dmaengine: add OMAP DMA engine driver
authorRussell King <rmk+kernel@arm.linux.org.uk>
Fri, 13 Apr 2012 11:10:24 +0000 (12:10 +0100)
committerRussell King <rmk+kernel@arm.linux.org.uk>
Tue, 31 Jul 2012 11:06:20 +0000 (12:06 +0100)
Tested-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
drivers/dma/Kconfig
drivers/dma/Makefile
drivers/dma/omap-dma.c [new file with mode: 0644]
include/linux/omap-dma.h [new file with mode: 0644]

index eb2b60e..8be3bf6 100644 (file)
@@ -261,6 +261,12 @@ config DMA_SA11X0
          SA-1110 SoCs.  This DMA engine can only be used with on-chip
          devices.
 
+config DMA_OMAP
+       tristate "OMAP DMA support"
+       depends on ARCH_OMAP
+       select DMA_ENGINE
+       select DMA_VIRTUAL_CHANNELS
+
 config DMA_ENGINE
        bool
 
index fc05f7d..ddc291a 100644 (file)
@@ -29,3 +29,4 @@ obj-$(CONFIG_PCH_DMA) += pch_dma.o
 obj-$(CONFIG_AMBA_PL08X) += amba-pl08x.o
 obj-$(CONFIG_EP93XX_DMA) += ep93xx_dma.o
 obj-$(CONFIG_DMA_SA11X0) += sa11x0-dma.o
+obj-$(CONFIG_DMA_OMAP) += omap-dma.o
diff --git a/drivers/dma/omap-dma.c b/drivers/dma/omap-dma.c
new file mode 100644 (file)
index 0000000..fe33ddd
--- /dev/null
@@ -0,0 +1,524 @@
+/*
+ * OMAP DMAengine support
+ *
+ * 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.
+ */
+#include <linux/dmaengine.h>
+#include <linux/dma-mapping.h>
+#include <linux/err.h>
+#include <linux/init.h>
+#include <linux/interrupt.h>
+#include <linux/list.h>
+#include <linux/module.h>
+#include <linux/omap-dma.h>
+#include <linux/platform_device.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+
+#include "virt-dma.h"
+#include <plat/dma.h>
+
+struct omap_dmadev {
+       struct dma_device ddev;
+       spinlock_t lock;
+       struct tasklet_struct task;
+       struct list_head pending;
+};
+
+struct omap_chan {
+       struct virt_dma_chan vc;
+       struct list_head node;
+
+       struct dma_slave_config cfg;
+       unsigned dma_sig;
+
+       int dma_ch;
+       struct omap_desc *desc;
+       unsigned sgidx;
+};
+
+struct omap_sg {
+       dma_addr_t addr;
+       uint32_t en;            /* number of elements (24-bit) */
+       uint32_t fn;            /* number of frames (16-bit) */
+};
+
+struct omap_desc {
+       struct virt_dma_desc vd;
+       enum dma_transfer_direction dir;
+       dma_addr_t dev_addr;
+
+       uint8_t es;             /* OMAP_DMA_DATA_TYPE_xxx */
+       uint8_t sync_mode;      /* OMAP_DMA_SYNC_xxx */
+       uint8_t sync_type;      /* OMAP_DMA_xxx_SYNC* */
+       uint8_t periph_port;    /* Peripheral port */
+
+       unsigned sglen;
+       struct omap_sg sg[0];
+};
+
+static const unsigned es_bytes[] = {
+       [OMAP_DMA_DATA_TYPE_S8] = 1,
+       [OMAP_DMA_DATA_TYPE_S16] = 2,
+       [OMAP_DMA_DATA_TYPE_S32] = 4,
+};
+
+static inline struct omap_dmadev *to_omap_dma_dev(struct dma_device *d)
+{
+       return container_of(d, struct omap_dmadev, ddev);
+}
+
+static inline struct omap_chan *to_omap_dma_chan(struct dma_chan *c)
+{
+       return container_of(c, struct omap_chan, vc.chan);
+}
+
+static inline struct omap_desc *to_omap_dma_desc(struct dma_async_tx_descriptor *t)
+{
+       return container_of(t, struct omap_desc, vd.tx);
+}
+
+static void omap_dma_desc_free(struct virt_dma_desc *vd)
+{
+       kfree(container_of(vd, struct omap_desc, vd));
+}
+
+static void omap_dma_start_sg(struct omap_chan *c, struct omap_desc *d,
+       unsigned idx)
+{
+       struct omap_sg *sg = d->sg + idx;
+
+       if (d->dir == DMA_DEV_TO_MEM)
+               omap_set_dma_dest_params(c->dma_ch, OMAP_DMA_PORT_EMIFF,
+                       OMAP_DMA_AMODE_POST_INC, sg->addr, 0, 0);
+       else
+               omap_set_dma_src_params(c->dma_ch, OMAP_DMA_PORT_EMIFF,
+                       OMAP_DMA_AMODE_POST_INC, sg->addr, 0, 0);
+
+       omap_set_dma_transfer_params(c->dma_ch, d->es, sg->en, sg->fn,
+               d->sync_mode, c->dma_sig, d->sync_type);
+
+       omap_start_dma(c->dma_ch);
+}
+
+static void omap_dma_start_desc(struct omap_chan *c)
+{
+       struct virt_dma_desc *vd = vchan_next_desc(&c->vc);
+       struct omap_desc *d;
+
+       if (!vd) {
+               c->desc = NULL;
+               return;
+       }
+
+       list_del(&vd->node);
+
+       c->desc = d = to_omap_dma_desc(&vd->tx);
+       c->sgidx = 0;
+
+       if (d->dir == DMA_DEV_TO_MEM)
+               omap_set_dma_src_params(c->dma_ch, d->periph_port,
+                       OMAP_DMA_AMODE_CONSTANT, d->dev_addr, 0, 0);
+       else
+               omap_set_dma_dest_params(c->dma_ch, d->periph_port,
+                       OMAP_DMA_AMODE_CONSTANT, d->dev_addr, 0, 0);
+
+       omap_dma_start_sg(c, d, 0);
+}
+
+static void omap_dma_callback(int ch, u16 status, void *data)
+{
+       struct omap_chan *c = data;
+       struct omap_desc *d;
+       unsigned long flags;
+
+       spin_lock_irqsave(&c->vc.lock, flags);
+       d = c->desc;
+       if (d) {
+               if (++c->sgidx < d->sglen) {
+                       omap_dma_start_sg(c, d, c->sgidx);
+               } else {
+                       omap_dma_start_desc(c);
+                       vchan_cookie_complete(&d->vd);
+               }
+       }
+       spin_unlock_irqrestore(&c->vc.lock, flags);
+}
+
+/*
+ * This callback schedules all pending channels.  We could be more
+ * clever here by postponing allocation of the real DMA channels to
+ * this point, and freeing them when our virtual channel becomes idle.
+ *
+ * We would then need to deal with 'all channels in-use'
+ */
+static void omap_dma_sched(unsigned long data)
+{
+       struct omap_dmadev *d = (struct omap_dmadev *)data;
+       LIST_HEAD(head);
+
+       spin_lock_irq(&d->lock);
+       list_splice_tail_init(&d->pending, &head);
+       spin_unlock_irq(&d->lock);
+
+       while (!list_empty(&head)) {
+               struct omap_chan *c = list_first_entry(&head,
+                       struct omap_chan, node);
+
+               spin_lock_irq(&c->vc.lock);
+               list_del_init(&c->node);
+               omap_dma_start_desc(c);
+               spin_unlock_irq(&c->vc.lock);
+       }
+}
+
+static int omap_dma_alloc_chan_resources(struct dma_chan *chan)
+{
+       struct omap_chan *c = to_omap_dma_chan(chan);
+
+       dev_info(c->vc.chan.device->dev, "allocating channel for %u\n", c->dma_sig);
+
+       return omap_request_dma(c->dma_sig, "DMA engine",
+               omap_dma_callback, c, &c->dma_ch);
+}
+
+static void omap_dma_free_chan_resources(struct dma_chan *chan)
+{
+       struct omap_chan *c = to_omap_dma_chan(chan);
+
+       vchan_free_chan_resources(&c->vc);
+       omap_free_dma(c->dma_ch);
+
+       dev_info(c->vc.chan.device->dev, "freeing channel for %u\n", c->dma_sig);
+}
+
+static enum dma_status omap_dma_tx_status(struct dma_chan *chan,
+       dma_cookie_t cookie, struct dma_tx_state *txstate)
+{
+       /*
+        * FIXME: do we need to return pending bytes?
+        * We have no users of that info at the moment...
+        */
+       return dma_cookie_status(chan, cookie, txstate);
+}
+
+static void omap_dma_issue_pending(struct dma_chan *chan)
+{
+       struct omap_chan *c = to_omap_dma_chan(chan);
+       unsigned long flags;
+
+       spin_lock_irqsave(&c->vc.lock, flags);
+       if (vchan_issue_pending(&c->vc) && !c->desc) {
+               struct omap_dmadev *d = to_omap_dma_dev(chan->device);
+               spin_lock(&d->lock);
+               if (list_empty(&c->node))
+                       list_add_tail(&c->node, &d->pending);
+               spin_unlock(&d->lock);
+               tasklet_schedule(&d->task);
+       }
+       spin_unlock_irqrestore(&c->vc.lock, flags);
+}
+
+static struct dma_async_tx_descriptor *omap_dma_prep_slave_sg(
+       struct dma_chan *chan, struct scatterlist *sgl, unsigned sglen,
+       enum dma_transfer_direction dir, unsigned long tx_flags, void *context)
+{
+       struct omap_chan *c = to_omap_dma_chan(chan);
+       enum dma_slave_buswidth dev_width;
+       struct scatterlist *sgent;
+       struct omap_desc *d;
+       dma_addr_t dev_addr;
+       unsigned i, j = 0, es, en, frame_bytes, sync_type;
+       u32 burst;
+
+       if (dir == DMA_DEV_TO_MEM) {
+               dev_addr = c->cfg.src_addr;
+               dev_width = c->cfg.src_addr_width;
+               burst = c->cfg.src_maxburst;
+               sync_type = OMAP_DMA_SRC_SYNC;
+       } else if (dir == DMA_MEM_TO_DEV) {
+               dev_addr = c->cfg.dst_addr;
+               dev_width = c->cfg.dst_addr_width;
+               burst = c->cfg.dst_maxburst;
+               sync_type = OMAP_DMA_DST_SYNC;
+       } else {
+               dev_err(chan->device->dev, "%s: bad direction?\n", __func__);
+               return NULL;
+       }
+
+       /* Bus width translates to the element size (ES) */
+       switch (dev_width) {
+       case DMA_SLAVE_BUSWIDTH_1_BYTE:
+               es = OMAP_DMA_DATA_TYPE_S8;
+               break;
+       case DMA_SLAVE_BUSWIDTH_2_BYTES:
+               es = OMAP_DMA_DATA_TYPE_S16;
+               break;
+       case DMA_SLAVE_BUSWIDTH_4_BYTES:
+               es = OMAP_DMA_DATA_TYPE_S32;
+               break;
+       default: /* not reached */
+               return NULL;
+       }
+
+       /* Now allocate and setup the descriptor. */
+       d = kzalloc(sizeof(*d) + sglen * sizeof(d->sg[0]), GFP_ATOMIC);
+       if (!d)
+               return NULL;
+
+       d->dir = dir;
+       d->dev_addr = dev_addr;
+       d->es = es;
+       d->sync_mode = OMAP_DMA_SYNC_FRAME;
+       d->sync_type = sync_type;
+       d->periph_port = OMAP_DMA_PORT_TIPB;
+
+       /*
+        * Build our scatterlist entries: each contains the address,
+        * the number of elements (EN) in each frame, and the number of
+        * frames (FN).  Number of bytes for this entry = ES * EN * FN.
+        *
+        * Burst size translates to number of elements with frame sync.
+        * Note: DMA engine defines burst to be the number of dev-width
+        * transfers.
+        */
+       en = burst;
+       frame_bytes = es_bytes[es] * en;
+       for_each_sg(sgl, sgent, sglen, i) {
+               d->sg[j].addr = sg_dma_address(sgent);
+               d->sg[j].en = en;
+               d->sg[j].fn = sg_dma_len(sgent) / frame_bytes;
+               j++;
+       }
+
+       d->sglen = j;
+
+       return vchan_tx_prep(&c->vc, &d->vd, tx_flags);
+}
+
+static int omap_dma_slave_config(struct omap_chan *c, struct dma_slave_config *cfg)
+{
+       if (cfg->src_addr_width == DMA_SLAVE_BUSWIDTH_8_BYTES ||
+           cfg->dst_addr_width == DMA_SLAVE_BUSWIDTH_8_BYTES)
+               return -EINVAL;
+
+       memcpy(&c->cfg, cfg, sizeof(c->cfg));
+
+       return 0;
+}
+
+static int omap_dma_terminate_all(struct omap_chan *c)
+{
+       struct omap_dmadev *d = to_omap_dma_dev(c->vc.chan.device);
+       unsigned long flags;
+       LIST_HEAD(head);
+
+       spin_lock_irqsave(&c->vc.lock, flags);
+
+       /* Prevent this channel being scheduled */
+       spin_lock(&d->lock);
+       list_del_init(&c->node);
+       spin_unlock(&d->lock);
+
+       /*
+        * Stop DMA activity: we assume the callback will not be called
+        * after omap_stop_dma() returns (even if it does, it will see
+        * c->desc is NULL and exit.)
+        */
+       if (c->desc) {
+               c->desc = NULL;
+               omap_stop_dma(c->dma_ch);
+       }
+
+       vchan_get_all_descriptors(&c->vc, &head);
+       spin_unlock_irqrestore(&c->vc.lock, flags);
+       vchan_dma_desc_free_list(&c->vc, &head);
+
+       return 0;
+}
+
+static int omap_dma_pause(struct omap_chan *c)
+{
+       /* FIXME: not supported by platform private API */
+       return -EINVAL;
+}
+
+static int omap_dma_resume(struct omap_chan *c)
+{
+       /* FIXME: not supported by platform private API */
+       return -EINVAL;
+}
+
+static int omap_dma_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
+       unsigned long arg)
+{
+       struct omap_chan *c = to_omap_dma_chan(chan);
+       int ret;
+
+       switch (cmd) {
+       case DMA_SLAVE_CONFIG:
+               ret = omap_dma_slave_config(c, (struct dma_slave_config *)arg);
+               break;
+
+       case DMA_TERMINATE_ALL:
+               ret = omap_dma_terminate_all(c);
+               break;
+
+       case DMA_PAUSE:
+               ret = omap_dma_pause(c);
+               break;
+
+       case DMA_RESUME:
+               ret = omap_dma_resume(c);
+               break;
+
+       default:
+               ret = -ENXIO;
+               break;
+       }
+
+       return ret;
+}
+
+static int omap_dma_chan_init(struct omap_dmadev *od, int dma_sig)
+{
+       struct omap_chan *c;
+
+       c = kzalloc(sizeof(*c), GFP_KERNEL);
+       if (!c)
+               return -ENOMEM;
+
+       c->dma_sig = dma_sig;
+       c->vc.desc_free = omap_dma_desc_free;
+       vchan_init(&c->vc, &od->ddev);
+       INIT_LIST_HEAD(&c->node);
+
+       od->ddev.chancnt++;
+
+       return 0;
+}
+
+static void omap_dma_free(struct omap_dmadev *od)
+{
+       tasklet_kill(&od->task);
+       while (!list_empty(&od->ddev.channels)) {
+               struct omap_chan *c = list_first_entry(&od->ddev.channels,
+                       struct omap_chan, vc.chan.device_node);
+
+               list_del(&c->vc.chan.device_node);
+               tasklet_kill(&c->vc.task);
+               kfree(c);
+       }
+       kfree(od);
+}
+
+static int omap_dma_probe(struct platform_device *pdev)
+{
+       struct omap_dmadev *od;
+       int rc, i;
+
+       od = kzalloc(sizeof(*od), GFP_KERNEL);
+       if (!od)
+               return -ENOMEM;
+
+       dma_cap_set(DMA_SLAVE, od->ddev.cap_mask);
+       od->ddev.device_alloc_chan_resources = omap_dma_alloc_chan_resources;
+       od->ddev.device_free_chan_resources = omap_dma_free_chan_resources;
+       od->ddev.device_tx_status = omap_dma_tx_status;
+       od->ddev.device_issue_pending = omap_dma_issue_pending;
+       od->ddev.device_prep_slave_sg = omap_dma_prep_slave_sg;
+       od->ddev.device_control = omap_dma_control;
+       od->ddev.dev = &pdev->dev;
+       INIT_LIST_HEAD(&od->ddev.channels);
+       INIT_LIST_HEAD(&od->pending);
+       spin_lock_init(&od->lock);
+
+       tasklet_init(&od->task, omap_dma_sched, (unsigned long)od);
+
+       for (i = 0; i < 127; i++) {
+               rc = omap_dma_chan_init(od, i);
+               if (rc) {
+                       omap_dma_free(od);
+                       return rc;
+               }
+       }
+
+       rc = dma_async_device_register(&od->ddev);
+       if (rc) {
+               pr_warn("OMAP-DMA: failed to register slave DMA engine device: %d\n",
+                       rc);
+               omap_dma_free(od);
+       } else {
+               platform_set_drvdata(pdev, od);
+       }
+
+       dev_info(&pdev->dev, "OMAP DMA engine driver\n");
+
+       return rc;
+}
+
+static int omap_dma_remove(struct platform_device *pdev)
+{
+       struct omap_dmadev *od = platform_get_drvdata(pdev);
+
+       dma_async_device_unregister(&od->ddev);
+       omap_dma_free(od);
+
+       return 0;
+}
+
+static struct platform_driver omap_dma_driver = {
+       .probe  = omap_dma_probe,
+       .remove = omap_dma_remove,
+       .driver = {
+               .name = "omap-dma-engine",
+               .owner = THIS_MODULE,
+       },
+};
+
+bool omap_dma_filter_fn(struct dma_chan *chan, void *param)
+{
+       if (chan->device->dev->driver == &omap_dma_driver.driver) {
+               struct omap_chan *c = to_omap_dma_chan(chan);
+               unsigned req = *(unsigned *)param;
+
+               return req == c->dma_sig;
+       }
+       return false;
+}
+EXPORT_SYMBOL_GPL(omap_dma_filter_fn);
+
+static struct platform_device *pdev;
+
+static const struct platform_device_info omap_dma_dev_info = {
+       .name = "omap-dma-engine",
+       .id = -1,
+       .dma_mask = DMA_BIT_MASK(32),
+};
+
+static int omap_dma_init(void)
+{
+       int rc = platform_driver_register(&omap_dma_driver);
+
+       if (rc == 0) {
+               pdev = platform_device_register_full(&omap_dma_dev_info);
+               if (IS_ERR(pdev)) {
+                       platform_driver_unregister(&omap_dma_driver);
+                       rc = PTR_ERR(pdev);
+               }
+       }
+       return rc;
+}
+subsys_initcall(omap_dma_init);
+
+static void __exit omap_dma_exit(void)
+{
+       platform_device_unregister(pdev);
+       platform_driver_unregister(&omap_dma_driver);
+}
+module_exit(omap_dma_exit);
+
+MODULE_AUTHOR("Russell King");
+MODULE_LICENSE("GPL");
diff --git a/include/linux/omap-dma.h b/include/linux/omap-dma.h
new file mode 100644 (file)
index 0000000..eb475a8
--- /dev/null
@@ -0,0 +1,22 @@
+/*
+ * OMAP DMA Engine support
+ *
+ * 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.
+ */
+#ifndef __LINUX_OMAP_DMA_H
+#define __LINUX_OMAP_DMA_H
+
+struct dma_chan;
+
+#if defined(CONFIG_DMA_OMAP) || defined(CONFIG_DMA_OMAP_MODULE)
+bool omap_dma_filter_fn(struct dma_chan *, void *);
+#else
+static inline bool omap_dma_filter_fn(struct dma_chan *c, void *d)
+{
+       return false;
+}
+#endif
+
+#endif