Revision 2be23c475af8ae4e25f8bab08d815b17593bd547 authored by Russell King on 08 September 2010, 15:27:56 UTC, committed by Russell King on 08 September 2010, 15:27:56 UTC
Dave Hylands reports: | We've observed a problem with dma_alloc_writecombine when the system | is under heavy load (heavy bus traffic). We've managed to reduce the | problem to the following snippet, which is run from a kthread in a | continuous loop: | | void *virtAddr; | dma_addr_t physAddr; | unsigned int numBytes = 256; | | for (;;) { | virtAddr = dma_alloc_writecombine(NULL, | numBytes, &physAddr, GFP_KERNEL); | if (virtAddr == NULL) { | printk(KERN_ERR "Running out of memory\n"); | break; | } | | /* access DMA memory allocated */ | tmp = virtAddr; | *tmp = 0x77; | | /* free DMA memory */ | dma_free_writecombine(NULL, | numBytes, virtAddr, physAddr); | | ...sleep here... | } | | By itself, the code will run forever with no issues. However, as we | increase our bus traffic (typically using DMA) then the *tmp = 0x77 | line will eventually cause a page fault. If we add a small delay (a | few microseconds) before the *tmp = 0x77, then we don't see a page | fault, even under heavy load. A dsb() is required after modifying the PTE entries to ensure that they will always be visible. Add this dsb(). Reported-by: Dave Hylands <dhylands@gmail.com> Tested-by: Dave Hylands <dhylands@gmail.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
1 parent a387f0f
orion_spi.c
/*
* orion_spi.c -- Marvell Orion SPI controller driver
*
* Author: Shadi Ammouri <shadi@marvell.com>
* Copyright (C) 2007-2008 Marvell Ltd.
*
* 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/init.h>
#include <linux/interrupt.h>
#include <linux/delay.h>
#include <linux/platform_device.h>
#include <linux/err.h>
#include <linux/io.h>
#include <linux/spi/spi.h>
#include <linux/spi/orion_spi.h>
#include <asm/unaligned.h>
#define DRIVER_NAME "orion_spi"
#define ORION_NUM_CHIPSELECTS 1 /* only one slave is supported*/
#define ORION_SPI_WAIT_RDY_MAX_LOOP 2000 /* in usec */
#define ORION_SPI_IF_CTRL_REG 0x00
#define ORION_SPI_IF_CONFIG_REG 0x04
#define ORION_SPI_DATA_OUT_REG 0x08
#define ORION_SPI_DATA_IN_REG 0x0c
#define ORION_SPI_INT_CAUSE_REG 0x10
#define ORION_SPI_IF_8_16_BIT_MODE (1 << 5)
#define ORION_SPI_CLK_PRESCALE_MASK 0x1F
struct orion_spi {
struct work_struct work;
/* Lock access to transfer list. */
spinlock_t lock;
struct list_head msg_queue;
struct spi_master *master;
void __iomem *base;
unsigned int max_speed;
unsigned int min_speed;
struct orion_spi_info *spi_info;
};
static struct workqueue_struct *orion_spi_wq;
static inline void __iomem *spi_reg(struct orion_spi *orion_spi, u32 reg)
{
return orion_spi->base + reg;
}
static inline void
orion_spi_setbits(struct orion_spi *orion_spi, u32 reg, u32 mask)
{
void __iomem *reg_addr = spi_reg(orion_spi, reg);
u32 val;
val = readl(reg_addr);
val |= mask;
writel(val, reg_addr);
}
static inline void
orion_spi_clrbits(struct orion_spi *orion_spi, u32 reg, u32 mask)
{
void __iomem *reg_addr = spi_reg(orion_spi, reg);
u32 val;
val = readl(reg_addr);
val &= ~mask;
writel(val, reg_addr);
}
static int orion_spi_set_transfer_size(struct orion_spi *orion_spi, int size)
{
if (size == 16) {
orion_spi_setbits(orion_spi, ORION_SPI_IF_CONFIG_REG,
ORION_SPI_IF_8_16_BIT_MODE);
} else if (size == 8) {
orion_spi_clrbits(orion_spi, ORION_SPI_IF_CONFIG_REG,
ORION_SPI_IF_8_16_BIT_MODE);
} else {
pr_debug("Bad bits per word value %d (only 8 or 16 are "
"allowed).\n", size);
return -EINVAL;
}
return 0;
}
static int orion_spi_baudrate_set(struct spi_device *spi, unsigned int speed)
{
u32 tclk_hz;
u32 rate;
u32 prescale;
u32 reg;
struct orion_spi *orion_spi;
orion_spi = spi_master_get_devdata(spi->master);
tclk_hz = orion_spi->spi_info->tclk;
/*
* the supported rates are: 4,6,8...30
* round up as we look for equal or less speed
*/
rate = DIV_ROUND_UP(tclk_hz, speed);
rate = roundup(rate, 2);
/* check if requested speed is too small */
if (rate > 30)
return -EINVAL;
if (rate < 4)
rate = 4;
/* Convert the rate to SPI clock divisor value. */
prescale = 0x10 + rate/2;
reg = readl(spi_reg(orion_spi, ORION_SPI_IF_CONFIG_REG));
reg = ((reg & ~ORION_SPI_CLK_PRESCALE_MASK) | prescale);
writel(reg, spi_reg(orion_spi, ORION_SPI_IF_CONFIG_REG));
return 0;
}
/*
* called only when no transfer is active on the bus
*/
static int
orion_spi_setup_transfer(struct spi_device *spi, struct spi_transfer *t)
{
struct orion_spi *orion_spi;
unsigned int speed = spi->max_speed_hz;
unsigned int bits_per_word = spi->bits_per_word;
int rc;
orion_spi = spi_master_get_devdata(spi->master);
if ((t != NULL) && t->speed_hz)
speed = t->speed_hz;
if ((t != NULL) && t->bits_per_word)
bits_per_word = t->bits_per_word;
rc = orion_spi_baudrate_set(spi, speed);
if (rc)
return rc;
return orion_spi_set_transfer_size(orion_spi, bits_per_word);
}
static void orion_spi_set_cs(struct orion_spi *orion_spi, int enable)
{
if (enable)
orion_spi_setbits(orion_spi, ORION_SPI_IF_CTRL_REG, 0x1);
else
orion_spi_clrbits(orion_spi, ORION_SPI_IF_CTRL_REG, 0x1);
}
static inline int orion_spi_wait_till_ready(struct orion_spi *orion_spi)
{
int i;
for (i = 0; i < ORION_SPI_WAIT_RDY_MAX_LOOP; i++) {
if (readl(spi_reg(orion_spi, ORION_SPI_INT_CAUSE_REG)))
return 1;
else
udelay(1);
}
return -1;
}
static inline int
orion_spi_write_read_8bit(struct spi_device *spi,
const u8 **tx_buf, u8 **rx_buf)
{
void __iomem *tx_reg, *rx_reg, *int_reg;
struct orion_spi *orion_spi;
orion_spi = spi_master_get_devdata(spi->master);
tx_reg = spi_reg(orion_spi, ORION_SPI_DATA_OUT_REG);
rx_reg = spi_reg(orion_spi, ORION_SPI_DATA_IN_REG);
int_reg = spi_reg(orion_spi, ORION_SPI_INT_CAUSE_REG);
/* clear the interrupt cause register */
writel(0x0, int_reg);
if (tx_buf && *tx_buf)
writel(*(*tx_buf)++, tx_reg);
else
writel(0, tx_reg);
if (orion_spi_wait_till_ready(orion_spi) < 0) {
dev_err(&spi->dev, "TXS timed out\n");
return -1;
}
if (rx_buf && *rx_buf)
*(*rx_buf)++ = readl(rx_reg);
return 1;
}
static inline int
orion_spi_write_read_16bit(struct spi_device *spi,
const u16 **tx_buf, u16 **rx_buf)
{
void __iomem *tx_reg, *rx_reg, *int_reg;
struct orion_spi *orion_spi;
orion_spi = spi_master_get_devdata(spi->master);
tx_reg = spi_reg(orion_spi, ORION_SPI_DATA_OUT_REG);
rx_reg = spi_reg(orion_spi, ORION_SPI_DATA_IN_REG);
int_reg = spi_reg(orion_spi, ORION_SPI_INT_CAUSE_REG);
/* clear the interrupt cause register */
writel(0x0, int_reg);
if (tx_buf && *tx_buf)
writel(__cpu_to_le16(get_unaligned((*tx_buf)++)), tx_reg);
else
writel(0, tx_reg);
if (orion_spi_wait_till_ready(orion_spi) < 0) {
dev_err(&spi->dev, "TXS timed out\n");
return -1;
}
if (rx_buf && *rx_buf)
put_unaligned(__le16_to_cpu(readl(rx_reg)), (*rx_buf)++);
return 1;
}
static unsigned int
orion_spi_write_read(struct spi_device *spi, struct spi_transfer *xfer)
{
struct orion_spi *orion_spi;
unsigned int count;
int word_len;
orion_spi = spi_master_get_devdata(spi->master);
word_len = spi->bits_per_word;
count = xfer->len;
if (word_len == 8) {
const u8 *tx = xfer->tx_buf;
u8 *rx = xfer->rx_buf;
do {
if (orion_spi_write_read_8bit(spi, &tx, &rx) < 0)
goto out;
count--;
} while (count);
} else if (word_len == 16) {
const u16 *tx = xfer->tx_buf;
u16 *rx = xfer->rx_buf;
do {
if (orion_spi_write_read_16bit(spi, &tx, &rx) < 0)
goto out;
count -= 2;
} while (count);
}
out:
return xfer->len - count;
}
static void orion_spi_work(struct work_struct *work)
{
struct orion_spi *orion_spi =
container_of(work, struct orion_spi, work);
spin_lock_irq(&orion_spi->lock);
while (!list_empty(&orion_spi->msg_queue)) {
struct spi_message *m;
struct spi_device *spi;
struct spi_transfer *t = NULL;
int par_override = 0;
int status = 0;
int cs_active = 0;
m = container_of(orion_spi->msg_queue.next, struct spi_message,
queue);
list_del_init(&m->queue);
spin_unlock_irq(&orion_spi->lock);
spi = m->spi;
/* Load defaults */
status = orion_spi_setup_transfer(spi, NULL);
if (status < 0)
goto msg_done;
list_for_each_entry(t, &m->transfers, transfer_list) {
if (par_override || t->speed_hz || t->bits_per_word) {
par_override = 1;
status = orion_spi_setup_transfer(spi, t);
if (status < 0)
break;
if (!t->speed_hz && !t->bits_per_word)
par_override = 0;
}
if (!cs_active) {
orion_spi_set_cs(orion_spi, 1);
cs_active = 1;
}
if (t->len)
m->actual_length +=
orion_spi_write_read(spi, t);
if (t->delay_usecs)
udelay(t->delay_usecs);
if (t->cs_change) {
orion_spi_set_cs(orion_spi, 0);
cs_active = 0;
}
}
msg_done:
if (cs_active)
orion_spi_set_cs(orion_spi, 0);
m->status = status;
m->complete(m->context);
spin_lock_irq(&orion_spi->lock);
}
spin_unlock_irq(&orion_spi->lock);
}
static int __init orion_spi_reset(struct orion_spi *orion_spi)
{
/* Verify that the CS is deasserted */
orion_spi_set_cs(orion_spi, 0);
return 0;
}
static int orion_spi_setup(struct spi_device *spi)
{
struct orion_spi *orion_spi;
orion_spi = spi_master_get_devdata(spi->master);
/* Fix ac timing if required. */
if (orion_spi->spi_info->enable_clock_fix)
orion_spi_setbits(orion_spi, ORION_SPI_IF_CONFIG_REG,
(1 << 14));
if ((spi->max_speed_hz == 0)
|| (spi->max_speed_hz > orion_spi->max_speed))
spi->max_speed_hz = orion_spi->max_speed;
if (spi->max_speed_hz < orion_spi->min_speed) {
dev_err(&spi->dev, "setup: requested speed too low %d Hz\n",
spi->max_speed_hz);
return -EINVAL;
}
/*
* baudrate & width will be set orion_spi_setup_transfer
*/
return 0;
}
static int orion_spi_transfer(struct spi_device *spi, struct spi_message *m)
{
struct orion_spi *orion_spi;
struct spi_transfer *t = NULL;
unsigned long flags;
m->actual_length = 0;
m->status = 0;
/* reject invalid messages and transfers */
if (list_empty(&m->transfers) || !m->complete)
return -EINVAL;
orion_spi = spi_master_get_devdata(spi->master);
list_for_each_entry(t, &m->transfers, transfer_list) {
unsigned int bits_per_word = spi->bits_per_word;
if (t->tx_buf == NULL && t->rx_buf == NULL && t->len) {
dev_err(&spi->dev,
"message rejected : "
"invalid transfer data buffers\n");
goto msg_rejected;
}
if ((t != NULL) && t->bits_per_word)
bits_per_word = t->bits_per_word;
if ((bits_per_word != 8) && (bits_per_word != 16)) {
dev_err(&spi->dev,
"message rejected : "
"invalid transfer bits_per_word (%d bits)\n",
bits_per_word);
goto msg_rejected;
}
/*make sure buffer length is even when working in 16 bit mode*/
if ((t != NULL) && (t->bits_per_word == 16) && (t->len & 1)) {
dev_err(&spi->dev,
"message rejected : "
"odd data length (%d) while in 16 bit mode\n",
t->len);
goto msg_rejected;
}
if (t->speed_hz && t->speed_hz < orion_spi->min_speed) {
dev_err(&spi->dev,
"message rejected : "
"device min speed (%d Hz) exceeds "
"required transfer speed (%d Hz)\n",
orion_spi->min_speed, t->speed_hz);
goto msg_rejected;
}
}
spin_lock_irqsave(&orion_spi->lock, flags);
list_add_tail(&m->queue, &orion_spi->msg_queue);
queue_work(orion_spi_wq, &orion_spi->work);
spin_unlock_irqrestore(&orion_spi->lock, flags);
return 0;
msg_rejected:
/* Message rejected and not queued */
m->status = -EINVAL;
if (m->complete)
m->complete(m->context);
return -EINVAL;
}
static int __init orion_spi_probe(struct platform_device *pdev)
{
struct spi_master *master;
struct orion_spi *spi;
struct resource *r;
struct orion_spi_info *spi_info;
int status = 0;
spi_info = pdev->dev.platform_data;
master = spi_alloc_master(&pdev->dev, sizeof *spi);
if (master == NULL) {
dev_dbg(&pdev->dev, "master allocation failed\n");
return -ENOMEM;
}
if (pdev->id != -1)
master->bus_num = pdev->id;
/* we support only mode 0, and no options */
master->mode_bits = 0;
master->setup = orion_spi_setup;
master->transfer = orion_spi_transfer;
master->num_chipselect = ORION_NUM_CHIPSELECTS;
dev_set_drvdata(&pdev->dev, master);
spi = spi_master_get_devdata(master);
spi->master = master;
spi->spi_info = spi_info;
spi->max_speed = DIV_ROUND_UP(spi_info->tclk, 4);
spi->min_speed = DIV_ROUND_UP(spi_info->tclk, 30);
r = platform_get_resource(pdev, IORESOURCE_MEM, 0);
if (r == NULL) {
status = -ENODEV;
goto out;
}
if (!request_mem_region(r->start, (r->end - r->start) + 1,
dev_name(&pdev->dev))) {
status = -EBUSY;
goto out;
}
spi->base = ioremap(r->start, SZ_1K);
INIT_WORK(&spi->work, orion_spi_work);
spin_lock_init(&spi->lock);
INIT_LIST_HEAD(&spi->msg_queue);
if (orion_spi_reset(spi) < 0)
goto out_rel_mem;
status = spi_register_master(master);
if (status < 0)
goto out_rel_mem;
return status;
out_rel_mem:
release_mem_region(r->start, (r->end - r->start) + 1);
out:
spi_master_put(master);
return status;
}
static int __exit orion_spi_remove(struct platform_device *pdev)
{
struct spi_master *master;
struct orion_spi *spi;
struct resource *r;
master = dev_get_drvdata(&pdev->dev);
spi = spi_master_get_devdata(master);
cancel_work_sync(&spi->work);
r = platform_get_resource(pdev, IORESOURCE_MEM, 0);
release_mem_region(r->start, (r->end - r->start) + 1);
spi_unregister_master(master);
return 0;
}
MODULE_ALIAS("platform:" DRIVER_NAME);
static struct platform_driver orion_spi_driver = {
.driver = {
.name = DRIVER_NAME,
.owner = THIS_MODULE,
},
.remove = __exit_p(orion_spi_remove),
};
static int __init orion_spi_init(void)
{
orion_spi_wq = create_singlethread_workqueue(
orion_spi_driver.driver.name);
if (orion_spi_wq == NULL)
return -ENOMEM;
return platform_driver_probe(&orion_spi_driver, orion_spi_probe);
}
module_init(orion_spi_init);
static void __exit orion_spi_exit(void)
{
flush_workqueue(orion_spi_wq);
platform_driver_unregister(&orion_spi_driver);
destroy_workqueue(orion_spi_wq);
}
module_exit(orion_spi_exit);
MODULE_DESCRIPTION("Orion SPI driver");
MODULE_AUTHOR("Shadi Ammouri <shadi@marvell.com>");
MODULE_LICENSE("GPL");
![swh spinner](/static/img/swh-spinner.gif)
Computing file changes ...