Hi Thierry,
Thanks for review. I will amend accordingly and submit a new patch.
JC
On 8/31/20 8:09 PM, Thierry Reding wrote:
> On Mon, Aug 31, 2020 at 12:40:37PM +0800, JC Kuo wrote:
>> This commit implements a register map which grants USB (UTMI and HSIC)
>> sleepwalk registers access to USB phy drivers. The USB sleepwalk logic
>> is in PMC hardware block but USB phy drivers have the best knowledge
>> of proper programming sequence. This approach prevents using custom
>> pmc APIs.
>>
>> Signed-off-by: JC Kuo <jc...@nvidia.com>
>> ---
>> drivers/soc/tegra/pmc.c | 89 +++++++++++++++++++++++++++++++++++++++++
>> 1 file changed, 89 insertions(+)
>
> Same comment as in earlier patches regarding the subject and "USB PHY"
> in the commit message.
>
>> diff --git a/drivers/soc/tegra/pmc.c b/drivers/soc/tegra/pmc.c
>> index d332e5d9abac..03317978915a 100644
>> --- a/drivers/soc/tegra/pmc.c
>> +++ b/drivers/soc/tegra/pmc.c
>> @@ -43,6 +43,7 @@
>> #include <linux/seq_file.h>
>> #include <linux/slab.h>
>> #include <linux/spinlock.h>
>> +#include <linux/regmap.h>
>>
>> #include <soc/tegra/common.h>
>> #include <soc/tegra/fuse.h>
>> @@ -102,6 +103,9 @@
>>
>> #define PMC_PWR_DET_VALUE 0xe4
>>
>> +#define PMC_USB_DEBOUNCE_DEL 0xec
>> +#define PMC_USB_AO 0xf0
>> +
>> #define PMC_SCRATCH41 0x140
>>
>> #define PMC_WAKE2_MASK 0x160
>> @@ -133,6 +137,13 @@
>> #define IO_DPD2_STATUS 0x1c4
>> #define SEL_DPD_TIM 0x1c8
>>
>> +#define PMC_UTMIP_UHSIC_TRIGGERS 0x1ec
>> +#define PMC_UTMIP_UHSIC_SAVED_STATE 0x1f0
>> +
>> +#define PMC_UTMIP_TERM_PAD_CFG 0x1f8
>> +#define PMC_UTMIP_UHSIC_SLEEP_CFG 0x1fc
>> +#define PMC_UTMIP_UHSIC_FAKE 0x218
>> +
>> #define PMC_SCRATCH54 0x258
>> #define PMC_SCRATCH54_DATA_SHIFT 8
>> #define PMC_SCRATCH54_ADDR_SHIFT 0
>> @@ -145,8 +156,18 @@
>> #define PMC_SCRATCH55_CHECKSUM_SHIFT 16
>> #define PMC_SCRATCH55_I2CSLV1_SHIFT 0
>>
>> +#define PMC_UTMIP_UHSIC_LINE_WAKEUP 0x26c
>> +
>> +#define PMC_UTMIP_BIAS_MASTER_CNTRL 0x270
>> +#define PMC_UTMIP_MASTER_CONFIG 0x274
>> +#define PMC_UTMIP_UHSIC2_TRIGGERS 0x27c
>> +#define PMC_UTMIP_MASTER2_CONFIG 0x29c
>> +
>> #define GPU_RG_CNTRL 0x2d4
>>
>> +#define PMC_UTMIP_PAD_CFG0 0x4c0
>> +#define PMC_UTMIP_UHSIC_SLEEP_CFG1 0x4d0
>> +#define PMC_UTMIP_SLEEPWALK_P3 0x4e0
>> /* Tegra186 and later */
>> #define WAKE_AOWAKE_CNTRL(x) (0x000 + ((x) << 2))
>> #define WAKE_AOWAKE_CNTRL_LEVEL (1 << 3)
>> @@ -334,6 +355,7 @@ struct tegra_pmc_soc {
>> const struct pmc_clk_init_data *pmc_clks_data;
>> unsigned int num_pmc_clks;
>> bool has_blink_output;
>> + bool has_usb_sleepwalk;
>> };
>>
>> static const char * const tegra186_reset_sources[] = {
>> @@ -2495,6 +2517,67 @@ static void tegra_pmc_clock_register(struct tegra_pmc
>> *pmc,
>> err);
>> }
>>
>> +#define regmap_reg(x) regmap_reg_range(x, x)
>
> This doesn't seem like a good idea. What if anyone ever thought it was a
> good idea to add this to the core regmap header? We'd get a naming
> conflict that would first have to get resolved.
>
>> +static const struct regmap_range pmc_usb_sleepwalk_ranges[] = {
>> + regmap_reg_range(PMC_USB_DEBOUNCE_DEL, PMC_USB_AO),
>> + regmap_reg_range(PMC_UTMIP_UHSIC_TRIGGERS, PMC_UTMIP_UHSIC_SAVED_STATE),
>> + regmap_reg_range(PMC_UTMIP_TERM_PAD_CFG, PMC_UTMIP_UHSIC_FAKE),
>> + regmap_reg(PMC_UTMIP_UHSIC_LINE_WAKEUP),
>> + regmap_reg_range(PMC_UTMIP_BIAS_MASTER_CNTRL, PMC_UTMIP_MASTER_CONFIG),
>> + regmap_reg_range(PMC_UTMIP_UHSIC2_TRIGGERS, PMC_UTMIP_MASTER2_CONFIG),
>> + regmap_reg_range(PMC_UTMIP_PAD_CFG0, PMC_UTMIP_UHSIC_SLEEP_CFG1),
>> + regmap_reg(PMC_UTMIP_SLEEPWALK_P3),
>> +};
>
> Since we only have two usages of the regmap_reg() macro, perhaps just
> use regmap_reg_range() with the same parameter used twice?
>
>> +
>> +static const struct regmap_access_table pmc_usb_sleepwalk_table = {
>> + .yes_ranges = pmc_usb_sleepwalk_ranges,
>> + .n_yes_ranges = ARRAY_SIZE(pmc_usb_sleepwalk_ranges),
>> +};
>> +
>> +static int tegra_pmc_regmap_readl(void *context, unsigned int reg, unsigned
>> int *val)
>
> s/reg/offset/ to make it clearer what this really is. Also: s/val/value/ to
> avoid potential confusion with "valid".
>
>> +{
>> + struct tegra_pmc *pmc = context;
>> +
>> + *val = tegra_pmc_readl(pmc, reg);
>> + return 0;
>> +}
>> +
>> +static int tegra_pmc_regmap_writel(void *context, unsigned int reg,
>> unsigned int val)
>> +{
>> + struct tegra_pmc *pmc = context;
>> +
>> + tegra_pmc_writel(pmc, val, reg);
>> + return 0;
>> +}
>
> Same here.
>
>> +
>> +static const struct regmap_config usb_sleepwalk_regmap_config = {
>> + .name = "usb_sleepwalk",
>> + .reg_bits = 32,
>> + .val_bits = 32,
>> + .reg_stride = 4,
>> + .fast_io = true,
>> + .rd_table = &pmc_usb_sleepwalk_table,
>> + .wr_table = &pmc_usb_sleepwalk_table,
>> + .reg_read = tegra_pmc_regmap_readl,
>> + .reg_write = tegra_pmc_regmap_writel,
>> +};
>> +
>> +static int tegra_pmc_regmap_init(struct tegra_pmc *pmc)
>> +{
>> + struct regmap *regmap;
>> +
>> + if (pmc->soc->has_usb_sleepwalk) {
>> + regmap = devm_regmap_init(pmc->dev, NULL, (__force void *)pmc,
>
> Do you really need that __force in there?
>
>> + &usb_sleepwalk_regmap_config);
>> + if (IS_ERR(regmap)) {
>> + dev_err(pmc->dev, "failed to allocate register map\n");
>
> Maybe output the error code here?
>
>> + return PTR_ERR(regmap);
>> + }
>> + }
>> +
>> + return 0;
>> +}
>> +
>> static int tegra_pmc_probe(struct platform_device *pdev)
>> {
>> void __iomem *base;
>> @@ -2613,6 +2696,10 @@ static int tegra_pmc_probe(struct platform_device
>> *pdev)
>> pmc->base = base;
>> mutex_unlock(&pmc->powergates_lock);
>>
>> + err = tegra_pmc_regmap_init(pmc);
>> + if (err < 0)
>> + goto cleanup_powergates;
>
> You could call this perhaps a little bit earlier to avoid having to
> clean up powergates? Since you register with devm_regmap_init() you
> won't have to clean this up manually.
>
> For that reason it makes sense as a general rule to initialize devm
> things before anything that's not managed (unless, of course, if it
> doesn't make any sense).
>
>> +
>> tegra_pmc_clock_register(pmc, pdev->dev.of_node);
>> platform_set_drvdata(pdev, pmc);
>>
>> @@ -2976,6 +3063,7 @@ static const struct tegra_pmc_soc tegra124_pmc_soc = {
>> .pmc_clks_data = tegra_pmc_clks_data,
>> .num_pmc_clks = ARRAY_SIZE(tegra_pmc_clks_data),
>> .has_blink_output = true,
>> + .has_usb_sleepwalk = true,
>> };
>>
>> static const char * const tegra210_powergates[] = {
>> @@ -3094,6 +3182,7 @@ static const struct tegra_pmc_soc tegra210_pmc_soc = {
>> .pmc_clks_data = tegra_pmc_clks_data,
>> .num_pmc_clks = ARRAY_SIZE(tegra_pmc_clks_data),
>> .has_blink_output = true,
>> + .has_usb_sleepwalk = true,
>> };
>>
>> #define TEGRA186_IO_PAD_TABLE(_pad)
>> \
>
> I'd prefer if we explicitly set this to false on SoC generations where
> we don't have sleepwalk support (or don't need to deal with it in the
> kernel). That avoids confusion as to whether this was simply forgotten
> or whether the omission was on purpose.
>
> Thierry
>
