Some drivers like ath10k can sometimg give an sg buffer with an offset whose alignment
is not compatible with the Amlogic DMA descriptor engine requirements.
Simply replace with dev_warn_once() to inform user this should be fixed to avoid
degraded performance.
This should be ultimately fixed in ath10k, but since it's only a performance issue
the warning should be removed.
Fixes: 79ed05e329c3 ("mmc: meson-gx: add support for descriptor chain mode")
Reported-by: Christian Hewitt <[email protected]>
Signed-off-by: Neil Armstrong <[email protected]>
---
drivers/mmc/host/meson-gx-mmc.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/drivers/mmc/host/meson-gx-mmc.c b/drivers/mmc/host/meson-gx-mmc.c
index b8b771b643cc..1c61f0f24c09 100644
--- a/drivers/mmc/host/meson-gx-mmc.c
+++ b/drivers/mmc/host/meson-gx-mmc.c
@@ -258,7 +258,9 @@ static void meson_mmc_get_transfer_mode(struct mmc_host *mmc,
for_each_sg(data->sg, sg, data->sg_len, i) {
/* check for 8 byte alignment */
if (sg->offset % 8) {
- WARN_ONCE(1, "unaligned scatterlist buffer\n");
+ dev_warn_once(mmc_dev(mmc),
+ "unaligned sg offset %u, disabling descriptor DMA for transfer\n",
+ sg->offset);
return;
}
}
--
2.25.1
The brcmfmac driver can generate a scatterlist from a skb with each packets
not aligned to the block size. This is not supported by the Amlogic Descriptor
dma engine where each descriptor must match a multiple of the block size.
The sg list is valid, since the sum of the sg buffers is a multiple of the
block size, but we must discard those when in SD_IO_RW_EXTENDED mode since
SDIO block mode can be used under the hood even with data->blocks == 1.
Those transfers are very rare, thus can be replaced by a bounce buffer
without real performance loss.
Fixes: 7412dee9f1fd ("mmc: meson-gx: replace WARN_ONCE with dev_warn_once about scatterlist size alignment in block mode")
Reported-by: Christian Hewitt <[email protected]>
Signed-off-by: Neil Armstrong <[email protected]>
---
drivers/mmc/host/meson-gx-mmc.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/mmc/host/meson-gx-mmc.c b/drivers/mmc/host/meson-gx-mmc.c
index 1c61f0f24c09..016a6106151a 100644
--- a/drivers/mmc/host/meson-gx-mmc.c
+++ b/drivers/mmc/host/meson-gx-mmc.c
@@ -236,7 +236,8 @@ static void meson_mmc_get_transfer_mode(struct mmc_host *mmc,
if (host->dram_access_quirk)
return;
- if (data->blocks > 1) {
+ /* SD_IO_RW_EXTENDED (CMD53) can also use block mode under the hood */
+ if (data->blocks > 1 || mrq->cmd->opcode == SD_IO_RW_EXTENDED) {
/*
* In block mode DMA descriptor format, "length" field indicates
* number of blocks and there is no way to pass DMA size that
--
2.25.1
Hi Neil,
On Mon, Apr 26, 2021 at 7:58 PM Neil Armstrong <[email protected]> wrote:
>
> Some drivers like ath10k can sometimg give an sg buffer with an offset whose alignment
sometimg -> sometimes
> is not compatible with the Amlogic DMA descriptor engine requirements.
>
> Simply replace with dev_warn_once() to inform user this should be fixed to avoid
> degraded performance.
>
> This should be ultimately fixed in ath10k, but since it's only a performance issue
> the warning should be removed.
>
> Fixes: 79ed05e329c3 ("mmc: meson-gx: add support for descriptor chain mode")
> Reported-by: Christian Hewitt <[email protected]>
> Signed-off-by: Neil Armstrong <[email protected]>
Acked-by: Martin Blumenstingl <[email protected]>
On Mon, 26 Apr 2021 at 19:56, Neil Armstrong <[email protected]> wrote:
>
> Some drivers like ath10k can sometimg give an sg buffer with an offset whose alignment
> is not compatible with the Amlogic DMA descriptor engine requirements.
>
> Simply replace with dev_warn_once() to inform user this should be fixed to avoid
> degraded performance.
>
> This should be ultimately fixed in ath10k, but since it's only a performance issue
> the warning should be removed.
>
> Fixes: 79ed05e329c3 ("mmc: meson-gx: add support for descriptor chain mode")
> Reported-by: Christian Hewitt <[email protected]>
> Signed-off-by: Neil Armstrong <[email protected]>
Applied for fixes and by adding a stable tag, thanks!
Kind regards
Uffe
> ---
> drivers/mmc/host/meson-gx-mmc.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/mmc/host/meson-gx-mmc.c b/drivers/mmc/host/meson-gx-mmc.c
> index b8b771b643cc..1c61f0f24c09 100644
> --- a/drivers/mmc/host/meson-gx-mmc.c
> +++ b/drivers/mmc/host/meson-gx-mmc.c
> @@ -258,7 +258,9 @@ static void meson_mmc_get_transfer_mode(struct mmc_host *mmc,
> for_each_sg(data->sg, sg, data->sg_len, i) {
> /* check for 8 byte alignment */
> if (sg->offset % 8) {
> - WARN_ONCE(1, "unaligned scatterlist buffer\n");
> + dev_warn_once(mmc_dev(mmc),
> + "unaligned sg offset %u, disabling descriptor DMA for transfer\n",
> + sg->offset);
> return;
> }
> }
> --
> 2.25.1
>
On Mon, 26 Apr 2021 at 19:56, Neil Armstrong <[email protected]> wrote:
>
> The brcmfmac driver can generate a scatterlist from a skb with each packets
> not aligned to the block size. This is not supported by the Amlogic Descriptor
> dma engine where each descriptor must match a multiple of the block size.
>
> The sg list is valid, since the sum of the sg buffers is a multiple of the
> block size, but we must discard those when in SD_IO_RW_EXTENDED mode since
> SDIO block mode can be used under the hood even with data->blocks == 1.
>
> Those transfers are very rare, thus can be replaced by a bounce buffer
> without real performance loss.
>
> Fixes: 7412dee9f1fd ("mmc: meson-gx: replace WARN_ONCE with dev_warn_once about scatterlist size alignment in block mode")
> Reported-by: Christian Hewitt <[email protected]>
> Signed-off-by: Neil Armstrong <[email protected]>
Applied for fixes and by adding a stable tag, thanks!
Kind regards
Uffe
> ---
> drivers/mmc/host/meson-gx-mmc.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/mmc/host/meson-gx-mmc.c b/drivers/mmc/host/meson-gx-mmc.c
> index 1c61f0f24c09..016a6106151a 100644
> --- a/drivers/mmc/host/meson-gx-mmc.c
> +++ b/drivers/mmc/host/meson-gx-mmc.c
> @@ -236,7 +236,8 @@ static void meson_mmc_get_transfer_mode(struct mmc_host *mmc,
> if (host->dram_access_quirk)
> return;
>
> - if (data->blocks > 1) {
> + /* SD_IO_RW_EXTENDED (CMD53) can also use block mode under the hood */
> + if (data->blocks > 1 || mrq->cmd->opcode == SD_IO_RW_EXTENDED) {
> /*
> * In block mode DMA descriptor format, "length" field indicates
> * number of blocks and there is no way to pass DMA size that
> --
> 2.25.1
>