Thread (20 messages) 20 messages, 2 authors, 2025-08-21

Re: [PATCH v6 01/10] ASoC: mediatek: common: modify mtk afe platform driver for mt8196

From: Chen-Yu Tsai <wenst@chromium.org>
Date: 2025-07-22 07:38:18
Also in: linux-devicetree, linux-gpio, linux-mediatek, linux-sound, lkml

Hi,

On Tue, Jul 8, 2025 at 7:33 PM Darren.Ye [off-list ref] wrote:
From: Darren Ye <redacted>

Mofify the pcm pointer interface to support 64-bit address access.
  ^ Modify

And the subject should say the same, or shorter:

    Support 64-bit addresses in PCM pointer callback
quoted hunk ↗ jump to hunk
Signed-off-by: Darren Ye <redacted>
Reviewed-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>
Tested-by: Louis-Alexis Eyraud <redacted>
---
 .../mediatek/common/mtk-afe-platform-driver.c | 47 ++++++++++++-------
 .../mediatek/common/mtk-afe-platform-driver.h |  2 +
 2 files changed, 33 insertions(+), 16 deletions(-)
diff --git a/sound/soc/mediatek/common/mtk-afe-platform-driver.c b/sound/soc/mediatek/common/mtk-afe-platform-driver.c
index 70fd05d5ff48..cab4ef035199 100644
--- a/sound/soc/mediatek/common/mtk-afe-platform-driver.c
+++ b/sound/soc/mediatek/common/mtk-afe-platform-driver.c
@@ -86,29 +86,44 @@ snd_pcm_uframes_t mtk_afe_pcm_pointer(struct snd_soc_component *component,
        const struct mtk_base_memif_data *memif_data = memif->data;
        struct regmap *regmap = afe->regmap;
        struct device *dev = afe->dev;
-       int reg_ofs_base = memif_data->reg_ofs_base;
-       int reg_ofs_cur = memif_data->reg_ofs_cur;
-       unsigned int hw_ptr = 0, hw_base = 0;
-       int ret, pcm_ptr_bytes;
-
-       ret = regmap_read(regmap, reg_ofs_cur, &hw_ptr);
-       if (ret || hw_ptr == 0) {
-               dev_err(dev, "%s hw_ptr err\n", __func__);
-               pcm_ptr_bytes = 0;
+       unsigned int hw_ptr_lower32 = 0, hw_ptr_upper32 = 0;
+       unsigned int hw_base_lower32 = 0, hw_base_upper32 = 0;
+       unsigned long long hw_ptr = 0, hw_base = 0;
+       int ret;
+       unsigned long long pcm_ptr_bytes = 0;
+
+       ret = regmap_read(regmap, memif_data->reg_ofs_cur, &hw_ptr_lower32);
+       if (ret || hw_ptr_lower32 == 0) {
I'm not sure how the hardware is, but I think hw_ptr_lower32 == 0
should no longer be a failure, since it could be 0x100000000 or
some other address with the lower 32-bits all zero. Instead the
check should be done once the addresses are put together.
+               dev_err(dev, "%s hw_ptr_lower32 err\n", __func__);
                goto POINTER_RETURN_FRAMES;
This is out of scope, but maybe this should just return zero directly
to simplify reading.
        }

-       ret = regmap_read(regmap, reg_ofs_base, &hw_base);
-       if (ret || hw_base == 0) {
-               dev_err(dev, "%s hw_ptr err\n", __func__);
-               pcm_ptr_bytes = 0;
-               goto POINTER_RETURN_FRAMES;
+       if (memif_data->reg_ofs_cur_msb) {
+               ret = regmap_read(regmap, memif_data->reg_ofs_cur_msb, &hw_ptr_upper32);
+               if (ret) {
+                       dev_err(dev, "%s hw_ptr_upper32 err\n", __func__);
+                       goto POINTER_RETURN_FRAMES;
+               }
        }

-       pcm_ptr_bytes = hw_ptr - hw_base;
+       ret = regmap_read(regmap, memif_data->reg_ofs_base, &hw_base_lower32);
+       if (ret || hw_base_lower32 == 0) {
Same here.
+               dev_err(dev, "%s hw_base_lower32 err\n", __func__);
+               goto POINTER_RETURN_FRAMES;
+       }
+       if (memif_data->reg_ofs_base_msb) {
+               ret = regmap_read(regmap, memif_data->reg_ofs_base_msb, &hw_base_upper32);
+               if (ret) {
+                       dev_err(dev, "%s hw_base_upper32 err\n", __func__);
+                       goto POINTER_RETURN_FRAMES;
+               }
+       }
+       hw_ptr = ((unsigned long long)hw_ptr_upper32 << 32) + hw_ptr_lower32;
+       hw_base = ((unsigned long long)hw_base_upper32 << 32) + hw_base_lower32;
Instead the check should be here. And to follow the original logic,
if either pointer value is zero, the function should return zero here
directly.


ChenYu
quoted hunk ↗ jump to hunk
 POINTER_RETURN_FRAMES:
-       return bytes_to_frames(substream->runtime, pcm_ptr_bytes);
+       pcm_ptr_bytes = MTK_ALIGN_16BYTES(hw_ptr - hw_base);
+       return bytes_to_frames(substream->runtime, (ssize_t)pcm_ptr_bytes);
 }
 EXPORT_SYMBOL_GPL(mtk_afe_pcm_pointer);
diff --git a/sound/soc/mediatek/common/mtk-afe-platform-driver.h b/sound/soc/mediatek/common/mtk-afe-platform-driver.h
index fcc923b88f12..71070b26f8f8 100644
--- a/sound/soc/mediatek/common/mtk-afe-platform-driver.h
+++ b/sound/soc/mediatek/common/mtk-afe-platform-driver.h
@@ -12,6 +12,8 @@
 #define AFE_PCM_NAME "mtk-afe-pcm"
 extern const struct snd_soc_component_driver mtk_afe_pcm_platform;

+#define MTK_ALIGN_16BYTES(x) ((x) & GENMASK_ULL(39, 4))
+
 struct mtk_base_afe;
 struct snd_pcm;
 struct snd_soc_component;
--
2.45.2
  
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help