[PATCH 4/4] atmel-mci: Add experimental DMA support
From: Haavard Skinnemoen
Date: Mon Sep 22 2008 - 12:39:35 EST
This adds support for DMA transfers through the generic DMA engine
framework with the DMA slave extensions.
The driver has been tested using mmc-block and ext3fs on several SD,
SDHC and MMC+ cards. Reads and writes work fine, with read transfer
rates up to 7.5 MiB/s on fast cards with debugging disabled.
Unfortunately, the driver has been known to lock up from time to time
with DMA enabled, so DMA support is currently optional and marked
EXPERIMENTAL. I didn't see any problems while testing it prior to
submission, but I'm still not 100% convinced this bug is gone.
Signed-off-by: Haavard Skinnemoen <haavard.skinnemoen@xxxxxxxxx>
---
arch/avr32/include/asm/atmel-mci.h | 3 +
arch/avr32/mach-at32ap/at32ap700x.c | 16 +++
drivers/mmc/host/Kconfig | 11 ++
drivers/mmc/host/atmel-mci.c | 246 ++++++++++++++++++++++++++++++++++-
4 files changed, 270 insertions(+), 6 deletions(-)
diff --git a/arch/avr32/include/asm/atmel-mci.h b/arch/avr32/include/asm/atmel-mci.h
index 27371c2..37f978c 100644
--- a/arch/avr32/include/asm/atmel-mci.h
+++ b/arch/avr32/include/asm/atmel-mci.h
@@ -1,6 +1,8 @@
#ifndef __ASM_AVR32_ATMEL_MCI_H
#define __ASM_AVR32_ATMEL_MCI_H
+struct dma_slave;
+
/**
* struct mci_slot_pdata - board-specific per-slot configuration
* @bus_width: Number of data lines wired up the slot
@@ -24,6 +26,7 @@ struct mci_slot_pdata {
* @slot: Per-slot configuration data.
*/
struct mci_platform_data {
+ struct dma_slave *dma_slave;
struct mci_slot_pdata slot[2];
};
diff --git a/arch/avr32/mach-at32ap/at32ap700x.c b/arch/avr32/mach-at32ap/at32ap700x.c
index 9967d5a..f1b9a3a 100644
--- a/arch/avr32/mach-at32ap/at32ap700x.c
+++ b/arch/avr32/mach-at32ap/at32ap700x.c
@@ -1273,6 +1273,7 @@ struct platform_device *__init
at32_add_device_mci(unsigned int id, struct mci_platform_data *data)
{
struct platform_device *pdev;
+ struct dw_dma_slave *dws;
if (id != 0 || !data)
return NULL;
@@ -1289,6 +1290,21 @@ at32_add_device_mci(unsigned int id, struct mci_platform_data *data)
ARRAY_SIZE(atmel_mci0_resource)))
goto fail;
+ if (data->dma_slave)
+ dws = kmemdup(to_dw_dma_slave(data->dma_slave),
+ sizeof(struct dw_dma_slave), GFP_KERNEL);
+ else
+ dws = kzalloc(sizeof(struct dw_dma_slave), GFP_KERNEL);
+
+ dws->slave.dev = &pdev->dev;
+ dws->slave.dma_dev = &dw_dmac0_device.dev;
+ dws->slave.reg_width = DMA_SLAVE_WIDTH_32BIT;
+ dws->cfg_hi = (DWC_CFGH_SRC_PER(0)
+ | DWC_CFGH_DST_PER(1));
+ dws->cfg_lo &= ~(DWC_CFGL_HS_DST_POL
+ | DWC_CFGL_HS_SRC_POL);
+
+ data->dma_slave = &dws->slave;
if (platform_device_add_data(pdev, data,
sizeof(struct mci_platform_data)))
diff --git a/drivers/mmc/host/Kconfig b/drivers/mmc/host/Kconfig
index ea8d7a3..1ce21d4 100644
--- a/drivers/mmc/host/Kconfig
+++ b/drivers/mmc/host/Kconfig
@@ -114,6 +114,17 @@ config MMC_ATMELMCI
If unsure, say N.
+config MMC_ATMELMCI_DMA
+ bool "Atmel MCI DMA support (EXPERIMENTAL)"
+ depends on MMC_ATMELMCI && DMA_ENGINE && EXPERIMENTAL
+ help
+ Say Y here to have the Atmel MCI driver use a DMA engine to
+ do data transfers and thus increase the throughput and
+ reduce the CPU utilization. Note that this is highly
+ experimental and may cause the driver to lock up.
+
+ If unsure, say N.
+
config MMC_IMX
tristate "Motorola i.MX Multimedia Card Interface support"
depends on ARCH_IMX
diff --git a/drivers/mmc/host/atmel-mci.c b/drivers/mmc/host/atmel-mci.c
index a7de55f..b35ab11 100644
--- a/drivers/mmc/host/atmel-mci.c
+++ b/drivers/mmc/host/atmel-mci.c
@@ -11,6 +11,8 @@
#include <linux/clk.h>
#include <linux/debugfs.h>
#include <linux/device.h>
+#include <linux/dmaengine.h>
+#include <linux/dma-mapping.h>
#include <linux/err.h>
#include <linux/gpio.h>
#include <linux/init.h>
@@ -33,6 +35,7 @@
#include "atmel-mci-regs.h"
#define ATMCI_DATA_ERROR_FLAGS (MCI_DCRCE | MCI_DTOE | MCI_OVRE | MCI_UNRE)
+#define ATMCI_DMA_THRESHOLD 16
enum {
EVENT_CMD_COMPLETE = 0,
@@ -50,6 +53,14 @@ enum atmel_mci_state {
STATE_DATA_ERROR,
};
+struct atmel_mci_dma {
+#ifdef CONFIG_MMC_ATMELMCI_DMA
+ struct dma_client client;
+ struct dma_chan *chan;
+ struct dma_async_tx_descriptor *data_desc;
+#endif
+};
+
/**
* struct atmel_mci - MMC controller state shared between all slots
* @lock: Spinlock protecting the queue and associated data.
@@ -121,6 +132,9 @@ struct atmel_mci {
struct mmc_command *cmd;
struct mmc_data *data;
+ struct atmel_mci_dma dma;
+ struct dma_chan *data_chan;
+
u32 cmd_status;
u32 data_status;
u32 stop_cmdr;
@@ -481,6 +495,138 @@ static void send_stop_cmd(struct atmel_mci *host, struct mmc_data *data)
mci_writel(host, IER, MCI_CMDRDY);
}
+#ifdef CONFIG_MMC_ATMELMCI_DMA
+static void atmci_dma_cleanup(struct atmel_mci *host)
+{
+ struct mmc_data *data = host->data;
+
+ dma_unmap_sg(&host->pdev->dev, data->sg, data->sg_len,
+ ((data->flags & MMC_DATA_WRITE)
+ ? DMA_TO_DEVICE : DMA_FROM_DEVICE));
+}
+
+static void atmci_stop_dma(struct atmel_mci *host)
+{
+ struct dma_chan *chan = host->data_chan;
+
+ if (chan) {
+ chan->device->device_terminate_all(chan);
+ atmci_dma_cleanup(host);
+ }
+}
+
+/* This function is called by the DMA driver from tasklet context. */
+static void atmci_dma_complete(void *arg)
+{
+ struct atmel_mci *host = arg;
+ struct mmc_data *data = host->data;
+
+ dev_vdbg(&host->pdev->dev, "DMA complete\n");
+
+ atmci_dma_cleanup(host);
+
+ /*
+ * If the card was removed, data will be NULL. No point trying
+ * to send the stop command or waiting for NBUSY in this case.
+ */
+ if (data) {
+ atmci_set_pending(host, EVENT_XFER_COMPLETE);
+ tasklet_schedule(&host->tasklet);
+
+ /*
+ * Regardless of what the documentation says, we have
+ * to wait for NOTBUSY even after block read
+ * operations.
+ *
+ * When the DMA transfer is complete, the controller
+ * may still be reading the CRC from the card, i.e.
+ * the data transfer is still in progress and we
+ * haven't seen all the potential error bits yet.
+ *
+ * The interrupt handler will schedule a different
+ * tasklet to finish things up when the data transfer
+ * is completely done.
+ *
+ * We may not complete the mmc request here anyway
+ * because the mmc layer may call back and cause us to
+ * violate the "don't submit new operations from the
+ * completion callback" rule of the dma engine
+ * framework.
+ */
+ mci_writel(host, IER, MCI_NOTBUSY);
+ }
+}
+
+static int
+atmci_submit_data_dma(struct atmel_mci *host, struct mmc_data *data)
+{
+ struct dma_chan *chan;
+ struct dma_async_tx_descriptor *desc;
+ struct scatterlist *sg;
+ unsigned int i;
+ enum dma_data_direction direction;
+
+ /*
+ * We don't do DMA on "complex" transfers, i.e. with
+ * non-word-aligned buffers or lengths. Also, we don't bother
+ * with all the DMA setup overhead for short transfers.
+ */
+ if (data->blocks * data->blksz < ATMCI_DMA_THRESHOLD)
+ return -EINVAL;
+ if (data->blksz & 3)
+ return -EINVAL;
+
+ for_each_sg(data->sg, sg, data->sg_len, i) {
+ if (sg->offset & 3 || sg->length & 3)
+ return -EINVAL;
+ }
+
+ /* If we don't have a channel, we can't do DMA */
+ chan = host->dma.chan;
+ if (chan) {
+ dma_chan_get(chan);
+ host->data_chan = chan;
+ }
+
+ if (!chan)
+ return -ENODEV;
+
+ if (data->flags & MMC_DATA_READ)
+ direction = DMA_FROM_DEVICE;
+ else
+ direction = DMA_TO_DEVICE;
+
+ desc = chan->device->device_prep_slave_sg(chan,
+ data->sg, data->sg_len, direction,
+ DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
+ if (!desc)
+ return -ENOMEM;
+
+ host->dma.data_desc = desc;
+ desc->callback = atmci_dma_complete;
+ desc->callback_param = host;
+ desc->tx_submit(desc);
+
+ /* Go! */
+ chan->device->device_issue_pending(chan);
+
+ return 0;
+}
+
+#else /* CONFIG_MMC_ATMELMCI_DMA */
+
+static int atmci_submit_data_dma(struct atmel_mci *host, struct mmc_data *data)
+{
+ return -ENOSYS;
+}
+
+static void atmci_stop_dma(struct atmel_mci *host)
+{
+
+}
+
+#endif /* CONFIG_MMC_ATMELMCI_DMA */
+
/*
* Returns a mask of interrupt flags to be enabled after the whole
* request has been prepared.
@@ -496,12 +642,15 @@ static u32 atmci_submit_data(struct atmel_mci *host, struct mmc_data *data)
host->data = data;
iflags = ATMCI_DATA_ERROR_FLAGS;
- host->sg = data->sg;
- host->pio_offset = 0;
- if (data->flags & MMC_DATA_READ)
- iflags |= MCI_RXRDY;
- else
- iflags |= MCI_TXRDY;
+ if (atmci_submit_data_dma(host, data)) {
+ host->data_chan = NULL;
+ host->sg = data->sg;
+ host->pio_offset = 0;
+ if (data->flags & MMC_DATA_READ)
+ iflags |= MCI_RXRDY;
+ else
+ iflags |= MCI_TXRDY;
+ }
return iflags;
}
@@ -767,6 +916,7 @@ static void atmci_command_complete(struct atmel_mci *host,
if (cmd->data) {
host->data = NULL;
+ atmci_stop_dma(host);
mci_writel(host, IDR, MCI_NOTBUSY
| MCI_TXRDY | MCI_RXRDY
| ATMCI_DATA_ERROR_FLAGS);
@@ -834,6 +984,7 @@ static void atmci_detect_change(unsigned long data)
/* fall through */
case STATE_SENDING_DATA:
mrq->data->error = -ENOMEDIUM;
+ atmci_stop_dma(host);
break;
case STATE_DATA_BUSY:
case STATE_DATA_ERROR:
@@ -912,6 +1063,7 @@ static void atmci_tasklet_func(unsigned long priv)
case STATE_SENDING_DATA:
if (atmci_test_and_clear_pending(host,
EVENT_DATA_ERROR)) {
+ atmci_stop_dma(host);
if (data->stop)
send_stop_cmd(host, data);
state = STATE_DATA_ERROR;
@@ -1197,6 +1349,60 @@ static irqreturn_t atmci_detect_interrupt(int irq, void *dev_id)
return IRQ_HANDLED;
}
+#ifdef CONFIG_MMC_ATMELMCI_DMA
+
+static inline struct atmel_mci *
+dma_client_to_atmel_mci(struct dma_client *client)
+{
+ return container_of(client, struct atmel_mci, dma.client);
+}
+
+static enum dma_state_client atmci_dma_event(struct dma_client *client,
+ struct dma_chan *chan, enum dma_state state)
+{
+ struct atmel_mci *host;
+ enum dma_state_client ret = DMA_NAK;
+
+ host = dma_client_to_atmel_mci(client);
+
+ switch (state) {
+ case DMA_RESOURCE_AVAILABLE:
+ spin_lock_bh(&host->lock);
+ if (!host->dma.chan) {
+ host->dma.chan = chan;
+ ret = DMA_ACK;
+ }
+ spin_unlock_bh(&host->lock);
+
+ if (ret == DMA_ACK)
+ dev_info(&host->pdev->dev,
+ "Using %s for DMA transfers\n",
+ chan->dev.bus_id);
+ break;
+
+ case DMA_RESOURCE_REMOVED:
+ spin_lock_bh(&host->lock);
+ if (host->dma.chan == chan) {
+ host->dma.chan = NULL;
+ ret = DMA_ACK;
+ }
+ spin_unlock_bh(&host->lock);
+
+ if (ret == DMA_ACK)
+ dev_info(&host->pdev->dev,
+ "Lost %s, falling back to PIO\n",
+ chan->dev.bus_id);
+ break;
+
+ default:
+ break;
+ }
+
+
+ return ret;
+}
+#endif /* CONFIG_MMC_ATMELMCI_DMA */
+
static int __init atmci_init_slot(struct atmel_mci *host,
struct mci_slot_pdata *slot_data, u32 sdc_reg)
{
@@ -1350,6 +1556,25 @@ static int __init atmci_probe(struct platform_device *pdev)
if (ret)
goto err_request_irq;
+#ifdef CONFIG_MMC_ATMELMCI_DMA
+ if (pdata->dma_slave) {
+ struct dma_slave *slave = pdata->dma_slave;
+
+ slave->tx_reg = regs->start + MCI_TDR;
+ slave->rx_reg = regs->start + MCI_RDR;
+
+ /* Try to grab a DMA channel */
+ host->dma.client.event_callback = atmci_dma_event;
+ dma_cap_set(DMA_SLAVE, host->dma.client.cap_mask);
+ host->dma.client.slave = slave;
+
+ dma_async_client_register(&host->dma.client);
+ dma_async_client_chan_request(&host->dma.client);
+ } else {
+ dev_notice(&pdev->dev, "DMA not available, using PIO\n");
+ }
+#endif /* CONFIG_MMC_ATMELMCI_DMA */
+
platform_set_drvdata(pdev, host);
/* We need at least one slot to succeed */
@@ -1371,6 +1596,10 @@ static int __init atmci_probe(struct platform_device *pdev)
return 0;
err_init_slot:
+#ifdef CONFIG_MMC_ATMELMCI_DMA
+ if (pdata->dma_slave)
+ dma_async_client_unregister(&host->dma.client);
+#endif
free_irq(irq, host);
err_request_irq:
iounmap(host->regs);
@@ -1397,6 +1626,11 @@ static int __exit atmci_remove(struct platform_device *pdev)
mci_readl(host, SR);
clk_disable(host->mck);
+#ifdef CONFIG_MMC_ATMELMCI_DMA
+ if (host->dma.client.slave)
+ dma_async_client_unregister(&host->dma.client);
+#endif
+
free_irq(platform_get_irq(pdev, 0), host);
iounmap(host->regs);
--
1.5.6.5
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/