Re: [PATCH v11 3/3] mmc: Add driver for LiteX's LiteSDCard interface

[Stafford Horne]

On Sun, Jan 09, 2022 at 06:20:03PM -0500, Gabriel Somlo wrote:
> LiteX (https://github.com/enjoy-digital/litex) is a SoC framework
> that targets FPGAs. LiteSDCard is a small footprint, configurable
> SDCard core commonly used in LiteX designs.
> 
> The driver was first written in May 2020 and has been maintained
> cooperatively by the LiteX community. Thanks to all contributors!
> 
> Co-developed-by: Kamil Rakoczy <krakoczy@xxxxxxxxxxxx>
> Signed-off-by: Kamil Rakoczy <krakoczy@xxxxxxxxxxxx>
> Co-developed-by: Maciej Dudek <mdudek@xxxxxxxxxxxxxxxxxxxxxxxx>
> Signed-off-by: Maciej Dudek <mdudek@xxxxxxxxxxxxxxxxxxxxxxxx>
> Co-developed-by: Paul Mackerras <paulus@xxxxxxxxxx>
> Signed-off-by: Paul Mackerras <paulus@xxxxxxxxxx>
> Signed-off-by: Gabriel Somlo <gsomlo@xxxxxxxxx>
> Reviewed-by: Joel Stanley <joel@xxxxxxxxx>
> Reviewed-by: Andy Shevchenko <andy.shevchenko@xxxxxxxxx>
> ---
> 
> New in v11:
>   - picked up r/b from Andy
> drivers/mmc/host/litex_mmc.c:
>   - defensive coding of litex_mmc_interrupt() return logic
>   - remove `dev` member of `struct litex_mmc_host`, only used during probe
> 
> >New in v10:
> >drivers/mmc/host/litex_mmc.c:
> >  - group `linux/mmc/*` includes by themselves
> >  - clean-up of `return` style (multiple locations throughout source)
> >  - create `mmc_free_host()` wrapper for use with 
> >    `devm_add_action_or_reset()`
> >  - use GFP_KERNEL with `dmam_alloc_coherent()`
> >
> >>New in v9:
> >>drivers/mmc/host/Kconfig:
> >>  - fix OF dependency
> >>drivers/mmc/host/litex_mmc.c:
> >>  - remove `linux/of.h` include, no longer needed since dropping
> >>    `of_match_ptr()`
> >>  - add `linux/mod_devicetable.h` include
> >>  - use devm_action_or_reset() to devm-ify mmc_alloc_host(), and obviate
> >>    the need to call mmc_free_host() explicitly during either probe()
> >>    error path or during remove()
> >>
> >>>New in v8:
> >>>  - remove `Cc:` lines from commit blurb
> >>>drivers/mmc/host/litex_mmc.c:
> >>>  - fix file header comment (for real, this time)
> >>>  - add explicit `bits.h` include
> >>>  - remove `of_match_ptr()` wrapper from around .of_match_table argument
> >>>  - fix devm ordering issues: use `devm_request_irq()`, which precludes
> >>>    the need to call `free_irq()` on `probe()` error path or from `remove()`
> >>>
> >>>>New in v7:
> >>>>drivers/mmc/host/Kconfig:
> >>>>  - added module name in LiteSDCard Kconfig entry
> >>>>drivers/mmc/host/litex_mmc.c:
> >>>>  - fixed comment formatting, ordering, and capitalization throughout
> >>>>    the entire file
> >>>>  - sorted header #include statements
> >>>>  - removed redundant parantheses in readx_poll_timeout() condition
> >>>>  - explicit handling of readx_poll_timeout() timeout scenarios
> >>>>  - dev_err() used in litex_mmc_sdcard_wait_done()
> >>>>  - use memcpy_fromio() to grab command response
> >>>>  - no need to apply 0xffff mask to a 32-bit value right-shifted by 16
> >>>>    (host->resp[3])
> >>>>  - use clamp() instead of min(max(...)...)
> >>>>  - reworked platform_get_irq_optional() error handling logic
> >>>>  - no need to explicitly zero host->irq, kzalloc() does that already
> >>>>  - added missing free_irq() in litex_mmc_probe() error path
> >>>>  - reordered calls inside litex_mmc_remove() (calling mmc_free_host()
> >>>>    before free_irq()
> >>>>
> >>>>>New in v6:
> >>>>>  - fix handling of deferred probe vs. platform_get_irq_optional()
> >>>>>  - don't #ifdef dma_set_mask_and_coherent(), since it automatically
> >>>>>    does the right thing on both 32- and 64-bit DMA capable arches
> >>>>>  - remove MMC_CAP2_FULL_PWR_CYCLE, add MMC_CAP2_NO_MMC to list of
> >>>>>    hardcoded capabilities during litex_mmc_probe()
> >>>>>  - hardcode mmc->ocr_avail to the full 2.7-3.6V range allowed by the
> >>>>>    SDCard spec (the LiteSDCard device doesn't accept software
> >>>>>    configuration)
> >>>>>
> >>>>>>New in v5:
> >>>>>>  - shorter #define constant names (cosmetic, make them less unwieldy)
> >>>>>>  - picked up reviewed-by Joel
> >>>>>>
> >>>>>>>New in v4:
> >>>>>>>  - struct litex_mmc_host fields re-ordered so that `pahole` reports
> >>>>>>>    no holes on either 32- or 64-bit builds
> >>>>>>>  - litex_mmc_set_bus_width() now encapsulates check for
> >>>>>>>    host->is_bus_width_set
> >>>>>>>  - litex_mmc_request() - factor out dma data setup into separate
> >>>>>>>    helper function: litex_mmc_do_dma()
> >>>>>>>
> >>>>>>>> New in v3:
> >>>>>>>>   - fixed function signature (no line split), and naming (litex_mmc_*)
> >>>>>>>>   - more informative MODULE_AUTHOR() entries
> >>>>>>>>     - also added matching "Copyright" entries in file header
> >>>>>>>>   - fixed description and dependencies in Kconfig
> >>>>>>>>   - removed magic constants
> >>>>>>>>   - removed litex_map_status(), have sdcard_wait_done() return *real*
> >>>>>>>>     error codes directly instead.
> >>>>>>>>   - streamlined litex_mmc_reponse_len()
> >>>>>>>>   - call litex_mmc_set_bus_width() only once, and ensure it returns
> >>>>>>>>     correct error code(s)
> >>>>>>>>   - use readx_poll_timeout() -- more concise -- instead of
> >>>>>>>>     read_poll_timeout()
> >>>>>>>>   - use dev_err() in litex_mmc_send_cmd() (instead of pr_err())
> >>>>>>>>   - litex_mmc_setclk() will update host->clock before returning
> >>>>>>>>   - separate irq initialization into its own function,
> >>>>>>>>     litex_mmc_irq_init()
> >>>>>>>>   - document rationale for f_min, f_max
> >>>>>>>>   - use dmam_alloc_coherent(), which simplifies cleanup significantly
> >>>>>>>>   - bump dma_set_mask_and_coherent() to 64-bits on suitable
> >>>>>>>>     architectures
> >>>>>>>>   - clock source picked up from dedicated DT clock reference property
> >>>>>>>>   - remove gpio card-detect logic (needs testing and a dt binding
> >>>>>>>>     example before being eligible for upstream inclusion)
> >>>>>>>>   - large `if (data) { ... }` block in litex_mmc_request() left as-is,
> >>>>>>>>     there are too many variables shared with the rest of the parent
> >>>>>>>>     function body to easily separate (e.g., `len`, `transfer`, `direct`).
> >>>>>>>>     If this is indeed a blocker, I can take another shot at refactoring
> >>>>>>>>     it in a future revision!
> 
>  drivers/mmc/host/Kconfig     |   9 +
>  drivers/mmc/host/Makefile    |   1 +
>  drivers/mmc/host/litex_mmc.c | 657 +++++++++++++++++++++++++++++++++++
>  3 files changed, 667 insertions(+)
>  create mode 100644 drivers/mmc/host/litex_mmc.c
> 
> diff --git a/drivers/mmc/host/Kconfig b/drivers/mmc/host/Kconfig
> index 5af8494c31b5..8f661f52b502 100644
> --- a/drivers/mmc/host/Kconfig
> +++ b/drivers/mmc/host/Kconfig
> @@ -1093,3 +1093,12 @@ config MMC_OWL
>  
>  config MMC_SDHCI_EXTERNAL_DMA
>  	bool
> +
> +config MMC_LITEX
> +	tristate "LiteX MMC Host Controller support"
> +	depends on ((PPC_MICROWATT || LITEX) && OF) || COMPILE_TEST
> +	help
> +	  This selects support for the MMC Host Controller found in LiteX SoCs.

See comment below.

> +	  To compile this driver as a module, choose M here: the
> +	  module will be called litex_mmc.

Can we add this here too?

	If unsure, say N.

> diff --git a/drivers/mmc/host/Makefile b/drivers/mmc/host/Makefile
> index ea36d379bd3c..4e4ceb32c4b4 100644
> --- a/drivers/mmc/host/Makefile
> +++ b/drivers/mmc/host/Makefile
> @@ -101,6 +101,7 @@ obj-$(CONFIG_MMC_CQHCI)			+= cqhci.o
>  cqhci-y					+= cqhci-core.o
>  cqhci-$(CONFIG_MMC_CRYPTO)		+= cqhci-crypto.o
>  obj-$(CONFIG_MMC_HSQ)			+= mmc_hsq.o
> +obj-$(CONFIG_MMC_LITEX)			+= litex_mmc.o
>  
>  ifeq ($(CONFIG_CB710_DEBUG),y)
>  	CFLAGS-cb710-mmc	+= -DDEBUG
> diff --git a/drivers/mmc/host/litex_mmc.c b/drivers/mmc/host/litex_mmc.c
> new file mode 100644
> index 000000000000..cc2848030e6b
> --- /dev/null
> +++ b/drivers/mmc/host/litex_mmc.c
> @@ -0,0 +1,657 @@
...
> +
> +	/* Initialize clock source */
> +	clk = devm_clk_get(dev, NULL);
> +	if (IS_ERR(clk))
> +		return dev_err_probe(dev, PTR_ERR(clk), "can't get clock\n");
> +	host->ref_clk = clk_get_rate(clk);

Note, we used to have:

	cpu = of_get_next_cpu_node(NULL);
	ret = of_property_read_u32(cpu, "clock-frequency", &host->freq);
	of_node_put(cpu);
	if (ret) {
		dev_err(&pdev->dev, "No \"clock-frequency\" property in DT\n");
		goto err_free_host;
	}

As we agree, this worked but was hacky.  The new use of devm_clk_get above uses
the COMMON_CLK framework.  Which is great but OpenRISC doesn't yet support
COMMON_CLK, so this driver no longer works on OpenRISC.

I think the ideal thing is to not change the driver code and I will just add
support for COMMON_CLK on openrisc (already done and tested).  But as we
discussed off e-mail I think we should change the config constraints to.

    depends on ((PPC_MICROWATT || LITEX) && OF && HAVE_CLK) || COMPILE_TEST

-Stafford