Skip to contentSkip to conversation

clk: qcom: rpmh: Set wake/sleep state for BCM clks

PATCH In review Stephen Boyd 3 participants 7 messages Apr 12, 2022 – May 17, 2022

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 clk: qcom: rpmh: Set wake/sleep state for BCM clks

Set the wake and sleep state for BCM clks here, not just the active state, as the active only state is dropped when CPUs go to deep idle. This ensures the clk is always on when the driver thinks it is on.

This was found by inspection, and could very well be incorrect if the RPMh hardware copies over the active only state to the sleep and wake states.

Cc Alex Elder <[email protected]>
Cc Taniya Das <[email protected]>
Fixes 04053f4d23a4 ("clk: qcom: clk-rpmh: Add IPA clock support")
---
 drivers/clk/qcom/clk-rpmh.c | 19 ++++++++++++-------
 1 file changed, 12 insertions(+), 7 deletions(-)
drivers/clk/qcom/clk-rpmh.c +12 -7
index aed907982344..29da1ffd10cf 100644
@@ -260,6 +260,7 @@ static int clk_rpmh_bcm_send_cmd(struct clk_rpmh *c, bool enable)
 	struct tcs_cmd cmd = { 0 };
 	u32 cmd_state;
 	int ret = 0;
+	enum rpmh_state state;
 
 	mutex_lock(&rpmh_clk_lock);
 	if (enable) {
@@ -274,15 +275,19 @@ static int clk_rpmh_bcm_send_cmd(struct clk_rpmh *c, bool enable)
 		cmd.addr = c->res_addr;
 		cmd.data = BCM_TCS_CMD(1, enable, 0, cmd_state);
 
-		ret = clk_rpmh_send(c, RPMH_ACTIVE_ONLY_STATE, &cmd, enable);
-		if (ret) {
-			dev_err(c->dev, "set active state of %s failed: (%d)\n",
-				c->res_name, ret);
-		} else {
-			c->last_sent_aggr_state = cmd_state;
+		for (state = RPMH_SLEEP_STATE; state <= RPMH_ACTIVE_ONLY_STATE; state++) {
+			ret = clk_rpmh_send(c, state, &cmd, enable);
+			if (ret) {
+				dev_err(c->dev, "set %s state of %s failed: (%d)\n",
+					!state ? "sleep" :
+					state == RPMH_WAKE_ONLY_STATE	?
+					"wake" : "active", c->res_name, ret);
+				goto out;
+			}
 		}
+		c->last_sent_aggr_state = cmd_state;
 	}
-
+out:
 	mutex_unlock(&rpmh_clk_lock);
 
 	return ret;

base-commit: 3123109284176b1532874591f7c81f3837bbdc17

Signature · 1 line
https://chromeos.dev
#1 lore raw
Quoted 7 lines · Stephen Boyd

Set the wake and sleep state for BCM clks here, not just the active state, as the active only state is dropped when CPUs go to deep idle. This ensures the clk is always on when the driver thinks it is on.

This was found by inspection, and could very well be incorrect if the RPMh hardware copies over the active only state to the sleep and wake states.

I ran with this code and found no obvious errors. And it looks reasonable to me.

However someone (probably Taniya) needs to comment on whether it's the right thing to do (based on Stephen's second paragraph above), or if not, why not. It might warrant an additional comment also.

I can add a "Tested-by" but I'm going to wait for some more input.

-Alex

Quoted 49 lines
Cc Alex Elder <[email protected]>
Cc Taniya Das <[email protected]>
Fixes 04053f4d23a4 ("clk: qcom: clk-rpmh: Add IPA clock support")
---
  drivers/clk/qcom/clk-rpmh.c | 19 ++++++++++++-------
  1 file changed, 12 insertions(+), 7 deletions(-)
drivers/clk/qcom/clk-rpmh.c +12 -7
index aed907982344..29da1ffd10cf 100644
@@ -260,6 +260,7 @@ static int clk_rpmh_bcm_send_cmd(struct clk_rpmh *c, bool enable)
  	struct tcs_cmd cmd = { 0 };
  	u32 cmd_state;
  	int ret = 0;
+	enum rpmh_state state;
 
  	mutex_lock(&rpmh_clk_lock);
  	if (enable) {
@@ -274,15 +275,19 @@ static int clk_rpmh_bcm_send_cmd(struct clk_rpmh *c, bool enable)
  		cmd.addr = c->res_addr;
  		cmd.data = BCM_TCS_CMD(1, enable, 0, cmd_state);
 
-		ret = clk_rpmh_send(c, RPMH_ACTIVE_ONLY_STATE, &cmd, enable);
-		if (ret) {
-			dev_err(c->dev, "set active state of %s failed: (%d)\n",
-				c->res_name, ret);
-		} else {
-			c->last_sent_aggr_state = cmd_state;
+		for (state = RPMH_SLEEP_STATE; state <= RPMH_ACTIVE_ONLY_STATE; state++) {
+			ret = clk_rpmh_send(c, state, &cmd, enable);
+			if (ret) {
+				dev_err(c->dev, "set %s state of %s failed: (%d)\n",
+					!state ? "sleep" :
+					state == RPMH_WAKE_ONLY_STATE	?
+					"wake" : "active", c->res_name, ret);
+				goto out;
+			}
  		}
+		c->last_sent_aggr_state = cmd_state;
  	}
-
+out:
  	mutex_unlock(&rpmh_clk_lock);
 
  	return ret;

base-commit: 3123109284176b1532874591f7c81f3837bbdc17

Reviewed-by on v1 1/1
#2 lore raw
In reply to Stephen Boyd
Quoted 8 lines · Stephen Boyd

Set the wake and sleep state for BCM clks here, not just the active state, as the active only state is dropped when CPUs go to deep idle. This ensures the clk is always on when the driver thinks it is on.

This was found by inspection, and could very well be incorrect if the RPMh hardware copies over the active only state to the sleep and wake states.

Taking another look at this patch and now it makes perfect sense to me. Sorry for not grasping the problem earlier.

Will you take this in fixes, or do you want me to pick it for 5.19?

Quoted 22 lines
Cc Alex Elder <[email protected]>
Cc Taniya Das <[email protected]>
Fixes 04053f4d23a4 ("clk: qcom: clk-rpmh: Add IPA clock support")
---
 drivers/clk/qcom/clk-rpmh.c | 19 ++++++++++++-------
 1 file changed, 12 insertions(+), 7 deletions(-)
drivers/clk/qcom/clk-rpmh.c +1 -0
index aed907982344..29da1ffd10cf 100644
@@ -260,6 +260,7 @@ static int clk_rpmh_bcm_send_cmd(struct clk_rpmh *c, bool enable)
 	struct tcs_cmd cmd = { 0 };
 	u32 cmd_state;
 	int ret = 0;
+	enum rpmh_state state;
 
 	mutex_lock(&rpmh_clk_lock);
 	if (enable) {
@@ -274,15 +275,19 @@ static int clk_rpmh_bcm_send_cmd(struct clk_rpmh *c, bool enable)
 		cmd.addr = c->res_addr;

drivers/clk/qcom/clk-rpmh.c static int clk_rpmh_bcm_send_cmd(struct clk_rpmh *c, bool enable) · line 279

Quoted 10 lines
drivers/clk/qcom/clk-rpmh.c +2 -6
 		cmd.data = BCM_TCS_CMD(1, enable, 0, cmd_state);
 
-		ret = clk_rpmh_send(c, RPMH_ACTIVE_ONLY_STATE, &cmd, enable);
-		if (ret) {
-			dev_err(c->dev, "set active state of %s failed: (%d)\n",
-				c->res_name, ret);
-		} else {
-			c->last_sent_aggr_state = cmd_state;
+		for (state = RPMH_SLEEP_STATE; state <= RPMH_ACTIVE_ONLY_STATE; state++) {
+			ret = clk_rpmh_send(c, state, &cmd, enable);

Nit. We only need to pass the positive enable on the last iteration here...

Regards,
Bjorn

Quoted 20 lines
drivers/clk/qcom/clk-rpmh.c +9 -1
+			if (ret) {
+				dev_err(c->dev, "set %s state of %s failed: (%d)\n",
+					!state ? "sleep" :
+					state == RPMH_WAKE_ONLY_STATE	?
+					"wake" : "active", c->res_name, ret);
+				goto out;
+			}
 		}
+		c->last_sent_aggr_state = cmd_state;
 	}
-
+out:
 	mutex_unlock(&rpmh_clk_lock);
 
 	return ret;

base-commit: 3123109284176b1532874591f7c81f3837bbdc17

Signature · 1 line
https://chromeos.dev
#3 lore raw

Quoting Bjorn Andersson (2022-05-04 09:53:54)

Quoted 11 lines · Stephen Boyd
Quoted 9 lines

Set the wake and sleep state for BCM clks here, not just the active state, as the active only state is dropped when CPUs go to deep idle. This ensures the clk is always on when the driver thinks it is on.

This was found by inspection, and could very well be incorrect if the RPMh hardware copies over the active only state to the sleep and wake states.

Quoted 8 lines

Taking another look at this patch and now it makes perfect sense to me. Sorry for not grasping the problem earlier.

Will you take this in fixes, or do you want me to pick it for 5.19?

I'm waiting for Taniya to reply. For all I know this has no effect because there's some sort of copy/paste from one state to another. Until then it doesn't seem like we should do anything.

#4 lore raw

Quoting Stephen Boyd (2022-05-06 16:24:15)

Quoted 18 lines

Quoting Bjorn Andersson (2022-05-04 09:53:54)

Quoted 17 lines · Stephen Boyd
Quoted 9 lines

Set the wake and sleep state for BCM clks here, not just the active state, as the active only state is dropped when CPUs go to deep idle. This ensures the clk is always on when the driver thinks it is on.

This was found by inspection, and could very well be incorrect if the RPMh hardware copies over the active only state to the sleep and wake states.

Taking another look at this patch and now it makes perfect sense to me. Sorry for not grasping the problem earlier.

Quoted 6 lines
Quoted 2 lines

Will you take this in fixes, or do you want me to pick it for 5.19?

I'm waiting for Taniya to reply. For all I know this has no effect because there's some sort of copy/paste from one state to another. Until then it doesn't seem like we should do anything.

Taniya told me that if there's no sleep or wake state set then active state remains even when the subsystem is in sleep. Not exactly copy/paste but at least it is consistent. We need a comment here so this doesn't come up again.

#5 lore raw
Quoted 9 lines · Stephen Boyd
Quoted 5 lines
Quoted 2 lines

Will you take this in fixes, or do you want me to pick it for 5.19?

I'm waiting for Taniya to reply. For all I know this has no effect because there's some sort of copy/paste from one state to another. Until then it doesn't seem like we should do anything.

Taniya told me that if there's no sleep or wake state set then active state remains even when the subsystem is in sleep. Not exactly copy/paste but at least it is consistent. We need a comment here so this doesn't come up again.

If I understand what you're saying here, your original patch is not necessary, but there should be a comment in the code that explains why that is the case. Is that right? And am I right to assume you will be sending out a patch with such a comment?

-Alex

#6 lore raw

Quoting Alex Elder (2022-05-12 08:15:31)

Quoted 11 lines · Stephen Boyd
Quoted 9 lines
Quoted 5 lines
Quoted 2 lines

Will you take this in fixes, or do you want me to pick it for 5.19?

I'm waiting for Taniya to reply. For all I know this has no effect because there's some sort of copy/paste from one state to another. Until then it doesn't seem like we should do anything.

Taniya told me that if there's no sleep or wake state set then active state remains even when the subsystem is in sleep. Not exactly copy/paste but at least it is consistent. We need a comment here so this doesn't come up again.

Quoted 5 lines

If I understand what you're saying here, your original patch is not necessary, but there should be a comment in the code that explains why that is the case. Is that right? And am I right to assume you will be sending out a patch with such a comment?

Yes.