Commit ef6fb2d6 authored by Sia Jee Heng's avatar Sia Jee Heng Committed by Vinod Koul

dmaengine: dw-axi-dmac: simplify descriptor management

Simplify and refactor the descriptor management by removing the redundant
Linked List Item (LLI) queue control logic from the AxiDMA driver.
The descriptor is split into virtual descriptor and hardware LLI so that
only hardware LLI memories are allocated from the DMA memory pool.

Up to 64 descriptors can be allocated within a PAGE_SIZE compare to 16
descriptors in previous version. This solves the problem where an
ALSA driver expects more than 16 DMA descriptors to run.
Signed-off-by: default avatarSia Jee Heng <jee.heng.sia@intel.com>
Tested-by: default avatarEugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
Reviewed-by: default avatarAndy Shevchenko <andriy.shevchenko@linux.intel.com>
Reviewed-by: default avatarEugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
Link: https://lore.kernel.org/r/20210125013255.25799-3-jee.heng.sia@intel.comSigned-off-by: default avatarVinod Koul <vkoul@kernel.org>
parent 8c70fb7e
......@@ -21,6 +21,7 @@
#include <linux/platform_device.h>
#include <linux/pm_runtime.h>
#include <linux/property.h>
#include <linux/slab.h>
#include <linux/types.h>
#include "dw-axi-dmac.h"
......@@ -195,43 +196,58 @@ static inline const char *axi_chan_name(struct axi_dma_chan *chan)
return dma_chan_name(&chan->vc.chan);
}
static struct axi_dma_desc *axi_desc_get(struct axi_dma_chan *chan)
static struct axi_dma_desc *axi_desc_alloc(u32 num)
{
struct dw_axi_dma *dw = chan->chip->dw;
struct axi_dma_desc *desc;
desc = kzalloc(sizeof(*desc), GFP_NOWAIT);
if (!desc)
return NULL;
desc->hw_desc = kcalloc(num, sizeof(*desc->hw_desc), GFP_NOWAIT);
if (!desc->hw_desc) {
kfree(desc);
return NULL;
}
return desc;
}
static struct axi_dma_lli *axi_desc_get(struct axi_dma_chan *chan,
dma_addr_t *addr)
{
struct dw_axi_dma *dw = chan->chip->dw;
struct axi_dma_lli *lli;
dma_addr_t phys;
desc = dma_pool_zalloc(dw->desc_pool, GFP_NOWAIT, &phys);
if (unlikely(!desc)) {
lli = dma_pool_zalloc(dw->desc_pool, GFP_NOWAIT, &phys);
if (unlikely(!lli)) {
dev_err(chan2dev(chan), "%s: not enough descriptors available\n",
axi_chan_name(chan));
return NULL;
}
atomic_inc(&chan->descs_allocated);
INIT_LIST_HEAD(&desc->xfer_list);
desc->vd.tx.phys = phys;
desc->chan = chan;
*addr = phys;
return desc;
return lli;
}
static void axi_desc_put(struct axi_dma_desc *desc)
{
struct axi_dma_chan *chan = desc->chan;
struct dw_axi_dma *dw = chan->chip->dw;
struct axi_dma_desc *child, *_next;
unsigned int descs_put = 0;
int count = atomic_read(&chan->descs_allocated);
struct axi_dma_hw_desc *hw_desc;
int descs_put;
list_for_each_entry_safe(child, _next, &desc->xfer_list, xfer_list) {
list_del(&child->xfer_list);
dma_pool_free(dw->desc_pool, child, child->vd.tx.phys);
descs_put++;
for (descs_put = 0; descs_put < count; descs_put++) {
hw_desc = &desc->hw_desc[descs_put];
dma_pool_free(dw->desc_pool, hw_desc->lli, hw_desc->llp);
}
dma_pool_free(dw->desc_pool, desc, desc->vd.tx.phys);
descs_put++;
kfree(desc->hw_desc);
kfree(desc);
atomic_sub(descs_put, &chan->descs_allocated);
dev_vdbg(chan2dev(chan), "%s: %d descs put, %d still allocated\n",
axi_chan_name(chan), descs_put,
......@@ -258,9 +274,9 @@ dma_chan_tx_status(struct dma_chan *dchan, dma_cookie_t cookie,
return ret;
}
static void write_desc_llp(struct axi_dma_desc *desc, dma_addr_t adr)
static void write_desc_llp(struct axi_dma_hw_desc *desc, dma_addr_t adr)
{
desc->lli.llp = cpu_to_le64(adr);
desc->lli->llp = cpu_to_le64(adr);
}
static void write_chan_llp(struct axi_dma_chan *chan, dma_addr_t adr)
......@@ -295,7 +311,7 @@ static void axi_chan_block_xfer_start(struct axi_dma_chan *chan,
DWAXIDMAC_HS_SEL_HW << CH_CFG_H_HS_SEL_SRC_POS);
axi_chan_iowrite32(chan, CH_CFG_H, reg);
write_chan_llp(chan, first->vd.tx.phys | lms);
write_chan_llp(chan, first->hw_desc[0].llp | lms);
irq_mask = DWAXIDMAC_IRQ_DMA_TRF | DWAXIDMAC_IRQ_ALL_ERR;
axi_chan_irq_sig_set(chan, irq_mask);
......@@ -378,67 +394,78 @@ static void dma_chan_free_chan_resources(struct dma_chan *dchan)
* transfer and completes the DMA transfer operation at the end of current
* block transfer.
*/
static void set_desc_last(struct axi_dma_desc *desc)
static void set_desc_last(struct axi_dma_hw_desc *desc)
{
u32 val;
val = le32_to_cpu(desc->lli.ctl_hi);
val = le32_to_cpu(desc->lli->ctl_hi);
val |= CH_CTL_H_LLI_LAST;
desc->lli.ctl_hi = cpu_to_le32(val);
desc->lli->ctl_hi = cpu_to_le32(val);
}
static void write_desc_sar(struct axi_dma_desc *desc, dma_addr_t adr)
static void write_desc_sar(struct axi_dma_hw_desc *desc, dma_addr_t adr)
{
desc->lli.sar = cpu_to_le64(adr);
desc->lli->sar = cpu_to_le64(adr);
}
static void write_desc_dar(struct axi_dma_desc *desc, dma_addr_t adr)
static void write_desc_dar(struct axi_dma_hw_desc *desc, dma_addr_t adr)
{
desc->lli.dar = cpu_to_le64(adr);
desc->lli->dar = cpu_to_le64(adr);
}
static void set_desc_src_master(struct axi_dma_desc *desc)
static void set_desc_src_master(struct axi_dma_hw_desc *desc)
{
u32 val;
/* Select AXI0 for source master */
val = le32_to_cpu(desc->lli.ctl_lo);
val = le32_to_cpu(desc->lli->ctl_lo);
val &= ~CH_CTL_L_SRC_MAST;
desc->lli.ctl_lo = cpu_to_le32(val);
desc->lli->ctl_lo = cpu_to_le32(val);
}
static void set_desc_dest_master(struct axi_dma_desc *desc)
static void set_desc_dest_master(struct axi_dma_hw_desc *hw_desc,
struct axi_dma_desc *desc)
{
u32 val;
/* Select AXI1 for source master if available */
val = le32_to_cpu(desc->lli.ctl_lo);
val = le32_to_cpu(hw_desc->lli->ctl_lo);
if (desc->chan->chip->dw->hdata->nr_masters > 1)
val |= CH_CTL_L_DST_MAST;
else
val &= ~CH_CTL_L_DST_MAST;
desc->lli.ctl_lo = cpu_to_le32(val);
hw_desc->lli->ctl_lo = cpu_to_le32(val);
}
static struct dma_async_tx_descriptor *
dma_chan_prep_dma_memcpy(struct dma_chan *dchan, dma_addr_t dst_adr,
dma_addr_t src_adr, size_t len, unsigned long flags)
{
struct axi_dma_desc *first = NULL, *desc = NULL, *prev = NULL;
struct axi_dma_chan *chan = dchan_to_axi_dma_chan(dchan);
size_t block_ts, max_block_ts, xfer_len;
u32 xfer_width, reg;
struct axi_dma_hw_desc *hw_desc = NULL;
struct axi_dma_desc *desc = NULL;
u32 xfer_width, reg, num;
u64 llp = 0;
u8 lms = 0; /* Select AXI0 master for LLI fetching */
dev_dbg(chan2dev(chan), "%s: memcpy: src: %pad dst: %pad length: %zd flags: %#lx",
axi_chan_name(chan), &src_adr, &dst_adr, len, flags);
max_block_ts = chan->chip->dw->hdata->block_size[chan->id];
xfer_width = axi_chan_get_xfer_width(chan, src_adr, dst_adr, len);
num = DIV_ROUND_UP(len, max_block_ts << xfer_width);
desc = axi_desc_alloc(num);
if (unlikely(!desc))
goto err_desc_get;
desc->chan = chan;
num = 0;
while (len) {
xfer_len = len;
hw_desc = &desc->hw_desc[num];
/*
* Take care for the alignment.
* Actually source and destination widths can be different, but
......@@ -457,13 +484,13 @@ dma_chan_prep_dma_memcpy(struct dma_chan *dchan, dma_addr_t dst_adr,
xfer_len = max_block_ts << xfer_width;
}
desc = axi_desc_get(chan);
if (unlikely(!desc))
hw_desc->lli = axi_desc_get(chan, &hw_desc->llp);
if (unlikely(!hw_desc->lli))
goto err_desc_get;
write_desc_sar(desc, src_adr);
write_desc_dar(desc, dst_adr);
desc->lli.block_ts_lo = cpu_to_le32(block_ts - 1);
write_desc_sar(hw_desc, src_adr);
write_desc_dar(hw_desc, dst_adr);
hw_desc->lli->block_ts_lo = cpu_to_le32(block_ts - 1);
reg = CH_CTL_H_LLI_VALID;
if (chan->chip->dw->hdata->restrict_axi_burst_len) {
......@@ -474,7 +501,7 @@ dma_chan_prep_dma_memcpy(struct dma_chan *dchan, dma_addr_t dst_adr,
CH_CTL_H_AWLEN_EN |
burst_len << CH_CTL_H_AWLEN_POS);
}
desc->lli.ctl_hi = cpu_to_le32(reg);
hw_desc->lli->ctl_hi = cpu_to_le32(reg);
reg = (DWAXIDMAC_BURST_TRANS_LEN_4 << CH_CTL_L_DST_MSIZE_POS |
DWAXIDMAC_BURST_TRANS_LEN_4 << CH_CTL_L_SRC_MSIZE_POS |
......@@ -482,62 +509,61 @@ dma_chan_prep_dma_memcpy(struct dma_chan *dchan, dma_addr_t dst_adr,
xfer_width << CH_CTL_L_SRC_WIDTH_POS |
DWAXIDMAC_CH_CTL_L_INC << CH_CTL_L_DST_INC_POS |
DWAXIDMAC_CH_CTL_L_INC << CH_CTL_L_SRC_INC_POS);
desc->lli.ctl_lo = cpu_to_le32(reg);
hw_desc->lli->ctl_lo = cpu_to_le32(reg);
set_desc_src_master(desc);
set_desc_dest_master(desc);
set_desc_src_master(hw_desc);
set_desc_dest_master(hw_desc, desc);
/* Manage transfer list (xfer_list) */
if (!first) {
first = desc;
} else {
list_add_tail(&desc->xfer_list, &first->xfer_list);
write_desc_llp(prev, desc->vd.tx.phys | lms);
}
prev = desc;
/* update the length and addresses for the next loop cycle */
len -= xfer_len;
dst_adr += xfer_len;
src_adr += xfer_len;
num++;
}
/* Total len of src/dest sg == 0, so no descriptor were allocated */
if (unlikely(!first))
if (unlikely(!desc))
return NULL;
/* Set end-of-link to the last link descriptor of list */
set_desc_last(desc);
set_desc_last(&desc->hw_desc[num - 1]);
/* Managed transfer list */
do {
hw_desc = &desc->hw_desc[--num];
write_desc_llp(hw_desc, llp | lms);
llp = hw_desc->llp;
} while (num);
return vchan_tx_prep(&chan->vc, &first->vd, flags);
return vchan_tx_prep(&chan->vc, &desc->vd, flags);
err_desc_get:
if (first)
axi_desc_put(first);
if (desc)
axi_desc_put(desc);
return NULL;
}
static void axi_chan_dump_lli(struct axi_dma_chan *chan,
struct axi_dma_desc *desc)
struct axi_dma_hw_desc *desc)
{
dev_err(dchan2dev(&chan->vc.chan),
"SAR: 0x%llx DAR: 0x%llx LLP: 0x%llx BTS 0x%x CTL: 0x%x:%08x",
le64_to_cpu(desc->lli.sar),
le64_to_cpu(desc->lli.dar),
le64_to_cpu(desc->lli.llp),
le32_to_cpu(desc->lli.block_ts_lo),
le32_to_cpu(desc->lli.ctl_hi),
le32_to_cpu(desc->lli.ctl_lo));
le64_to_cpu(desc->lli->sar),
le64_to_cpu(desc->lli->dar),
le64_to_cpu(desc->lli->llp),
le32_to_cpu(desc->lli->block_ts_lo),
le32_to_cpu(desc->lli->ctl_hi),
le32_to_cpu(desc->lli->ctl_lo));
}
static void axi_chan_list_dump_lli(struct axi_dma_chan *chan,
struct axi_dma_desc *desc_head)
{
struct axi_dma_desc *desc;
int count = atomic_read(&chan->descs_allocated);
int i;
axi_chan_dump_lli(chan, desc_head);
list_for_each_entry(desc, &desc_head->xfer_list, xfer_list)
axi_chan_dump_lli(chan, desc);
for (i = 0; i < count; i++)
axi_chan_dump_lli(chan, &desc_head->hw_desc[i]);
}
static noinline void axi_chan_handle_err(struct axi_dma_chan *chan, u32 status)
......@@ -872,7 +898,7 @@ static int dw_probe(struct platform_device *pdev)
/* Lli address must be aligned to a 64-byte boundary */
dw->desc_pool = dmam_pool_create(KBUILD_MODNAME, chip->dev,
sizeof(struct axi_dma_desc), 64, 0);
sizeof(struct axi_dma_lli), 64, 0);
if (!dw->desc_pool) {
dev_err(chip->dev, "No memory for descriptors dma pool\n");
return -ENOMEM;
......
......@@ -41,6 +41,7 @@ struct axi_dma_chan {
struct virt_dma_chan vc;
struct axi_dma_desc *desc;
/* these other elements are all protected by vc.lock */
bool is_paused;
};
......@@ -80,12 +81,16 @@ struct __packed axi_dma_lli {
__le32 reserved_hi;
};
struct axi_dma_hw_desc {
struct axi_dma_lli *lli;
dma_addr_t llp;
};
struct axi_dma_desc {
struct axi_dma_lli lli;
struct axi_dma_hw_desc *hw_desc;
struct virt_dma_desc vd;
struct axi_dma_chan *chan;
struct list_head xfer_list;
};
static inline struct device *dchan2dev(struct dma_chan *dchan)
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment