wlcore: Fix bringing up wlan0 again if powered down briefly

At least HiKey board can fail to bring up wireless interface again if the interface is brought down and up with no delay inbetween.

This can be done tested with:

# while true; do ifconfig wlan0 down; ifconfig wlan0 up; done

According to Ricardo Salveti <[email protected]>, we need to wait between 30 - 40 ms on HiKey. This seems to happen when we
get -EBUSY returned by pm_runtime_put() basedon the check in rpm_check_suspend_allowed(). But as it still unclear if part of
the delay needed is because of capacitance, let's always do a
at least 50 ms msleep even if no -EBUSY is returned.

Cc Anders Roxell <[email protected]>

Cc Eyal Reizer <[email protected]>

Cc John Stultz <[email protected]>

Cc Ricardo Salveti <[email protected]>

Cc Ulf Hansson <[email protected]>

Reported-by Ricardo Salveti <[email protected]>

Fixes 60f36637bbbd ("wlcore: sdio: allow pm to handle sdio power")

Signed-off-by Tony Lindgren <[email protected]>

---

Uffe, do you have some better ideas on how to fix this issue?

---
 drivers/net/wireless/ti/wlcore/sdio.c | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)

drivers/net/wireless/ti/wlcore/sdio.c +12 -2

@@ -174,7 +174,7 @@ static int wl12xx_sdio_power_off(struct wl12xx_sdio_glue *glue)
 {
        struct sdio_func *func = dev_to_sdio_func(glue->dev);
        struct mmc_card *card = func->card;
-       int error;
+       int error, retries = 5;
 
        sdio_claim_host(func);
        sdio_disable_func(func);

@@ -188,6 +188,17 @@ static int wl12xx_sdio_power_off(struct wl12xx_sdio_glue *glue)
                return error;
        }
 
+       /* Wait a bit to ensure the card gets power off. Otherwise
+        * bringing interface down and up again may not power off the
+        * card inbetween. And then firmware load will fail on trying
+        * to bring the card up again. We need to wait between 30 - 40
+        * ms for this based on measurements on HiKey board.
+        */
+       do {
+               msleep(50);
+       } while (error == -EBUSY && !pm_runtime_suspended(&card->dev) &&
+                retries--);
+
        return 0;
 }
 
--

2.19.2

At least HiKey board can fail to bring up wireless interface again if the interface is brought down and up with no delay inbetween.

This can be done tested with:

# while true; do ifconfig wlan0 down; ifconfig wlan0 up; done

According to Ricardo Salveti <[email protected]>, we need to wait between 30 - 40 ms on HiKey. This seems to happen when we
get -EBUSY returned by pm_runtime_put() basedon the check in rpm_check_suspend_allowed(). But as it still unclear if part of
the delay needed is because of capacitance, let's always do a
at least 50 ms msleep even if no -EBUSY is returned.

Cc Anders Roxell <[email protected]>

Cc Eyal Reizer <[email protected]>

Cc John Stultz <[email protected]>

Cc Ricardo Salveti <[email protected]>

Cc Ulf Hansson <[email protected]>

Reported-by Ricardo Salveti <[email protected]>

Fixes 60f36637bbbd ("wlcore: sdio: allow pm to handle sdio power")

Signed-off-by Tony Lindgren <[email protected]>

---

Uffe, do you have some better ideas on how to fix this issue?

---
 drivers/net/wireless/ti/wlcore/sdio.c | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)

drivers/net/wireless/ti/wlcore/sdio.c +2 -1

@@ -174,7 +174,7 @@ static int wl12xx_sdio_power_off(struct wl12xx_sdio_glue *glue)
 {
        struct sdio_func *func = dev_to_sdio_func(glue->dev);
        struct mmc_card *card = func->card;
-       int error;
+       int error, retries = 5;
 
        sdio_claim_host(func);
        sdio_disable_func(func);

@@ -188,6 +188,17 @@ static int wl12xx_sdio_power_off(struct wl12xx_sdio_glue *glue)
                return error;
        }
 
+       /* Wait a bit to ensure the card gets power off. Otherwise

Instead, it looks like what you need, is a way to keep track of whether the SDIO card, became power cycled or if it stayed powered on, when "ifconfig wlan0 up" is done. In case of a power cycle, you need to re-program the firmware, right?

Would it be possible to re-program the firmware, even if the SDIO card stayed powered-on?

In regards to delays needed due to a capacitance. If that really is the case, that shall be the managed by the mmc core, as it's there the power cycle sequence is being done. As a matter of fact, you should be able to use drivers/mmc/core/pwrseq_simple.c and extend "power-off-delay-us" in the DTB for Hikey if that is needed.

Hi,

Quoted 4 lines · Ulf Hansson

Instead, it looks like what you need, is a way to keep track of whether the SDIO card, became power cycled or if it stayed powered on, when "ifconfig wlan0 up" is done. In case of a power cycle, you need to re-program the firmware, right?

Yeah mostly. But we also need to ensure things do get powered down properly after ifconfig wlan0 down :) IMO after ifconfig wlan0 down returns, there should be no waiting needed.

Quoted 2 lines

Would it be possible to re-program the firmware, even if the SDIO card stayed powered-on?

That might help for some cases, but the problem of how to ensure the card is powered down after ifconfig wlan0 down returns still exists.

Quoted 6 lines

In regards to delays needed due to a capacitance. If that really is the case, that shall be the managed by the mmc core, as it's there the power cycle sequence is being done. As a matter of fact, you should be able to use drivers/mmc/core/pwrseq_simple.c and extend "power-off-delay-us" in the DTB for Hikey if that is needed.

Hmm I do have a mostly done wlcore pwrseq driver here somewhere that I have not had a chance to finish again. But that's not going to work for a fix and needs updating of the related dts files too.

Quoted 13 lines · Tony Lindgren

Hi,

Quoted 4 lines · Ulf Hansson

Instead, it looks like what you need, is a way to keep track of whether the SDIO card, became power cycled or if it stayed powered on, when "ifconfig wlan0 up" is done. In case of a power cycle, you need to re-program the firmware, right?

Yeah mostly. But we also need to ensure things do get powered down properly after ifconfig wlan0 down :) IMO after ifconfig wlan0 down returns, there should be no waiting needed.

Quoted 5 lines

Quoted 2 lines

Would it be possible to re-program the firmware, even if the SDIO card stayed powered-on?

That might help for some cases, but the problem of how to ensure the card is powered down after ifconfig wlan0 down returns still exists.

Well, does the SDIO card really have to be powered down before "ifconfig wlan0 down" returns? If so, why?

An option would be to call pm_runtime_get_sync() at "ifconfig wlan0 up" (I assume you already do that) and then re-program the firmware, even if the card hasn't been power cycled. Wouldn't that work?

Or perhaps this is the problem you are encountering, that the FW can't be reprogrammed, unless the card has been power cycled? You can easily test that, by simply bumping the runtime usage count for the card device via sysfs, before doing the "ifconfig wlan0 up/down" thingy.

Quoted 2 lines

Quoted 2 lines

Quoted 2 lines

Instead, it looks like what you need, is a way to keep track of whether the SDIO card, became power cycled or if it stayed powered

on,

Quoted 10 lines

Quoted 10 lines

Quoted 2 lines

when "ifconfig wlan0 up" is done. In case of a power cycle, you need to re-program the firmware, right?

Yeah mostly. But we also need to ensure things do get powered down properly after ifconfig wlan0 down :) IMO after ifconfig wlan0 down returns, there should be no waiting needed.

Quoted 2 lines

Would it be possible to re-program the firmware, even if the SDIO card stayed powered-on?

Quoted 5 lines

Quoted 2 lines

That might help for some cases, but the problem of how to ensure the card is powered down after ifconfig wlan0 down returns still exists.

Well, does the SDIO card really have to be powered down before "ifconfig wlan0 down" returns? If so, why?

Good question. Eyal, any comments what should happen here from the wlcore hardware point of view?

Quoted 3 lines

An option would be to call pm_runtime_get_sync() at "ifconfig wlan0 up" (I assume you already do that) and then re-program the firmware, even if the card hasn't been power cycled. Wouldn't that work?

Ricardo, care to test and see if the problem comes back if you keep PM runtime enabled?

Quoted 4 lines

Or perhaps this is the problem you are encountering, that the FW can't be reprogrammed, unless the card has been power cycled? You can easily test that, by simply bumping the runtime usage count for the card device via sysfs, before doing the "ifconfig wlan0 up/down" thingy.

Ricardo, care to play with this too?

You are ok as long as the wlan_enable pin Does go down for a sufficient amount of time turning the wl18xx device off.
The firmware can only be downloaded once after power on.
In between down/up you have to make sure the wlan_enable is fully going through on->off->on and the wl18xx device is fully reset.
On power on the firmware is loaded by the driver and this will fail in case the reset didn't happen

Hi,

Quoted 6 lines · Reizer, Eyal

You are ok as long as the wlan_enable pin Does go down for a sufficient amount of time turning the wl18xx device off.
The firmware can only be downloaded once after power on.
In between down/up you have to make sure the wlan_enable is fully going through on->off->on and the wl18xx device is fully reset.
On power on the firmware is loaded by the driver and this will fail in case the reset didn't happen

OK thanks for confirming that it's the enable pin that needs to toggle.

Sounds like I need to get the wlcore pwrseq changes done and posted as the long term solution.

And for a short term fix, we should just add msleep(50) for now with updated patch description.

Does that that sound OK to everybody?

Quoted 11 lines · Tony Lindgren

Hi,

Quoted 6 lines · Reizer, Eyal

You are ok as long as the wlan_enable pin Does go down for a sufficient amount of time turning the wl18xx device off.
The firmware can only be downloaded once after power on.
In between down/up you have to make sure the wlan_enable is fully going through on->off->on and the wl18xx device is fully reset.
On power on the firmware is loaded by the driver and this will fail in case the reset didn't happen

Quoted 4 lines

OK thanks for confirming that it's the enable pin that needs to toggle.

Sounds like I need to get the wlcore pwrseq changes done and posted as the long term solution.

Just to make sure we are talking about the same things here. The pwrseq thingy, is already implemented in the mmc core. When it comes to Hikey, there is already a pwrseq node deployed in the DTB. So, we should be fine.

Here are the MMC pwrseq bindings:
Documentation/devicetree/bindings/mmc/mmc-pwrseq-simple.txt Documentation/devicetree/bindings/mmc/mmc.txt

Here is the Hikey DTS file:
arch/arm64/boot/dts/hisilicon/hi6220-hikey.dts

Quoted 4 lines

And for a short term fix, we should just add msleep(50) for now with updated patch description.

Does that that sound OK to everybody?

Unfortunate, that doesn't work. Because, if user space via sysfs has prevented runtime suspend, the SDIO card never becomes power off with a call to pm_runtime_put*(), not even after waiting 50 ms.

If we need a short term solution, I think there are two options.

1) Call pm_runtime_force_suspend() instead of pm_runtime_put*() at "down". This makes sure the card becomes powered off. At "up", call pm_runtime_force_resume() instead of pm_runtime_get_sync(). Because the runtime PM usage count it > 1, at "up", pm_runtime_force_resume() will power up the card, rather than deferring it.

The problem with 1), is if there is an ACPI PM domain attached to the SDIO card device. Then this doesn't work. I can't tell if that is ever the case here.

2) At "up", after the call to pm_runtime_get_sync(), add a call to mmc_hw_reset(). This forces the mmc core to power cycle and re-init the SDIO card. This may in some cases be a waste, as the card may already have been power cycled, but at least we should now always be able to re-program the firmware.

If there is no rush, I think we may consider to move away from using runtime PM to control power for SDIO cards. Because, what we seem to need, is a simple interface (reference counted) to power-on/off SDIO cards.

Hi,

Quoted 26 lines · Ulf Hansson

Quoted 11 lines · Tony Lindgren

Hi,

Quoted 6 lines · Reizer, Eyal

You are ok as long as the wlan_enable pin Does go down for a sufficient amount of time turning the wl18xx device off.
The firmware can only be downloaded once after power on.
In between down/up you have to make sure the wlan_enable is fully going through on->off->on and the wl18xx device is fully reset.
On power on the firmware is loaded by the driver and this will fail in case the reset didn't happen

Quoted 4 lines

OK thanks for confirming that it's the enable pin that needs to toggle.

Sounds like I need to get the wlcore pwrseq changes done and posted as the long term solution.

Just to make sure we are talking about the same things here. The pwrseq thingy, is already implemented in the mmc core. When it comes to Hikey, there is already a pwrseq node deployed in the DTB. So, we should be fine.

Here are the MMC pwrseq bindings:
Documentation/devicetree/bindings/mmc/mmc-pwrseq-simple.txt Documentation/devicetree/bindings/mmc/mmc.txt

Quoted 2 lines

Here is the Hikey DTS file:
arch/arm64/boot/dts/hisilicon/hi6220-hikey.dts

Oh so it seems. And looking at my earlier pwrseq_wlcore.c hacks, I only added pinctrl handling to work around a GPIO glitch on some SoCs for deeper idle states. So I don't think we need a pwrseq_wlcore.c as I've found a better way to deal with the GPIO glitch by implementing gpiochip for the pinctrl driver.

Quoted 8 lines

Quoted 4 lines

And for a short term fix, we should just add msleep(50) for now with updated patch description.

Does that that sound OK to everybody?

Unfortunate, that doesn't work. Because, if user space via sysfs has prevented runtime suspend, the SDIO card never becomes power off with a call to pm_runtime_put*(), not even after waiting 50 ms.

Yes you're right.

Quoted 11 lines

If we need a short term solution, I think there are two options.

1) Call pm_runtime_force_suspend() instead of pm_runtime_put*() at "down". This makes sure the card becomes powered off. At "up", call pm_runtime_force_resume() instead of pm_runtime_get_sync(). Because the runtime PM usage count it > 1, at "up", pm_runtime_force_resume() will power up the card, rather than deferring it.

The problem with 1), is if there is an ACPI PM domain attached to the SDIO card device. Then this doesn't work. I can't tell if that is ever the case here.

Hmm..

Quoted 5 lines

2) At "up", after the call to pm_runtime_get_sync(), add a call to mmc_hw_reset(). This forces the mmc core to power cycle and re-init the SDIO card. This may in some cases be a waste, as the card may already have been power cycled, but at least we should now always be able to re-program the firmware.

Option 2 sounds more generic to me. Do you have some test patch for this?

Sounds like you're thinking about adding this to the MMC framework?

If the only issue is if the card must be powered off when ifconfig wlan0 returns, I think that's a cosmetic issue. For example, switching to wlcore test mode after ifconfig wlan0 down might need the card powered down, but then again the test mode resets things anyway. Eyal?

Quoted 5 lines

If there is no rush, I think we may consider to move away from using runtime PM to control power for SDIO cards. Because, what we seem to need, is a simple interface (reference counted) to power-on/off SDIO cards.

Well we should fix the regression in a minimal way first though. And to me it sounds like option 2 above should do the trick, not sure if we need something beyond that.

However, if this is about an out-of-band IRQ line, instead of using the SDIO IRQs on DAT1, I think management of that, belongs in the wlan (or gpiochip) layers.

We have to distinguish if "down" has a strict requirement to power off the wlan-chip.

For example, in "flight mode", is it okay to leave the wlan chip powered on and thus potentially also having its radio part active?

Quoted 4 lines · Ulf Hansson

However, if this is about an out-of-band IRQ line, instead of using the SDIO IRQs on DAT1, I think management of that, belongs in the wlan (or gpiochip) layers.

Yes this can be handled at the gpiochip layer.

Quoted 3 lines

We have to distinguish if "down" has a strict requirement to power off the wlan-chip.

Quoted 2 lines

For example, in "flight mode", is it okay to leave the wlan chip powered on and thus potentially also having its radio part active?

I don't know. If unsure, we should at least have that possibility available for future versions if needed. It might also help a bit with PM measurements :)

Quoted 11 lines · Tony Lindgren

Quoted 4 lines · Ulf Hansson

However, if this is about an out-of-band IRQ line, instead of using the SDIO IRQs on DAT1, I think management of that, belongs in the wlan (or gpiochip) layers.

Yes this can be handled at the gpiochip layer.

Quoted 3 lines

We have to distinguish if "down" has a strict requirement to power off the wlan-chip.

Quoted 6 lines

Quoted 2 lines

For example, in "flight mode", is it okay to leave the wlan chip powered on and thus potentially also having its radio part active?

I don't know. If unsure, we should at least have that possibility available for future versions if needed. It might also help a bit with PM measurements :)

I don't know if there are any written rules but I have preferred that in wireless drivers "ifconfig down" means that all power is turned off on the device, at least this way user space can achive maximum power savings. And there's also a simple way to start the firmware from clean state, we all know how buggy firmwares are so I think that's important as well.

Quoted 14 lines

Quoted 3 lines

Quoted 3 lines

Quoted 3 lines

Instead, it looks like what you need, is a way to keep track of whether the SDIO card, became power cycled or if it stayed powered

on,

Quoted 10 lines

Quoted 10 lines

Quoted 2 lines

when "ifconfig wlan0 up" is done. In case of a power cycle, you need to re-program the firmware, right?

Yeah mostly. But we also need to ensure things do get powered down properly after ifconfig wlan0 down :) IMO after ifconfig wlan0 down returns, there should be no waiting needed.

Quoted 2 lines

Would it be possible to re-program the firmware, even if the SDIO card stayed powered-on?

Quoted 9 lines

Quoted 5 lines

Quoted 2 lines

That might help for some cases, but the problem of how to ensure the card is powered down after ifconfig wlan0 down returns still exists.

Well, does the SDIO card really have to be powered down before "ifconfig wlan0 down" returns? If so, why?

Good question. Eyal, any comments what should happen here from the wlcore hardware point of view?

You are ok as long as the wlan_enable pin Does go down for a sufficient amount of time turning the wl18xx device off.
The firmware can only be downloaded once after power on.
In between down/up you have to make sure the wlan_enable is fully going through on->off->on and the wl18xx device is fully reset.
On power on the firmware is loaded by the driver and this will fail in case the reset didn't happen

.

Quoted 4 lines

Quoted 3 lines

An option would be to call pm_runtime_get_sync() at "ifconfig wlan0 up" (I assume you already do that) and then re-program the firmware, even if the card hasn't been power cycled. Wouldn't that work?

No, this wouldn't work in case the wlan_enable pin didn't reset the chip as part of the "ifconfig wlan0 down" Command.

Quoted 3 lines

Ricardo, care to test and see if the problem comes back if you keep PM runtime enabled?

Quoted 5 lines

Quoted 4 lines

Or perhaps this is the problem you are encountering, that the FW can't be reprogrammed, unless the card has been power cycled? You can easily test that, by simply bumping the runtime usage count for the card device via sysfs, before doing the "ifconfig wlan0 up/down" thingy.

Correct. The firmware can't be reprogrammed , unless the card has been power cycled.