linux/sound/soc/intel/skylake/skl-sst-cldma.c
Jeeja KP b7d0254c51 ASoC: Intel: Skylake: Fix module load when module size > DMA buffer size
When module size > DMA buffer size, driver copies first chunk and waits
for the BDL complete interrupt. BDL complete interrupt never occurs and
wait time expires as module load IPC is not send to start the DMA from
DSP.

To fix the above issue need to follow the below steps:
1. After copying the first chunk, send the module load IPC to start the
DMA.
2. Wait for the BDL interrupt. Once interrupt is received, copy the
next chunk.
3. Continue step 2 till all bytes are copied.
4. When all the bytes are copied (bytes_left = 0), wait for module load
IPC response
5. Handled module load IPC response messages, check the load module IPC
response and wake up the thread to complete module load.

Signed-off-by: Jeeja KP <jeeja.kp@intel.com>
Acked-by: Vinod Koul <vinod.koul@intel.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
2017-03-15 17:28:18 +00:00

376 lines
11 KiB
C

/*
* skl-sst-cldma.c - Code Loader DMA handler
*
* Copyright (C) 2015, Intel Corporation.
* Author: Subhransu S. Prusty <subhransu.s.prusty@intel.com>
* ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as version 2, as
* published by the Free Software Foundation.
*
* 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.
*/
#include <linux/device.h>
#include <linux/mm.h>
#include <linux/delay.h>
#include "../common/sst-dsp.h"
#include "../common/sst-dsp-priv.h"
static void skl_cldma_int_enable(struct sst_dsp *ctx)
{
sst_dsp_shim_update_bits_unlocked(ctx, SKL_ADSP_REG_ADSPIC,
SKL_ADSPIC_CL_DMA, SKL_ADSPIC_CL_DMA);
}
void skl_cldma_int_disable(struct sst_dsp *ctx)
{
sst_dsp_shim_update_bits_unlocked(ctx,
SKL_ADSP_REG_ADSPIC, SKL_ADSPIC_CL_DMA, 0);
}
static void skl_cldma_stream_run(struct sst_dsp *ctx, bool enable)
{
unsigned char val;
int timeout;
sst_dsp_shim_update_bits_unlocked(ctx,
SKL_ADSP_REG_CL_SD_CTL,
CL_SD_CTL_RUN_MASK, CL_SD_CTL_RUN(enable));
udelay(3);
timeout = 300;
do {
/* waiting for hardware to report that the stream Run bit set */
val = sst_dsp_shim_read(ctx, SKL_ADSP_REG_CL_SD_CTL) &
CL_SD_CTL_RUN_MASK;
if (enable && val)
break;
else if (!enable && !val)
break;
udelay(3);
} while (--timeout);
if (timeout == 0)
dev_err(ctx->dev, "Failed to set Run bit=%d enable=%d\n", val, enable);
}
static void skl_cldma_stream_clear(struct sst_dsp *ctx)
{
/* make sure Run bit is cleared before setting stream register */
skl_cldma_stream_run(ctx, 0);
sst_dsp_shim_update_bits(ctx, SKL_ADSP_REG_CL_SD_CTL,
CL_SD_CTL_IOCE_MASK, CL_SD_CTL_IOCE(0));
sst_dsp_shim_update_bits(ctx, SKL_ADSP_REG_CL_SD_CTL,
CL_SD_CTL_FEIE_MASK, CL_SD_CTL_FEIE(0));
sst_dsp_shim_update_bits(ctx, SKL_ADSP_REG_CL_SD_CTL,
CL_SD_CTL_DEIE_MASK, CL_SD_CTL_DEIE(0));
sst_dsp_shim_update_bits(ctx, SKL_ADSP_REG_CL_SD_CTL,
CL_SD_CTL_STRM_MASK, CL_SD_CTL_STRM(0));
sst_dsp_shim_write(ctx, SKL_ADSP_REG_CL_SD_BDLPL, CL_SD_BDLPLBA(0));
sst_dsp_shim_write(ctx, SKL_ADSP_REG_CL_SD_BDLPU, 0);
sst_dsp_shim_write(ctx, SKL_ADSP_REG_CL_SD_CBL, 0);
sst_dsp_shim_write(ctx, SKL_ADSP_REG_CL_SD_LVI, 0);
}
/* Code loader helper APIs */
static void skl_cldma_setup_bdle(struct sst_dsp *ctx,
struct snd_dma_buffer *dmab_data,
u32 **bdlp, int size, int with_ioc)
{
u32 *bdl = *bdlp;
ctx->cl_dev.frags = 0;
while (size > 0) {
phys_addr_t addr = virt_to_phys(dmab_data->area +
(ctx->cl_dev.frags * ctx->cl_dev.bufsize));
bdl[0] = cpu_to_le32(lower_32_bits(addr));
bdl[1] = cpu_to_le32(upper_32_bits(addr));
bdl[2] = cpu_to_le32(ctx->cl_dev.bufsize);
size -= ctx->cl_dev.bufsize;
bdl[3] = (size || !with_ioc) ? 0 : cpu_to_le32(0x01);
bdl += 4;
ctx->cl_dev.frags++;
}
}
/*
* Setup controller
* Configure the registers to update the dma buffer address and
* enable interrupts.
* Note: Using the channel 1 for transfer
*/
static void skl_cldma_setup_controller(struct sst_dsp *ctx,
struct snd_dma_buffer *dmab_bdl, unsigned int max_size,
u32 count)
{
skl_cldma_stream_clear(ctx);
sst_dsp_shim_write(ctx, SKL_ADSP_REG_CL_SD_BDLPL,
CL_SD_BDLPLBA(dmab_bdl->addr));
sst_dsp_shim_write(ctx, SKL_ADSP_REG_CL_SD_BDLPU,
CL_SD_BDLPUBA(dmab_bdl->addr));
sst_dsp_shim_write(ctx, SKL_ADSP_REG_CL_SD_CBL, max_size);
sst_dsp_shim_write(ctx, SKL_ADSP_REG_CL_SD_LVI, count - 1);
sst_dsp_shim_update_bits(ctx, SKL_ADSP_REG_CL_SD_CTL,
CL_SD_CTL_IOCE_MASK, CL_SD_CTL_IOCE(1));
sst_dsp_shim_update_bits(ctx, SKL_ADSP_REG_CL_SD_CTL,
CL_SD_CTL_FEIE_MASK, CL_SD_CTL_FEIE(1));
sst_dsp_shim_update_bits(ctx, SKL_ADSP_REG_CL_SD_CTL,
CL_SD_CTL_DEIE_MASK, CL_SD_CTL_DEIE(1));
sst_dsp_shim_update_bits(ctx, SKL_ADSP_REG_CL_SD_CTL,
CL_SD_CTL_STRM_MASK, CL_SD_CTL_STRM(FW_CL_STREAM_NUMBER));
}
static void skl_cldma_setup_spb(struct sst_dsp *ctx,
unsigned int size, bool enable)
{
if (enable)
sst_dsp_shim_update_bits_unlocked(ctx,
SKL_ADSP_REG_CL_SPBFIFO_SPBFCCTL,
CL_SPBFIFO_SPBFCCTL_SPIBE_MASK,
CL_SPBFIFO_SPBFCCTL_SPIBE(1));
sst_dsp_shim_write_unlocked(ctx, SKL_ADSP_REG_CL_SPBFIFO_SPIB, size);
}
static void skl_cldma_cleanup_spb(struct sst_dsp *ctx)
{
sst_dsp_shim_update_bits_unlocked(ctx,
SKL_ADSP_REG_CL_SPBFIFO_SPBFCCTL,
CL_SPBFIFO_SPBFCCTL_SPIBE_MASK,
CL_SPBFIFO_SPBFCCTL_SPIBE(0));
sst_dsp_shim_write_unlocked(ctx, SKL_ADSP_REG_CL_SPBFIFO_SPIB, 0);
}
static void skl_cldma_cleanup(struct sst_dsp *ctx)
{
skl_cldma_cleanup_spb(ctx);
skl_cldma_stream_clear(ctx);
ctx->dsp_ops.free_dma_buf(ctx->dev, &ctx->cl_dev.dmab_data);
ctx->dsp_ops.free_dma_buf(ctx->dev, &ctx->cl_dev.dmab_bdl);
}
int skl_cldma_wait_interruptible(struct sst_dsp *ctx)
{
int ret = 0;
if (!wait_event_timeout(ctx->cl_dev.wait_queue,
ctx->cl_dev.wait_condition,
msecs_to_jiffies(SKL_WAIT_TIMEOUT))) {
dev_err(ctx->dev, "%s: Wait timeout\n", __func__);
ret = -EIO;
goto cleanup;
}
dev_dbg(ctx->dev, "%s: Event wake\n", __func__);
if (ctx->cl_dev.wake_status != SKL_CL_DMA_BUF_COMPLETE) {
dev_err(ctx->dev, "%s: DMA Error\n", __func__);
ret = -EIO;
}
cleanup:
ctx->cl_dev.wake_status = SKL_CL_DMA_STATUS_NONE;
return ret;
}
static void skl_cldma_stop(struct sst_dsp *ctx)
{
skl_cldma_stream_run(ctx, false);
}
static void skl_cldma_fill_buffer(struct sst_dsp *ctx, unsigned int size,
const void *curr_pos, bool intr_enable, bool trigger)
{
dev_dbg(ctx->dev, "Size: %x, intr_enable: %d\n", size, intr_enable);
dev_dbg(ctx->dev, "buf_pos_index:%d, trigger:%d\n",
ctx->cl_dev.dma_buffer_offset, trigger);
dev_dbg(ctx->dev, "spib position: %d\n", ctx->cl_dev.curr_spib_pos);
/*
* Check if the size exceeds buffer boundary. If it exceeds
* max_buffer size, then copy till buffer size and then copy
* remaining buffer from the start of ring buffer.
*/
if (ctx->cl_dev.dma_buffer_offset + size > ctx->cl_dev.bufsize) {
unsigned int size_b = ctx->cl_dev.bufsize -
ctx->cl_dev.dma_buffer_offset;
memcpy(ctx->cl_dev.dmab_data.area + ctx->cl_dev.dma_buffer_offset,
curr_pos, size_b);
size -= size_b;
curr_pos += size_b;
ctx->cl_dev.dma_buffer_offset = 0;
}
memcpy(ctx->cl_dev.dmab_data.area + ctx->cl_dev.dma_buffer_offset,
curr_pos, size);
if (ctx->cl_dev.curr_spib_pos == ctx->cl_dev.bufsize)
ctx->cl_dev.dma_buffer_offset = 0;
else
ctx->cl_dev.dma_buffer_offset = ctx->cl_dev.curr_spib_pos;
ctx->cl_dev.wait_condition = false;
if (intr_enable)
skl_cldma_int_enable(ctx);
ctx->cl_dev.ops.cl_setup_spb(ctx, ctx->cl_dev.curr_spib_pos, trigger);
if (trigger)
ctx->cl_dev.ops.cl_trigger(ctx, true);
}
/*
* The CL dma doesn't have any way to update the transfer status until a BDL
* buffer is fully transferred
*
* So Copying is divided in two parts.
* 1. Interrupt on buffer done where the size to be transferred is more than
* ring buffer size.
* 2. Polling on fw register to identify if data left to transferred doesn't
* fill the ring buffer. Caller takes care of polling the required status
* register to identify the transfer status.
* 3. if wait flag is set, waits for DBL interrupt to copy the next chunk till
* bytes_left is 0.
* if wait flag is not set, doesn't wait for BDL interrupt. after ccopying
* the first chunk return the no of bytes_left to be copied.
*/
static int
skl_cldma_copy_to_buf(struct sst_dsp *ctx, const void *bin,
u32 total_size, bool wait)
{
int ret = 0;
bool start = true;
unsigned int excess_bytes;
u32 size;
unsigned int bytes_left = total_size;
const void *curr_pos = bin;
if (total_size <= 0)
return -EINVAL;
dev_dbg(ctx->dev, "%s: Total binary size: %u\n", __func__, bytes_left);
while (bytes_left) {
if (bytes_left > ctx->cl_dev.bufsize) {
/*
* dma transfers only till the write pointer as
* updated in spib
*/
if (ctx->cl_dev.curr_spib_pos == 0)
ctx->cl_dev.curr_spib_pos = ctx->cl_dev.bufsize;
size = ctx->cl_dev.bufsize;
skl_cldma_fill_buffer(ctx, size, curr_pos, true, start);
if (wait) {
start = false;
ret = skl_cldma_wait_interruptible(ctx);
if (ret < 0) {
skl_cldma_stop(ctx);
return ret;
}
}
} else {
skl_cldma_int_disable(ctx);
if ((ctx->cl_dev.curr_spib_pos + bytes_left)
<= ctx->cl_dev.bufsize) {
ctx->cl_dev.curr_spib_pos += bytes_left;
} else {
excess_bytes = bytes_left -
(ctx->cl_dev.bufsize -
ctx->cl_dev.curr_spib_pos);
ctx->cl_dev.curr_spib_pos = excess_bytes;
}
size = bytes_left;
skl_cldma_fill_buffer(ctx, size,
curr_pos, false, start);
}
bytes_left -= size;
curr_pos = curr_pos + size;
if (!wait)
return bytes_left;
}
return bytes_left;
}
void skl_cldma_process_intr(struct sst_dsp *ctx)
{
u8 cl_dma_intr_status;
cl_dma_intr_status =
sst_dsp_shim_read_unlocked(ctx, SKL_ADSP_REG_CL_SD_STS);
if (!(cl_dma_intr_status & SKL_CL_DMA_SD_INT_COMPLETE))
ctx->cl_dev.wake_status = SKL_CL_DMA_ERR;
else
ctx->cl_dev.wake_status = SKL_CL_DMA_BUF_COMPLETE;
ctx->cl_dev.wait_condition = true;
wake_up(&ctx->cl_dev.wait_queue);
}
int skl_cldma_prepare(struct sst_dsp *ctx)
{
int ret;
u32 *bdl;
ctx->cl_dev.bufsize = SKL_MAX_BUFFER_SIZE;
/* Allocate cl ops */
ctx->cl_dev.ops.cl_setup_bdle = skl_cldma_setup_bdle;
ctx->cl_dev.ops.cl_setup_controller = skl_cldma_setup_controller;
ctx->cl_dev.ops.cl_setup_spb = skl_cldma_setup_spb;
ctx->cl_dev.ops.cl_cleanup_spb = skl_cldma_cleanup_spb;
ctx->cl_dev.ops.cl_trigger = skl_cldma_stream_run;
ctx->cl_dev.ops.cl_cleanup_controller = skl_cldma_cleanup;
ctx->cl_dev.ops.cl_copy_to_dmabuf = skl_cldma_copy_to_buf;
ctx->cl_dev.ops.cl_stop_dma = skl_cldma_stop;
/* Allocate buffer*/
ret = ctx->dsp_ops.alloc_dma_buf(ctx->dev,
&ctx->cl_dev.dmab_data, ctx->cl_dev.bufsize);
if (ret < 0) {
dev_err(ctx->dev, "Alloc buffer for base fw failed: %x\n", ret);
return ret;
}
/* Setup Code loader BDL */
ret = ctx->dsp_ops.alloc_dma_buf(ctx->dev,
&ctx->cl_dev.dmab_bdl, PAGE_SIZE);
if (ret < 0) {
dev_err(ctx->dev, "Alloc buffer for blde failed: %x\n", ret);
ctx->dsp_ops.free_dma_buf(ctx->dev, &ctx->cl_dev.dmab_data);
return ret;
}
bdl = (u32 *)ctx->cl_dev.dmab_bdl.area;
/* Allocate BDLs */
ctx->cl_dev.ops.cl_setup_bdle(ctx, &ctx->cl_dev.dmab_data,
&bdl, ctx->cl_dev.bufsize, 1);
ctx->cl_dev.ops.cl_setup_controller(ctx, &ctx->cl_dev.dmab_bdl,
ctx->cl_dev.bufsize, ctx->cl_dev.frags);
ctx->cl_dev.curr_spib_pos = 0;
ctx->cl_dev.dma_buffer_offset = 0;
init_waitqueue_head(&ctx->cl_dev.wait_queue);
return ret;
}