Re: [PATCH v2 3/3] mmc: sdhci-of-aspeed: Sync capabilities from device tree to ast2600 SoC registers

[Andrew Jeffery]

On Mon, 3 May 2021, at 20:22, Steven Lee wrote:
> The 05/03/2021 13:04, Andrew Jeffery wrote:
> > Hi Steven,
> > 
> > On Mon, 3 May 2021, at 11:13, Steven Lee wrote:
> > > Sync Capbility Registers(SDIO140, SDIO144, SDIO240, SDIO244) of ast2600
> > > SoC from the device tree.
> > > The bit 26(Voltage Support 1.8v) of SDIO140/SDIO240 is set to 1 if
> > > "mmc-hs200-1_8v" or "sd-uhs-sdr104" is added in the device tree.
> > > The bit 1(SDR104 Supported) of SDR144/SDR244 is set to 1 if "sd-uhs-sdr104"
> > > is added in the device tree.
> > > "timing-phase" is synced to SDIO0F4(Colock Phase Control)
> > > 
> > > Signed-off-by: Steven Lee <steven_lee@xxxxxxxxxxxxxx>
> > > ---
> > >  drivers/mmc/host/sdhci-of-aspeed.c | 107 ++++++++++++++++++++++++++---
> > >  1 file changed, 98 insertions(+), 9 deletions(-)
> > > 
> > > diff --git a/drivers/mmc/host/sdhci-of-aspeed.c 
> > > b/drivers/mmc/host/sdhci-of-aspeed.c
> > > index 7d8692e90996..2d755bac777a 100644
> > > --- a/drivers/mmc/host/sdhci-of-aspeed.c
> > > +++ b/drivers/mmc/host/sdhci-of-aspeed.c
> > > @@ -13,6 +13,7 @@
> > >  #include <linux/of.h>
> > >  #include <linux/of_platform.h>
> > >  #include <linux/platform_device.h>
> > > +#include <linux/reset.h>
> > >  #include <linux/spinlock.h>
> > >  
> > >  #include "sdhci-pltfm.h"
> > > @@ -30,10 +31,18 @@
> > >  #define   ASPEED_SDC_S0_PHASE_IN_EN	BIT(2)
> > >  #define   ASPEED_SDC_S0_PHASE_OUT_EN	GENMASK(1, 0)
> > >  #define   ASPEED_SDC_PHASE_MAX		31
> > > +#define ASPEED_SDC_CAP1_1_8V           BIT(26)
> > > +#define ASPEED_SDC_CAP2_SDR104         BIT(1)
> > > +#define PROBE_AFTER_ASSET_DEASSERT     0x1
> > > +
> > > +struct aspeed_sdc_info {
> > > +	u32 flag;
> > > +};
> > >  
> > >  struct aspeed_sdc {
> > >  	struct clk *clk;
> > >  	struct resource *res;
> > > +	struct reset_control *rst;
> > >  
> > >  	spinlock_t lock;
> > >  	void __iomem *regs;
> > > @@ -72,6 +81,44 @@ struct aspeed_sdhci {
> > >  	const struct aspeed_sdhci_phase_desc *phase_desc;
> > >  };
> > >  
> > > +struct aspeed_sdc_info ast2600_sdc_info = {
> > > +	.flag = PROBE_AFTER_ASSET_DEASSERT
> > > +};
> > > +
> > > +/*
> > > + * The function sets the mirror register for updating
> > > + * capbilities of the current slot.
> > > + *
> > > + *   slot | cap_idx | caps_reg | mirror_reg
> > > + *   -----|---------|----------|------------
> > > + *     0  |    0    | SDIO140  |   SDIO10
> > > + *     0  |    1    | SDIO144  |   SDIO14
> > > + *     1  |    0    | SDIO240  |   SDIO20
> > > + *     1  |    1    | SDIO244  |   SDIO24
> > > + */
> > > +static void aspeed_sdc_set_slot_capability(struct sdhci_host *host,
> > > +					   struct aspeed_sdc *sdc,
> > > +					   u32 reg_val,
> > > +					   u8 slot,
> > > +					   u8 cap_idx)
> > 
> > Having thought about this some more now we have code, I wonder if we can get
> > rid of `cap_idx` as a parameter. Something like:
> > 
> > static void aspeed_sdc_set_slot_capability(struct sdhci_host *host,
> >     struct aspeed_sdc *sdc, int capability, bool enable, u8 slot);
> > 
> > From there, instead of
> > 
> > #define ASPEED_SDC_CAP1_1_8V           BIT(26)
> > #define ASPEED_SDC_CAP2_SDR104         BIT(1)
> > 
> > We do
> > 
> > /* SDIO{10,20} */
> > #define ASPEED_SDC_CAP1_1_8V           (0 * 32 + 26)
> > /* SDIO{14,24} */
> > #define ASPEED_SDC_CAP2_SDR104         (1 * 32 + 1)
> > 
> > Then in the implementation of aspeed_sdc_set_slot_capability() we 
> > derive cap_idx and reg_val:
> > 
> > u8 reg_val = enable * BIT(capability % 32);
> > u8 cap_reg = capability / 32;
> > 
> > That way we get rid of the 0 and 1 magic values for cap_idx when 
> > invoking aspeed_sdc_set_slot_capability() and the caller can't
> > accidentally pass the wrong value.
> > 
> 
> Thanks for the detailed suggestion, I will modify the function as you
> suggested.

Great!

> 
> > > +{
> > > +	u8 caps_reg_offset;
> > > +	u32 caps_reg;
> > > +	u32 mirror_reg_offset;
> > > +	u32 caps_val;
> > > +
> > > +	if (cap_idx > 1 || slot > 1)
> > > +		return;
> > > +
> > > +	caps_reg_offset = (cap_idx == 0) ? 0 : 4;
> > > +	caps_reg = 0x40 + caps_reg_offset;
> > > +	caps_val = sdhci_readl(host, caps_reg);
> > 
> > Hmm, I think you used sdhci_readl() because I commented on that last 
> > time. If the global-area registers are truly mirrored we could read 
> > from them as well right? In which case we could just use 
> > readl(sdc->regs + mirror_reg_offset)? If so we can drop the host 
> > parameter and (incorporating my suggestion above) just have:
> > 
> > static void aspeed_sdc_set_slot_capability(struct aspeed_sdc *sdc,
> >     int capability, bool enable, u8 slot);
> > 
> > Sorry if I've sort of flip-flopped on that, but I think originally you 
> > were still reading from the SDHCI (read-only) address?
> > 
> 
> Yes, mirror registers are used to update the capability register, it returns
> zero if we read the mirror register.
> The test result is as follows:
> 
> # devmem 0x1e740010 32             // Read SDIO010(Mirror of SDIO140)
> 0x00000000
> 
> # devmem 0x1e740140 32             // Read capability
> 0x07FC0080
> 
> # devmem 0x1e740010 32 0x07fb0080  // Write mirror
> 
> # devmem 0x1e740010 32             // Read mirror
> 0x00000000
> 
> # devmem 0x1e740140 32             // Read capability
> 0x07FB0080

Ah well, I guess we continue to pass the struct sdhci_host pointer then.

> 
> > > +	caps_val |= reg_val;
> > > +	mirror_reg_offset = (slot == 0) ? 0x10 : 0x20;
> > > +	mirror_reg_offset += caps_reg_offset;
> > > +	writel(caps_val, sdc->regs + mirror_reg_offset);
> > > +}
> > > +
> > >  static void aspeed_sdc_configure_8bit_mode(struct aspeed_sdc *sdc,
> > >  					   struct aspeed_sdhci *sdhci,
> > >  					   bool bus8)
> > > @@ -329,9 +376,11 @@ static int aspeed_sdhci_probe(struct platform_device *pdev)
> > >  {
> > >  	const struct aspeed_sdhci_pdata *aspeed_pdata;
> > >  	struct sdhci_pltfm_host *pltfm_host;
> > > +	struct device_node *np = pdev->dev.of_node;
> > >  	struct aspeed_sdhci *dev;
> > >  	struct sdhci_host *host;
> > >  	struct resource *res;
> > > +	u32 reg_val;
> > >  	int slot;
> > >  	int ret;
> > >  
> > > @@ -372,6 +421,21 @@ static int aspeed_sdhci_probe(struct platform_device *pdev)
> > >  
> > >  	sdhci_get_of_property(pdev);
> > >  
> > > +	if (of_property_read_bool(np, "mmc-hs200-1_8v") ||
> > > +	    of_property_read_bool(np, "sd-uhs-sdr104"))
> > > +		aspeed_sdc_set_slot_capability(host,
> > > +					       dev->parent,
> > > +					       ASPEED_SDC_CAP1_1_8V,
> > > +					       slot,
> > > +					       0);
> > 
> > See the discussion above about reworking aspeed_sdc_set_slot_capability.
> > 
> 
> Will do it.
> 
> > > +
> > > +	if (of_property_read_bool(np, "sd-uhs-sdr104"))
> > > +		aspeed_sdc_set_slot_capability(host,
> > > +					       dev->parent,
> > > +					       ASPEED_SDC_CAP2_SDR104,
> > > +					       slot,
> > > +					       1);
> > 
> > Again here.
> > 
> 
> Will do it.
> 
> > > +
> > >  	pltfm_host->clk = devm_clk_get(&pdev->dev, NULL);
> > >  	if (IS_ERR(pltfm_host->clk))
> > >  		return PTR_ERR(pltfm_host->clk);
> > > @@ -476,12 +540,25 @@ static struct platform_driver aspeed_sdhci_driver = {
> > >  	.remove		= aspeed_sdhci_remove,
> > >  };
> > >  
> > > +static const struct of_device_id aspeed_sdc_of_match[] = {
> > > +	{ .compatible = "aspeed,ast2400-sd-controller", },
> > > +	{ .compatible = "aspeed,ast2500-sd-controller", },
> > > +	{ .compatible = "aspeed,ast2600-sd-controller", .data = &ast2600_sdc_info},
> > > +	{ }
> > > +};
> > > +
> > > +MODULE_DEVICE_TABLE(of, aspeed_sdc_of_match);
> > > +
> > >  static int aspeed_sdc_probe(struct platform_device *pdev)
> > >  
> > >  {
> > >  	struct device_node *parent, *child;
> > >  	struct aspeed_sdc *sdc;
> > > +	const struct of_device_id *match = NULL;
> > > +	const struct aspeed_sdc_info *info = NULL;
> > > +
> > >  	int ret;
> > > +	u32 timing_phase;
> > >  
> > >  	sdc = devm_kzalloc(&pdev->dev, sizeof(*sdc), GFP_KERNEL);
> > >  	if (!sdc)
> > > @@ -489,6 +566,23 @@ static int aspeed_sdc_probe(struct platform_device *pdev)
> > >  
> > >  	spin_lock_init(&sdc->lock);
> > >  
> > > +	match = of_match_device(aspeed_sdc_of_match, &pdev->dev);
> > > +	if (!match)
> > > +		return -ENODEV;
> > > +
> > > +	if (match->data)
> > > +		info = match->data;
> > > +
> > > +	if (info) {
> > > +		if (info->flag & PROBE_AFTER_ASSET_DEASSERT) {
> > > +			sdc->rst = devm_reset_control_get(&pdev->dev, NULL);
> > > +			if (!IS_ERR(sdc->rst)) {
> > > +				reset_control_assert(sdc->rst);
> > > +				reset_control_deassert(sdc->rst);
> > > +			}
> > > +		}
> > > +	}
> > 
> > I think this should be a separate patch.
> > 
> > From the code it seems that this is necessary for just the 2600? Where 
> > is this documented?
> > 
> 
> Yes it is just for 2600. The patch is suggested by our chip designer and
> is used for cleaning up MMC controller.
> Currently, there is no document describes this changes.
> 
> And I have a question regarding the "separate patch", does it mean I should
> create another patch set or I can add a new patch in the current
> patch set?

Depends what you mean by this :)

It's kind-of awkward to send another patch as part of the existing v2 
of the series, as you'll wind up with what is effectively patch 4/3. 
It's less confusing to just send a v3 with all 4 patches.

However, in general if patches don't depend on each other it's good to 
send them as separate series, that way the series can be applied in any 
order.

> 
> > > +
> > >  	sdc->clk = devm_clk_get(&pdev->dev, NULL);
> > >  	if (IS_ERR(sdc->clk))
> > >  		return PTR_ERR(sdc->clk);
> > > @@ -506,6 +600,10 @@ static int aspeed_sdc_probe(struct platform_device *pdev)
> > >  		goto err_clk;
> > >  	}
> > >  
> > > +	if (!of_property_read_u32(pdev->dev.of_node,
> > > +				  "timing-phase", &timing_phase))
> > > +		writel(timing_phase, sdc->regs + ASPEED_SDC_PHASE);
> > 
> > I asked on v1 that you use the phase support already in the bindings 
> > and in the driver. The example you added in the binding document[1] 
> > used the existing devicetree properties but it seems you haven't fixed 
> > the code.
> > 
> > Please drop your phase implementation from the patch.
> > 
> 
> Sorry, I misunderstand the comment in the v1 patch. I thought that I should use
> the exists ASPEED_SDC_PHASE for timing-phase.

Ah!

> 
> Now I think I understand what you mean in the previous review.
> I will remove the implementation you mentioned and add the following setting in
> the device tree to verify again.
> 
>  clk-phase-mmc-hs200 = <N>, <N>;

Right, that's what I was suggesting. We have support for most of the 
other speeds and as well (not just HS200, just that HS200 is what 
Rainier cares about), see:

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/mmc/core/host.c?h=v5.12#n181

Cheers,

Andrew