[s3c24xx] glamo-mmc: Limit clock rate.

git-svn-id: svn://svn.openwrt.org/openwrt/trunk@17047 3c298f89-4303-0410-b956-a3cf2f4a3e73
master
Lars-Peter Clausen 2009-07-30 09:07:59 +00:00
parent 3ae7d9a0f9
commit 7b7f645d26
1 changed files with 89 additions and 15 deletions

View File

@ -18,7 +18,6 @@
#include <linux/irq.h>
#include <linux/delay.h>
#include <linux/interrupt.h>
#include <linux/spinlock.h>
#include <linux/workqueue.h>
#include <linux/crc7.h>
#include <linux/scatterlist.h>
@ -53,6 +52,7 @@ struct glamo_mci_host {
struct timer_list disable_timer;
struct work_struct irq_work;
struct work_struct read_work;
unsigned clk_enabled : 1;
};
@ -74,7 +74,7 @@ static void glamo_mci_send_command(struct glamo_mci_host *host,
* for example
*/
static int sd_max_clk = 50000000;
static int sd_max_clk = 21000000;
module_param(sd_max_clk, int, 0644);
/*
@ -216,7 +216,6 @@ static int glamo_mci_set_card_clock(struct glamo_mci_host *host, int freq)
static void glamo_mci_request_done(struct glamo_mci_host *host, struct
mmc_request *mrq) {
mod_timer(&host->disable_timer, jiffies + HZ / 16);
mmc_request_done(host->mmc, mrq);
}
@ -227,12 +226,17 @@ static void glamo_mci_irq_worker(struct work_struct *work)
irq_work);
struct mmc_command *cmd;
uint16_t status;
if (!host->mrq || !host->mrq->cmd)
return;
cmd = host->mrq->cmd;
#if 0
if (cmd->data->flags & MMC_DATA_READ) {
return;
}
#endif
status = glamo_reg_read(host, GLAMO_REG_MMC_RB_STAT1);
dev_dbg(&host->pdev->dev, "status = 0x%04x\n", status);
@ -240,20 +244,20 @@ static void glamo_mci_irq_worker(struct work_struct *work)
if (status & GLAMO_STAT1_MMC_RB_DRDY)
status &= ~GLAMO_STAT1_MMC_DTOUT;
if (status & (GLAMO_STAT1_MMC_RTOUT |
GLAMO_STAT1_MMC_DTOUT))
if (status & (GLAMO_STAT1_MMC_RTOUT | GLAMO_STAT1_MMC_DTOUT))
cmd->error = -ETIMEDOUT;
if (status & (GLAMO_STAT1_MMC_BWERR |
GLAMO_STAT1_MMC_BRERR))
if (status & (GLAMO_STAT1_MMC_BWERR | GLAMO_STAT1_MMC_BRERR)) {
cmd->error = -EILSEQ;
}
if (cmd->error) {
dev_info(&host->pdev->dev, "Error after cmd: 0x%x\n", status);
goto done;
}
/* issue STOP if we have been given one to use */
if (host->mrq->stop)
if (host->mrq->stop) {
glamo_mci_send_command(host, host->mrq->stop);
}
if (cmd->data->flags & MMC_DATA_READ)
do_pio_read(host, cmd->data);
@ -263,10 +267,70 @@ done:
glamo_mci_request_done(host, cmd->mrq);
}
static void glamo_mci_read_worker(struct work_struct *work)
{
struct glamo_mci_host *host = container_of(work, struct glamo_mci_host,
read_work);
struct mmc_command *cmd;
uint16_t status;
uint16_t blocks_ready;
size_t data_read = 0;
size_t data_ready;
struct scatterlist *sg;
u16 __iomem *from_ptr = host->data_base;
void *sg_pointer;
cmd = host->mrq->cmd;
sg = cmd->data->sg;
do {
status = glamo_reg_read(host, GLAMO_REG_MMC_RB_STAT1);
if (status & (GLAMO_STAT1_MMC_RTOUT | GLAMO_STAT1_MMC_DTOUT))
cmd->error = -ETIMEDOUT;
if (status & (GLAMO_STAT1_MMC_BWERR | GLAMO_STAT1_MMC_BRERR))
cmd->error = -EILSEQ;
if (cmd->error) {
dev_info(&host->pdev->dev, "Error after cmd: 0x%x\n", status);
goto done;
}
blocks_ready = glamo_reg_read(host, GLAMO_REG_MMC_RB_BLKCNT);
data_ready = blocks_ready * cmd->data->blksz;
if (data_ready == data_read)
yield();
while(sg && data_read + sg->length <= data_ready) {
sg_pointer = page_address(sg_page(sg)) + sg->offset;
memcpy(sg_pointer, from_ptr, sg->length);
from_ptr += sg->length >> 1;
data_read += sg->length;
sg = sg_next(sg);
}
} while(sg);
cmd->data->bytes_xfered = data_read;
do {
status = glamo_reg_read(host, GLAMO_REG_MMC_RB_STAT1);
} while (!(status & GLAMO_STAT1_MMC_IDLE));
if (host->mrq->stop)
glamo_mci_send_command(host, host->mrq->stop);
do {
status = glamo_reg_read(host, GLAMO_REG_MMC_RB_STAT1);
} while (!(status & GLAMO_STAT1_MMC_IDLE));
done:
host->mrq = NULL;
glamo_mci_request_done(host, cmd->mrq);
}
static irqreturn_t glamo_mci_irq(int irq, void *devid)
{
struct glamo_mci_host *host = (struct glamo_mci_host*)devid;
schedule_work(&host->irq_work);
return IRQ_HANDLED;
@ -391,7 +455,7 @@ static void glamo_mci_send_command(struct glamo_mci_host *host,
break;
}
if (triggers_int)
if (cmd->data)
host->mrq = cmd->mrq;
/* always largest timeout */
@ -430,8 +494,7 @@ static void glamo_mci_send_command(struct glamo_mci_host *host,
GLAMO_STAT1_MMC_DTOUT)) ||
(timeout == 0)) {
cmd->error = -ETIMEDOUT;
} else if (status & (GLAMO_STAT1_MMC_BWERR |
GLAMO_STAT1_MMC_BRERR)) {
} else if (status & (GLAMO_STAT1_MMC_BWERR | GLAMO_STAT1_MMC_BRERR)) {
cmd->error = -EILSEQ;
}
@ -451,6 +514,15 @@ static void glamo_mci_send_command(struct glamo_mci_host *host,
((readw(&reg_resp[2])) << 24);
}
}
#if 0
/* We'll only get an interrupt when all data has been transfered.
By starting to copy data when it's avaiable we can increase throughput by
up to 30%. */
if (cmd->data && (cmd->data->flags & MMC_DATA_READ))
schedule_work(&host->read_work);
#endif
}
static int glamo_mci_prepare_pio(struct glamo_mci_host *host,
@ -513,9 +585,11 @@ static void glamo_mci_send_request(struct mmc_host *mmc, struct mmc_request *mrq
struct glamo_mci_host *host = mmc_priv(mmc);
struct mmc_command *cmd = mrq->cmd;
glamo_mci_clock_enable(host);
host->request_counter++;
if (cmd->data) {
if(glamo_mci_prepare_pio(host, cmd->data)) {
cmd->error = -EIO;
cmd->data->error = -EIO;
goto done;
}
@ -526,7 +600,6 @@ static void glamo_mci_send_request(struct mmc_host *mmc, struct mmc_request *mrq
cmd->opcode, cmd->arg, cmd->data, cmd->mrq->stop,
cmd->flags);
glamo_mci_clock_enable(host);
glamo_mci_send_command(host, cmd);
/*
@ -664,6 +737,7 @@ static int glamo_mci_probe(struct platform_device *pdev)
host->clk_enabled = 0;
INIT_WORK(&host->irq_work, glamo_mci_irq_worker);
INIT_WORK(&host->read_work, glamo_mci_read_worker);
host->regulator = regulator_get(pdev->dev.parent, "SD_3V3");
if (!host->regulator) {
@ -744,7 +818,7 @@ static int glamo_mci_probe(struct platform_device *pdev)
MMC_CAP_MMC_HIGHSPEED |
MMC_CAP_SD_HIGHSPEED;
mmc->f_min = host->clk_rate / 256;
mmc->f_max = host->clk_rate;
mmc->f_max = sd_max_clk;
mmc->max_blk_count = (1 << 16) - 1; /* GLAMO_REG_MMC_RB_BLKCNT */
mmc->max_blk_size = (1 << 12) - 1; /* GLAMO_REG_MMC_RB_BLKLEN */