Skip to contentSkip to conversation

net: ipa: non-setup bug fixes

PATCH In review Alex Elder 2 participants 6 messages Oct 6, 2020 – Oct 9, 2020

Subsystems (3)
lore
Share & tools
lore

Alex Elder

author
#0 lore raw

cover net: ipa: non-setup bug fixes

This series fixes two bugs that occur if the IPA driver has not completed its setup phase of initialization when an event occurs.

One event is the crash of the modem. If the modem crashes, ipa_modem_crashed() is called, and it performs various activities intended to put the IPA hardware into a good state before the modem reboots. But if the IPA setup has not completed, resources used for this cleanup will not be properly initialized. So we must skip doing this activity if we have not completed setup.

Similarly, if a system suspend is initiated but IPA setup has not completed, the processing done in ipa_endpoint_suspend() should be avoided. And a subsequent system resume should also avoid resuming endpoints that have not been initialized by IPA setup.

-Alex

Alex Elder (2):
net: ipa: only clear hardware state if setup has completed
net: ipa: skip suspend/resume activities if not set up

 drivers/net/ipa/ipa_endpoint.c | 6 ++++++
 drivers/net/ipa/ipa_modem.c    | 3 +++
 2 files changed, 9 insertions(+)
Signature · 1 line
2.20.1
author
#1 lore raw

PATCH 1/2 net: ipa: only clear hardware state if setup has completed

In the setup phase of initialization, GSI firmware gets loaded and initialized, and the AP command TX and default RX endpoints get set up. Until that happens, IPA commands must not be issued to the hardware.

If the modem crashes, we stop endpoint channels and perform a number of other activities to clear modem-related IPA state, including zeroing filter and routing tables (using IPA commands). This is a bug if setup is not complete. We should also not be performing the other cleanup activity in that case either.

Fix this by returning immediately when handling a modem crash if we haven't completed setup.

Fixes a646d6ec9098 ("soc: qcom: ipa: modem and microcontroller")
---
 drivers/net/ipa/ipa_modem.c | 3 +++
 1 file changed, 3 insertions(+)
drivers/net/ipa/ipa_modem.c +3 -0
index e34fe2d77324e..dd5b89c5cb2d4 100644
@@ -285,6 +285,9 @@ static void ipa_modem_crashed(struct ipa *ipa)
 	struct device *dev = &ipa->pdev->dev;
 	int ret;
 
+	if (!ipa->setup_complete)
+		return;
+
 	ipa_endpoint_modem_pause_all(ipa, true);
 
 	ipa_endpoint_modem_hol_block_clear_all(ipa);
Signature · 1 line
2.20.1
maintainer
#2 lore raw

Re: [PATCH net 1/2] net: ipa: only clear hardware state if setup has completed

Quoted 23 lines · On Tue, 6 Oct 2020 16:30:46 -0500 Alex Elder wrote:

In the setup phase of initialization, GSI firmware gets loaded and initialized, and the AP command TX and default RX endpoints get set up. Until that happens, IPA commands must not be issued to the hardware.

If the modem crashes, we stop endpoint channels and perform a number of other activities to clear modem-related IPA state, including zeroing filter and routing tables (using IPA commands). This is a bug if setup is not complete. We should also not be performing the other cleanup activity in that case either.

Fix this by returning immediately when handling a modem crash if we haven't completed setup.

Fixes a646d6ec9098 ("soc: qcom: ipa: modem and microcontroller")
drivers/net/ipa/ipa_modem.c +0 -0
index e34fe2d77324e..dd5b89c5cb2d4 100644

drivers/net/ipa/ipa_modem.c static void ipa_modem_crashed(struct ipa *ipa) · line 293

Quoted 9 lines
drivers/net/ipa/ipa_modem.c +3 -0
@@ -285,6 +285,9 @@ static void ipa_modem_crashed(struct ipa *ipa)
 	struct device *dev = &ipa->pdev->dev;
 	int ret;
 
+	if (!ipa->setup_complete)
+		return;
+
 	ipa_endpoint_modem_pause_all(ipa, true);
 
 	ipa_endpoint_modem_hol_block_clear_all(ipa);

The only call site already checks setup_complete, so this is not needed, no?

$ git grep -C2 ipa_modem_crashed\(
drivers/net/ipa/ipa_modem.c-
drivers/net/ipa/ipa_modem.c-/* Treat a "clean" modem stop the same as a crash */
drivers/net/ipa/ipa_modem.c:static void ipa_modem_crashed(struct ipa *ipa)
drivers/net/ipa/ipa_modem.c-{
drivers/net/ipa/ipa_modem.c-    struct device *dev = &ipa->pdev->dev;
--
drivers/net/ipa/ipa_modem.c-                     notify_data->crashed ? "crashed" : "stopping");
drivers/net/ipa/ipa_modem.c-            if (ipa->setup_complete)
drivers/net/ipa/ipa_modem.c:                    ipa_modem_crashed(ipa);
drivers/net/ipa/ipa_modem.c-            break;
drivers/net/ipa/ipa_modem.c-
author
#3 lore raw

Re: [PATCH net 1/2] net: ipa: only clear hardware state if setup has completed

Quoted 32 lines · Jakub Kicinski
Quoted 31 lines · On Tue, 6 Oct 2020 16:30:46 -0500 Alex Elder wrote:

In the setup phase of initialization, GSI firmware gets loaded and initialized, and the AP command TX and default RX endpoints get set up. Until that happens, IPA commands must not be issued to the hardware.

If the modem crashes, we stop endpoint channels and perform a number of other activities to clear modem-related IPA state, including zeroing filter and routing tables (using IPA commands). This is a bug if setup is not complete. We should also not be performing the other cleanup activity in that case either.

Fix this by returning immediately when handling a modem crash if we haven't completed setup.

Fixes a646d6ec9098 ("soc: qcom: ipa: modem and microcontroller")
drivers/net/ipa/ipa_modem.c +3 -0
index e34fe2d77324e..dd5b89c5cb2d4 100644
@@ -285,6 +285,9 @@ static void ipa_modem_crashed(struct ipa *ipa)
  	struct device *dev = &ipa->pdev->dev;
  	int ret;
 
+	if (!ipa->setup_complete)
+		return;
+
  	ipa_endpoint_modem_pause_all(ipa, true);
 
Quoted 4 lines

drivers/net/ipa/ipa_modem.c static void ipa_modem_crashed(struct ipa *ipa) · line 293

Quoted 1 line
drivers/net/ipa/ipa_modem.c +0 -0
  	ipa_endpoint_modem_hol_block_clear_all(ipa);

The only call site already checks setup_complete, so this is not needed, no?

Wow, you are correct.

I was mainly focused on the fact that the ipa_modem_crashed() call could happen asynchronously, and that it called ipa_table_reset() which requires IPA immediate commands. I should have followed it back to its callers.

I agree with you, this is not needed. The other patch is definitely needed though.

Would you like me to re-post that single patch, or is it acceptable as-is?

Thank you for reviewing this and catching my mistake.

-Alex

Quoted 13 lines
$ git grep -C2 ipa_modem_crashed\(
drivers/net/ipa/ipa_modem.c-
drivers/net/ipa/ipa_modem.c-/* Treat a "clean" modem stop the same as a crash */
drivers/net/ipa/ipa_modem.c:static void ipa_modem_crashed(struct ipa *ipa)
drivers/net/ipa/ipa_modem.c-{
drivers/net/ipa/ipa_modem.c-    struct device *dev = &ipa->pdev->dev;
--
drivers/net/ipa/ipa_modem.c-                     notify_data->crashed ? "crashed" : "stopping");
drivers/net/ipa/ipa_modem.c-            if (ipa->setup_complete)
drivers/net/ipa/ipa_modem.c:                    ipa_modem_crashed(ipa);
drivers/net/ipa/ipa_modem.c-            break;
drivers/net/ipa/ipa_modem.c-
maintainer
#4 lore raw

Re: [PATCH net 1/2] net: ipa: only clear hardware state if setup has completed

Quoted 10 lines · On Fri, 9 Oct 2020 14:39:45 -0500 Alex Elder wrote:
Quoted 2 lines

The only call site already checks setup_complete, so this is not needed, no?

Wow, you are correct.

I was mainly focused on the fact that the ipa_modem_crashed() call could happen asynchronously, and that it called ipa_table_reset() which requires IPA immediate commands. I should have followed it back to its callers.

Quoted 5 lines

I agree with you, this is not needed. The other patch is definitely needed though.

Would you like me to re-post that single patch, or is it acceptable as-is?

Please repost, thank you!

author
#5 lore raw

PATCH 2/2 net: ipa: skip suspend/resume activities if not set up

In reply to Alex Elder

When processing a system suspend request we suspend modem endpoints if they are enabled, and call ipa_cmd_tag_process() (which issues IPA commands) to ensure the IPA pipeline is cleared. It is an error to attempt to issue an IPA command before setup is complete, so this is clearly a bug. But we also shouldn't suspend or resume any endpoints that have not been set up.

Have ipa_endpoint_suspend() and ipa_endpoint_resume() immediately return if setup hasn't completed, to avoid any attempt to configure endpoints or issue IPA commands in that case.

Fixes 84f9bd12d46d ("soc: qcom: ipa: IPA endpoints")
---
 drivers/net/ipa/ipa_endpoint.c | 6 ++++++
 1 file changed, 6 insertions(+)
drivers/net/ipa/ipa_endpoint.c +6 -0
index b7efd7c95e9c8..ed60fa5bcdaca 100644
@@ -1471,6 +1471,9 @@ void ipa_endpoint_resume_one(struct ipa_endpoint *endpoint)
 
 void ipa_endpoint_suspend(struct ipa *ipa)
 {
+	if (!ipa->setup_complete)
+		return;
+
 	if (ipa->modem_netdev)
 		ipa_modem_suspend(ipa->modem_netdev);
 
@@ -1482,6 +1485,9 @@ void ipa_endpoint_suspend(struct ipa *ipa)
 
 void ipa_endpoint_resume(struct ipa *ipa)
 {
+	if (!ipa->setup_complete)
+		return;
+
 	ipa_endpoint_resume_one(ipa->name_map[IPA_ENDPOINT_AP_COMMAND_TX]);
 	ipa_endpoint_resume_one(ipa->name_map[IPA_ENDPOINT_AP_LAN_RX]);
 
Signature · 1 line
2.20.1