Re: [PATCH v2 0/3] Add SDIO userspace passthrough support

[Shawn Lin <shawn.lin@rock-chips.com>]

Hi Ulf

在 2026/09/21 星期一 15:18, Shawn Lin 写道:
> SDIO cards are normally driven by function drivers (brcmfmac,
> mwifiex, rtw88_8723bs, ...), but bring-up, debugging and factory
> test frequently need to poke the SDIO register space directly:
> CCCR/FBR in function 0, or vendor registers before any function
> driver is bound, and some deployments, notably lightweight IoT
> boards, want to handle the SDIO function entirely from userspace.
> Today the only way is to write a tiny kernel driver per device.
> 

Gentle ping...


> So this series adds raw userspace access to the SDIO register
> space as well a small reference client. Meanwhile, adding a new
> documentation about how to write a proper SDIO function driver,
> because I have recently been working on SDIO function drivers
> and kept referring to the existing drivers for guidance. It turns
> out that the knowledge is scattered and there is no unified document
> describing how an SDIO function driver is supposed to be written,
> despite the MMC core having clear expectations about how the APIs
> are used.
> 
> In particular, the interrupt model has many pitfalls:
> sdio_claim_irq()/sdio_release_irq() must be called with the host
> claimed, the handler runs with the host already claimed and must
> not sleep or claim the bus itself, and sdio_release_irq() must be
> called from ->remove() because it is the last chance before the
> core removes the function devices from mmc_remove_host(). Recent
> fixes around the SDIO IRQ work and host removal and the discussions
> on the mailing list showed how easily these contracts get broken.
> 
> Also worth documenting is the buffer requirement of the CMD53 based
> helpers: the core builds a scatterlist directly from the caller's
> buffer and the host driver maps it for DMA, so vmalloc() or stack
> allocated buffers are not acceptable, while the CMD52 byte accessors
> are safe with stack buffers.
> 
> Changes-in-v2:
> - drop RFC as I don't include mmc-utils to be sdio usersapce tool
> - add sdio doc
> 
> Shawn Lin (3):
>    mmc: core: Add SDIO userspace passthrough character device
>    tools/sdio: add sdio-raw userspace client
>    docs: mmc: add introduction to writing SDIO function drivers
> 
>   Documentation/driver-api/mmc/index.rst             |   1 +
>   .../driver-api/mmc/sdio-function-drivers.rst       | 211 ++++++
>   MAINTAINERS                                        |   2 +
>   drivers/mmc/Kconfig                                |  11 +
>   drivers/mmc/core/Makefile                          |   1 +
>   drivers/mmc/core/sdio_bus.c                        |  12 +-
>   drivers/mmc/core/sdio_cdev.c                       | 189 ++++++
>   drivers/mmc/core/sdio_cdev.h                       |  21 +
>   include/linux/mmc/sdio_func.h                      |   4 +
>   include/uapi/linux/mmc/sdio_ioctl.h                |  48 ++
>   tools/Makefile                                     |  14 +-
>   tools/sdio/Build                                   |   1 +
>   tools/sdio/Makefile                                |  66 ++
>   tools/sdio/gen-sdio-ids.sh                         |  59 ++
>   tools/sdio/sdio-raw.c                              | 714 +++++++++++++++++++++
>   15 files changed, 1346 insertions(+), 8 deletions(-)
>   create mode 100644 Documentation/driver-api/mmc/sdio-function-drivers.rst
>   create mode 100644 drivers/mmc/core/sdio_cdev.c
>   create mode 100644 drivers/mmc/core/sdio_cdev.h
>   create mode 100644 include/uapi/linux/mmc/sdio_ioctl.h
>   create mode 100644 tools/sdio/Build
>   create mode 100644 tools/sdio/Makefile
>   create mode 100644 tools/sdio/gen-sdio-ids.sh
>   create mode 100644 tools/sdio/sdio-raw.c
>