Re: [PATCH v3 0/2] Configuring DMA threshold value for DW-MMC controllers

Re: [PATCH v3 0/2] Configuring DMA threshold value for DW-MMC controllers

[Kaustabh Chakraborty]

On 2026-04-15 20:32 +05:30, Kaustabh Chakraborty wrote:
> In Samsung Exynos 7870 devices with Broadcom Wi-Fi, it has been observed
> that small sized DMA transfers are unreliable and are not written
> properly, which renders the cache incoherent.
>
> Experimental observations say that DMA transfer sizes of somewhere
> around 64 to 512 are intolerable. We must thus implement a mechanism to
> fall back to PIO transfer in this case. One such approach, which this
> series implements is allowing the DMA transfer threshold, which is
> already defined in the driver, to be configurable.
>
> Note that this patch is likely to be labelled as a workaround. These
> smaller transfers seem to be successful from downstream kernels,
> however efforts to figure out how so went in vain. It is also very
> possible that the downstream Broadcom Wi-Fi SDIO driver uses PIO
> transfers as well.
>
> Signed-off-by: Kaustabh Chakraborty <kauschluss@disroot.org>

Hi, bump! What's the status of this patch series? Both patches have been
reviewed.

Re: [PATCH v3 0/2] Configuring DMA threshold value for DW-MMC controllers

[Ulf Hansson]

On Wed, 15 Apr 2026 at 17:03, Kaustabh Chakraborty
<kauschluss@disroot.org> wrote:
>
> In Samsung Exynos 7870 devices with Broadcom Wi-Fi, it has been observed
> that small sized DMA transfers are unreliable and are not written
> properly, which renders the cache incoherent.
>
> Experimental observations say that DMA transfer sizes of somewhere
> around 64 to 512 are intolerable. We must thus implement a mechanism to
> fall back to PIO transfer in this case. One such approach, which this
> series implements is allowing the DMA transfer threshold, which is
> already defined in the driver, to be configurable.
>
> Note that this patch is likely to be labelled as a workaround. These
> smaller transfers seem to be successful from downstream kernels,
> however efforts to figure out how so went in vain. It is also very
> possible that the downstream Broadcom Wi-Fi SDIO driver uses PIO
> transfers as well.
>
> Signed-off-by: Kaustabh Chakraborty <kauschluss@disroot.org>
> ---
> Changes in v3:
> - Move host->dma_threshold default to dw_mci_alloc_host() (Shawn Lin)
> - Move comment to document struct dw_mci::dma_threshold (Shawn Lin)
> - Link to v2: https://lore.kernel.org/r/20260414-dwmmc-dma-thr-v2-0-4058078f5361@disroot.org

The v3 series applied for next, thanks!

Kind regards
Uffe


>
> Changes in v2:
> - Remove dt-binding to set DMA threshold (Krzysztof Kozlowski)
> - Add comment to describe struct dw_mci::dma_threshold (Shawn Lin)
> - Set DMA threshold in Exynos 7870 DW-MMC driver (Krzysztof Kozlowski)
> - Link to v1: https://lore.kernel.org/r/20260412-dwmmc-dma-thr-v1-0-75a2f658eee3@disroot.org
>
> ---
> Kaustabh Chakraborty (2):
>       mmc: dw_mmc: implement option for configuring DMA threshold
>       mmc: dw_mmc: exynos: increase DMA threshold value for exynos7870
>
>  drivers/mmc/host/dw_mmc-exynos.c | 1 +
>  drivers/mmc/host/dw_mmc.c        | 4 ++--
>  drivers/mmc/host/dw_mmc.h        | 2 ++
>  3 files changed, 5 insertions(+), 2 deletions(-)
> ---
> base-commit: 1c7cc4904160c6fc6377564140062d68a3dc93a0
> change-id: 20260412-dwmmc-dma-thr-1090d8285ea7
>
> Best regards,
> --
> Kaustabh Chakraborty <kauschluss@disroot.org>
>