]> git.karo-electronics.de Git - linux-beck.git/commitdiff
dmaengine: Driver support for FSL RaidEngine device.
authorXuelin Shi <xuelin.shi@freescale.com>
Tue, 3 Mar 2015 06:26:22 +0000 (14:26 +0800)
committerVinod Koul <vinod.koul@intel.com>
Thu, 2 Apr 2015 10:40:27 +0000 (16:10 +0530)
The RaidEngine is a new FSL hardware used for Raid5/6 acceration.
This patch enables the RaidEngine functionality and provides
hardware offloading capability for memcpy, xor and pq computation.
It works with async_tx.

Signed-off-by: Harninder Rai <harninder.rai@freescale.com>
Signed-off-by: Xuelin Shi <xuelin.shi@freescale.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
drivers/dma/Kconfig
drivers/dma/Makefile
drivers/dma/fsl_raid.c [new file with mode: 0644]
drivers/dma/fsl_raid.h [new file with mode: 0644]

index 4be766f43aa908ebd3ccbe7eaae09fb7b3214ff2..b674683de24bac34d867c67f0c2f35dc9ed3cef3 100644 (file)
@@ -125,6 +125,17 @@ config FSL_DMA
          EloPlus is on mpc85xx and mpc86xx and Pxxx parts, and the Elo3 is on
          some Txxx and Bxxx parts.
 
+config FSL_RAID
+        tristate "Freescale RAID engine Support"
+        depends on FSL_SOC && !ASYNC_TX_ENABLE_CHANNEL_SWITCH
+        select DMA_ENGINE
+        select DMA_ENGINE_RAID
+        ---help---
+          Enable support for Freescale RAID Engine. RAID Engine is
+          available on some QorIQ SoCs (like P5020/P5040). It has
+          the capability to offload memcpy, xor and pq computation
+         for raid5/6.
+
 config MPC512X_DMA
        tristate "Freescale MPC512x built-in DMA engine support"
        depends on PPC_MPC512x || PPC_MPC831x
index 1dab9ef196b0ff14567b122bf293a6aac50561c1..345ec4758b9d9558dd9926b5d844582a7ea7d6f1 100644 (file)
@@ -45,6 +45,7 @@ obj-$(CONFIG_DMA_JZ4780) += dma-jz4780.o
 obj-$(CONFIG_TI_CPPI41) += cppi41.o
 obj-$(CONFIG_K3_DMA) += k3dma.o
 obj-$(CONFIG_MOXART_DMA) += moxart-dma.o
+obj-$(CONFIG_FSL_RAID) += fsl_raid.o
 obj-$(CONFIG_FSL_EDMA) += fsl-edma.o
 obj-$(CONFIG_QCOM_BAM_DMA) += qcom_bam_dma.o
 obj-y += xilinx/
diff --git a/drivers/dma/fsl_raid.c b/drivers/dma/fsl_raid.c
new file mode 100644 (file)
index 0000000..12778bd
--- /dev/null
@@ -0,0 +1,904 @@
+/*
+ * drivers/dma/fsl_raid.c
+ *
+ * Freescale RAID Engine device driver
+ *
+ * Author:
+ *     Harninder Rai <harninder.rai@freescale.com>
+ *     Naveen Burmi <naveenburmi@freescale.com>
+ *
+ * Rewrite:
+ *     Xuelin Shi <xuelin.shi@freescale.com>
+ *
+ * Copyright (c) 2010-2014 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * Theory of operation:
+ *
+ * General capabilities:
+ *     RAID Engine (RE) block is capable of offloading XOR, memcpy and P/Q
+ *     calculations required in RAID5 and RAID6 operations. RE driver
+ *     registers with Linux's ASYNC layer as dma driver. RE hardware
+ *     maintains strict ordering of the requests through chained
+ *     command queueing.
+ *
+ * Data flow:
+ *     Software RAID layer of Linux (MD layer) maintains RAID partitions,
+ *     strips, stripes etc. It sends requests to the underlying ASYNC layer
+ *     which further passes it to RE driver. ASYNC layer decides which request
+ *     goes to which job ring of RE hardware. For every request processed by
+ *     RAID Engine, driver gets an interrupt unless coalescing is set. The
+ *     per job ring interrupt handler checks the status register for errors,
+ *     clears the interrupt and leave the post interrupt processing to the irq
+ *     thread.
+ */
+#include <linux/interrupt.h>
+#include <linux/module.h>
+#include <linux/of_irq.h>
+#include <linux/of_address.h>
+#include <linux/of_platform.h>
+#include <linux/dma-mapping.h>
+#include <linux/dmapool.h>
+#include <linux/dmaengine.h>
+#include <linux/io.h>
+#include <linux/spinlock.h>
+#include <linux/slab.h>
+
+#include "dmaengine.h"
+#include "fsl_raid.h"
+
+#define FSL_RE_MAX_XOR_SRCS    16
+#define FSL_RE_MAX_PQ_SRCS     16
+#define FSL_RE_MIN_DESCS       256
+#define FSL_RE_MAX_DESCS       (4 * FSL_RE_MIN_DESCS)
+#define FSL_RE_FRAME_FORMAT    0x1
+#define FSL_RE_MAX_DATA_LEN    (1024*1024)
+
+#define to_fsl_re_dma_desc(tx) container_of(tx, struct fsl_re_desc, async_tx)
+
+/* Add descriptors into per chan software queue - submit_q */
+static dma_cookie_t fsl_re_tx_submit(struct dma_async_tx_descriptor *tx)
+{
+       struct fsl_re_desc *desc;
+       struct fsl_re_chan *re_chan;
+       dma_cookie_t cookie;
+       unsigned long flags;
+
+       desc = to_fsl_re_dma_desc(tx);
+       re_chan = container_of(tx->chan, struct fsl_re_chan, chan);
+
+       spin_lock_irqsave(&re_chan->desc_lock, flags);
+       cookie = dma_cookie_assign(tx);
+       list_add_tail(&desc->node, &re_chan->submit_q);
+       spin_unlock_irqrestore(&re_chan->desc_lock, flags);
+
+       return cookie;
+}
+
+/* Copy descriptor from per chan software queue into hardware job ring */
+static void fsl_re_issue_pending(struct dma_chan *chan)
+{
+       struct fsl_re_chan *re_chan;
+       int avail;
+       struct fsl_re_desc *desc, *_desc;
+       unsigned long flags;
+
+       re_chan = container_of(chan, struct fsl_re_chan, chan);
+
+       spin_lock_irqsave(&re_chan->desc_lock, flags);
+       avail = FSL_RE_SLOT_AVAIL(
+               in_be32(&re_chan->jrregs->inbring_slot_avail));
+
+       list_for_each_entry_safe(desc, _desc, &re_chan->submit_q, node) {
+               if (!avail)
+                       break;
+
+               list_move_tail(&desc->node, &re_chan->active_q);
+
+               memcpy(&re_chan->inb_ring_virt_addr[re_chan->inb_count],
+                      &desc->hwdesc, sizeof(struct fsl_re_hw_desc));
+
+               re_chan->inb_count = (re_chan->inb_count + 1) &
+                                               FSL_RE_RING_SIZE_MASK;
+               out_be32(&re_chan->jrregs->inbring_add_job, FSL_RE_ADD_JOB(1));
+               avail--;
+       }
+       spin_unlock_irqrestore(&re_chan->desc_lock, flags);
+}
+
+static void fsl_re_desc_done(struct fsl_re_desc *desc)
+{
+       dma_async_tx_callback callback;
+       void *callback_param;
+
+       dma_cookie_complete(&desc->async_tx);
+
+       callback = desc->async_tx.callback;
+       callback_param = desc->async_tx.callback_param;
+       if (callback)
+               callback(callback_param);
+
+       dma_descriptor_unmap(&desc->async_tx);
+}
+
+static void fsl_re_cleanup_descs(struct fsl_re_chan *re_chan)
+{
+       struct fsl_re_desc *desc, *_desc;
+       unsigned long flags;
+
+       spin_lock_irqsave(&re_chan->desc_lock, flags);
+       list_for_each_entry_safe(desc, _desc, &re_chan->ack_q, node) {
+               if (async_tx_test_ack(&desc->async_tx))
+                       list_move_tail(&desc->node, &re_chan->free_q);
+       }
+       spin_unlock_irqrestore(&re_chan->desc_lock, flags);
+
+       fsl_re_issue_pending(&re_chan->chan);
+}
+
+static void fsl_re_dequeue(unsigned long data)
+{
+       struct fsl_re_chan *re_chan;
+       struct fsl_re_desc *desc, *_desc;
+       struct fsl_re_hw_desc *hwdesc;
+       unsigned long flags;
+       unsigned int count, oub_count;
+       int found;
+
+       re_chan = dev_get_drvdata((struct device *)data);
+
+       fsl_re_cleanup_descs(re_chan);
+
+       spin_lock_irqsave(&re_chan->desc_lock, flags);
+       count = FSL_RE_SLOT_FULL(in_be32(&re_chan->jrregs->oubring_slot_full));
+       while (count--) {
+               found = 0;
+               hwdesc = &re_chan->oub_ring_virt_addr[re_chan->oub_count];
+               list_for_each_entry_safe(desc, _desc, &re_chan->active_q,
+                                        node) {
+                       /* compare the hw dma addr to find the completed */
+                       if (desc->hwdesc.lbea32 == hwdesc->lbea32 &&
+                           desc->hwdesc.addr_low == hwdesc->addr_low) {
+                               found = 1;
+                               break;
+                       }
+               }
+
+               if (found) {
+                       fsl_re_desc_done(desc);
+                       list_move_tail(&desc->node, &re_chan->ack_q);
+               } else {
+                       dev_err(re_chan->dev,
+                               "found hwdesc not in sw queue, discard it\n");
+               }
+
+               oub_count = (re_chan->oub_count + 1) & FSL_RE_RING_SIZE_MASK;
+               re_chan->oub_count = oub_count;
+
+               out_be32(&re_chan->jrregs->oubring_job_rmvd,
+                        FSL_RE_RMVD_JOB(1));
+       }
+       spin_unlock_irqrestore(&re_chan->desc_lock, flags);
+}
+
+/* Per Job Ring interrupt handler */
+static irqreturn_t fsl_re_isr(int irq, void *data)
+{
+       struct fsl_re_chan *re_chan;
+       u32 irqstate, status;
+
+       re_chan = dev_get_drvdata((struct device *)data);
+
+       irqstate = in_be32(&re_chan->jrregs->jr_interrupt_status);
+       if (!irqstate)
+               return IRQ_NONE;
+
+       /*
+        * There's no way in upper layer (read MD layer) to recover from
+        * error conditions except restart everything. In long term we
+        * need to do something more than just crashing
+        */
+       if (irqstate & FSL_RE_ERROR) {
+               status = in_be32(&re_chan->jrregs->jr_status);
+               dev_err(re_chan->dev, "chan error irqstate: %x, status: %x\n",
+                       irqstate, status);
+       }
+
+       /* Clear interrupt */
+       out_be32(&re_chan->jrregs->jr_interrupt_status, FSL_RE_CLR_INTR);
+
+       tasklet_schedule(&re_chan->irqtask);
+
+       return IRQ_HANDLED;
+}
+
+static enum dma_status fsl_re_tx_status(struct dma_chan *chan,
+                                       dma_cookie_t cookie,
+                                       struct dma_tx_state *txstate)
+{
+       return dma_cookie_status(chan, cookie, txstate);
+}
+
+static void fill_cfd_frame(struct fsl_re_cmpnd_frame *cf, u8 index,
+                          size_t length, dma_addr_t addr, bool final)
+{
+       u32 efrl = length & FSL_RE_CF_LENGTH_MASK;
+
+       efrl |= final << FSL_RE_CF_FINAL_SHIFT;
+       cf[index].efrl32 = efrl;
+       cf[index].addr_high = upper_32_bits(addr);
+       cf[index].addr_low = lower_32_bits(addr);
+}
+
+static struct fsl_re_desc *fsl_re_init_desc(struct fsl_re_chan *re_chan,
+                                           struct fsl_re_desc *desc,
+                                           void *cf, dma_addr_t paddr)
+{
+       desc->re_chan = re_chan;
+       desc->async_tx.tx_submit = fsl_re_tx_submit;
+       dma_async_tx_descriptor_init(&desc->async_tx, &re_chan->chan);
+       INIT_LIST_HEAD(&desc->node);
+
+       desc->hwdesc.fmt32 = FSL_RE_FRAME_FORMAT << FSL_RE_HWDESC_FMT_SHIFT;
+       desc->hwdesc.lbea32 = upper_32_bits(paddr);
+       desc->hwdesc.addr_low = lower_32_bits(paddr);
+       desc->cf_addr = cf;
+       desc->cf_paddr = paddr;
+
+       desc->cdb_addr = (void *)(cf + FSL_RE_CF_DESC_SIZE);
+       desc->cdb_paddr = paddr + FSL_RE_CF_DESC_SIZE;
+
+       return desc;
+}
+
+static struct fsl_re_desc *fsl_re_chan_alloc_desc(struct fsl_re_chan *re_chan,
+                                                 unsigned long flags)
+{
+       struct fsl_re_desc *desc = NULL;
+       void *cf;
+       dma_addr_t paddr;
+       unsigned long lock_flag;
+
+       fsl_re_cleanup_descs(re_chan);
+
+       spin_lock_irqsave(&re_chan->desc_lock, lock_flag);
+       if (!list_empty(&re_chan->free_q)) {
+               /* take one desc from free_q */
+               desc = list_first_entry(&re_chan->free_q,
+                                       struct fsl_re_desc, node);
+               list_del(&desc->node);
+
+               desc->async_tx.flags = flags;
+       }
+       spin_unlock_irqrestore(&re_chan->desc_lock, lock_flag);
+
+       if (!desc) {
+               desc = kzalloc(sizeof(*desc), GFP_NOWAIT);
+               if (!desc)
+                       return NULL;
+
+               cf = dma_pool_alloc(re_chan->re_dev->cf_desc_pool, GFP_NOWAIT,
+                                   &paddr);
+               if (!cf) {
+                       kfree(desc);
+                       return NULL;
+               }
+
+               desc = fsl_re_init_desc(re_chan, desc, cf, paddr);
+               desc->async_tx.flags = flags;
+
+               spin_lock_irqsave(&re_chan->desc_lock, lock_flag);
+               re_chan->alloc_count++;
+               spin_unlock_irqrestore(&re_chan->desc_lock, lock_flag);
+       }
+
+       return desc;
+}
+
+static struct dma_async_tx_descriptor *fsl_re_prep_dma_genq(
+               struct dma_chan *chan, dma_addr_t dest, dma_addr_t *src,
+               unsigned int src_cnt, const unsigned char *scf, size_t len,
+               unsigned long flags)
+{
+       struct fsl_re_chan *re_chan;
+       struct fsl_re_desc *desc;
+       struct fsl_re_xor_cdb *xor;
+       struct fsl_re_cmpnd_frame *cf;
+       u32 cdb;
+       unsigned int i, j;
+       unsigned int save_src_cnt = src_cnt;
+       int cont_q = 0;
+
+       re_chan = container_of(chan, struct fsl_re_chan, chan);
+       if (len > FSL_RE_MAX_DATA_LEN) {
+               dev_err(re_chan->dev, "genq tx length %lu, max length %d\n",
+                       len, FSL_RE_MAX_DATA_LEN);
+               return NULL;
+       }
+
+       desc = fsl_re_chan_alloc_desc(re_chan, flags);
+       if (desc <= 0)
+               return NULL;
+
+       if (scf && (flags & DMA_PREP_CONTINUE)) {
+               cont_q = 1;
+               src_cnt += 1;
+       }
+
+       /* Filling xor CDB */
+       cdb = FSL_RE_XOR_OPCODE << FSL_RE_CDB_OPCODE_SHIFT;
+       cdb |= (src_cnt - 1) << FSL_RE_CDB_NRCS_SHIFT;
+       cdb |= FSL_RE_BLOCK_SIZE << FSL_RE_CDB_BLKSIZE_SHIFT;
+       cdb |= FSL_RE_INTR_ON_ERROR << FSL_RE_CDB_ERROR_SHIFT;
+       cdb |= FSL_RE_DATA_DEP << FSL_RE_CDB_DEPEND_SHIFT;
+       xor = desc->cdb_addr;
+       xor->cdb32 = cdb;
+
+       if (scf) {
+               /* compute q = src0*coef0^src1*coef1^..., * is GF(8) mult */
+               for (i = 0; i < save_src_cnt; i++)
+                       xor->gfm[i] = scf[i];
+               if (cont_q)
+                       xor->gfm[i++] = 1;
+       } else {
+               /* compute P, that is XOR all srcs */
+               for (i = 0; i < src_cnt; i++)
+                       xor->gfm[i] = 1;
+       }
+
+       /* Filling frame 0 of compound frame descriptor with CDB */
+       cf = desc->cf_addr;
+       fill_cfd_frame(cf, 0, sizeof(*xor), desc->cdb_paddr, 0);
+
+       /* Fill CFD's 1st frame with dest buffer */
+       fill_cfd_frame(cf, 1, len, dest, 0);
+
+       /* Fill CFD's rest of the frames with source buffers */
+       for (i = 2, j = 0; j < save_src_cnt; i++, j++)
+               fill_cfd_frame(cf, i, len, src[j], 0);
+
+       if (cont_q)
+               fill_cfd_frame(cf, i++, len, dest, 0);
+
+       /* Setting the final bit in the last source buffer frame in CFD */
+       cf[i - 1].efrl32 |= 1 << FSL_RE_CF_FINAL_SHIFT;
+
+       return &desc->async_tx;
+}
+
+/*
+ * Prep function for P parity calculation.In RAID Engine terminology,
+ * XOR calculation is called GenQ calculation done through GenQ command
+ */
+static struct dma_async_tx_descriptor *fsl_re_prep_dma_xor(
+               struct dma_chan *chan, dma_addr_t dest, dma_addr_t *src,
+               unsigned int src_cnt, size_t len, unsigned long flags)
+{
+       /* NULL let genq take all coef as 1 */
+       return fsl_re_prep_dma_genq(chan, dest, src, src_cnt, NULL, len, flags);
+}
+
+/*
+ * Prep function for P/Q parity calculation.In RAID Engine terminology,
+ * P/Q calculation is called GenQQ done through GenQQ command
+ */
+static struct dma_async_tx_descriptor *fsl_re_prep_dma_pq(
+               struct dma_chan *chan, dma_addr_t *dest, dma_addr_t *src,
+               unsigned int src_cnt, const unsigned char *scf, size_t len,
+               unsigned long flags)
+{
+       struct fsl_re_chan *re_chan;
+       struct fsl_re_desc *desc;
+       struct fsl_re_pq_cdb *pq;
+       struct fsl_re_cmpnd_frame *cf;
+       u32 cdb;
+       u8 *p;
+       int gfmq_len, i, j;
+       unsigned int save_src_cnt = src_cnt;
+
+       re_chan = container_of(chan, struct fsl_re_chan, chan);
+       if (len > FSL_RE_MAX_DATA_LEN) {
+               dev_err(re_chan->dev, "pq tx length is %lu, max length is %d\n",
+                       len, FSL_RE_MAX_DATA_LEN);
+               return NULL;
+       }
+
+       /*
+        * RE requires at least 2 sources, if given only one source, we pass the
+        * second source same as the first one.
+        * With only one source, generating P is meaningless, only generate Q.
+        */
+       if (src_cnt == 1) {
+               struct dma_async_tx_descriptor *tx;
+               dma_addr_t dma_src[2];
+               unsigned char coef[2];
+
+               dma_src[0] = *src;
+               coef[0] = *scf;
+               dma_src[1] = *src;
+               coef[1] = 0;
+               tx = fsl_re_prep_dma_genq(chan, dest[1], dma_src, 2, coef, len,
+                                         flags);
+               if (tx)
+                       desc = to_fsl_re_dma_desc(tx);
+
+               return tx;
+       }
+
+       /*
+        * During RAID6 array creation, Linux's MD layer gets P and Q
+        * calculated separately in two steps. But our RAID Engine has
+        * the capability to calculate both P and Q with a single command
+        * Hence to merge well with MD layer, we need to provide a hook
+        * here and call re_jq_prep_dma_genq() function
+        */
+
+       if (flags & DMA_PREP_PQ_DISABLE_P)
+               return fsl_re_prep_dma_genq(chan, dest[1], src, src_cnt,
+                               scf, len, flags);
+
+       if (flags & DMA_PREP_CONTINUE)
+               src_cnt += 3;
+
+       desc = fsl_re_chan_alloc_desc(re_chan, flags);
+       if (desc <= 0)
+               return NULL;
+
+       /* Filling GenQQ CDB */
+       cdb = FSL_RE_PQ_OPCODE << FSL_RE_CDB_OPCODE_SHIFT;
+       cdb |= (src_cnt - 1) << FSL_RE_CDB_NRCS_SHIFT;
+       cdb |= FSL_RE_BLOCK_SIZE << FSL_RE_CDB_BLKSIZE_SHIFT;
+       cdb |= FSL_RE_BUFFER_OUTPUT << FSL_RE_CDB_BUFFER_SHIFT;
+       cdb |= FSL_RE_DATA_DEP << FSL_RE_CDB_DEPEND_SHIFT;
+
+       pq = desc->cdb_addr;
+       pq->cdb32 = cdb;
+
+       p = pq->gfm_q1;
+       /* Init gfm_q1[] */
+       for (i = 0; i < src_cnt; i++)
+               p[i] = 1;
+
+       /* Align gfm[] to 32bit */
+       gfmq_len = ALIGN(src_cnt, 4);
+
+       /* Init gfm_q2[] */
+       p += gfmq_len;
+       for (i = 0; i < src_cnt; i++)
+               p[i] = scf[i];
+
+       /* Filling frame 0 of compound frame descriptor with CDB */
+       cf = desc->cf_addr;
+       fill_cfd_frame(cf, 0, sizeof(struct fsl_re_pq_cdb), desc->cdb_paddr, 0);
+
+       /* Fill CFD's 1st & 2nd frame with dest buffers */
+       for (i = 1, j = 0; i < 3; i++, j++)
+               fill_cfd_frame(cf, i, len, dest[j], 0);
+
+       /* Fill CFD's rest of the frames with source buffers */
+       for (i = 3, j = 0; j < save_src_cnt; i++, j++)
+               fill_cfd_frame(cf, i, len, src[j], 0);
+
+       /* PQ computation continuation */
+       if (flags & DMA_PREP_CONTINUE) {
+               if (src_cnt - save_src_cnt == 3) {
+                       p[save_src_cnt] = 0;
+                       p[save_src_cnt + 1] = 0;
+                       p[save_src_cnt + 2] = 1;
+                       fill_cfd_frame(cf, i++, len, dest[0], 0);
+                       fill_cfd_frame(cf, i++, len, dest[1], 0);
+                       fill_cfd_frame(cf, i++, len, dest[1], 0);
+               } else {
+                       dev_err(re_chan->dev, "PQ tx continuation error!\n");
+                       return NULL;
+               }
+       }
+
+       /* Setting the final bit in the last source buffer frame in CFD */
+       cf[i - 1].efrl32 |= 1 << FSL_RE_CF_FINAL_SHIFT;
+
+       return &desc->async_tx;
+}
+
+/*
+ * Prep function for memcpy. In RAID Engine, memcpy is done through MOVE
+ * command. Logic of this function will need to be modified once multipage
+ * support is added in Linux's MD/ASYNC Layer
+ */
+static struct dma_async_tx_descriptor *fsl_re_prep_dma_memcpy(
+               struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
+               size_t len, unsigned long flags)
+{
+       struct fsl_re_chan *re_chan;
+       struct fsl_re_desc *desc;
+       size_t length;
+       struct fsl_re_cmpnd_frame *cf;
+       struct fsl_re_move_cdb *move;
+       u32 cdb;
+
+       re_chan = container_of(chan, struct fsl_re_chan, chan);
+
+       if (len > FSL_RE_MAX_DATA_LEN) {
+               dev_err(re_chan->dev, "cp tx length is %lu, max length is %d\n",
+                       len, FSL_RE_MAX_DATA_LEN);
+               return NULL;
+       }
+
+       desc = fsl_re_chan_alloc_desc(re_chan, flags);
+       if (desc <= 0)
+               return NULL;
+
+       /* Filling move CDB */
+       cdb = FSL_RE_MOVE_OPCODE << FSL_RE_CDB_OPCODE_SHIFT;
+       cdb |= FSL_RE_BLOCK_SIZE << FSL_RE_CDB_BLKSIZE_SHIFT;
+       cdb |= FSL_RE_INTR_ON_ERROR << FSL_RE_CDB_ERROR_SHIFT;
+       cdb |= FSL_RE_DATA_DEP << FSL_RE_CDB_DEPEND_SHIFT;
+
+       move = desc->cdb_addr;
+       move->cdb32 = cdb;
+
+       /* Filling frame 0 of CFD with move CDB */
+       cf = desc->cf_addr;
+       fill_cfd_frame(cf, 0, sizeof(*move), desc->cdb_paddr, 0);
+
+       length = min_t(size_t, len, FSL_RE_MAX_DATA_LEN);
+
+       /* Fill CFD's 1st frame with dest buffer */
+       fill_cfd_frame(cf, 1, length, dest, 0);
+
+       /* Fill CFD's 2nd frame with src buffer */
+       fill_cfd_frame(cf, 2, length, src, 1);
+
+       return &desc->async_tx;
+}
+
+static int fsl_re_alloc_chan_resources(struct dma_chan *chan)
+{
+       struct fsl_re_chan *re_chan;
+       struct fsl_re_desc *desc;
+       void *cf;
+       dma_addr_t paddr;
+       int i;
+
+       re_chan = container_of(chan, struct fsl_re_chan, chan);
+       for (i = 0; i < FSL_RE_MIN_DESCS; i++) {
+               desc = kzalloc(sizeof(*desc), GFP_KERNEL);
+               if (!desc)
+                       break;
+
+               cf = dma_pool_alloc(re_chan->re_dev->cf_desc_pool, GFP_KERNEL,
+                                   &paddr);
+               if (!cf) {
+                       kfree(desc);
+                       break;
+               }
+
+               INIT_LIST_HEAD(&desc->node);
+               fsl_re_init_desc(re_chan, desc, cf, paddr);
+
+               list_add_tail(&desc->node, &re_chan->free_q);
+               re_chan->alloc_count++;
+       }
+       return re_chan->alloc_count;
+}
+
+static void fsl_re_free_chan_resources(struct dma_chan *chan)
+{
+       struct fsl_re_chan *re_chan;
+       struct fsl_re_desc *desc;
+
+       re_chan = container_of(chan, struct fsl_re_chan, chan);
+       while (re_chan->alloc_count--) {
+               desc = list_first_entry(&re_chan->free_q,
+                                       struct fsl_re_desc,
+                                       node);
+
+               list_del(&desc->node);
+               dma_pool_free(re_chan->re_dev->cf_desc_pool, desc->cf_addr,
+                             desc->cf_paddr);
+               kfree(desc);
+       }
+
+       if (!list_empty(&re_chan->free_q))
+               dev_err(re_chan->dev, "chan resource cannot be cleaned!\n");
+}
+
+int fsl_re_chan_probe(struct platform_device *ofdev,
+                     struct device_node *np, u8 q, u32 off)
+{
+       struct device *dev, *chandev;
+       struct fsl_re_drv_private *re_priv;
+       struct fsl_re_chan *chan;
+       struct dma_device *dma_dev;
+       u32 ptr;
+       u32 status;
+       int ret = 0, rc;
+       struct platform_device *chan_ofdev;
+
+       dev = &ofdev->dev;
+       re_priv = dev_get_drvdata(dev);
+       dma_dev = &re_priv->dma_dev;
+
+       chan = devm_kzalloc(dev, sizeof(*chan), GFP_KERNEL);
+       if (!chan)
+               return -ENOMEM;
+
+       /* create platform device for chan node */
+       chan_ofdev = of_platform_device_create(np, NULL, dev);
+       if (!chan_ofdev) {
+               dev_err(dev, "Not able to create ofdev for jr %d\n", q);
+               ret = -EINVAL;
+               goto err_free;
+       }
+
+       /* read reg property from dts */
+       rc = of_property_read_u32(np, "reg", &ptr);
+       if (rc) {
+               dev_err(dev, "Reg property not found in jr %d\n", q);
+               ret = -ENODEV;
+               goto err_free;
+       }
+
+       chan->jrregs = (struct fsl_re_chan_cfg *)((u8 *)re_priv->re_regs +
+                       off + ptr);
+
+       /* read irq property from dts */
+       chan->irq = irq_of_parse_and_map(np, 0);
+       if (chan->irq == NO_IRQ) {
+               dev_err(dev, "No IRQ defined for JR %d\n", q);
+               ret = -ENODEV;
+               goto err_free;
+       }
+
+       snprintf(chan->name, sizeof(chan->name), "re_jr%02d", q);
+
+       chandev = &chan_ofdev->dev;
+       tasklet_init(&chan->irqtask, fsl_re_dequeue, (unsigned long)chandev);
+
+       ret = request_irq(chan->irq, fsl_re_isr, 0, chan->name, chandev);
+       if (ret) {
+               dev_err(dev, "Unable to register interrupt for JR %d\n", q);
+               ret = -EINVAL;
+               goto err_free;
+       }
+
+       re_priv->re_jrs[q] = chan;
+       chan->chan.device = dma_dev;
+       chan->chan.private = chan;
+       chan->dev = chandev;
+       chan->re_dev = re_priv;
+
+       spin_lock_init(&chan->desc_lock);
+       INIT_LIST_HEAD(&chan->ack_q);
+       INIT_LIST_HEAD(&chan->active_q);
+       INIT_LIST_HEAD(&chan->submit_q);
+       INIT_LIST_HEAD(&chan->free_q);
+
+       chan->inb_ring_virt_addr = dma_pool_alloc(chan->re_dev->hw_desc_pool,
+               GFP_KERNEL, &chan->inb_phys_addr);
+       if (!chan->inb_ring_virt_addr) {
+               dev_err(dev, "No dma memory for inb_ring_virt_addr\n");
+               ret = -ENOMEM;
+               goto err_free;
+       }
+
+       chan->oub_ring_virt_addr = dma_pool_alloc(chan->re_dev->hw_desc_pool,
+               GFP_KERNEL, &chan->oub_phys_addr);
+       if (!chan->oub_ring_virt_addr) {
+               dev_err(dev, "No dma memory for oub_ring_virt_addr\n");
+               ret = -ENOMEM;
+               goto err_free_1;
+       }
+
+       /* Program the Inbound/Outbound ring base addresses and size */
+       out_be32(&chan->jrregs->inbring_base_h,
+                chan->inb_phys_addr & FSL_RE_ADDR_BIT_MASK);
+       out_be32(&chan->jrregs->oubring_base_h,
+                chan->oub_phys_addr & FSL_RE_ADDR_BIT_MASK);
+       out_be32(&chan->jrregs->inbring_base_l,
+                chan->inb_phys_addr >> FSL_RE_ADDR_BIT_SHIFT);
+       out_be32(&chan->jrregs->oubring_base_l,
+                chan->oub_phys_addr >> FSL_RE_ADDR_BIT_SHIFT);
+       out_be32(&chan->jrregs->inbring_size,
+                FSL_RE_RING_SIZE << FSL_RE_RING_SIZE_SHIFT);
+       out_be32(&chan->jrregs->oubring_size,
+                FSL_RE_RING_SIZE << FSL_RE_RING_SIZE_SHIFT);
+
+       /* Read LIODN value from u-boot */
+       status = in_be32(&chan->jrregs->jr_config_1) & FSL_RE_REG_LIODN_MASK;
+
+       /* Program the CFG reg */
+       out_be32(&chan->jrregs->jr_config_1,
+                FSL_RE_CFG1_CBSI | FSL_RE_CFG1_CBS0 | status);
+
+       dev_set_drvdata(chandev, chan);
+
+       /* Enable RE/CHAN */
+       out_be32(&chan->jrregs->jr_command, FSL_RE_ENABLE);
+
+       return 0;
+
+err_free_1:
+       dma_pool_free(chan->re_dev->hw_desc_pool, chan->inb_ring_virt_addr,
+                     chan->inb_phys_addr);
+err_free:
+       return ret;
+}
+
+/* Probe function for RAID Engine */
+static int fsl_re_probe(struct platform_device *ofdev)
+{
+       struct fsl_re_drv_private *re_priv;
+       struct device_node *np;
+       struct device_node *child;
+       u32 off;
+       u8 ridx = 0;
+       struct dma_device *dma_dev;
+       struct resource *res;
+       int rc;
+       struct device *dev = &ofdev->dev;
+
+       re_priv = devm_kzalloc(dev, sizeof(*re_priv), GFP_KERNEL);
+       if (!re_priv)
+               return -ENOMEM;
+
+       res = platform_get_resource(ofdev, IORESOURCE_MEM, 0);
+       if (!res)
+               return -ENODEV;
+
+       /* IOMAP the entire RAID Engine region */
+       re_priv->re_regs = devm_ioremap(dev, res->start, resource_size(res));
+       if (!re_priv->re_regs)
+               return -EBUSY;
+
+       /* Program the RE mode */
+       out_be32(&re_priv->re_regs->global_config, FSL_RE_NON_DPAA_MODE);
+
+       /* Program Galois Field polynomial */
+       out_be32(&re_priv->re_regs->galois_field_config, FSL_RE_GFM_POLY);
+
+       dev_info(dev, "version %x, mode %x, gfp %x\n",
+                in_be32(&re_priv->re_regs->re_version_id),
+                in_be32(&re_priv->re_regs->global_config),
+                in_be32(&re_priv->re_regs->galois_field_config));
+
+       dma_dev = &re_priv->dma_dev;
+       dma_dev->dev = dev;
+       INIT_LIST_HEAD(&dma_dev->channels);
+       dma_set_mask(dev, DMA_BIT_MASK(40));
+
+       dma_dev->device_alloc_chan_resources = fsl_re_alloc_chan_resources;
+       dma_dev->device_tx_status = fsl_re_tx_status;
+       dma_dev->device_issue_pending = fsl_re_issue_pending;
+
+       dma_dev->max_xor = FSL_RE_MAX_XOR_SRCS;
+       dma_dev->device_prep_dma_xor = fsl_re_prep_dma_xor;
+       dma_cap_set(DMA_XOR, dma_dev->cap_mask);
+
+       dma_dev->max_pq = FSL_RE_MAX_PQ_SRCS;
+       dma_dev->device_prep_dma_pq = fsl_re_prep_dma_pq;
+       dma_cap_set(DMA_PQ, dma_dev->cap_mask);
+
+       dma_dev->device_prep_dma_memcpy = fsl_re_prep_dma_memcpy;
+       dma_cap_set(DMA_MEMCPY, dma_dev->cap_mask);
+
+       dma_dev->device_free_chan_resources = fsl_re_free_chan_resources;
+
+       re_priv->total_chans = 0;
+
+       re_priv->cf_desc_pool = dmam_pool_create("fsl_re_cf_desc_pool", dev,
+                                       FSL_RE_CF_CDB_SIZE,
+                                       FSL_RE_CF_CDB_ALIGN, 0);
+
+       if (!re_priv->cf_desc_pool) {
+               dev_err(dev, "No memory for fsl re_cf desc pool\n");
+               return -ENOMEM;
+       }
+
+       re_priv->hw_desc_pool = dmam_pool_create("fsl_re_hw_desc_pool", dev,
+                       sizeof(struct fsl_re_hw_desc) * FSL_RE_RING_SIZE,
+                       FSL_RE_FRAME_ALIGN, 0);
+       if (!re_priv->hw_desc_pool) {
+               dev_err(dev, "No memory for fsl re_hw desc pool\n");
+               return -ENOMEM;
+       }
+
+       dev_set_drvdata(dev, re_priv);
+
+       /* Parse Device tree to find out the total number of JQs present */
+       for_each_compatible_node(np, NULL, "fsl,raideng-v1.0-job-queue") {
+               rc = of_property_read_u32(np, "reg", &off);
+               if (rc) {
+                       dev_err(dev, "Reg property not found in JQ node\n");
+                       return -ENODEV;
+               }
+               /* Find out the Job Rings present under each JQ */
+               for_each_child_of_node(np, child) {
+                       rc = of_device_is_compatible(child,
+                                            "fsl,raideng-v1.0-job-ring");
+                       if (rc) {
+                               fsl_re_chan_probe(ofdev, child, ridx++, off);
+                               re_priv->total_chans++;
+                       }
+               }
+       }
+
+       dma_async_device_register(dma_dev);
+
+       return 0;
+}
+
+static void fsl_re_remove_chan(struct fsl_re_chan *chan)
+{
+       dma_pool_free(chan->re_dev->hw_desc_pool, chan->inb_ring_virt_addr,
+                     chan->inb_phys_addr);
+
+       dma_pool_free(chan->re_dev->hw_desc_pool, chan->oub_ring_virt_addr,
+                     chan->oub_phys_addr);
+}
+
+static int fsl_re_remove(struct platform_device *ofdev)
+{
+       struct fsl_re_drv_private *re_priv;
+       struct device *dev;
+       int i;
+
+       dev = &ofdev->dev;
+       re_priv = dev_get_drvdata(dev);
+
+       /* Cleanup chan related memory areas */
+       for (i = 0; i < re_priv->total_chans; i++)
+               fsl_re_remove_chan(re_priv->re_jrs[i]);
+
+       /* Unregister the driver */
+       dma_async_device_unregister(&re_priv->dma_dev);
+
+       return 0;
+}
+
+static struct of_device_id fsl_re_ids[] = {
+       { .compatible = "fsl,raideng-v1.0", },
+       {}
+};
+
+static struct platform_driver fsl_re_driver = {
+       .driver = {
+               .name = "fsl-raideng",
+               .owner = THIS_MODULE,
+               .of_match_table = fsl_re_ids,
+       },
+       .probe = fsl_re_probe,
+       .remove = fsl_re_remove,
+};
+
+module_platform_driver(fsl_re_driver);
+
+MODULE_AUTHOR("Harninder Rai <harninder.rai@freescale.com>");
+MODULE_LICENSE("GPL v2");
+MODULE_DESCRIPTION("Freescale RAID Engine Device Driver");
diff --git a/drivers/dma/fsl_raid.h b/drivers/dma/fsl_raid.h
new file mode 100644 (file)
index 0000000..69d743c
--- /dev/null
@@ -0,0 +1,306 @@
+/*
+ * drivers/dma/fsl_raid.h
+ *
+ * Freescale RAID Engine device driver
+ *
+ * Author:
+ *     Harninder Rai <harninder.rai@freescale.com>
+ *     Naveen Burmi <naveenburmi@freescale.com>
+ *
+ * Rewrite:
+ *     Xuelin Shi <xuelin.shi@freescale.com>
+
+ * Copyright (c) 2010-2012 Freescale Semiconductor, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of Freescale Semiconductor nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") as published by the Free Software
+ * Foundation, either version 2 of that License or (at your option) any
+ * later version.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ */
+
+#define FSL_RE_MAX_CHANS               4
+#define FSL_RE_DPAA_MODE               BIT(30)
+#define FSL_RE_NON_DPAA_MODE           BIT(31)
+#define FSL_RE_GFM_POLY                        0x1d000000
+#define FSL_RE_ADD_JOB(x)              ((x) << 16)
+#define FSL_RE_RMVD_JOB(x)             ((x) << 16)
+#define FSL_RE_CFG1_CBSI               0x08000000
+#define FSL_RE_CFG1_CBS0               0x00080000
+#define FSL_RE_SLOT_FULL_SHIFT         8
+#define FSL_RE_SLOT_FULL(x)            ((x) >> FSL_RE_SLOT_FULL_SHIFT)
+#define FSL_RE_SLOT_AVAIL_SHIFT                8
+#define FSL_RE_SLOT_AVAIL(x)           ((x) >> FSL_RE_SLOT_AVAIL_SHIFT)
+#define FSL_RE_PQ_OPCODE               0x1B
+#define FSL_RE_XOR_OPCODE              0x1A
+#define FSL_RE_MOVE_OPCODE             0x8
+#define FSL_RE_FRAME_ALIGN             16
+#define FSL_RE_BLOCK_SIZE              0x3 /* 4096 bytes */
+#define FSL_RE_CACHEABLE_IO            0x0
+#define FSL_RE_BUFFER_OUTPUT           0x0
+#define FSL_RE_INTR_ON_ERROR           0x1
+#define FSL_RE_DATA_DEP                        0x1
+#define FSL_RE_ENABLE_DPI              0x0
+#define FSL_RE_RING_SIZE               0x400
+#define FSL_RE_RING_SIZE_MASK          (FSL_RE_RING_SIZE - 1)
+#define FSL_RE_RING_SIZE_SHIFT         8
+#define FSL_RE_ADDR_BIT_SHIFT          4
+#define FSL_RE_ADDR_BIT_MASK           (BIT(FSL_RE_ADDR_BIT_SHIFT) - 1)
+#define FSL_RE_ERROR                   0x40000000
+#define FSL_RE_INTR                    0x80000000
+#define FSL_RE_CLR_INTR                        0x80000000
+#define FSL_RE_PAUSE                   0x80000000
+#define FSL_RE_ENABLE                  0x80000000
+#define FSL_RE_REG_LIODN_MASK          0x00000FFF
+
+#define FSL_RE_CDB_OPCODE_MASK         0xF8000000
+#define FSL_RE_CDB_OPCODE_SHIFT                27
+#define FSL_RE_CDB_EXCLEN_MASK         0x03000000
+#define FSL_RE_CDB_EXCLEN_SHIFT                24
+#define FSL_RE_CDB_EXCLQ1_MASK         0x00F00000
+#define FSL_RE_CDB_EXCLQ1_SHIFT                20
+#define FSL_RE_CDB_EXCLQ2_MASK         0x000F0000
+#define FSL_RE_CDB_EXCLQ2_SHIFT                16
+#define FSL_RE_CDB_BLKSIZE_MASK                0x0000C000
+#define FSL_RE_CDB_BLKSIZE_SHIFT       14
+#define FSL_RE_CDB_CACHE_MASK          0x00003000
+#define FSL_RE_CDB_CACHE_SHIFT         12
+#define FSL_RE_CDB_BUFFER_MASK         0x00000800
+#define FSL_RE_CDB_BUFFER_SHIFT                11
+#define FSL_RE_CDB_ERROR_MASK          0x00000400
+#define FSL_RE_CDB_ERROR_SHIFT         10
+#define FSL_RE_CDB_NRCS_MASK           0x0000003C
+#define FSL_RE_CDB_NRCS_SHIFT          6
+#define FSL_RE_CDB_DEPEND_MASK         0x00000008
+#define FSL_RE_CDB_DEPEND_SHIFT                3
+#define FSL_RE_CDB_DPI_MASK            0x00000004
+#define FSL_RE_CDB_DPI_SHIFT           2
+
+/*
+ * the largest cf block is 19*sizeof(struct cmpnd_frame), which is 304 bytes.
+ * here 19 = 1(cdb)+2(dest)+16(src), align to 64bytes, that is 320 bytes.
+ * the largest cdb block: struct pq_cdb which is 180 bytes, adding to cf block
+ * 320+180=500, align to 64bytes, that is 512 bytes.
+ */
+#define FSL_RE_CF_DESC_SIZE            320
+#define FSL_RE_CF_CDB_SIZE             512
+#define FSL_RE_CF_CDB_ALIGN            64
+
+struct fsl_re_ctrl {
+       /* General Configuration Registers */
+       __be32 global_config;   /* Global Configuration Register */
+       u8     rsvd1[4];
+       __be32 galois_field_config; /* Galois Field Configuration Register */
+       u8     rsvd2[4];
+       __be32 jq_wrr_config;   /* WRR Configuration register */
+       u8     rsvd3[4];
+       __be32 crc_config;      /* CRC Configuration register */
+       u8     rsvd4[228];
+       __be32 system_reset;    /* System Reset Register */
+       u8     rsvd5[252];
+       __be32 global_status;   /* Global Status Register */
+       u8     rsvd6[832];
+       __be32 re_liodn_base;   /* LIODN Base Register */
+       u8     rsvd7[1712];
+       __be32 re_version_id;   /* Version ID register of RE */
+       __be32 re_version_id_2; /* Version ID 2 register of RE */
+       u8     rsvd8[512];
+       __be32 host_config;     /* Host I/F Configuration Register */
+};
+
+struct fsl_re_chan_cfg {
+       /* Registers for JR interface */
+       __be32 jr_config_0;     /* Job Queue Configuration 0 Register */
+       __be32 jr_config_1;     /* Job Queue Configuration 1 Register */
+       __be32 jr_interrupt_status; /* Job Queue Interrupt Status Register */
+       u8     rsvd1[4];
+       __be32 jr_command;      /* Job Queue Command Register */
+       u8     rsvd2[4];
+       __be32 jr_status;       /* Job Queue Status Register */
+       u8     rsvd3[228];
+
+       /* Input Ring */
+       __be32 inbring_base_h;  /* Inbound Ring Base Address Register - High */
+       __be32 inbring_base_l;  /* Inbound Ring Base Address Register - Low */
+       __be32 inbring_size;    /* Inbound Ring Size Register */
+       u8     rsvd4[4];
+       __be32 inbring_slot_avail; /* Inbound Ring Slot Available Register */
+       u8     rsvd5[4];
+       __be32 inbring_add_job; /* Inbound Ring Add Job Register */
+       u8     rsvd6[4];
+       __be32 inbring_cnsmr_indx; /* Inbound Ring Consumer Index Register */
+       u8     rsvd7[220];
+
+       /* Output Ring */
+       __be32 oubring_base_h;  /* Outbound Ring Base Address Register - High */
+       __be32 oubring_base_l;  /* Outbound Ring Base Address Register - Low */
+       __be32 oubring_size;    /* Outbound Ring Size Register */
+       u8     rsvd8[4];
+       __be32 oubring_job_rmvd; /* Outbound Ring Job Removed Register */
+       u8     rsvd9[4];
+       __be32 oubring_slot_full; /* Outbound Ring Slot Full Register */
+       u8     rsvd10[4];
+       __be32 oubring_prdcr_indx; /* Outbound Ring Producer Index */
+};
+
+/*
+ * Command Descriptor Block (CDB) for unicast move command.
+ * In RAID Engine terms, memcpy is done through move command
+ */
+struct fsl_re_move_cdb {
+       __be32 cdb32;
+};
+
+/* Data protection/integrity related fields */
+#define FSL_RE_DPI_APPS_MASK           0xC0000000
+#define FSL_RE_DPI_APPS_SHIFT          30
+#define FSL_RE_DPI_REF_MASK            0x30000000
+#define FSL_RE_DPI_REF_SHIFT           28
+#define FSL_RE_DPI_GUARD_MASK          0x0C000000
+#define FSL_RE_DPI_GUARD_SHIFT         26
+#define FSL_RE_DPI_ATTR_MASK           0x03000000
+#define FSL_RE_DPI_ATTR_SHIFT          24
+#define FSL_RE_DPI_META_MASK           0x0000FFFF
+
+struct fsl_re_dpi {
+       __be32 dpi32;
+       __be32 ref;
+};
+
+/*
+ * CDB for GenQ command. In RAID Engine terminology, XOR is
+ * done through this command
+ */
+struct fsl_re_xor_cdb {
+       __be32 cdb32;
+       u8 gfm[16];
+       struct fsl_re_dpi dpi_dest_spec;
+       struct fsl_re_dpi dpi_src_spec[16];
+};
+
+/* CDB for no-op command */
+struct fsl_re_noop_cdb {
+       __be32 cdb32;
+};
+
+/*
+ * CDB for GenQQ command. In RAID Engine terminology, P/Q is
+ * done through this command
+ */
+struct fsl_re_pq_cdb {
+       __be32 cdb32;
+       u8 gfm_q1[16];
+       u8 gfm_q2[16];
+       struct fsl_re_dpi dpi_dest_spec[2];
+       struct fsl_re_dpi dpi_src_spec[16];
+};
+
+/* Compound frame */
+#define FSL_RE_CF_ADDR_HIGH_MASK       0x000000FF
+#define FSL_RE_CF_EXT_MASK             0x80000000
+#define FSL_RE_CF_EXT_SHIFT            31
+#define FSL_RE_CF_FINAL_MASK           0x40000000
+#define FSL_RE_CF_FINAL_SHIFT          30
+#define FSL_RE_CF_LENGTH_MASK          0x000FFFFF
+#define FSL_RE_CF_BPID_MASK            0x00FF0000
+#define FSL_RE_CF_BPID_SHIFT           16
+#define FSL_RE_CF_OFFSET_MASK          0x00001FFF
+
+struct fsl_re_cmpnd_frame {
+       __be32 addr_high;
+       __be32 addr_low;
+       __be32 efrl32;
+       __be32 rbro32;
+};
+
+/* Frame descriptor */
+#define FSL_RE_HWDESC_LIODN_MASK       0x3F000000
+#define FSL_RE_HWDESC_LIODN_SHIFT      24
+#define FSL_RE_HWDESC_BPID_MASK                0x00FF0000
+#define FSL_RE_HWDESC_BPID_SHIFT       16
+#define FSL_RE_HWDESC_ELIODN_MASK      0x0000F000
+#define FSL_RE_HWDESC_ELIODN_SHIFT     12
+#define FSL_RE_HWDESC_FMT_SHIFT                29
+#define FSL_RE_HWDESC_FMT_MASK         (0x3 << FSL_RE_HWDESC_FMT_SHIFT)
+
+struct fsl_re_hw_desc {
+       __be32 lbea32;
+       __be32 addr_low;
+       __be32 fmt32;
+       __be32 status;
+};
+
+/* Raid Engine device private data */
+struct fsl_re_drv_private {
+       u8 total_chans;
+       struct dma_device dma_dev;
+       struct fsl_re_ctrl *re_regs;
+       struct fsl_re_chan *re_jrs[FSL_RE_MAX_CHANS];
+       struct dma_pool *cf_desc_pool;
+       struct dma_pool *hw_desc_pool;
+};
+
+/* Per job ring data structure */
+struct fsl_re_chan {
+       char name[16];
+       spinlock_t desc_lock; /* queue lock */
+       struct list_head ack_q;  /* wait to acked queue */
+       struct list_head active_q; /* already issued on hw, not completed */
+       struct list_head submit_q;
+       struct list_head free_q; /* alloc available queue */
+       struct device *dev;
+       struct fsl_re_drv_private *re_dev;
+       struct dma_chan chan;
+       struct fsl_re_chan_cfg *jrregs;
+       int irq;
+       struct tasklet_struct irqtask;
+       u32 alloc_count;
+
+       /* hw descriptor ring for inbound queue*/
+       dma_addr_t inb_phys_addr;
+       struct fsl_re_hw_desc *inb_ring_virt_addr;
+       u32 inb_count;
+
+       /* hw descriptor ring for outbound queue */
+       dma_addr_t oub_phys_addr;
+       struct fsl_re_hw_desc *oub_ring_virt_addr;
+       u32 oub_count;
+};
+
+/* Async transaction descriptor */
+struct fsl_re_desc {
+       struct dma_async_tx_descriptor async_tx;
+       struct list_head node;
+       struct fsl_re_hw_desc hwdesc;
+       struct fsl_re_chan *re_chan;
+
+       /* hwdesc will point to cf_addr */
+       void *cf_addr;
+       dma_addr_t cf_paddr;
+
+       void *cdb_addr;
+       dma_addr_t cdb_paddr;
+       int status;
+};