Re: [PATCH v2 2/4] soc: mediatek: Add support of WAY_EN operations

Hi Angelo,

Thanks for your review, I fixed most, comments inline.

On Mon, Jul 25, 2022 at 11:55:27AM +0200, AngeloGioacchino Del Regno wrote:

Il 25/07/22 10:18, Markus Schneider-Pargmann ha scritto:

quoted

From: Alexandre Bailon <redacted>

This updates the power domain to support WAY_EN operations. These
operations enable a path between different units of the chip and are
labeled as 'way_en' in the register descriptions.

This operation is required by the mt8365 for the MM power domain.

Signed-off-by: Alexandre Bailon <redacted>
Signed-off-by: Fabien Parent <redacted>
Signed-off-by: Markus Schneider-Pargmann <msp@baylibre.com>
---

Notes:
     Changes in v2:
     - some minor style fixes.
     - Renamed 'wayen' to 'way_en' to clarify the meaning
     - Updated commit message

  drivers/soc/mediatek/mtk-pm-domains.c | 64 +++++++++++++++++++++------
  drivers/soc/mediatek/mtk-pm-domains.h | 28 +++++++-----
  2 files changed, 68 insertions(+), 24 deletions(-)

diff --git a/drivers/soc/mediatek/mtk-pm-domains.c b/drivers/soc/mediatek/mtk-pm-domains.c
index 5ced254b082b..d0eae2227813 100644
--- a/drivers/soc/mediatek/mtk-pm-domains.c
+++ b/drivers/soc/mediatek/mtk-pm-domains.c

@@ -44,6 +44,7 @@ struct scpsys_domain {
  	struct clk_bulk_data *subsys_clks;
  	struct regmap *infracfg;
  	struct regmap *smi;
+	struct regmap *infracfg_nao;

What does "nao" mean?

I couldn't find the meaning of nao right now. It is the name of the
infracfg node in the datasheet. The normal one is called 'infracfg_ao'
the other one 'infracfg_nao' as far as I can see.

Besides, please move that before *infracfg to at least keep the same type members
alphabetically sorted..

quoted

  	struct regulator *supply;
  };

@@ -116,23 +117,38 @@ static int scpsys_sram_disable(struct scpsys_domain *pd)
  					MTK_POLL_TIMEOUT);
  }
-static int _scpsys_bus_protect_enable(const struct scpsys_bus_prot_data *bpd, struct regmap *regmap)
+static int _scpsys_bus_protect_enable(const struct scpsys_bus_prot_data *bpd,
+				      struct regmap *regmap, struct regmap *infracfg_nao)
  {
  	int i, ret;
  	for (i = 0; i < SPM_MAX_BUS_PROT_DATA; i++) {
-		u32 val, mask = bpd[i].bus_prot_mask;
+		u32 mask = bpd[i].bus_prot_mask;
+		u32 val = mask, sta_mask = mask;

You have modified the macros to use sta_mask as mask, so, why are you doing
that distinction in here between the two? You can simply keep assigning

		u32 mask = bpd[1].bus_prot_mask;
		u32 sta_mask = bpd[1].bus_prot_sta_mask;

quoted

+		struct regmap *ack_regmap = regmap;

Double assignment. You're reassigning this if way_en == true.

quoted

  		if (!mask)
  			break;
+		if (bpd[i].way_en) {
+			if (!infracfg_nao)
+				return -ENODEV;
+
+			val = 0;
+			sta_mask = bpd[i].bus_prot_sta_mask;
+			ack_regmap = infracfg_nao;
+		}

		if (bpd[i].way_en) {
			ack_regmap = regmap_nao;
			val = 0;
		} else {
			ack_regmap = regmap;
			val = mask;
		}

quoted

+
  		if (bpd[i].bus_prot_reg_update)
-			regmap_set_bits(regmap, bpd[i].bus_prot_set, mask);
+			regmap_update_bits(regmap, bpd[i].bus_prot_set, mask, val);
  		else
  			regmap_write(regmap, bpd[i].bus_prot_set, mask);
-		ret = regmap_read_poll_timeout(regmap, bpd[i].bus_prot_sta,
-					       val, (val & mask) == mask,
+		if (bpd[i].ignore_clr_ack)
+			continue;

You're adding that ignore_clr_ack here in the bus prot enablement function
which wasn't here before... and I didn't check carefully, but I think that
this is wrong: as the name says, it's to "ignore CLEAR ack", we're not doing
any clearing here, we're not in bus_protect_disable.

If you're really sure that this is not a mistake, you should guard it for way_en.

We are clearing bits here if way_en is true and bus_prot_reg_update is
true as well. Then val=0 and regmap_update_bits(..., mask, val) will
clear the bits given in mask. And yes either way_en or val==0 should
probably be checked here. Thanks.

quoted

+
+		ret = regmap_read_poll_timeout(ack_regmap, bpd[i].bus_prot_sta,
+					       val, (val & sta_mask) == sta_mask,
  					       MTK_POLL_DELAY_US, MTK_POLL_TIMEOUT);
  		if (ret)
  			return ret;

@@ -145,34 +161,49 @@ static int scpsys_bus_protect_enable(struct scpsys_domain *pd)
  {
  	int ret;
-	ret = _scpsys_bus_protect_enable(pd->data->bp_infracfg, pd->infracfg);
+	ret = _scpsys_bus_protect_enable(pd->data->bp_infracfg,
+					 pd->infracfg, pd->infracfg_nao);
  	if (ret)
  		return ret;
-	return _scpsys_bus_protect_enable(pd->data->bp_smi, pd->smi);
+	return _scpsys_bus_protect_enable(pd->data->bp_smi, pd->smi, NULL);
  }
+#define mask_cond(way_en, val, mask) \
+	((way_en && ((val & mask) == mask)) || (!way_en && !(val & mask)))
+
  static int _scpsys_bus_protect_disable(const struct scpsys_bus_prot_data *bpd,
-				       struct regmap *regmap)
+				       struct regmap *regmap, struct regmap *infracfg_nao)
  {
  	int i, ret;
  	for (i = SPM_MAX_BUS_PROT_DATA - 1; i >= 0; i--) {
-		u32 val, mask = bpd[i].bus_prot_mask;
+		u32 val = 0, mask = bpd[i].bus_prot_mask;
+		u32 sta_mask = mask;
+		struct regmap *ack_regmap = regmap;
  		if (!mask)
  			continue;
+		if (bpd[i].way_en) {
+			if (!infracfg_nao)
+				return -ENODEV;
+
+			val = mask;
+			sta_mask = bpd[i].bus_prot_sta_mask;
+			ack_regmap = infracfg_nao;
+		}
+
  		if (bpd[i].bus_prot_reg_update)
-			regmap_clear_bits(regmap, bpd[i].bus_prot_clr, mask);
+			regmap_update_bits(regmap, bpd[i].bus_prot_clr, mask, val);
  		else
  			regmap_write(regmap, bpd[i].bus_prot_clr, mask);
  		if (bpd[i].ignore_clr_ack)
  			continue;
-		ret = regmap_read_poll_timeout(regmap, bpd[i].bus_prot_sta,
-					       val, !(val & mask),
+		ret = regmap_read_poll_timeout(ack_regmap, bpd[i].bus_prot_sta,
+					       val, mask_cond(bpd[i].way_en, val, sta_mask),

"I don't know why", my brain still keeps telling me that using different functions
for the WAY_EN (en/dis) is just better.

This commit seems to be overcomplicating two "easy" en/dis functions.

Looking at the code again, I think you are right. I redesigned basically
this whole patch, and I think it is easier to understand now.

quoted

  					       MTK_POLL_DELAY_US, MTK_POLL_TIMEOUT);
  		if (ret)
  			return ret;

@@ -185,11 +216,12 @@ static int scpsys_bus_protect_disable(struct scpsys_domain *pd)
  {
  	int ret;
-	ret = _scpsys_bus_protect_disable(pd->data->bp_smi, pd->smi);
+	ret = _scpsys_bus_protect_disable(pd->data->bp_smi, pd->smi, NULL);
  	if (ret)
  		return ret;
-	return _scpsys_bus_protect_disable(pd->data->bp_infracfg, pd->infracfg);
+	return _scpsys_bus_protect_disable(pd->data->bp_infracfg,
+			pd->infracfg, pd->infracfg_nao);
  }
  static int scpsys_regulator_enable(struct regulator *supply)

@@ -363,6 +395,10 @@ generic_pm_domain *scpsys_add_one_domain(struct scpsys *scpsys, struct device_no
  			return ERR_CAST(pd->smi);
  	}
+	pd->infracfg_nao = syscon_regmap_lookup_by_phandle_optional(node, "mediatek,infracfg_nao");
+	if (IS_ERR(pd->infracfg_nao))
+		return ERR_CAST(pd->infracfg_nao);
+

I think that we should enforce a check here:

pd->infracfg_nao = syscon_regmap_lookup_by_phandle(node, "mediatek,infracfg_nao");
if (IS_ERR(pd->infracfg_nao)) {
	/* checking if infracfg_nao != NULL at every pwoeron/poweroff is largely
	 * suboptimal, as if it't present once, it's present always (!)
	 */
	if (we have WAY_EN)
		return ERR_CAST ...
	pd->infracfg_nao = NULL;
}

Yes, I added another check that enforces .bp_smi not having any way_en
configuration.

Thanks,
Markus