Now that cs_dsp uses regmap_raw_write() instead of regmap_raw_write_async()
it doesn't need to keep multiple DMA-safe buffers of every chunk of data
it wrote.

See commit fe08b7d5085a ("firmware: cs_dsp: Remove async regmap writes")

Only one write can be in progress at a time, so one DMA-safe buffer can be
re-used. The single DMA-safe buffer is reallocated if the next write chunk
is larger. Reallocation size is rounded up to reduce the amount of churn.
PAGE_SIZE is used as a convenient size multiple. Typically for .wmfw files
the first chunk is the largest.

A DMA-safe intermediate buffer is used because we can't assume that the
bus underlying regmap can accept non-DMA-safe buffers.

Note that a chunk from the firmware file cannot simply be split into
arbitrarily-sized chunks to avoid buffer reallocation. The data in the
firmware file is preformatted to be written directly to the device as one
block. It already takes account of alignment, write-bursts and write
length requirements of the target hardware, so that the cs_dsp driver can
treat it as an opaque blob.

Signed-off-by: Richard Fitzgerald <rf@opensource.cirrus.com>
---
 drivers/firmware/cirrus/cs_dsp.c | 103 ++++++++++---------------------
 1 file changed, 34 insertions(+), 69 deletions(-)

diff --git a/drivers/firmware/cirrus/cs_dsp.c b/drivers/firmware/cirrus/cs_dsp.c
index 36a5aefa16e7..6444ef1f7f32 100644
--- a/drivers/firmware/cirrus/cs_dsp.c
+++ b/drivers/firmware/cirrus/cs_dsp.c
@@ -13,6 +13,7 @@
 #include <linux/ctype.h>
 #include <linux/debugfs.h>
 #include <linux/delay.h>
+#include <linux/math.h>
 #include <linux/minmax.h>
 #include <linux/module.h>
 #include <linux/moduleparam.h>
@@ -317,44 +318,6 @@ struct cs_dsp_alg_region_list_item {
 	struct cs_dsp_alg_region alg_region;
 };
 
-struct cs_dsp_buf {
-	struct list_head list;
-	void *buf;
-};
-
-static struct cs_dsp_buf *cs_dsp_buf_alloc(const void *src, size_t len,
-					   struct list_head *list)
-{
-	struct cs_dsp_buf *buf = kzalloc(sizeof(*buf), GFP_KERNEL);
-
-	if (buf == NULL)
-		return NULL;
-
-	buf->buf = vmalloc(len);
-	if (!buf->buf) {
-		kfree(buf);
-		return NULL;
-	}
-	memcpy(buf->buf, src, len);
-
-	if (list)
-		list_add_tail(&buf->list, list);
-
-	return buf;
-}
-
-static void cs_dsp_buf_free(struct list_head *list)
-{
-	while (!list_empty(list)) {
-		struct cs_dsp_buf *buf = list_first_entry(list,
-							  struct cs_dsp_buf,
-							  list);
-		list_del(&buf->list);
-		vfree(buf->buf);
-		kfree(buf);
-	}
-}
-
 /**
  * cs_dsp_mem_region_name() - Return a name string for a memory type
  * @type: the memory type to match
@@ -1481,7 +1444,9 @@ static int cs_dsp_load(struct cs_dsp *dsp, const struct firmware *firmware,
 	const struct wmfw_region *region;
 	const struct cs_dsp_region *mem;
 	const char *region_name;
-	struct cs_dsp_buf *buf;
+	u8 *buf __free(kfree) = NULL;
+	size_t buf_len = 0;
+	size_t region_len;
 	unsigned int reg;
 	int regions = 0;
 	int ret, offset, type;
@@ -1601,23 +1566,23 @@ static int cs_dsp_load(struct cs_dsp *dsp, const struct firmware *firmware,
 			   region_name);
 
 		if (reg) {
-			buf = cs_dsp_buf_alloc(region->data,
-					       le32_to_cpu(region->len),
-					       &buf_list);
-			if (!buf) {
-				cs_dsp_err(dsp, "Out of memory\n");
-				ret = -ENOMEM;
-				goto out_fw;
+			region_len = le32_to_cpu(region->len);
+			if (region_len > buf_len) {
+				buf_len = round_up(region_len, PAGE_SIZE);
+				kfree(buf);
+				buf = kmalloc(buf_len, GFP_KERNEL | GFP_DMA);
+				if (!buf) {
+					ret = -ENOMEM;
+					goto out_fw;
+				}
 			}
 
-			ret = regmap_raw_write(regmap, reg, buf->buf,
-					       le32_to_cpu(region->len));
+			memcpy(buf, region->data, region_len);
+			ret = regmap_raw_write(regmap, reg, buf, region_len);
 			if (ret != 0) {
 				cs_dsp_err(dsp,
-					   "%s.%d: Failed to write %d bytes at %d in %s: %d\n",
-					   file, regions,
-					   le32_to_cpu(region->len), offset,
-					   region_name, ret);
+					   "%s.%d: Failed to write %zu bytes at %d in %s: %d\n",
+					   file, regions, region_len, offset, region_name, ret);
 				goto out_fw;
 			}
 		}
@@ -1634,8 +1599,6 @@ static int cs_dsp_load(struct cs_dsp *dsp, const struct firmware *firmware,
 
 	ret = 0;
 out_fw:
-	cs_dsp_buf_free(&buf_list);
-
 	if (ret == -EOVERFLOW)
 		cs_dsp_err(dsp, "%s: file content overflows file data\n", file);
 
@@ -2167,7 +2130,9 @@ static int cs_dsp_load_coeff(struct cs_dsp *dsp, const struct firmware *firmware
 	struct cs_dsp_alg_region *alg_region;
 	const char *region_name;
 	int ret, pos, blocks, type, offset, reg, version;
-	struct cs_dsp_buf *buf;
+	u8 *buf __free(kfree) = NULL;
+	size_t buf_len = 0;
+	size_t region_len;
 
 	if (!firmware)
 		return 0;
@@ -2309,20 +2274,22 @@ static int cs_dsp_load_coeff(struct cs_dsp *dsp, const struct firmware *firmware
 		}
 
 		if (reg) {
-			buf = cs_dsp_buf_alloc(blk->data,
-					       le32_to_cpu(blk->len),
-					       &buf_list);
-			if (!buf) {
-				cs_dsp_err(dsp, "Out of memory\n");
-				ret = -ENOMEM;
-				goto out_fw;
+			region_len = le32_to_cpu(blk->len);
+			if (region_len > buf_len) {
+				buf_len = round_up(region_len, PAGE_SIZE);
+				kfree(buf);
+				buf = kmalloc(buf_len, GFP_KERNEL | GFP_DMA);
+				if (!buf) {
+					ret = -ENOMEM;
+					goto out_fw;
+				}
 			}
 
-			cs_dsp_dbg(dsp, "%s.%d: Writing %d bytes at %x\n",
-				   file, blocks, le32_to_cpu(blk->len),
-				   reg);
-			ret = regmap_raw_write(regmap, reg, buf->buf,
-					       le32_to_cpu(blk->len));
+			memcpy(buf, blk->data, region_len);
+
+			cs_dsp_dbg(dsp, "%s.%d: Writing %zu bytes at %x\n",
+				   file, blocks, region_len, reg);
+			ret = regmap_raw_write(regmap, reg, buf, region_len);
 			if (ret != 0) {
 				cs_dsp_err(dsp,
 					   "%s.%d: Failed to write to %x in %s: %d\n",
@@ -2342,8 +2309,6 @@ static int cs_dsp_load_coeff(struct cs_dsp *dsp, const struct firmware *firmware
 
 	ret = 0;
 out_fw:
-	cs_dsp_buf_free(&buf_list);
-
 	if (ret == -EOVERFLOW)
 		cs_dsp_err(dsp, "%s: file content overflows file data\n", file);
 
-- 
2.47.3

On Wed, 26 Nov 2025 13:15:01 +0000, Richard Fitzgerald wrote:
> Now that cs_dsp uses regmap_raw_write() instead of regmap_raw_write_async()
> it doesn't need to keep multiple DMA-safe buffers of every chunk of data
> it wrote.
> 
> See commit fe08b7d5085a ("firmware: cs_dsp: Remove async regmap writes")
> 
> Only one write can be in progress at a time, so one DMA-safe buffer can be
> re-used. The single DMA-safe buffer is reallocated if the next write chunk
> is larger. Reallocation size is rounded up to reduce the amount of churn.
> PAGE_SIZE is used as a convenient size multiple. Typically for .wmfw files
> the first chunk is the largest.
> 
> [...]

Applied to

   https://git.kernel.org/pub/scm/linux/kernel/git/broonie/sound.git for-next

Thanks!

[1/1] firmware: cs_dsp: Remove redundant download buffer allocator
      commit: 900baa6e7bb08ab3d24388cf945c7be063ac1b89

All being well this means that it will be integrated into the linux-next
tree (usually sometime in the next 24 hours) and sent to Linus during
the next merge window (or sooner if it is a bug fix), however if
problems are discovered then the patch may be dropped or reverted.

You may get further e-mails resulting from automated or manual testing
and review of the tree, please engage with people reporting problems and
send followup patches addressing any issues that are reported if needed.

If any updates are required or you are submitting further changes they
should be sent as incremental updates against current git, existing
patches will not be replaced.

Please add any relevant lists and maintainers to the CCs when replying
to this mail.

Thanks,
Mark