linux/arch/blackfin/kernel/bfin_dma_5xx.c
Michael Hennerich d1401e1dc2 Blackfin: fix DMA/cache bug when resuming from suspend to RAM
The dma_memcpy() function takes care of flushing different caches for us.
Normally this is what we want, but when resuming from mem, we don't yet
have caches enabled.  If these functions happen to be placed into L1 mem
(which is what we're trying to relocate), then things aren't going to
work.  So define a non-cache dma_memcpy() variant to utilize in situations
like this.

Signed-off-by: Michael Hennerich <michael.hennerich@analog.com>
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
2010-08-06 12:55:50 -04:00

525 lines
14 KiB
C

/*
* bfin_dma_5xx.c - Blackfin DMA implementation
*
* Copyright 2004-2008 Analog Devices Inc.
*
* Licensed under the GPL-2 or later.
*/
#include <linux/errno.h>
#include <linux/interrupt.h>
#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/param.h>
#include <linux/proc_fs.h>
#include <linux/sched.h>
#include <linux/seq_file.h>
#include <linux/spinlock.h>
#include <asm/blackfin.h>
#include <asm/cacheflush.h>
#include <asm/dma.h>
#include <asm/uaccess.h>
#include <asm/early_printk.h>
/*
* To make sure we work around 05000119 - we always check DMA_DONE bit,
* never the DMA_RUN bit
*/
struct dma_channel dma_ch[MAX_DMA_CHANNELS];
EXPORT_SYMBOL(dma_ch);
static int __init blackfin_dma_init(void)
{
int i;
printk(KERN_INFO "Blackfin DMA Controller\n");
for (i = 0; i < MAX_DMA_CHANNELS; i++) {
atomic_set(&dma_ch[i].chan_status, 0);
dma_ch[i].regs = dma_io_base_addr[i];
}
/* Mark MEMDMA Channel 0 as requested since we're using it internally */
request_dma(CH_MEM_STREAM0_DEST, "Blackfin dma_memcpy");
request_dma(CH_MEM_STREAM0_SRC, "Blackfin dma_memcpy");
#if defined(CONFIG_DEB_DMA_URGENT)
bfin_write_EBIU_DDRQUE(bfin_read_EBIU_DDRQUE()
| DEB1_URGENT | DEB2_URGENT | DEB3_URGENT);
#endif
return 0;
}
arch_initcall(blackfin_dma_init);
#ifdef CONFIG_PROC_FS
static int proc_dma_show(struct seq_file *m, void *v)
{
int i;
for (i = 0; i < MAX_DMA_CHANNELS; ++i)
if (dma_channel_active(i))
seq_printf(m, "%2d: %s\n", i, dma_ch[i].device_id);
return 0;
}
static int proc_dma_open(struct inode *inode, struct file *file)
{
return single_open(file, proc_dma_show, NULL);
}
static const struct file_operations proc_dma_operations = {
.open = proc_dma_open,
.read = seq_read,
.llseek = seq_lseek,
.release = single_release,
};
static int __init proc_dma_init(void)
{
return proc_create("dma", 0, NULL, &proc_dma_operations) != NULL;
}
late_initcall(proc_dma_init);
#endif
/**
* request_dma - request a DMA channel
*
* Request the specific DMA channel from the system if it's available.
*/
int request_dma(unsigned int channel, const char *device_id)
{
pr_debug("request_dma() : BEGIN\n");
if (device_id == NULL)
printk(KERN_WARNING "request_dma(%u): no device_id given\n", channel);
#if defined(CONFIG_BF561) && ANOMALY_05000182
if (channel >= CH_IMEM_STREAM0_DEST && channel <= CH_IMEM_STREAM1_DEST) {
if (get_cclk() > 500000000) {
printk(KERN_WARNING
"Request IMDMA failed due to ANOMALY 05000182\n");
return -EFAULT;
}
}
#endif
if (atomic_cmpxchg(&dma_ch[channel].chan_status, 0, 1)) {
pr_debug("DMA CHANNEL IN USE\n");
return -EBUSY;
}
#ifdef CONFIG_BF54x
if (channel >= CH_UART2_RX && channel <= CH_UART3_TX) {
unsigned int per_map;
per_map = dma_ch[channel].regs->peripheral_map & 0xFFF;
if (strncmp(device_id, "BFIN_UART", 9) == 0)
dma_ch[channel].regs->peripheral_map = per_map |
((channel - CH_UART2_RX + 0xC)<<12);
else
dma_ch[channel].regs->peripheral_map = per_map |
((channel - CH_UART2_RX + 0x6)<<12);
}
#endif
dma_ch[channel].device_id = device_id;
dma_ch[channel].irq = 0;
/* This is to be enabled by putting a restriction -
* you have to request DMA, before doing any operations on
* descriptor/channel
*/
pr_debug("request_dma() : END\n");
return 0;
}
EXPORT_SYMBOL(request_dma);
int set_dma_callback(unsigned int channel, irq_handler_t callback, void *data)
{
int ret;
unsigned int irq;
BUG_ON(channel >= MAX_DMA_CHANNELS || !callback ||
!atomic_read(&dma_ch[channel].chan_status));
irq = channel2irq(channel);
ret = request_irq(irq, callback, 0, dma_ch[channel].device_id, data);
if (ret)
return ret;
dma_ch[channel].irq = irq;
dma_ch[channel].data = data;
return 0;
}
EXPORT_SYMBOL(set_dma_callback);
/**
* clear_dma_buffer - clear DMA fifos for specified channel
*
* Set the Buffer Clear bit in the Configuration register of specific DMA
* channel. This will stop the descriptor based DMA operation.
*/
static void clear_dma_buffer(unsigned int channel)
{
dma_ch[channel].regs->cfg |= RESTART;
SSYNC();
dma_ch[channel].regs->cfg &= ~RESTART;
}
void free_dma(unsigned int channel)
{
pr_debug("freedma() : BEGIN\n");
BUG_ON(channel >= MAX_DMA_CHANNELS ||
!atomic_read(&dma_ch[channel].chan_status));
/* Halt the DMA */
disable_dma(channel);
clear_dma_buffer(channel);
if (dma_ch[channel].irq)
free_irq(dma_ch[channel].irq, dma_ch[channel].data);
/* Clear the DMA Variable in the Channel */
atomic_set(&dma_ch[channel].chan_status, 0);
pr_debug("freedma() : END\n");
}
EXPORT_SYMBOL(free_dma);
#ifdef CONFIG_PM
# ifndef MAX_DMA_SUSPEND_CHANNELS
# define MAX_DMA_SUSPEND_CHANNELS MAX_DMA_CHANNELS
# endif
int blackfin_dma_suspend(void)
{
int i;
for (i = 0; i < MAX_DMA_CHANNELS; ++i) {
if (dma_ch[i].regs->cfg & DMAEN) {
printk(KERN_ERR "DMA Channel %d failed to suspend\n", i);
return -EBUSY;
}
if (i < MAX_DMA_SUSPEND_CHANNELS)
dma_ch[i].saved_peripheral_map = dma_ch[i].regs->peripheral_map;
}
return 0;
}
void blackfin_dma_resume(void)
{
int i;
for (i = 0; i < MAX_DMA_CHANNELS; ++i) {
dma_ch[i].regs->cfg = 0;
if (i < MAX_DMA_SUSPEND_CHANNELS)
dma_ch[i].regs->peripheral_map = dma_ch[i].saved_peripheral_map;
}
}
#endif
/**
* blackfin_dma_early_init - minimal DMA init
*
* Setup a few DMA registers so we can safely do DMA transfers early on in
* the kernel booting process. Really this just means using dma_memcpy().
*/
void __init blackfin_dma_early_init(void)
{
early_shadow_stamp();
bfin_write_MDMA_S0_CONFIG(0);
bfin_write_MDMA_S1_CONFIG(0);
}
void __init early_dma_memcpy(void *pdst, const void *psrc, size_t size)
{
unsigned long dst = (unsigned long)pdst;
unsigned long src = (unsigned long)psrc;
struct dma_register *dst_ch, *src_ch;
early_shadow_stamp();
/* We assume that everything is 4 byte aligned, so include
* a basic sanity check
*/
BUG_ON(dst % 4);
BUG_ON(src % 4);
BUG_ON(size % 4);
src_ch = 0;
/* Find an avalible memDMA channel */
while (1) {
if (src_ch == (struct dma_register *)MDMA_S0_NEXT_DESC_PTR) {
dst_ch = (struct dma_register *)MDMA_D1_NEXT_DESC_PTR;
src_ch = (struct dma_register *)MDMA_S1_NEXT_DESC_PTR;
} else {
dst_ch = (struct dma_register *)MDMA_D0_NEXT_DESC_PTR;
src_ch = (struct dma_register *)MDMA_S0_NEXT_DESC_PTR;
}
if (!bfin_read16(&src_ch->cfg))
break;
else if (bfin_read16(&dst_ch->irq_status) & DMA_DONE) {
bfin_write16(&src_ch->cfg, 0);
break;
}
}
/* Force a sync in case a previous config reset on this channel
* occurred. This is needed so subsequent writes to DMA registers
* are not spuriously lost/corrupted.
*/
__builtin_bfin_ssync();
/* Destination */
bfin_write32(&dst_ch->start_addr, dst);
bfin_write16(&dst_ch->x_count, size >> 2);
bfin_write16(&dst_ch->x_modify, 1 << 2);
bfin_write16(&dst_ch->irq_status, DMA_DONE | DMA_ERR);
/* Source */
bfin_write32(&src_ch->start_addr, src);
bfin_write16(&src_ch->x_count, size >> 2);
bfin_write16(&src_ch->x_modify, 1 << 2);
bfin_write16(&src_ch->irq_status, DMA_DONE | DMA_ERR);
/* Enable */
bfin_write16(&src_ch->cfg, DMAEN | WDSIZE_32);
bfin_write16(&dst_ch->cfg, WNR | DI_EN | DMAEN | WDSIZE_32);
/* Since we are atomic now, don't use the workaround ssync */
__builtin_bfin_ssync();
}
void __init early_dma_memcpy_done(void)
{
early_shadow_stamp();
while ((bfin_read_MDMA_S0_CONFIG() && !(bfin_read_MDMA_D0_IRQ_STATUS() & DMA_DONE)) ||
(bfin_read_MDMA_S1_CONFIG() && !(bfin_read_MDMA_D1_IRQ_STATUS() & DMA_DONE)))
continue;
bfin_write_MDMA_D0_IRQ_STATUS(DMA_DONE | DMA_ERR);
bfin_write_MDMA_D1_IRQ_STATUS(DMA_DONE | DMA_ERR);
/*
* Now that DMA is done, we would normally flush cache, but
* i/d cache isn't running this early, so we don't bother,
* and just clear out the DMA channel for next time
*/
bfin_write_MDMA_S0_CONFIG(0);
bfin_write_MDMA_S1_CONFIG(0);
bfin_write_MDMA_D0_CONFIG(0);
bfin_write_MDMA_D1_CONFIG(0);
__builtin_bfin_ssync();
}
/**
* __dma_memcpy - program the MDMA registers
*
* Actually program MDMA0 and wait for the transfer to finish. Disable IRQs
* while programming registers so that everything is fully configured. Wait
* for DMA to finish with IRQs enabled. If interrupted, the initial DMA_DONE
* check will make sure we don't clobber any existing transfer.
*/
static void __dma_memcpy(u32 daddr, s16 dmod, u32 saddr, s16 smod, size_t cnt, u32 conf)
{
static DEFINE_SPINLOCK(mdma_lock);
unsigned long flags;
spin_lock_irqsave(&mdma_lock, flags);
/* Force a sync in case a previous config reset on this channel
* occurred. This is needed so subsequent writes to DMA registers
* are not spuriously lost/corrupted. Do it under irq lock and
* without the anomaly version (because we are atomic already).
*/
__builtin_bfin_ssync();
if (bfin_read_MDMA_S0_CONFIG())
while (!(bfin_read_MDMA_D0_IRQ_STATUS() & DMA_DONE))
continue;
if (conf & DMA2D) {
/* For larger bit sizes, we've already divided down cnt so it
* is no longer a multiple of 64k. So we have to break down
* the limit here so it is a multiple of the incoming size.
* There is no limitation here in terms of total size other
* than the hardware though as the bits lost in the shift are
* made up by MODIFY (== we can hit the whole address space).
* X: (2^(16 - 0)) * 1 == (2^(16 - 1)) * 2 == (2^(16 - 2)) * 4
*/
u32 shift = abs(dmod) >> 1;
size_t ycnt = cnt >> (16 - shift);
cnt = 1 << (16 - shift);
bfin_write_MDMA_D0_Y_COUNT(ycnt);
bfin_write_MDMA_S0_Y_COUNT(ycnt);
bfin_write_MDMA_D0_Y_MODIFY(dmod);
bfin_write_MDMA_S0_Y_MODIFY(smod);
}
bfin_write_MDMA_D0_START_ADDR(daddr);
bfin_write_MDMA_D0_X_COUNT(cnt);
bfin_write_MDMA_D0_X_MODIFY(dmod);
bfin_write_MDMA_D0_IRQ_STATUS(DMA_DONE | DMA_ERR);
bfin_write_MDMA_S0_START_ADDR(saddr);
bfin_write_MDMA_S0_X_COUNT(cnt);
bfin_write_MDMA_S0_X_MODIFY(smod);
bfin_write_MDMA_S0_IRQ_STATUS(DMA_DONE | DMA_ERR);
bfin_write_MDMA_S0_CONFIG(DMAEN | conf);
bfin_write_MDMA_D0_CONFIG(WNR | DI_EN | DMAEN | conf);
spin_unlock_irqrestore(&mdma_lock, flags);
SSYNC();
while (!(bfin_read_MDMA_D0_IRQ_STATUS() & DMA_DONE))
if (bfin_read_MDMA_S0_CONFIG())
continue;
else
return;
bfin_write_MDMA_D0_IRQ_STATUS(DMA_DONE | DMA_ERR);
bfin_write_MDMA_S0_CONFIG(0);
bfin_write_MDMA_D0_CONFIG(0);
}
/**
* _dma_memcpy - translate C memcpy settings into MDMA settings
*
* Handle all the high level steps before we touch the MDMA registers. So
* handle direction, tweaking of sizes, and formatting of addresses.
*/
static void *_dma_memcpy(void *pdst, const void *psrc, size_t size)
{
u32 conf, shift;
s16 mod;
unsigned long dst = (unsigned long)pdst;
unsigned long src = (unsigned long)psrc;
if (size == 0)
return NULL;
if (dst % 4 == 0 && src % 4 == 0 && size % 4 == 0) {
conf = WDSIZE_32;
shift = 2;
} else if (dst % 2 == 0 && src % 2 == 0 && size % 2 == 0) {
conf = WDSIZE_16;
shift = 1;
} else {
conf = WDSIZE_8;
shift = 0;
}
/* If the two memory regions have a chance of overlapping, make
* sure the memcpy still works as expected. Do this by having the
* copy run backwards instead.
*/
mod = 1 << shift;
if (src < dst) {
mod *= -1;
dst += size + mod;
src += size + mod;
}
size >>= shift;
if (size > 0x10000)
conf |= DMA2D;
__dma_memcpy(dst, mod, src, mod, size, conf);
return pdst;
}
/**
* dma_memcpy - DMA memcpy under mutex lock
*
* Do not check arguments before starting the DMA memcpy. Break the transfer
* up into two pieces. The first transfer is in multiples of 64k and the
* second transfer is the piece smaller than 64k.
*/
void *dma_memcpy(void *pdst, const void *psrc, size_t size)
{
unsigned long dst = (unsigned long)pdst;
unsigned long src = (unsigned long)psrc;
if (bfin_addr_dcacheable(src))
blackfin_dcache_flush_range(src, src + size);
if (bfin_addr_dcacheable(dst))
blackfin_dcache_invalidate_range(dst, dst + size);
return dma_memcpy_nocache(pdst, psrc, size);
}
EXPORT_SYMBOL(dma_memcpy);
/**
* dma_memcpy_nocache - DMA memcpy under mutex lock
* - No cache flush/invalidate
*
* Do not check arguments before starting the DMA memcpy. Break the transfer
* up into two pieces. The first transfer is in multiples of 64k and the
* second transfer is the piece smaller than 64k.
*/
void *dma_memcpy_nocache(void *pdst, const void *psrc, size_t size)
{
size_t bulk, rest;
bulk = size & ~0xffff;
rest = size - bulk;
if (bulk)
_dma_memcpy(pdst, psrc, bulk);
_dma_memcpy(pdst + bulk, psrc + bulk, rest);
return pdst;
}
EXPORT_SYMBOL(dma_memcpy_nocache);
/**
* safe_dma_memcpy - DMA memcpy w/argument checking
*
* Verify arguments are safe before heading to dma_memcpy().
*/
void *safe_dma_memcpy(void *dst, const void *src, size_t size)
{
if (!access_ok(VERIFY_WRITE, dst, size))
return NULL;
if (!access_ok(VERIFY_READ, src, size))
return NULL;
return dma_memcpy(dst, src, size);
}
EXPORT_SYMBOL(safe_dma_memcpy);
static void _dma_out(unsigned long addr, unsigned long buf, unsigned short len,
u16 size, u16 dma_size)
{
blackfin_dcache_flush_range(buf, buf + len * size);
__dma_memcpy(addr, 0, buf, size, len, dma_size);
}
static void _dma_in(unsigned long addr, unsigned long buf, unsigned short len,
u16 size, u16 dma_size)
{
blackfin_dcache_invalidate_range(buf, buf + len * size);
__dma_memcpy(buf, size, addr, 0, len, dma_size);
}
#define MAKE_DMA_IO(io, bwl, isize, dmasize, cnst) \
void dma_##io##s##bwl(unsigned long addr, cnst void *buf, unsigned short len) \
{ \
_dma_##io(addr, (unsigned long)buf, len, isize, WDSIZE_##dmasize); \
} \
EXPORT_SYMBOL(dma_##io##s##bwl)
MAKE_DMA_IO(out, b, 1, 8, const);
MAKE_DMA_IO(in, b, 1, 8, );
MAKE_DMA_IO(out, w, 2, 16, const);
MAKE_DMA_IO(in, w, 2, 16, );
MAKE_DMA_IO(out, l, 4, 32, const);
MAKE_DMA_IO(in, l, 4, 32, );