1
linux/drivers/firewire/core-iso.c
Clemens Ladisch 13882a82ee firewire: optimize iso queueing by setting wake only after the last packet
When queueing iso packets, the run time is dominated by the two
MMIO accesses that set the DMA context's wake bit.  Because most
drivers submit packets in batches, we can save much time by
removing all but the last wakeup.

The internal kernel API is changed to require a call to
fw_iso_context_queue_flush() after a batch of queued packets.
The user space API does not change, so one call to
FW_CDEV_IOC_QUEUE_ISO must specify multiple packets to take
advantage of this optimization.

In my measurements, this patch reduces the time needed to queue
fifty skip packets from userspace to one sixth on a 2.5 GHz CPU,
or to one third at 800 MHz.

Signed-off-by: Clemens Ladisch <clemens@ladisch.de>
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2011-05-10 22:53:45 +02:00

373 lines
9.6 KiB
C

/*
* Isochronous I/O functionality:
* - Isochronous DMA context management
* - Isochronous bus resource management (channels, bandwidth), client side
*
* Copyright (C) 2006 Kristian Hoegsberg <krh@bitplanet.net>
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software Foundation,
* Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
*/
#include <linux/dma-mapping.h>
#include <linux/errno.h>
#include <linux/firewire.h>
#include <linux/firewire-constants.h>
#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/slab.h>
#include <linux/spinlock.h>
#include <linux/vmalloc.h>
#include <asm/byteorder.h>
#include "core.h"
/*
* Isochronous DMA context management
*/
int fw_iso_buffer_init(struct fw_iso_buffer *buffer, struct fw_card *card,
int page_count, enum dma_data_direction direction)
{
int i, j;
dma_addr_t address;
buffer->page_count = page_count;
buffer->direction = direction;
buffer->pages = kmalloc(page_count * sizeof(buffer->pages[0]),
GFP_KERNEL);
if (buffer->pages == NULL)
goto out;
for (i = 0; i < buffer->page_count; i++) {
buffer->pages[i] = alloc_page(GFP_KERNEL | GFP_DMA32 | __GFP_ZERO);
if (buffer->pages[i] == NULL)
goto out_pages;
address = dma_map_page(card->device, buffer->pages[i],
0, PAGE_SIZE, direction);
if (dma_mapping_error(card->device, address)) {
__free_page(buffer->pages[i]);
goto out_pages;
}
set_page_private(buffer->pages[i], address);
}
return 0;
out_pages:
for (j = 0; j < i; j++) {
address = page_private(buffer->pages[j]);
dma_unmap_page(card->device, address,
PAGE_SIZE, direction);
__free_page(buffer->pages[j]);
}
kfree(buffer->pages);
out:
buffer->pages = NULL;
return -ENOMEM;
}
EXPORT_SYMBOL(fw_iso_buffer_init);
int fw_iso_buffer_map(struct fw_iso_buffer *buffer, struct vm_area_struct *vma)
{
unsigned long uaddr;
int i, err;
uaddr = vma->vm_start;
for (i = 0; i < buffer->page_count; i++) {
err = vm_insert_page(vma, uaddr, buffer->pages[i]);
if (err)
return err;
uaddr += PAGE_SIZE;
}
return 0;
}
void fw_iso_buffer_destroy(struct fw_iso_buffer *buffer,
struct fw_card *card)
{
int i;
dma_addr_t address;
for (i = 0; i < buffer->page_count; i++) {
address = page_private(buffer->pages[i]);
dma_unmap_page(card->device, address,
PAGE_SIZE, buffer->direction);
__free_page(buffer->pages[i]);
}
kfree(buffer->pages);
buffer->pages = NULL;
}
EXPORT_SYMBOL(fw_iso_buffer_destroy);
/* Convert DMA address to offset into virtually contiguous buffer. */
size_t fw_iso_buffer_lookup(struct fw_iso_buffer *buffer, dma_addr_t completed)
{
int i;
dma_addr_t address;
ssize_t offset;
for (i = 0; i < buffer->page_count; i++) {
address = page_private(buffer->pages[i]);
offset = (ssize_t)completed - (ssize_t)address;
if (offset > 0 && offset <= PAGE_SIZE)
return (i << PAGE_SHIFT) + offset;
}
return 0;
}
struct fw_iso_context *fw_iso_context_create(struct fw_card *card,
int type, int channel, int speed, size_t header_size,
fw_iso_callback_t callback, void *callback_data)
{
struct fw_iso_context *ctx;
ctx = card->driver->allocate_iso_context(card,
type, channel, header_size);
if (IS_ERR(ctx))
return ctx;
ctx->card = card;
ctx->type = type;
ctx->channel = channel;
ctx->speed = speed;
ctx->header_size = header_size;
ctx->callback.sc = callback;
ctx->callback_data = callback_data;
return ctx;
}
EXPORT_SYMBOL(fw_iso_context_create);
void fw_iso_context_destroy(struct fw_iso_context *ctx)
{
ctx->card->driver->free_iso_context(ctx);
}
EXPORT_SYMBOL(fw_iso_context_destroy);
int fw_iso_context_start(struct fw_iso_context *ctx,
int cycle, int sync, int tags)
{
return ctx->card->driver->start_iso(ctx, cycle, sync, tags);
}
EXPORT_SYMBOL(fw_iso_context_start);
int fw_iso_context_set_channels(struct fw_iso_context *ctx, u64 *channels)
{
return ctx->card->driver->set_iso_channels(ctx, channels);
}
int fw_iso_context_queue(struct fw_iso_context *ctx,
struct fw_iso_packet *packet,
struct fw_iso_buffer *buffer,
unsigned long payload)
{
return ctx->card->driver->queue_iso(ctx, packet, buffer, payload);
}
EXPORT_SYMBOL(fw_iso_context_queue);
void fw_iso_context_queue_flush(struct fw_iso_context *ctx)
{
ctx->card->driver->flush_queue_iso(ctx);
}
EXPORT_SYMBOL(fw_iso_context_queue_flush);
int fw_iso_context_stop(struct fw_iso_context *ctx)
{
return ctx->card->driver->stop_iso(ctx);
}
EXPORT_SYMBOL(fw_iso_context_stop);
/*
* Isochronous bus resource management (channels, bandwidth), client side
*/
static int manage_bandwidth(struct fw_card *card, int irm_id, int generation,
int bandwidth, bool allocate)
{
int try, new, old = allocate ? BANDWIDTH_AVAILABLE_INITIAL : 0;
__be32 data[2];
/*
* On a 1394a IRM with low contention, try < 1 is enough.
* On a 1394-1995 IRM, we need at least try < 2.
* Let's just do try < 5.
*/
for (try = 0; try < 5; try++) {
new = allocate ? old - bandwidth : old + bandwidth;
if (new < 0 || new > BANDWIDTH_AVAILABLE_INITIAL)
return -EBUSY;
data[0] = cpu_to_be32(old);
data[1] = cpu_to_be32(new);
switch (fw_run_transaction(card, TCODE_LOCK_COMPARE_SWAP,
irm_id, generation, SCODE_100,
CSR_REGISTER_BASE + CSR_BANDWIDTH_AVAILABLE,
data, 8)) {
case RCODE_GENERATION:
/* A generation change frees all bandwidth. */
return allocate ? -EAGAIN : bandwidth;
case RCODE_COMPLETE:
if (be32_to_cpup(data) == old)
return bandwidth;
old = be32_to_cpup(data);
/* Fall through. */
}
}
return -EIO;
}
static int manage_channel(struct fw_card *card, int irm_id, int generation,
u32 channels_mask, u64 offset, bool allocate)
{
__be32 bit, all, old;
__be32 data[2];
int channel, ret = -EIO, retry = 5;
old = all = allocate ? cpu_to_be32(~0) : 0;
for (channel = 0; channel < 32; channel++) {
if (!(channels_mask & 1 << channel))
continue;
ret = -EBUSY;
bit = cpu_to_be32(1 << (31 - channel));
if ((old & bit) != (all & bit))
continue;
data[0] = old;
data[1] = old ^ bit;
switch (fw_run_transaction(card, TCODE_LOCK_COMPARE_SWAP,
irm_id, generation, SCODE_100,
offset, data, 8)) {
case RCODE_GENERATION:
/* A generation change frees all channels. */
return allocate ? -EAGAIN : channel;
case RCODE_COMPLETE:
if (data[0] == old)
return channel;
old = data[0];
/* Is the IRM 1394a-2000 compliant? */
if ((data[0] & bit) == (data[1] & bit))
continue;
/* 1394-1995 IRM, fall through to retry. */
default:
if (retry) {
retry--;
channel--;
} else {
ret = -EIO;
}
}
}
return ret;
}
static void deallocate_channel(struct fw_card *card, int irm_id,
int generation, int channel)
{
u32 mask;
u64 offset;
mask = channel < 32 ? 1 << channel : 1 << (channel - 32);
offset = channel < 32 ? CSR_REGISTER_BASE + CSR_CHANNELS_AVAILABLE_HI :
CSR_REGISTER_BASE + CSR_CHANNELS_AVAILABLE_LO;
manage_channel(card, irm_id, generation, mask, offset, false);
}
/**
* fw_iso_resource_manage() - Allocate or deallocate a channel and/or bandwidth
*
* In parameters: card, generation, channels_mask, bandwidth, allocate
* Out parameters: channel, bandwidth
* This function blocks (sleeps) during communication with the IRM.
*
* Allocates or deallocates at most one channel out of channels_mask.
* channels_mask is a bitfield with MSB for channel 63 and LSB for channel 0.
* (Note, the IRM's CHANNELS_AVAILABLE is a big-endian bitfield with MSB for
* channel 0 and LSB for channel 63.)
* Allocates or deallocates as many bandwidth allocation units as specified.
*
* Returns channel < 0 if no channel was allocated or deallocated.
* Returns bandwidth = 0 if no bandwidth was allocated or deallocated.
*
* If generation is stale, deallocations succeed but allocations fail with
* channel = -EAGAIN.
*
* If channel allocation fails, no bandwidth will be allocated either.
* If bandwidth allocation fails, no channel will be allocated either.
* But deallocations of channel and bandwidth are tried independently
* of each other's success.
*/
void fw_iso_resource_manage(struct fw_card *card, int generation,
u64 channels_mask, int *channel, int *bandwidth,
bool allocate)
{
u32 channels_hi = channels_mask; /* channels 31...0 */
u32 channels_lo = channels_mask >> 32; /* channels 63...32 */
int irm_id, ret, c = -EINVAL;
spin_lock_irq(&card->lock);
irm_id = card->irm_node->node_id;
spin_unlock_irq(&card->lock);
if (channels_hi)
c = manage_channel(card, irm_id, generation, channels_hi,
CSR_REGISTER_BASE + CSR_CHANNELS_AVAILABLE_HI,
allocate);
if (channels_lo && c < 0) {
c = manage_channel(card, irm_id, generation, channels_lo,
CSR_REGISTER_BASE + CSR_CHANNELS_AVAILABLE_LO,
allocate);
if (c >= 0)
c += 32;
}
*channel = c;
if (allocate && channels_mask != 0 && c < 0)
*bandwidth = 0;
if (*bandwidth == 0)
return;
ret = manage_bandwidth(card, irm_id, generation, *bandwidth, allocate);
if (ret < 0)
*bandwidth = 0;
if (allocate && ret < 0) {
if (c >= 0)
deallocate_channel(card, irm_id, generation, c);
*channel = ret;
}
}
EXPORT_SYMBOL(fw_iso_resource_manage);