan_idle(busy, iop_chan);
+
+       if (cookie > 0) {
+               iop_chan->completed_cookie = cookie;
+               pr_debug("\tcompleted cookie %d\n", cookie);
+       }
+}
+
+static void
+iop_adma_slot_cleanup(struct iop_adma_chan *iop_chan)
+{
+       spin_lock_bh(&iop_chan->lock);
+       __iop_adma_slot_cleanup(iop_chan);
+       spin_unlock_bh(&iop_chan->lock);
+}
+
+static void iop_adma_tasklet(unsigned long data)
+{
+       struct iop_adma_chan *chan = (struct iop_adma_chan *) data;
+       __iop_adma_slot_cleanup(chan);
+}
+
+static struct iop_adma_desc_slot *
+iop_adma_alloc_slots(struct iop_adma_chan *iop_chan, int num_slots,
+                       int slots_per_op)
+{
+       struct iop_adma_desc_slot *iter, *_iter, *alloc_start = NULL;
+       struct list_head chain = LIST_HEAD_INIT(chain);
+       int slots_found, retry = 0;
+
+       /* start search from the last allocated descrtiptor
+        * if a contiguous allocation can not be found start searching
+        * from the beginning of the list
+        */
+retry:
+       slots_found = 0;
+       if (retry == 0)
+               iter = iop_chan->last_used;
+       else
+               iter = list_entry(&iop_chan->all_slots,
+                       struct iop_adma_desc_slot,
+                       slot_node);
+
+       list_for_each_entry_safe_continue(
+               iter, _iter, &iop_chan->all_slots, slot_node) {
+               prefetch(_iter);
+               prefetch(&_iter->async_tx);
+               if (iter->slots_per_op) {
+                       /* give up after finding the first busy slot
+                        * on the second pass through the list
+                        */
+                       if (retry)
+                               break;
+
+                       slots_found = 0;
+                       continue;
+               }
+
+               /* start the allocation if the slot is correctly aligned */
+               if (!slots_found++) {
+                       if (iop_desc_is_aligned(iter, slots_per_op))
+                               alloc_start = iter;
+                       else {
+                               slots_found = 0;
+                               continue;
+                       }
+               }
+
+               if (slots_found == num_slots) {
+                       struct iop_adma_desc_slot *alloc_tail = NULL;
+                       struct iop_adma_desc_slot *last_used = NULL;
+                       iter = alloc_start;
+                       while (num_slots) {
+                               int i;
+                               dev_dbg(iop_chan->device->common.dev,
+                                       "allocated slot: %d "
+                                       "(desc %p phys: %#x) slots_per_op %d\n",
+                                       iter->idx, iter->hw_desc,
+                                       iter->async_tx.phys, slots_per_op);
+
+                               /* pre-ack all but the last descriptor */
+                               if (num_slots != slots_per_op)
+                                       iter->async_tx.ack = 1;
+                               else
+                                       iter->async_tx.ack = 0;
+
+                               list_add_tail(&iter->chain_node, &chain);
+                               alloc_tail = iter;
+                               iter->async_tx.cookie = 0;
+                               iter->slot_cnt = num_slots;
+                               iter->xor_check_result = NULL;
+                               for (i = 0; i < slots_per_op; i++) {
+                                       iter->slots_per_op = slots_per_op - i;
+                                       last_used = iter;
+                                       iter = list_entry(iter->slot_node.next,
+                                               struct iop_adma_desc_slot,
+                                               slot_node);
+                               }
+                               num_slots -= slots_per_op;
+                       }
+                       alloc_tail->group_head = alloc_start;
+                       alloc_tail->async_tx.cookie = -EBUSY;
+                       list_splice(&chain, &alloc_tail->async_tx.tx_list);
+                       iop_chan->last_used = last_used;
+                       iop_desc_clear_next_desc(alloc_start);
+                       iop_desc_clear_next_desc(alloc_tail);
+                       return alloc_tail;
+               }
+       }
+       if (!retry++)
+               goto retry;
+
+       /* try to free some slots if the allocation fails */
+       tasklet_schedule(&iop_chan->irq_tasklet);
+
+       return NULL;
+}
+
+static dma_cookie_t
+iop_desc_assign_cookie(struct iop_adma_chan *iop_chan,
+       struct iop_adma_desc_slot *desc)
+{
+       dma_cookie_t cookie = iop_chan->common.cookie;
+       cookie++;
+       if (cookie < 0)
+               cookie = 1;
+       iop_chan->common.cookie = desc->async_tx.cookie = cookie;
+       return cookie;
+}
+
+static void iop_adma_check_threshold(struct iop_adma_chan *iop_chan)
+{
+       dev_dbg(iop_chan->device->common.dev, "pending: %d\n",
+               iop_chan->pending);
+
+       if (iop_chan->pending >= IOP_ADMA_THRESHOLD) {
+               iop_chan->pending = 0;
+               iop_chan_append(iop_chan);
+       }
+}
+
+static dma_cookie_t
+iop_adma_tx_submit(struct dma_async_tx_descriptor *tx)
+{
+       struct iop_adma_desc_slot *sw_desc = tx_to_iop_adma_slot(tx);
+       struct iop_adma_chan *iop_chan = to_iop_adma_chan(tx->chan);
+       struct iop_adma_desc_slot *grp_start, *old_chain_tail;
+       int slot_cnt;
+       int slots_per_op;
+       dma_cookie_t cookie;
+
+       grp_start = sw_desc->group_head;
+       slot_cnt = grp_start->slot_cnt;
+       slots_per_op = grp_start->slots_per_op;
+
+       spin_lock_bh(&iop_chan->lock);
+       cookie = iop_desc_assign_cookie(iop_chan, sw_desc);
+
+       old_chain_tail = list_entry(iop_chan->chain.prev,
+               struct iop_adma_desc_slot, chain_node);
+       list_splice_init(&sw_desc->async_tx.tx_list,
+                        &old_chain_tail->chain_node);
+
+       /* fix up the hardware chain */
+       iop_desc_set_next_desc(old_chain_tail, grp_start->async_tx.phys);
+
+       /* 1/ don't add pre-chained descriptors
+        * 2/ dummy read to flush next_desc write
+        */
+       BUG_ON(iop_desc_get_next_desc(sw_desc));
+
+       /* increment the pending count by the number of slots
+        * memcpy operations have a 1:1 (slot:operation) relation
+        * other operations are heavier and will pop the threshold
+        * more often.
+        */
+       iop_chan->pending += slot_cnt;
+       iop_adma_check_threshold(iop_chan);
+       spin_unlock_bh(&iop_chan->lock);
+
+       dev_dbg(iop_chan->device->common.dev, "%s cookie: %d slot: %d\n",
+               __FUNCTION__, sw_desc->async_tx.cookie, sw_desc->idx);
+
+       return cookie;
+}
+
+static void
+iop_adma_set_dest(dma_addr_t addr, struct dma_async_tx_descriptor *tx,
+       int index)
+{
+       struct iop_adma_desc_slot *sw_desc = tx_to_iop_adma_slot(tx);
+       struct iop_adma_chan *iop_chan = to_iop_adma_chan(tx->chan);
+
+       /* to do: support transfers lengths > IOP_ADMA_MAX_BYTE_COUNT */
+       iop_desc_set_dest_addr(sw_desc->group_head, iop_chan, addr);
+}
+
+static void iop_chan_start_null_memcpy(struct iop_adma_chan *iop_chan);
+static void iop_chan_start_null_xor(struct iop_adma_chan *iop_chan);
+
+/* returns the number of allocated descriptors */
+static int iop_adma_alloc_chan_resources(struct dma_chan *chan)
+{
+       char *hw_desc;
+       int idx;
+       struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
+       struct iop_adma_desc_slot *slot = NULL;
+       int init = iop_chan->slots_allocated ? 0 : 1;
+       struct iop_adma_platform_data *plat_data =
+               iop_chan->device->pdev->dev.platform_data;
+       int num_descs_in_pool = plat_data->pool_size/IOP_ADMA_SLOT_SIZE;
+
+       /* Allocate descriptor slots */
+       do {
+               idx = iop_chan->slots_allocated;
+               if (idx == num_descs_in_pool)
+                       break;
+
+               slot = kzalloc(sizeof(*slot), GFP_KERNEL);
+               if (!slot) {
+                       printk(KERN_INFO "IOP ADMA Channel only initialized"
+                               " %d descriptor slots", idx);
+                       break;
+               }
+               hw_desc = (char *) iop_chan->device->dma_desc_pool_virt;
+               slot->hw_desc = (void *) &hw_desc[idx * IOP_ADMA_SLOT_SIZE];
+
+               dma_async_tx_descriptor_init(&slot->async_tx, chan);
+               slot->async_tx.tx_submit = iop_adma_tx_submit;
+               slot->async_tx.tx_set_dest = iop_adma_set_dest;
+               INIT_LIST_HEAD(&slot->chain_node);
+               INIT_LIST_HEAD(&slot->slot_node);
+               INIT_LIST_HEAD(&slot->async_tx.tx_list);
+               hw_desc = (char *) iop_chan->device->dma_desc_pool;
+               slot->async_tx.phys =
+                       (dma_addr_t) &hw_desc[idx * IOP_ADMA_SLOT_SIZE];
+               slot->idx = idx;
+
+               spin_lock_bh(&iop_chan->lock);
+               iop_chan->slots_allocated++;
+               list_add_tail(&slot->slot_node, &iop_chan->all_slots);
+               spin_unlock_bh(&iop_chan->lock);
+       } while (iop_chan->slots_allocated < num_descs_in_pool);
+
+       if (idx && !iop_chan->last_used)
+               iop_chan->last_used = list_entry(iop_chan->all_slots.next,
+                                       struct iop_adma_desc_slot,
+                                       slot_node);
+
+       dev_dbg(iop_chan->device->common.dev,
+               "allocated %d descriptor slots last_used: %p\n",
+               iop_chan->slots_allocated, iop_chan->last_used);
+
+       /* initialize the channel and the chain with a null operation */
+       if (init) {
+               if (dma_has_cap(DMA_MEMCPY,
+                       iop_chan->device->common.cap_mask))
+                       iop_chan_start_null_memcpy(iop_chan);
+               else if (dma_has_cap(DMA_XOR,
+                       iop_chan->device->common.cap_mask))
+                       iop_chan_start_null_xor(iop_chan);
+               else
+                       BUG();
+       }
+
+       return (idx > 0) ? idx : -ENOMEM;
+}
+
+static struct dma_async_tx_descriptor *
+iop_adma_prep_dma_interrupt(struct dma_chan *chan)
+{
+       struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
+       struct iop_adma_desc_slot *sw_desc, *grp_start;
+       int slot_cnt, slots_per_op;
+
+       dev_dbg(iop_chan->device->common.dev, "%s\n", __FUNCTION__);
+
+       spin_lock_bh(&iop_chan->lock);
+       slot_cnt = iop_chan_interrupt_slot_count(&slots_per_op, iop_chan);
+       sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
+       if (sw_desc) {
+               grp_start = sw_desc->group_head;
+               iop_desc_init_interrupt(grp_start, iop_chan);
+               grp_start->unmap_len = 0;
+       }
+       spin_unlock_bh(&iop_chan->lock);
+
+       return sw_desc ? &sw_desc->async_tx : NULL;
+}
+
+static void
+iop_adma_memcpy_set_src(dma_addr_t addr, struct dma_async_tx_descriptor *tx,
+       int index)
+{
+       struct iop_adma_desc_slot *sw_desc = tx_to_iop_adma_slot(tx);
+       struct iop_adma_desc_slot *grp_start = sw_desc->group_head;
+
+       iop_desc_set_memcpy_src_addr(grp_start, addr);
+}
+
+static struct dma_async_tx_descriptor *
+iop_adma_prep_dma_memcpy(struct dma_chan *chan, size_t len, int int_en)
+{
+       struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
+       struct iop_adma_desc_slot *sw_desc, *grp_start;
+       int slot_cnt, slots_per_op;
+
+       if (unlikely(!len))
+               return NULL;
+       BUG_ON(unlikely(len > IOP_ADMA_MAX_BYTE_COUNT));
+
+       dev_dbg(iop_chan->device->common.dev, "%s len: %u\n",
+               __FUNCTION__, len);
+
+       spin_lock_bh(&iop_chan->lock);
+       slot_cnt = iop_chan_memcpy_slot_count(len, &slots_per_op);
+       sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
+       if (sw_desc) {
+               grp_start = sw_desc->group_head;
+               iop_desc_init_memcpy(grp_start, int_en);
+               iop_desc_set_byte_count(grp_start, iop_chan, len);
+               sw_desc->unmap_src_cnt = 1;
+               sw_desc->unmap_len = len;
+               sw_desc->async_tx.tx_set_src = iop_adma_memcpy_set_src;
+       }
+       spin_unlock_bh(&iop_chan->lock);
+
+       return sw_desc ? &sw_desc->async_tx : NULL;
+}
+
+static struct dma_async_tx_descriptor *
+iop_adma_prep_dma_memset(struct dma_chan *chan, int value, size_t len,
+       int int_en)
+{
+       struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
+       struct iop_adma_desc_slot *sw_desc, *grp_start;
+       int slot_cnt, slots_per_op;
+
+       if (unlikely(!len))
+               return NULL;
+       BUG_ON(unlikely(len > IOP_ADMA_MAX_BYTE_COUNT));
+
+       dev_dbg(iop_chan->device->common.dev, "%s len: %u\n",
+               __FUNCTION__, len);
+
+       spin_lock_bh(&iop_chan->lock);
+       slot_cnt = iop_chan_memset_slot_count(len, &slots_per_op);
+       sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
+       if (sw_desc) {
+               grp_start = sw_desc->group_head;
+               iop_desc_init_memset(grp_start, int_en);
+               iop_desc_set_byte_count(grp_start, iop_chan, len);
+               iop_desc_set_block_fill_val(grp_start, value);
+               sw_desc->unmap_src_cnt = 1;
+               sw_desc->unmap_len = len;
+       }
+       spin_unlock_bh(&iop_chan->lock);
+
+       return sw_desc ? &sw_desc->async_tx : NULL;
+}
+
+static void
+iop_adma_xor_set_src(dma_addr_t addr, struct dma_async_tx_descriptor *tx,
+       int index)
+{
+       struct iop_adma_desc_slot *sw_desc = tx_to_iop_adma_slot(tx);
+       struct iop_adma_desc_slot *grp_start = sw_desc->group_head;
+
+       iop_desc_set_xor_src_addr(grp_start, index, addr);
+}
+
+static struct dma_async_tx_descriptor *
+iop_adma_prep_dma_xor(struct dma_chan *chan, unsigned int src_cnt, size_t len,
+       int int_en)
+{
+       struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
+       struct iop_adma_desc_slot *sw_desc, *grp_start;
+       int slot_cnt, slots_per_op;
+
+       if (unlikely(!len))
+               return NULL;
+       BUG_ON(unlikely(len > IOP_ADMA_XOR_MAX_BYTE_COUNT));
+
+       dev_dbg(iop_chan->device->common.dev,
+               "%s src_cnt: %d len: %u int_en: %d\n",
+               __FUNCTION__, src_cnt, len, int_en);
+
+       spin_lock_bh(&iop_chan->lock);
+       slot_cnt = iop_chan_xor_slot_count(len, src_cnt, &slots_per_op);
+       sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
+       if (sw_desc) {
+               grp_start = sw_desc->group_head;
+               iop_desc_init_xor(grp_start, src_cnt, int_en);
+               iop_desc_set_byte_count(grp_start, iop_chan, len);
+               sw_desc->unmap_src_cnt = src_cnt;
+               sw_desc->unmap_len = len;
+               sw_desc->async_tx.tx_set_src = iop_adma_xor_set_src;
+       }
+       spin_unlock_bh(&iop_chan->lock);
+
+       return sw_desc ? &sw_desc->async_tx : NULL;
+}
+
+static void
+iop_adma_xor_zero_sum_set_src(dma_addr_t addr,
+                               struct dma_async_tx_descriptor *tx,
+                               int index)
+{
+       struct iop_adma_desc_slot *sw_desc = tx_to_iop_adma_slot(tx);
+       struct iop_adma_desc_slot *grp_start = sw_desc->group_head;
+
+       iop_desc_set_zero_sum_src_addr(grp_start, index, addr);
+}
+
+static struct dma_async_tx_descriptor *
+iop_adma_prep_dma_zero_sum(struct dma_chan *chan, unsigned int src_cnt,
+       size_t len, u32 *result, int int_en)
+{
+       struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
+       struct iop_adma_desc_slot *sw_desc, *grp_start;
+       int slot_cnt, slots_per_op;
+
+       if (unlikely(!len))
+               return NULL;
+
+       dev_dbg(iop_chan->device->common.dev, "%s src_cnt: %d len: %u\n",
+               __FUNCTION__, src_cnt, len);
+
+       spin_lock_bh(&iop_chan->lock);
+       slot_cnt = iop_chan_zero_sum_slot_count(len, src_cnt, &slots_per_op);
+       sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
+       if (sw_desc) {
+               grp_start = sw_desc->group_head;
+               iop_desc_init_zero_sum(grp_start, src_cnt, int_en);
+               iop_desc_set_zero_sum_byte_count(grp_start, len);
+               grp_start->xor_check_result = result;
+               pr_debug("\t%s: grp_start->xor_check_result: %p\n",
+                       __FUNCTION__, grp_start->xor_check_result);
+               sw_desc->unmap_src_cnt = src_cnt;
+               sw_desc->unmap_len = len;
+               sw_desc->async_tx.tx_set_src = iop_adma_xor_zero_sum_set_src;
+       }
+       spin_unlock_bh(&iop_chan->lock);
+
+       return sw_desc ? &sw_desc->async_tx : NULL;
+}
+
+static void iop_adma_dependency_added(struct dma_chan *chan)
+{
+       struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
+       tasklet_schedule(&iop_chan->irq_tasklet);
+}
+
+static void iop_adma_free_chan_resources(struct dma_chan *chan)
+{
+       struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
+       struct iop_adma_desc_slot *iter, *_iter;
+       int in_use_descs = 0;
+
+       iop_adma_slot_cleanup(iop_chan);
+
+       spin_lock_bh(&iop_chan->lock);
+       list_for_each_entry_safe(iter, _iter, &iop_chan->chain,
+                                       chain_node) {
+               in_use_descs++;
+               list_del(&iter->chain_node);
+       }
+       list_for_each_entry_safe_reverse(
+               iter, _iter, &iop_chan->all_slots, slot_node) {
+               list_del(&iter->slot_node);
+               kfree(iter);
+               iop_chan->slots_allocated--;
+       }
+       iop_chan->last_used = NULL;
+
+       dev_dbg(iop_chan->device->common.dev, "%s slots_allocated %d\n",
+               __FUNCTION__, iop_chan->slots_allocated);
+       spin_unlock_bh(&iop_chan->lock);
+
+       /* one is ok since we left it on there on purpose */
+       if (in_use_descs > 1)
+               printk(KERN_ERR "IOP: Freeing %d in use descriptors!\n",
+                       in_use_descs - 1);
+}
+
+/**
+ * iop_adma_is_complete - poll the status of an ADMA transaction
+ * @chan: ADMA channel handle
+ * @cookie: ADMA transaction identifier
+ */
+static enum dma_status iop_adma_is_complete(struct dma_chan *chan,
+                                       dma_cookie_t cookie,
+                                       dma_cookie_t *done,
+                                       dma_cookie_t *used)
+{
+       struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
+       dma_cookie_t last_used;
+       dma_cookie_t last_complete;
+       enum dma_status ret;
+
+       last_used = chan->cookie;
+       last_complete = iop_chan->completed_cookie;
+
+       if (done)
+               *done = last_complete;
+       if (used)
+               *used = last_used;
+
+       ret = dma_async_is_complete(cookie, last_complete, last_used);
+       if (ret == DMA_SUCCESS)
+               return ret;
+
+       iop_adma_slot_cleanup(iop_chan);
+
+       last_used = chan->cookie;
+       last_complete = iop_chan->completed_cookie;
+
+       if (done)
+               *done = last_complete;
+       if (used)
+               *used = last_used;
+
+       return dma_async_is_complete(cookie, last_complete, last_used);
+}
+
+static irqreturn_t iop_adma_eot_handler(int irq, void *data)
+{
+       struct iop_adma_chan *chan = data;
+
+       dev_dbg(chan->device->common.dev, "%s\n", __FUNCTION__);
+
+       tasklet_schedule(&chan->irq_tasklet);
+
+       iop_adma_device_clear_eot_status(chan);
+
+       return IRQ_HANDLED;
+}
+
+static irqreturn_t iop_adma_eoc_handler(int irq, void *data)
+{
+       struct iop_adma_chan *chan = data;
+
+       dev_dbg(chan->device->common.dev, "%s\n", __FUNCTION__);
+
+       tasklet_schedule(&chan->irq_tasklet);
+
+       iop_adma_device_clear_eoc_status(chan);
+
+       return IRQ_HANDLED;
+}
+
+static irqreturn_t iop_adma_err_handler(int irq, void *data)
+{
+       struct iop_adma_chan *chan = data;
+       unsigned long status = iop_chan_get_status(chan);
+
+       dev_printk(KERN_ERR, chan->device->common.dev,
+               "error ( %s%s%s%s%s%s%s)\n",
+               iop_is_err_int_parity(status, chan) ? "int_parity " : "",
+               iop_is_err_mcu_abort(status, chan) ? "mcu_abort " : "",
+               iop_is_err_int_tabort(status, chan) ? "int_tabort " : "",
+               iop_is_err_int_mabort(status, chan) ? "int_mabort " : "",
+               iop_is_err_pci_tabort(status, chan) ? "pci_tabort " : "",
+               iop_is_err_pci_mabort(status, chan) ? "pci_mabort " : "",
+               iop_is_err_split_tx(status, chan) ? "split_tx " : "");
+
+       iop_adma_device_clear_err_status(chan);
+
+       BUG();
+
+       return IRQ_HANDLED;
+}
+
+static void iop_adma_issue_pending(struct dma_chan *chan)
+{
+       struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
+
+       if (iop_chan->pending) {
+               iop_chan->pending = 0;
+               iop_chan_append(iop_chan);
+       }
+}
+
+/*
+ * Perform a transaction to verify the HW works.
+ */
+#define IOP_ADMA_TEST_SIZE 2000
+
+static int __devinit iop_adma_memcpy_self_test(struct iop_adma_device *device)
+{
+       int i;
+       void *src, *dest;
+       dma_addr_t src_dma, dest_dma;
+       struct dma_chan *dma_chan;
+       dma_cookie_t cookie;
+       struct dma_async_tx_descriptor *tx;
+       int err = 0;
+       struct iop_adma_chan *iop_chan;
+
+       dev_dbg(device->common.dev, "%s\n", __FUNCTION__);
+
+       src = kzalloc(sizeof(u8) * IOP_ADMA_TEST_SIZE, GFP_KERNEL);
+       if (!src)
+               return -ENOMEM;
+       dest = kzalloc(sizeof(u8) * IOP_ADMA_TEST_SIZE, GFP_KERNEL);
+       if (!dest) {
+               kfree(src);
+               return -ENOMEM;
+       }
+
+       /* Fill in src buffer */
+       for (i = 0; i < IOP_ADMA_TEST_SIZE; i++)
+               ((u8 *) src)[i] = (u8)i;
+
+       memset(dest, 0, IOP_ADMA_TEST_SIZE);
+
+       /* Start copy, using first DMA channel */
+       dma_chan = container_of(device->common.channels.next,
+                               struct dma_chan,
+                               device_node);
+       if (iop_adma_alloc_chan_resources(dma_chan) < 1) {
+               err = -ENODEV;
+               goto out;
+       }
+
+       tx = iop_adma_prep_dma_memcpy(dma_chan, IOP_ADMA_TEST_SIZE, 1);
+       dest_dma = dma_map_single(dma_chan->device->dev, dest,
+                               IOP_ADMA_TEST_SIZE, DMA_FROM_DEVICE);
+       iop_adma_set_dest(dest_dma, tx, 0);
+       src_dma = dma_map_single(dma_chan->device->dev, src,
+                               IOP_ADMA_TEST_SIZE, DMA_TO_DEVICE);
+       iop_adma_memcpy_set_src(src_dma, tx, 0);
+
+       cookie = iop_adma_tx_submit(tx);
+       iop_adma_issue_pending(dma_chan);
+       async_tx_ack(tx);
+       msleep(1);
+
+       if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) !=
+                       DMA_SUCCESS) {
+               dev_printk(KERN_ERR, dma_chan->device->dev,
+                       "Self-test copy timed out, disabling\n");
+               err = -ENODEV;
+               goto free_resources;
+       }
+
+       iop_chan = to_iop_adma_chan(dma_chan);
+       dma_sync_single_for_cpu(&iop_chan->device->pdev->dev, dest_dma,
+               IOP_ADMA_TEST_SIZE, DMA_FROM_DEVICE);
+       if (memcmp(src, dest, IOP_ADMA_TEST_SIZE)) {
+               dev_printk(KERN_ERR, dma_chan->device->dev,
+                       "Self-test copy failed compare, disabling\n");
+               err = -ENODEV;
+               goto free_resources;
+       }
+
+free_resources:
+       iop_adma_free_chan_resources(dma_chan);
+out:
+       kfree(src);
+       kfree(dest);
+       return err;
+}
+
+#define IOP_ADMA_NUM_SRC_TEST 4 /* must be <= 15 */
+static int __devinit
+iop_adma_xor_zero_sum_self_test(struct iop_adma_device *device)
+{
+       int i, src_idx;
+       struct page *dest;
+       struct page *xor_srcs[IOP_ADMA_NUM_SRC_TEST];
+       struct page *zero_sum_srcs[IOP_ADMA_NUM_SRC_TEST + 1];
+       dma_addr_t dma_addr, dest_dma;
+       struct dma_async_tx_descriptor *tx;
+       struct dma_chan *dma_chan;
+       dma_cookie_t cookie;
+       u8 cmp_byte = 0;
+       u32 cmp_word;
+       u32 zero_sum_result;
+       int err = 0;
+       struct iop_adma_chan *iop_chan;
+
+       dev_dbg(device->common.dev, "%s\n", __FUNCTION__);
+
+       for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++) {
+               xor_srcs[src_idx] = alloc_page(GFP_KERNEL);
+               if (!xor_srcs[src_idx])
+                       while (src_idx--) {
+                               __free_page(xor_srcs[src_idx]);
+                               return -ENOMEM;
+                       }
+       }
+
+       dest = alloc_page(GFP_KERNEL);
+       if (!dest)
+               while (src_idx--) {
+                       __free_page(xor_srcs[src_idx]);
+                       return -ENOMEM;
+               }
+
+       /* Fill in src buffers */
+       for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++) {
+               u8 *ptr = page_address(xor_srcs[src_idx]);
+               for (i = 0; i < PAGE_SIZE; i++)
+                       ptr[i] = (1 << src_idx);
+       }
+
+       for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++)
+               cmp_byte ^= (u8) (1 << src_idx);
+
+       cmp_word = (cmp_byte << 24) | (cmp_byte << 16) |
+                       (cmp_byte << 8) | cmp_byte;
+
+       memset(page_address(dest), 0, PAGE_SIZE);
+
+       dma_chan = container_of(device->common.channels.next,
+                               struct dma_chan,
+                               device_node);
+       if (iop_adma_alloc_chan_resources(dma_chan) < 1) {
+               err = -ENODEV;
+               goto out;
+       }
+
+       /* test xor */
+       tx = iop_adma_prep_dma_xor(dma_chan, IOP_ADMA_NUM_SRC_TEST,
+                               PAGE_SIZE, 1);
+       dest_dma = dma_map_page(dma_chan->device->dev, dest, 0,
+                               PAGE_SIZE, DMA_FROM_DEVICE);
+       iop_adma_set_dest(dest_dma, tx, 0);
+
+       for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++) {
+               dma_addr = dma_map_page(dma_chan->device->dev, xor_srcs[i], 0,
+                       PAGE_SIZE, DMA_TO_DEVICE);
+               iop_adma_xor_set_src(dma_addr, tx, i);
+       }
+
+       cookie = iop_adma_tx_submit(tx);
+       iop_adma_issue_pending(dma_chan);
+       async_tx_ack(tx);
+       msleep(8);
+
+       if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) !=
+               DMA_SUCCESS) {
+               dev_printk(KERN_ERR, dma_chan->device->dev,
+                       "Self-test xor timed out, disabling\n");
+               err = -ENODEV;
+               goto free_resources;
+       }
+
+       iop_chan = to_iop_adma_chan(dma_chan);
+       dma_sync_single_for_cpu(&iop_chan->device->pdev->dev, dest_dma,
+               PAGE_SIZE, DMA_FROM_DEVICE);
+       for (i = 0; i < (PAGE_SIZE / sizeof(u32)); i++) {
+               u32 *ptr = page_address(dest);
+               if (ptr[i] != cmp_word) {
+                       dev_printk(KERN_ERR, dma_chan->device->dev,
+                               "Self-test xor failed compare, disabling\n");
+                       err = -ENODEV;
+                       goto free_resources;
+               }
+       }
+       dma_sync_single_for_device(&iop_chan->device->pdev->dev, dest_dma,
+               PAGE_SIZE, DMA_TO_DEVICE);
+
+       /* skip zero sum if the capability is not present */
+       if (!dma_has_cap(DMA_ZERO_SUM, dma_chan->device->cap_mask))
+               goto free_resources;
+
+       /* zero sum the sources with the destintation page */
+       for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++)
+               zero_sum_srcs[i] = xor_srcs[i];
+       zero_sum_srcs[i] = dest;
+
+       zero_sum_result = 1;
+
+       tx = iop_adma_prep_dma_zero_sum(dma_chan, IOP_ADMA_NUM_SRC_TEST + 1,
+               PAGE_SIZE, &zero_sum_result, 1);
+       for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 1; i++) {
+               dma_addr = dma_map_page(dma_chan->device->dev, zero_sum_srcs[i],
+                       0, PAGE_SIZE, DMA_TO_DEVICE);
+               iop_adma_xor_zero_sum_set_src(dma_addr, tx, i);
+       }
+
+       cookie = iop_adma_tx_submit(tx);
+       iop_adma_issue_pending(dma_chan);
+       async_tx_ack(tx);
+       msleep(8);
+
+       if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) {
+               dev_printk(KERN_ERR, dma_chan->device->dev,
+                       "Self-test zero sum timed out, disabling\n");
+               err = -ENODEV;
+               goto free_resources;
+       }
+
+       if (zero_sum_result != 0) {
+               dev_printk(KERN_ERR, dma_chan->device->dev,
+                       "Self-test zero sum failed compare, disabling\n");
+               err = -ENODEV;
+               goto free_resources;
+       }
+
+       /* test memset */
+       tx = iop_adma_prep_dma_memset(dma_chan, 0, PAGE_SIZE, 1);
+       dma_addr = dma_map_page(dma_chan->device->dev, dest, 0,
+                       PAGE_SIZE, DMA_FROM_DEVICE);
+       iop_adma_set_dest(dma_addr, tx, 0);
+
+       cookie = iop_adma_tx_submit(tx);
+       iop_adma_issue_pending(dma_chan);
+       async_tx_ack(tx);
+       msleep(8);
+
+       if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) {
+               dev_printk(KERN_ERR, dma_chan->device->dev,
+                       "Self-test memset timed out, disabling\n");
+               err = -ENODEV;
+               goto free_resources;
+       }
+
+       for (i = 0; i < PAGE_SIZE/sizeof(u32); i++) {
+               u32 *ptr = page_address(dest);
+               if (ptr[i]) {
+                       dev_printk(KERN_ERR, dma_chan->device->dev,
+                               "Self-test memset failed compare, disabling\n");
+                       err = -ENODEV;
+                       goto free_resources;
+               }
+       }
+
+       /* test for non-zero parity sum */
+       zero_sum_result = 0;
+       tx = iop_adma_prep_dma_zero_sum(dma_chan, IOP_ADMA_NUM_SRC_TEST + 1,
+               PAGE_SIZE, &zero_sum_result, 1);
+       for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 1; i++) {
+               dma_addr = dma_map_page(dma_chan->device->dev, zero_sum_srcs[i],
+                       0, PAGE_SIZE, DMA_TO_DEVICE);
+               iop_adma_xor_zero_sum_set_src(dma_addr, tx, i);
+       }
+
+       cookie = iop_adma_tx_submit(tx);
+       iop_adma_issue_pending(dma_chan);
+       async_tx_ack(tx);
+       msleep(8);
+
+       if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) {
+               dev_printk(KERN_ERR, dma_chan->device->dev,
+                       "Self-test non-zero sum timed out, disabling\n");
+               err = -ENODEV;
+               goto free_resources;
+       }
+
+       if (zero_sum_result != 1) {
+               dev_printk(KERN_ERR, dma_chan->device->dev,
+                       "Self-test non-zero sum failed compare, disabling\n");
+               err = -ENODEV;
+               goto free_resources;
+       }
+
+free_resources:
+       iop_adma_free_chan_resources(dma_chan);
+out:
+       src_idx = IOP_ADMA_NUM_SRC_TEST;
+       while (src_idx--)
+               __free_page(xor_srcs[src_idx]);
+       __free_page(dest);
+       return err;
+}
+
+static int __devexit iop_adma_remove(struct platform_device *dev)
+{
+       struct iop_adma_device *device = platform_get_drvdata(dev);
+       struct dma_chan *chan, *_chan;
+       struct iop_adma_chan *iop_chan;
+       int i;
+       struct iop_adma_platform_data *plat_data = dev->dev.platform_data;
+
+       dma_async_device_unregister(&device->common);
+
+       for (i = 0; i < 3; i++) {
+               unsigned int irq;
+               irq = platform_get_irq(dev, i);
+               free_irq(irq, device);
+       }
+
+       dma_free_coherent(&dev->dev, plat_data->pool_size,
+                       device->dma_desc_pool_virt, device->dma_desc_pool);
+
+       do {
+               struct resource *res;
+               res = platform_get_resource(dev, IORESOURCE_MEM, 0);
+               release_mem_region(res->start, res->end - res->start);
+       } while (0);
+
+       list_for_each_entry_safe(chan, _chan, &device->common.channels,
+                               device_node) {
+               iop_chan = to_iop_adma_chan(chan);
+               list_del(&chan->device_node);
+               kfree(iop_chan);
+       }
+       kfree(device);
+
+       return 0;
+}
+
+static int __devinit iop_adma_probe(struct platform_device *pdev)
+{
+       struct resource *res;
+       int ret = 0, i;
+       struct iop_adma_device *adev;
+       struct iop_adma_chan *iop_chan;
+       struct dma_device *dma_dev;
+       struct iop_adma_platform_data *plat_data = pdev->dev.platform_data;
+
+       res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       if (!res)
+               return -ENODEV;
+
+       if (!devm_request_mem_region(&pdev->dev, res->start,
+                               res->end - res->start, pdev->name))
+               return -EBUSY;
+
+       adev = kzalloc(sizeof(*adev), GFP_KERNEL);
+       if (!adev)
+               return -ENOMEM;
+       dma_dev = &adev->common;
+
+       /* allocate coherent memory for hardware descriptors
+        * note: writecombine gives slightly better performance, but
+        * requires that we explicitly flush the writes
+        */
+       if ((adev->dma_desc_pool_virt = dma_alloc_writecombine(&pdev->dev,
+                                       plat_data->pool_size,
+                                       &adev->dma_desc_pool,
+                                       GFP_KERNEL)) == NULL) {
+               ret = -ENOMEM;
+               goto err_free_adev;
+       }
+
+       dev_dbg(&pdev->dev, "%s: allocted descriptor pool virt %p phys %p\n",
+               __FUNCTION__, adev->dma_desc_pool_virt,
+               (void *) adev->dma_desc_pool);
+
+       adev->id = plat_data->hw_id;
+
+       /* discover transaction capabilites from the platform data */
+       dma_dev->cap_mask = plat_data->cap_mask;
+
+       adev->pdev = pdev;
+       platform_set_drvdata(pdev, adev);
+
+       INIT_LIST_HEAD(&dma_dev->channels);
+
+       /* set base routines */
+       dma_dev->device_alloc_chan_resources = iop_adma_alloc_chan_resources;
+       dma_dev->device_free_chan_resources = iop_adma_free_chan_resources;
+       dma_dev->device_is_tx_complete = iop_adma_is_complete;
+       dma_dev->device_issue_pending = iop_adma_issue_pending;
+       dma_dev->device_dependency_added = iop_adma_dependency_added;
+       dma_dev->dev = &pdev->dev;
+
+       /* set prep routines based on capability */
+       if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask))
+               dma_dev->device_prep_dma_memcpy = iop_adma_prep_dma_memcpy;
+       if (dma_has_cap(DMA_MEMSET, dma_dev->cap_mask))
+               dma_dev->device_prep_dma_memset = iop_adma_prep_dma_memset;
+       if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
+               dma_dev->max_xor = iop_adma_get_max_xor();
+               dma_dev->device_prep_dma_xor = iop_adma_prep_dma_xor;
+       }
+       if (dma_has_cap(DMA_ZERO_SUM, dma_dev->cap_mask))
+               dma_dev->device_prep_dma_zero_sum =
+                       iop_adma_prep_dma_zero_sum;
+       if (dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask))
+               dma_dev->device_prep_dma_interrupt =
+                       iop_adma_prep_dma_interrupt;
+
+       iop_chan = kzalloc(sizeof(*iop_chan), GFP_KERNEL);
+       if (!iop_chan) {
+               ret = -ENOMEM;
+               goto err_free_dma;
+       }
+       iop_chan->device = adev;
+
+       iop_chan->mmr_base = devm_ioremap(&pdev->dev, res->start,
+                                       res->end - res->start);
+       if (!iop_chan->mmr_base) {
+               ret = -ENOMEM;
+               goto err_free_iop_chan;
+       }
+       tasklet_init(&iop_chan->irq_tasklet, iop_adma_tasklet, (unsigned long)
+               iop_chan);
+
+       /* clear errors before enabling interrupts */
+       iop_adma_device_clear_err_status(iop_chan);
+
+       for (i = 0; i < 3; i++) {
+               irq_handler_t handler[] = { iop_adma_eot_handler,
+                                       iop_adma_eoc_handler,
+                                       iop_adma_err_handler };
+               int irq = platform_get_irq(pdev, i);
+               if (irq < 0) {
+                       ret = -ENXIO;
+                       goto err_free_iop_chan;
+               } else {
+                       ret = devm_request_irq(&pdev->dev, irq,
+                                       handler[i], 0, pdev->name, iop_chan);
+                       if (ret)
+                               goto err_free_iop_chan;
+               }
+       }
+
+       spin_lock_init(&iop_chan->lock);
+       init_timer(&iop_chan->cleanup_watchdog);
+       iop_chan->cleanup_watchdog.data = (unsigned long) iop_chan;
+       iop_chan->cleanup_watchdog.function = iop_adma_tasklet;
+       INIT_LIST_HEAD(&iop_chan->chain);
+       INIT_LIST_HEAD(&iop_chan->all_slots);
+       INIT_RCU_HEAD(&iop_chan->common.rcu);
+       iop_chan->common.device = dma_dev;
+       list_add_tail(&iop_chan->common.device_node, &dma_dev->channels);
+
+       if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask)) {
+               ret = iop_adma_memcpy_self_test(adev);
+               dev_dbg(&pdev->dev, "memcpy self test returned %d\n", ret);
+               if (ret)
+                       goto err_free_iop_chan;
+       }
+
+       if (dma_has_cap(DMA_XOR, dma_dev->cap_mask) ||
+               dma_has_cap(DMA_MEMSET, dma_dev->cap_mask)) {
+               ret = iop_adma_xor_zero_sum_self_test(adev);
+               dev_dbg(&pdev->dev, "xor self test returned %d\n", ret);
+               if (ret)
+                       goto err_free_iop_chan;
+       }
+
+       dev_printk(KERN_INFO, &pdev->dev, "Intel(R) IOP: "
+         "( %s%s%s%s%s%s%s%s%s%s)\n",
+         dma_has_cap(DMA_PQ_XOR, dma_dev->cap_mask) ? "pq_xor " : "",
+         dma_has_cap(DMA_PQ_UPDATE, dma_dev->cap_mask) ? "pq_update " : "",
+         dma_has_cap(DMA_PQ_ZERO_SUM, dma_dev->cap_mask) ? "pq_zero_sum " : "",
+         dma_has_cap(DMA_XOR, dma_dev->cap_mask) ? "xor " : "",
+         dma_has_cap(DMA_DUAL_XOR, dma_dev->cap_mask) ? "dual_xor " : "",
+         dma_has_cap(DMA_ZERO_SUM, dma_dev->cap_mask) ? "xor_zero_sum " : "",
+         dma_has_cap(DMA_MEMSET, dma_dev->cap_mask)  ? "fill " : "",
+         dma_has_cap(DMA_MEMCPY_CRC32C, dma_dev->cap_mask) ? "cpy+crc " : "",
+         dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask) ? "cpy " : "",
+         dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask) ? "intr " : "");
+
+       dma_async_device_register(dma_dev);
+       goto out;
+
+ err_free_iop_chan:
+       kfree(iop_chan);
+ err_free_dma:
+       dma_free_coherent(&adev->pdev->dev, plat_data->pool_size,
+                       adev->dma_desc_pool_virt, adev->dma_desc_pool);
+ err_free_adev:
+       kfree(adev);
+ out:
+       return ret;
+}
+
+static void iop_chan_start_null_memcpy(struct iop_adma_chan *iop_chan)
+{
+       struct iop_adma_desc_slot *sw_desc, *grp_start;
+       dma_cookie_t cookie;
+       int slot_cnt, slots_per_op;
+
+       dev_dbg(iop_chan->device->common.dev, "%s\n", __FUNCTION__);
+
+       spin_lock_bh(&iop_chan->lock);
+       slot_cnt = iop_chan_memcpy_slot_count(0, &slots_per_op);
+       sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
+       if (sw_desc) {
+               grp_start = sw_desc->group_head;
+
+               list_splice_init(&sw_desc->async_tx.tx_list, &iop_chan->chain);
+               sw_desc->async_tx.ack = 1;
+               iop_desc_init_memcpy(grp_start, 0);
+               iop_desc_set_byte_count(grp_start, iop_chan, 0);
+               iop_desc_set_dest_addr(grp_start, iop_chan, 0);
+               iop_desc_set_memcpy_src_addr(grp_start, 0);
+
+               cookie = iop_chan->common.cookie;
+               cookie++;
+               if (cookie <= 1)
+                       cookie = 2;
+
+               /* initialize the completed cookie to be less than
+                * the most recently used cookie
+                */
+               iop_chan->completed_cookie = cookie - 1;
+               iop_chan->common.cookie = sw_desc->async_tx.cookie = cookie;
+
+               /* channel should not be busy */
+               BUG_ON(iop_chan_is_busy(iop_chan));
+
+               /* clear any prior error-status bits */
+               iop_adma_device_clear_err_status(iop_chan);
+
+               /* disable operation */
+               iop_chan_disable(iop_chan);
+
+               /* set the descriptor address */
+               iop_chan_set_next_descriptor(iop_chan, sw_desc->async_tx.phys);
+
+               /* 1/ don't add pre-chained descriptors
+                * 2/ dummy read to flush next_desc write
+                */
+               BUG_ON(iop_desc_get_next_desc(sw_desc));
+
+               /* run the descriptor */
+               iop_chan_enable(iop_chan);
+       } else
+               dev_printk(KERN_ERR, iop_chan->device->common.dev,
+                        "failed to allocate null descriptor\n");
+       spin_unlock_bh(&iop_chan->lock);
+}
+
+static void iop_chan_start_null_xor(struct iop_adma_chan *iop_chan)
+{
+       struct iop_adma_desc_slot *sw_desc, *grp_start;
+       dma_cookie_t cookie;
+       int slot_cnt, slots_per_op;
+
+       dev_dbg(iop_chan->device->common.dev, "%s\n", __FUNCTION__);
+
+       spin_lock_bh(&iop_chan->lock);
+       slot_cnt = iop_chan_xor_slot_count(0, 2, &slots_per_op);
+       sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
+       if (sw_desc) {
+               grp_start = sw_desc->group_head;
+               list_splice_init(&sw_desc->async_tx.tx_list, &iop_chan->chain);
+               sw_desc->async_tx.ack = 1;
+               iop_desc_init_null_xor(grp_start, 2, 0);
+               iop_desc_set_byte_count(grp_start, iop_chan, 0);
+               iop_desc_set_dest_addr(grp_start, iop_chan, 0);
+               iop_desc_set_xor_src_addr(grp_start, 0, 0);
+               iop_desc_set_xor_src_addr(grp_start, 1, 0);
+
+               cookie = iop_chan->common.cookie;
+               cookie++;
+               if (cookie <= 1)
+                       cookie = 2;
+
+               /* initialize the completed cookie to be less than
+                * the most recently used cookie
+                */
+               iop_chan->completed_cookie = cookie - 1;
+               iop_chan->common.cookie = sw_desc->async_tx.cookie = cookie;
+
+               /* channel should not be busy */
+               BUG_ON(iop_chan_is_busy(iop_chan));
+
+               /* clear any prior error-status bits */
+               iop_adma_device_clear_err_status(iop_chan);
+
+               /* disable operation */
+               iop_chan_disable(iop_chan);
+
+               /* set the descriptor address */
+               iop_chan_set_next_descriptor(iop_chan, sw_desc->async_tx.phys);
+
+               /* 1/ don't add pre-chained descriptors
+                * 2/ dummy read to flush next_desc write
+                */
+               BUG_ON(iop_desc_get_next_desc(sw_desc));
+
+               /* run the descriptor */
+               iop_chan_enable(iop_chan);
+       } else
+               dev_printk(KERN_ERR, iop_chan->device->common.dev,
+                       "failed to allocate null descriptor\n");
+       spin_unlock_bh(&iop_chan->lock);
+}
+
+static struct platform_driver iop_adma_driver = {
+       .probe          = iop_adma_probe,
+       .remove         = iop_adma_remove,
+       .driver         = {
+               .owner  = THIS_MODULE,
+               .name   = "iop-adma",
+       },
+};
+
+static int __init iop_adma_init (void)
+{
+       /* it's currently unsafe to unload this module */
+       /* if forced, worst case is that rmmod hangs */
+       __unsafe(THIS_MODULE);
+
+       return platform_driver_register(&iop_adma_driver);
+}
+
+static void __exit iop_adma_exit (void)
+{
+       platform_driver_unregister(&iop_adma_driver);
+       return;
+}
+
+module_init(iop_adma_init);
+module_exit(iop_adma_exit);
+
+MODULE_AUTHOR("Intel Corporation");
+MODULE_DESCRIPTION("IOP ADMA Engine Driver");
+MODULE_LICENSE("GPL");
diff --git a/include/asm-arm/hardware/iop_adma.h b/include/asm-arm/hardware/iop_adma.h
new file mode 100644 (file)
index 0000000..ca8e71f
--- /dev/null
@@ -0,0 +1,118 @@
+/*
+ * Copyright © 2006, Intel Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ */
+#ifndef IOP_ADMA_H
+#define IOP_ADMA_H
+#include <linux/types.h>
+#include <linux/dmaengine.h>
+#include <linux/interrupt.h>
+
+#define IOP_ADMA_SLOT_SIZE 32
+#define IOP_ADMA_THRESHOLD 4
+
+/**
+ * struct iop_adma_device - internal representation of an ADMA device
+ * @pdev: Platform device
+ * @id: HW ADMA Device selector
+ * @dma_desc_pool: base of DMA descriptor region (DMA address)
+ * @dma_desc_pool_virt: base of DMA descriptor region (CPU address)
+ * @common: embedded struct dma_device
+ */
+struct iop_adma_device {
+       struct platform_device *pdev;
+       int id;
+       dma_addr_t dma_desc_pool;
+       void *dma_desc_pool_virt;
+       struct dma_device common;
+};
+
+/**
+ * struct iop_adma_chan - internal representation of an ADMA device
+ * @pending: allows batching of hardware operations
+ * @completed_cookie: identifier for the most recently completed operation
+ * @lock: serializes enqueue/dequeue operations to the slot pool
+ * @mmr_base: memory mapped register base
+ * @chain: device chain view of the descriptors
+ * @device: parent device
+ * @common: common dmaengine channel object members
+ * @last_used: place holder for allocation to continue from where it left off
+ * @all_slots: complete domain of slots usable by the channel
+ * @cleanup_watchdog: workaround missed interrupts on iop3xx
+ * @slots_allocated: records the actual size of the descriptor slot pool
+ * @irq_tasklet: bottom half where iop_adma_slot_cleanup runs
+ */
+struct iop_adma_chan {
+       int pending;
+       dma_cookie_t completed_cookie;
+       spinlock_t lock; /* protects the descriptor slot pool */
+       void __iomem *mmr_base;
+       struct list_head chain;
+       struct iop_adma_device *device;
+       struct dma_chan common;
+       struct iop_adma_desc_slot *last_used;
+       struct list_head all_slots;
+       struct timer_list cleanup_watchdog;
+       int slots_allocated;
+       struct tasklet_struct irq_tasklet;
+};
+
+/**
+ * struct iop_adma_desc_slot - IOP-ADMA software descriptor
+ * @slot_node: node on the iop_adma_chan.all_slots list
+ * @chain_node: node on the op_adma_chan.chain list
+ * @hw_desc: virtual address of the hardware descriptor chain
+ * @phys: hardware address of the hardware descriptor chain
+ * @group_head: first operation in a transaction
+ * @slot_cnt: total slots used in an transaction (group of operations)
+ * @slots_per_op: number of slots per operation
+ * @idx: pool index
+ * @unmap_src_cnt: number of xor sources
+ * @unmap_len: transaction bytecount
+ * @async_tx: support for the async_tx api
+ * @group_list: list of slots that make up a multi-descriptor transaction
+ *     for example transfer lengths larger than the supported hw max
+ * @xor_check_result: result of zero sum
+ * @crc32_result: result crc calculation
+ */
+struct iop_adma_desc_slot {
+       struct list_head slot_node;
+       struct list_head chain_node;
+       void *hw_desc;
+       struct iop_adma_desc_slot *group_head;
+       u16 slot_cnt;
+       u16 slots_per_op;
+       u16 idx;
+       u16 unmap_src_cnt;
+       size_t unmap_len;
+       struct dma_async_tx_descriptor async_tx;
+       union {
+               u32 *xor_check_result;
+               u32 *crc32_result;
+       };
+};
+
+struct iop_adma_platform_data {
+       int hw_id;
+       dma_cap_mask_t cap_mask;
+       size_t pool_size;
+};
+
+#define to_iop_sw_desc(addr_hw_desc) \
+       container_of(addr_hw_desc, struct iop_adma_desc_slot, hw_desc)
+#define iop_hw_desc_slot_idx(hw_desc, idx) \
+       ( (void *) (((unsigned long) hw_desc) + ((idx) << 5)) )
+#endif