dma: Add timb-dma
authorRichard Röjfors <richard.rojfors@pelagicore.com>
Thu, 25 Mar 2010 18:44:21 +0000 (19:44 +0100)
committerDan Williams <dan.j.williams@intel.com>
Fri, 26 Mar 2010 00:18:43 +0000 (17:18 -0700)
Adds the support for the DMA engine withing the timberdale FPGA.

The DMA channels are strict device to host, or host to device
and can not be used for generic memcpy.

Signed-off-by: Richard Röjfors <richard.rojfors@pelagicore.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
drivers/dma/Kconfig
drivers/dma/Makefile
drivers/dma/timb_dma.c [new file with mode: 0644]
include/linux/timb_dma.h [new file with mode: 0644]

index c27f80e..a2fcb2e 100644 (file)
@@ -149,6 +149,13 @@ config AMCC_PPC440SPE_ADMA
        help
          Enable support for the AMCC PPC440SPe RAID engines.
 
+config TIMB_DMA
+       tristate "Timberdale FPGA DMA support"
+       depends on MFD_TIMBERDALE || HAS_IOMEM
+       select DMA_ENGINE
+       help
+         Enable support for the Timberdale FPGA DMA engine.
+
 config ARCH_HAS_ASYNC_TX_FIND_CHANNEL
        bool
 
index 22bba3d..40c627d 100644 (file)
@@ -20,3 +20,4 @@ obj-$(CONFIG_TXX9_DMAC) += txx9dmac.o
 obj-$(CONFIG_SH_DMAE) += shdma.o
 obj-$(CONFIG_COH901318) += coh901318.o coh901318_lli.o
 obj-$(CONFIG_AMCC_PPC440SPE_ADMA) += ppc4xx/
+obj-$(CONFIG_TIMB_DMA) += timb_dma.o
diff --git a/drivers/dma/timb_dma.c b/drivers/dma/timb_dma.c
new file mode 100644 (file)
index 0000000..4dd7102
--- /dev/null
@@ -0,0 +1,853 @@
+/*
+ * timb_dma.c timberdale FPGA DMA driver
+ * Copyright (c) 2010 Intel Corporation
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that 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., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+/* Supports:
+ * Timberdale FPGA DMA engine
+ */
+
+#include <linux/dmaengine.h>
+#include <linux/dma-mapping.h>
+#include <linux/init.h>
+#include <linux/interrupt.h>
+#include <linux/io.h>
+#include <linux/module.h>
+#include <linux/platform_device.h>
+
+#include <linux/timb_dma.h>
+
+#define DRIVER_NAME "timb-dma"
+
+/* Global DMA registers */
+#define TIMBDMA_ACR            0x34
+#define TIMBDMA_32BIT_ADDR     0x01
+
+#define TIMBDMA_ISR            0x080000
+#define TIMBDMA_IPR            0x080004
+#define TIMBDMA_IER            0x080008
+
+/* Channel specific registers */
+/* RX instances base addresses are 0x00, 0x40, 0x80 ...
+ * TX instances base addresses are 0x18, 0x58, 0x98 ...
+ */
+#define TIMBDMA_INSTANCE_OFFSET                0x40
+#define TIMBDMA_INSTANCE_TX_OFFSET     0x18
+
+/* RX registers, relative the instance base */
+#define TIMBDMA_OFFS_RX_DHAR   0x00
+#define TIMBDMA_OFFS_RX_DLAR   0x04
+#define TIMBDMA_OFFS_RX_LR     0x0C
+#define TIMBDMA_OFFS_RX_BLR    0x10
+#define TIMBDMA_OFFS_RX_ER     0x14
+#define TIMBDMA_RX_EN          0x01
+/* bytes per Row, video specific register
+ * which is placed after the TX registers...
+ */
+#define TIMBDMA_OFFS_RX_BPRR   0x30
+
+/* TX registers, relative the instance base */
+#define TIMBDMA_OFFS_TX_DHAR   0x00
+#define TIMBDMA_OFFS_TX_DLAR   0x04
+#define TIMBDMA_OFFS_TX_BLR    0x0C
+#define TIMBDMA_OFFS_TX_LR     0x14
+
+
+#define TIMB_DMA_DESC_SIZE     8
+
+struct timb_dma_desc {
+       struct list_head                desc_node;
+       struct dma_async_tx_descriptor  txd;
+       u8                              *desc_list;
+       unsigned int                    desc_list_len;
+       bool                            interrupt;
+};
+
+struct timb_dma_chan {
+       struct dma_chan         chan;
+       void __iomem            *membase;
+       spinlock_t              lock; /* Used for mutual exclusion */
+       dma_cookie_t            last_completed_cookie;
+       bool                    ongoing;
+       struct list_head        active_list;
+       struct list_head        queue;
+       struct list_head        free_list;
+       unsigned int            bytes_per_line;
+       enum dma_data_direction direction;
+       unsigned int            descs; /* Descriptors to allocate */
+       unsigned int            desc_elems; /* number of elems per descriptor */
+};
+
+struct timb_dma {
+       struct dma_device       dma;
+       void __iomem            *membase;
+       struct tasklet_struct   tasklet;
+       struct timb_dma_chan    channels[0];
+};
+
+static struct device *chan2dev(struct dma_chan *chan)
+{
+       return &chan->dev->device;
+}
+static struct device *chan2dmadev(struct dma_chan *chan)
+{
+       return chan2dev(chan)->parent->parent;
+}
+
+static struct timb_dma *tdchantotd(struct timb_dma_chan *td_chan)
+{
+       int id = td_chan->chan.chan_id;
+       return (struct timb_dma *)((u8 *)td_chan -
+               id * sizeof(struct timb_dma_chan) - sizeof(struct timb_dma));
+}
+
+/* Must be called with the spinlock held */
+static void __td_enable_chan_irq(struct timb_dma_chan *td_chan)
+{
+       int id = td_chan->chan.chan_id;
+       struct timb_dma *td = tdchantotd(td_chan);
+       u32 ier;
+
+       /* enable interrupt for this channel */
+       ier = ioread32(td->membase + TIMBDMA_IER);
+       ier |= 1 << id;
+       dev_dbg(chan2dev(&td_chan->chan), "Enabling irq: %d, IER: 0x%x\n", id,
+               ier);
+       iowrite32(ier, td->membase + TIMBDMA_IER);
+}
+
+/* Should be called with the spinlock held */
+static bool __td_dma_done_ack(struct timb_dma_chan *td_chan)
+{
+       int id = td_chan->chan.chan_id;
+       struct timb_dma *td = (struct timb_dma *)((u8 *)td_chan -
+               id * sizeof(struct timb_dma_chan) - sizeof(struct timb_dma));
+       u32 isr;
+       bool done = false;
+
+       dev_dbg(chan2dev(&td_chan->chan), "Checking irq: %d, td: %p\n", id, td);
+
+       isr = ioread32(td->membase + TIMBDMA_ISR) & (1 << id);
+       if (isr) {
+               iowrite32(isr, td->membase + TIMBDMA_ISR);
+               done = true;
+       }
+
+       return done;
+}
+
+static void __td_unmap_desc(struct timb_dma_chan *td_chan, const u8 *dma_desc,
+       bool single)
+{
+       dma_addr_t addr;
+       int len;
+
+       addr = (dma_desc[7] << 24) | (dma_desc[6] << 16) | (dma_desc[5] << 8) |
+               dma_desc[4];
+
+       len = (dma_desc[3] << 8) | dma_desc[2];
+
+       if (single)
+               dma_unmap_single(chan2dev(&td_chan->chan), addr, len,
+                       td_chan->direction);
+       else
+               dma_unmap_page(chan2dev(&td_chan->chan), addr, len,
+                       td_chan->direction);
+}
+
+static void __td_unmap_descs(struct timb_dma_desc *td_desc, bool single)
+{
+       struct timb_dma_chan *td_chan = container_of(td_desc->txd.chan,
+               struct timb_dma_chan, chan);
+       u8 *descs;
+
+       for (descs = td_desc->desc_list; ; descs += TIMB_DMA_DESC_SIZE) {
+               __td_unmap_desc(td_chan, descs, single);
+               if (descs[0] & 0x02)
+                       break;
+       }
+}
+
+static int td_fill_desc(struct timb_dma_chan *td_chan, u8 *dma_desc,
+       struct scatterlist *sg, bool last)
+{
+       if (sg_dma_len(sg) > USHORT_MAX) {
+               dev_err(chan2dev(&td_chan->chan), "Too big sg element\n");
+               return -EINVAL;
+       }
+
+       /* length must be word aligned */
+       if (sg_dma_len(sg) % sizeof(u32)) {
+               dev_err(chan2dev(&td_chan->chan), "Incorrect length: %d\n",
+                       sg_dma_len(sg));
+               return -EINVAL;
+       }
+
+       dev_dbg(chan2dev(&td_chan->chan), "desc: %p, addr: %p\n",
+               dma_desc, (void *)(int)sg_dma_address(sg));
+
+       dma_desc[7] = (sg_dma_address(sg) >> 24) & 0xff;
+       dma_desc[6] = (sg_dma_address(sg) >> 16) & 0xff;
+       dma_desc[5] = (sg_dma_address(sg) >> 8) & 0xff;
+       dma_desc[4] = (sg_dma_address(sg) >> 0) & 0xff;
+
+       dma_desc[3] = (sg_dma_len(sg) >> 8) & 0xff;
+       dma_desc[2] = (sg_dma_len(sg) >> 0) & 0xff;
+
+       dma_desc[1] = 0x00;
+       dma_desc[0] = 0x21 | (last ? 0x02 : 0); /* tran, valid */
+
+       return 0;
+}
+
+/* Must be called with the spinlock held */
+static void __td_start_dma(struct timb_dma_chan *td_chan)
+{
+       struct timb_dma_desc *td_desc;
+
+       if (td_chan->ongoing) {
+               dev_err(chan2dev(&td_chan->chan),
+                       "Transfer already ongoing\n");
+               return;
+       }
+
+       td_desc = list_entry(td_chan->active_list.next, struct timb_dma_desc,
+               desc_node);
+
+       dev_dbg(chan2dev(&td_chan->chan),
+               "td_chan: %p, chan: %d, membase: %p\n",
+               td_chan, td_chan->chan.chan_id, td_chan->membase);
+
+       if (td_chan->direction == DMA_FROM_DEVICE) {
+
+               /* descriptor address */
+               iowrite32(0, td_chan->membase + TIMBDMA_OFFS_RX_DHAR);
+               iowrite32(td_desc->txd.phys, td_chan->membase +
+                       TIMBDMA_OFFS_RX_DLAR);
+               /* Bytes per line */
+               iowrite32(td_chan->bytes_per_line, td_chan->membase +
+                       TIMBDMA_OFFS_RX_BPRR);
+               /* enable RX */
+               iowrite32(TIMBDMA_RX_EN, td_chan->membase + TIMBDMA_OFFS_RX_ER);
+       } else {
+               /* address high */
+               iowrite32(0, td_chan->membase + TIMBDMA_OFFS_TX_DHAR);
+               iowrite32(td_desc->txd.phys, td_chan->membase +
+                       TIMBDMA_OFFS_TX_DLAR);
+       }
+
+       td_chan->ongoing = true;
+
+       if (td_desc->interrupt)
+               __td_enable_chan_irq(td_chan);
+}
+
+static void __td_finish(struct timb_dma_chan *td_chan)
+{
+       dma_async_tx_callback           callback;
+       void                            *param;
+       struct dma_async_tx_descriptor  *txd;
+       struct timb_dma_desc            *td_desc;
+
+       /* can happen if the descriptor is canceled */
+       if (list_empty(&td_chan->active_list))
+               return;
+
+       td_desc = list_entry(td_chan->active_list.next, struct timb_dma_desc,
+               desc_node);
+       txd = &td_desc->txd;
+
+       dev_dbg(chan2dev(&td_chan->chan), "descriptor %u complete\n",
+               txd->cookie);
+
+       /* make sure to stop the transfer */
+       if (td_chan->direction == DMA_FROM_DEVICE)
+               iowrite32(0, td_chan->membase + TIMBDMA_OFFS_RX_ER);
+/* Currently no support for stopping DMA transfers
+       else
+               iowrite32(0, td_chan->membase + TIMBDMA_OFFS_TX_DLAR);
+*/
+       td_chan->last_completed_cookie = txd->cookie;
+       td_chan->ongoing = false;
+
+       callback = txd->callback;
+       param = txd->callback_param;
+
+       list_move(&td_desc->desc_node, &td_chan->free_list);
+
+       if (!(txd->flags & DMA_COMPL_SKIP_SRC_UNMAP))
+               __td_unmap_descs(td_desc,
+                       txd->flags & DMA_COMPL_SRC_UNMAP_SINGLE);
+
+       /*
+        * The API requires that no submissions are done from a
+        * callback, so we don't need to drop the lock here
+        */
+       if (callback)
+               callback(param);
+}
+
+static u32 __td_ier_mask(struct timb_dma *td)
+{
+       int i;
+       u32 ret = 0;
+
+       for (i = 0; i < td->dma.chancnt; i++) {
+               struct timb_dma_chan *td_chan = td->channels + i;
+               if (td_chan->ongoing) {
+                       struct timb_dma_desc *td_desc =
+                               list_entry(td_chan->active_list.next,
+                               struct timb_dma_desc, desc_node);
+                       if (td_desc->interrupt)
+                               ret |= 1 << i;
+               }
+       }
+
+       return ret;
+}
+
+static void __td_start_next(struct timb_dma_chan *td_chan)
+{
+       struct timb_dma_desc *td_desc;
+
+       BUG_ON(list_empty(&td_chan->queue));
+       BUG_ON(td_chan->ongoing);
+
+       td_desc = list_entry(td_chan->queue.next, struct timb_dma_desc,
+               desc_node);
+
+       dev_dbg(chan2dev(&td_chan->chan), "%s: started %u\n",
+               __func__, td_desc->txd.cookie);
+
+       list_move(&td_desc->desc_node, &td_chan->active_list);
+       __td_start_dma(td_chan);
+}
+
+static dma_cookie_t td_tx_submit(struct dma_async_tx_descriptor *txd)
+{
+       struct timb_dma_desc *td_desc = container_of(txd, struct timb_dma_desc,
+               txd);
+       struct timb_dma_chan *td_chan = container_of(txd->chan,
+               struct timb_dma_chan, chan);
+       dma_cookie_t cookie;
+
+       spin_lock_bh(&td_chan->lock);
+
+       cookie = txd->chan->cookie;
+       if (++cookie < 0)
+               cookie = 1;
+       txd->chan->cookie = cookie;
+       txd->cookie = cookie;
+
+       if (list_empty(&td_chan->active_list)) {
+               dev_dbg(chan2dev(txd->chan), "%s: started %u\n", __func__,
+                       txd->cookie);
+               list_add_tail(&td_desc->desc_node, &td_chan->active_list);
+               __td_start_dma(td_chan);
+       } else {
+               dev_dbg(chan2dev(txd->chan), "tx_submit: queued %u\n",
+                       txd->cookie);
+
+               list_add_tail(&td_desc->desc_node, &td_chan->queue);
+       }
+
+       spin_unlock_bh(&td_chan->lock);
+
+       return cookie;
+}
+
+static struct timb_dma_desc *td_alloc_init_desc(struct timb_dma_chan *td_chan)
+{
+       struct dma_chan *chan = &td_chan->chan;
+       struct timb_dma_desc *td_desc;
+       int err;
+
+       td_desc = kzalloc(sizeof(struct timb_dma_desc), GFP_KERNEL);
+       if (!td_desc) {
+               dev_err(chan2dev(chan), "Failed to alloc descriptor\n");
+               goto err;
+       }
+
+       td_desc->desc_list_len = td_chan->desc_elems * TIMB_DMA_DESC_SIZE;
+
+       td_desc->desc_list = kzalloc(td_desc->desc_list_len, GFP_KERNEL);
+       if (!td_desc->desc_list) {
+               dev_err(chan2dev(chan), "Failed to alloc descriptor\n");
+               goto err;
+       }
+
+       dma_async_tx_descriptor_init(&td_desc->txd, chan);
+       td_desc->txd.tx_submit = td_tx_submit;
+       td_desc->txd.flags = DMA_CTRL_ACK;
+
+       td_desc->txd.phys = dma_map_single(chan2dmadev(chan),
+               td_desc->desc_list, td_desc->desc_list_len, DMA_TO_DEVICE);
+
+       err = dma_mapping_error(chan2dmadev(chan), td_desc->txd.phys);
+       if (err) {
+               dev_err(chan2dev(chan), "DMA mapping error: %d\n", err);
+               goto err;
+       }
+
+       return td_desc;
+err:
+       kfree(td_desc->desc_list);
+       kfree(td_desc);
+
+       return NULL;
+
+}
+
+static void td_free_desc(struct timb_dma_desc *td_desc)
+{
+       dev_dbg(chan2dev(td_desc->txd.chan), "Freeing desc: %p\n", td_desc);
+       dma_unmap_single(chan2dmadev(td_desc->txd.chan), td_desc->txd.phys,
+               td_desc->desc_list_len, DMA_TO_DEVICE);
+
+       kfree(td_desc->desc_list);
+       kfree(td_desc);
+}
+
+static void td_desc_put(struct timb_dma_chan *td_chan,
+       struct timb_dma_desc *td_desc)
+{
+       dev_dbg(chan2dev(&td_chan->chan), "Putting desc: %p\n", td_desc);
+
+       spin_lock_bh(&td_chan->lock);
+       list_add(&td_desc->desc_node, &td_chan->free_list);
+       spin_unlock_bh(&td_chan->lock);
+}
+
+static struct timb_dma_desc *td_desc_get(struct timb_dma_chan *td_chan)
+{
+       struct timb_dma_desc *td_desc, *_td_desc;
+       struct timb_dma_desc *ret = NULL;
+
+       spin_lock_bh(&td_chan->lock);
+       list_for_each_entry_safe(td_desc, _td_desc, &td_chan->free_list,
+               desc_node) {
+               if (async_tx_test_ack(&td_desc->txd)) {
+                       list_del(&td_desc->desc_node);
+                       ret = td_desc;
+                       break;
+               }
+               dev_dbg(chan2dev(&td_chan->chan), "desc %p not ACKed\n",
+                       td_desc);
+       }
+       spin_unlock_bh(&td_chan->lock);
+
+       return ret;
+}
+
+static int td_alloc_chan_resources(struct dma_chan *chan)
+{
+       struct timb_dma_chan *td_chan =
+               container_of(chan, struct timb_dma_chan, chan);
+       int i;
+
+       dev_dbg(chan2dev(chan), "%s: entry\n", __func__);
+
+       BUG_ON(!list_empty(&td_chan->free_list));
+       for (i = 0; i < td_chan->descs; i++) {
+               struct timb_dma_desc *td_desc = td_alloc_init_desc(td_chan);
+               if (!td_desc) {
+                       if (i)
+                               break;
+                       else {
+                               dev_err(chan2dev(chan),
+                                       "Couldnt allocate any descriptors\n");
+                               return -ENOMEM;
+                       }
+               }
+
+               td_desc_put(td_chan, td_desc);
+       }
+
+       spin_lock_bh(&td_chan->lock);
+       td_chan->last_completed_cookie = 1;
+       chan->cookie = 1;
+       spin_unlock_bh(&td_chan->lock);
+
+       return 0;
+}
+
+static void td_free_chan_resources(struct dma_chan *chan)
+{
+       struct timb_dma_chan *td_chan =
+               container_of(chan, struct timb_dma_chan, chan);
+       struct timb_dma_desc *td_desc, *_td_desc;
+       LIST_HEAD(list);
+
+       dev_dbg(chan2dev(chan), "%s: Entry\n", __func__);
+
+       /* check that all descriptors are free */
+       BUG_ON(!list_empty(&td_chan->active_list));
+       BUG_ON(!list_empty(&td_chan->queue));
+
+       spin_lock_bh(&td_chan->lock);
+       list_splice_init(&td_chan->free_list, &list);
+       spin_unlock_bh(&td_chan->lock);
+
+       list_for_each_entry_safe(td_desc, _td_desc, &list, desc_node) {
+               dev_dbg(chan2dev(chan), "%s: Freeing desc: %p\n", __func__,
+                       td_desc);
+               td_free_desc(td_desc);
+       }
+}
+
+static enum dma_status td_is_tx_complete(struct dma_chan *chan,
+       dma_cookie_t cookie, dma_cookie_t *done, dma_cookie_t *used)
+{
+       struct timb_dma_chan *td_chan =
+               container_of(chan, struct timb_dma_chan, chan);
+       dma_cookie_t            last_used;
+       dma_cookie_t            last_complete;
+       int                     ret;
+
+       dev_dbg(chan2dev(chan), "%s: Entry\n", __func__);
+
+       last_complete = td_chan->last_completed_cookie;
+       last_used = chan->cookie;
+
+       ret = dma_async_is_complete(cookie, last_complete, last_used);
+
+       if (done)
+               *done = last_complete;
+       if (used)
+               *used = last_used;
+
+       dev_dbg(chan2dev(chan),
+               "%s: exit, ret: %d, last_complete: %d, last_used: %d\n",
+               __func__, ret, last_complete, last_used);
+
+       return ret;
+}
+
+static void td_issue_pending(struct dma_chan *chan)
+{
+       struct timb_dma_chan *td_chan =
+               container_of(chan, struct timb_dma_chan, chan);
+
+       dev_dbg(chan2dev(chan), "%s: Entry\n", __func__);
+       spin_lock_bh(&td_chan->lock);
+
+       if (!list_empty(&td_chan->active_list))
+               /* transfer ongoing */
+               if (__td_dma_done_ack(td_chan))
+                       __td_finish(td_chan);
+
+       if (list_empty(&td_chan->active_list) && !list_empty(&td_chan->queue))
+               __td_start_next(td_chan);
+
+       spin_unlock_bh(&td_chan->lock);
+}
+
+static struct dma_async_tx_descriptor *td_prep_slave_sg(struct dma_chan *chan,
+       struct scatterlist *sgl, unsigned int sg_len,
+       enum dma_data_direction direction, unsigned long flags)
+{
+       struct timb_dma_chan *td_chan =
+               container_of(chan, struct timb_dma_chan, chan);
+       struct timb_dma_desc *td_desc;
+       struct scatterlist *sg;
+       unsigned int i;
+       unsigned int desc_usage = 0;
+
+       if (!sgl || !sg_len) {
+               dev_err(chan2dev(chan), "%s: No SG list\n", __func__);
+               return NULL;
+       }
+
+       /* even channels are for RX, odd for TX */
+       if (td_chan->direction != direction) {
+               dev_err(chan2dev(chan),
+                       "Requesting channel in wrong direction\n");
+               return NULL;
+       }
+
+       td_desc = td_desc_get(td_chan);
+       if (!td_desc) {
+               dev_err(chan2dev(chan), "Not enough descriptors available\n");
+               return NULL;
+       }
+
+       td_desc->interrupt = (flags & DMA_PREP_INTERRUPT) != 0;
+
+       for_each_sg(sgl, sg, sg_len, i) {
+               int err;
+               if (desc_usage > td_desc->desc_list_len) {
+                       dev_err(chan2dev(chan), "No descriptor space\n");
+                       return NULL;
+               }
+
+               err = td_fill_desc(td_chan, td_desc->desc_list + desc_usage, sg,
+                       i == (sg_len - 1));
+               if (err) {
+                       dev_err(chan2dev(chan), "Failed to update desc: %d\n",
+                               err);
+                       td_desc_put(td_chan, td_desc);
+                       return NULL;
+               }
+               desc_usage += TIMB_DMA_DESC_SIZE;
+       }
+
+       dma_sync_single_for_device(chan2dmadev(chan), td_desc->txd.phys,
+               td_desc->desc_list_len, DMA_TO_DEVICE);
+
+       return &td_desc->txd;
+}
+
+static void td_terminate_all(struct dma_chan *chan)
+{
+       struct timb_dma_chan *td_chan =
+               container_of(chan, struct timb_dma_chan, chan);
+       struct timb_dma_desc *td_desc, *_td_desc;
+
+       dev_dbg(chan2dev(chan), "%s: Entry\n", __func__);
+
+       /* first the easy part, put the queue into the free list */
+       spin_lock_bh(&td_chan->lock);
+       list_for_each_entry_safe(td_desc, _td_desc, &td_chan->queue,
+               desc_node)
+               list_move(&td_desc->desc_node, &td_chan->free_list);
+
+       /* now tear down the runnning */
+       __td_finish(td_chan);
+       spin_unlock_bh(&td_chan->lock);
+}
+
+static void td_tasklet(unsigned long data)
+{
+       struct timb_dma *td = (struct timb_dma *)data;
+       u32 isr;
+       u32 ipr;
+       u32 ier;
+       int i;
+
+       isr = ioread32(td->membase + TIMBDMA_ISR);
+       ipr = isr & __td_ier_mask(td);
+
+       /* ack the interrupts */
+       iowrite32(ipr, td->membase + TIMBDMA_ISR);
+
+       for (i = 0; i < td->dma.chancnt; i++)
+               if (ipr & (1 << i)) {
+                       struct timb_dma_chan *td_chan = td->channels + i;
+                       spin_lock(&td_chan->lock);
+                       __td_finish(td_chan);
+                       if (!list_empty(&td_chan->queue))
+                               __td_start_next(td_chan);
+                       spin_unlock(&td_chan->lock);
+               }
+
+       ier = __td_ier_mask(td);
+       iowrite32(ier, td->membase + TIMBDMA_IER);
+}
+
+
+static irqreturn_t td_irq(int irq, void *devid)
+{
+       struct timb_dma *td = devid;
+       u32 ipr = ioread32(td->membase + TIMBDMA_IPR);
+
+       if (ipr) {
+               /* disable interrupts, will be re-enabled in tasklet */
+               iowrite32(0, td->membase + TIMBDMA_IER);
+
+               tasklet_schedule(&td->tasklet);
+
+               return IRQ_HANDLED;
+       } else
+               return IRQ_NONE;
+}
+
+
+static int __devinit td_probe(struct platform_device *pdev)
+{
+       struct timb_dma_platform_data *pdata = pdev->dev.platform_data;
+       struct timb_dma *td;
+       struct resource *iomem;
+       int irq;
+       int err;
+       int i;
+
+       if (!pdata) {
+               dev_err(&pdev->dev, "No platform data\n");
+               return -EINVAL;
+       }
+
+       iomem = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       if (!iomem)
+               return -EINVAL;
+
+       irq = platform_get_irq(pdev, 0);
+       if (irq < 0)
+               return irq;
+
+       if (!request_mem_region(iomem->start, resource_size(iomem),
+               DRIVER_NAME))
+               return -EBUSY;
+
+       td  = kzalloc(sizeof(struct timb_dma) +
+               sizeof(struct timb_dma_chan) * pdata->nr_channels, GFP_KERNEL);
+       if (!td) {
+               err = -ENOMEM;
+               goto err_release_region;
+       }
+
+       dev_dbg(&pdev->dev, "Allocated TD: %p\n", td);
+
+       td->membase = ioremap(iomem->start, resource_size(iomem));
+       if (!td->membase) {
+               dev_err(&pdev->dev, "Failed to remap I/O memory\n");
+               err = -ENOMEM;
+               goto err_free_mem;
+       }
+
+       /* 32bit addressing */
+       iowrite32(TIMBDMA_32BIT_ADDR, td->membase + TIMBDMA_ACR);
+
+       /* disable and clear any interrupts */
+       iowrite32(0x0, td->membase + TIMBDMA_IER);
+       iowrite32(0xFFFFFFFF, td->membase + TIMBDMA_ISR);
+
+       tasklet_init(&td->tasklet, td_tasklet, (unsigned long)td);
+
+       err = request_irq(irq, td_irq, IRQF_SHARED, DRIVER_NAME, td);
+       if (err) {
+               dev_err(&pdev->dev, "Failed to request IRQ\n");
+               goto err_tasklet_kill;
+       }
+
+       td->dma.device_alloc_chan_resources     = td_alloc_chan_resources;
+       td->dma.device_free_chan_resources      = td_free_chan_resources;
+       td->dma.device_is_tx_complete           = td_is_tx_complete;
+       td->dma.device_issue_pending            = td_issue_pending;
+
+       dma_cap_set(DMA_SLAVE, td->dma.cap_mask);
+       dma_cap_set(DMA_PRIVATE, td->dma.cap_mask);
+       td->dma.device_prep_slave_sg = td_prep_slave_sg;
+       td->dma.device_terminate_all = td_terminate_all;
+
+       td->dma.dev = &pdev->dev;
+
+       INIT_LIST_HEAD(&td->dma.channels);
+
+       for (i = 0; i < pdata->nr_channels; i++, td->dma.chancnt++) {
+               struct timb_dma_chan *td_chan = &td->channels[i];
+               struct timb_dma_platform_data_channel *pchan =
+                       pdata->channels + i;
+
+               /* even channels are RX, odd are TX */
+               if (((i % 2) && pchan->rx) || (!(i % 2) && !pchan->rx)) {
+                       dev_err(&pdev->dev, "Wrong channel configuration\n");
+                       err = -EINVAL;
+                       goto err_tasklet_kill;
+               }
+
+               td_chan->chan.device = &td->dma;
+               td_chan->chan.cookie = 1;
+               td_chan->chan.chan_id = i;
+               spin_lock_init(&td_chan->lock);
+               INIT_LIST_HEAD(&td_chan->active_list);
+               INIT_LIST_HEAD(&td_chan->queue);
+               INIT_LIST_HEAD(&td_chan->free_list);
+
+               td_chan->descs = pchan->descriptors;
+               td_chan->desc_elems = pchan->descriptor_elements;
+               td_chan->bytes_per_line = pchan->bytes_per_line;
+               td_chan->direction = pchan->rx ? DMA_FROM_DEVICE :
+                       DMA_TO_DEVICE;
+
+               td_chan->membase = td->membase +
+                       (i / 2) * TIMBDMA_INSTANCE_OFFSET +
+                       (pchan->rx ? 0 : TIMBDMA_INSTANCE_TX_OFFSET);
+
+               dev_dbg(&pdev->dev, "Chan: %d, membase: %p\n",
+                       i, td_chan->membase);
+
+               list_add_tail(&td_chan->chan.device_node, &td->dma.channels);
+       }
+
+       err = dma_async_device_register(&td->dma);
+       if (err) {
+               dev_err(&pdev->dev, "Failed to register async device\n");
+               goto err_free_irq;
+       }
+
+       platform_set_drvdata(pdev, td);
+
+       dev_dbg(&pdev->dev, "Probe result: %d\n", err);
+       return err;
+
+err_free_irq:
+       free_irq(irq, td);
+err_tasklet_kill:
+       tasklet_kill(&td->tasklet);
+       iounmap(td->membase);
+err_free_mem:
+       kfree(td);
+err_release_region:
+       release_mem_region(iomem->start, resource_size(iomem));
+
+       return err;
+
+}
+
+static int __devexit td_remove(struct platform_device *pdev)
+{
+       struct timb_dma *td = platform_get_drvdata(pdev);
+       struct resource *iomem = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       int irq = platform_get_irq(pdev, 0);
+
+       dma_async_device_unregister(&td->dma);
+       free_irq(irq, td);
+       tasklet_kill(&td->tasklet);
+       iounmap(td->membase);
+       kfree(td);
+       release_mem_region(iomem->start, resource_size(iomem));
+
+       platform_set_drvdata(pdev, NULL);
+
+       dev_dbg(&pdev->dev, "Removed...\n");
+       return 0;
+}
+
+static struct platform_driver td_driver = {
+       .driver = {
+               .name   = DRIVER_NAME,
+               .owner  = THIS_MODULE,
+       },
+       .probe  = td_probe,
+       .remove = __exit_p(td_remove),
+};
+
+static int __init td_init(void)
+{
+       return platform_driver_register(&td_driver);
+}
+module_init(td_init);
+
+static void __exit td_exit(void)
+{
+       platform_driver_unregister(&td_driver);
+}
+module_exit(td_exit);
+
+MODULE_LICENSE("GPL v2");
+MODULE_DESCRIPTION("Timberdale DMA controller driver");
+MODULE_AUTHOR("Pelagicore AB <info@pelagicore.com>");
+MODULE_ALIAS("platform:"DRIVER_NAME);
diff --git a/include/linux/timb_dma.h b/include/linux/timb_dma.h
new file mode 100644 (file)
index 0000000..bb043e9
--- /dev/null
@@ -0,0 +1,55 @@
+/*
+ * timb_dma.h timberdale FPGA DMA driver defines
+ * Copyright (c) 2010 Intel Corporation
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that 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., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+/* Supports:
+ * Timberdale FPGA DMA engine
+ */
+
+#ifndef _LINUX_TIMB_DMA_H
+#define _LINUX_TIMB_DMA_H
+
+/**
+ * struct timb_dma_platform_data_channel - Description of each individual
+ *     DMA channel for the timberdale DMA driver
+ * @rx:                        true if this channel handles data in the direction to
+ *     the CPU.
+ * @bytes_per_line:    Number of bytes per line, this is specific for channels
+ *     handling video data. For other channels this shall be left to 0.
+ * @descriptors:       Number of descriptors to allocate for this channel.
+ * @descriptor_elements: Number of elements in each descriptor.
+ *
+ */
+struct timb_dma_platform_data_channel {
+       bool rx;
+       unsigned int bytes_per_line;
+       unsigned int descriptors;
+       unsigned int descriptor_elements;
+};
+
+/**
+ * struct timb_dma_platform_data - Platform data of the timberdale DMA driver
+ * @nr_channels:       Number of defined channels in the channels array.
+ * @channels:          Definition of the each channel.
+ *
+ */
+struct timb_dma_platform_data {
+       unsigned nr_channels;
+       struct timb_dma_platform_data_channel channels[32];
+};
+
+#endif