PCI: pciehp: Fix re-enabling the slot marked for safe removal
diff mbox series

Message ID 20190301161251.26774-1-s.miroshnichenko@yadro.com
State Superseded
Delegated to: Bjorn Helgaas
Headers show
Series
  • PCI: pciehp: Fix re-enabling the slot marked for safe removal
Related show

Commit Message

Sergey Miroshnichenko March 1, 2019, 4:12 p.m. UTC
During the safe removal procedure, a Data Link Layer State Changed event
may occur after pciehp_power_off_slot(), and it is handled when the slot is
already set to OFF_STATE. This results in re-enabling the device and makes
it impossible to actually safely remove it.

Add a check for a Presence Detect Changed bit to filter out this interrupt.

It is still possible to re-enable the device if it remains in the slot
after pressing the Attention Button by pressing it again.

Signed-off-by: Sergey Miroshnichenko <s.miroshnichenko@yadro.com>
Cc: Lukas Wunner <lukas@wunner.de>
---
 drivers/pci/hotplug/pciehp_ctrl.c | 8 ++++++++
 1 file changed, 8 insertions(+)

Comments

Lukas Wunner March 5, 2019, 11:01 a.m. UTC | #1
On Fri, Mar 01, 2019 at 07:12:51PM +0300, Sergey Miroshnichenko wrote:
> During the safe removal procedure, a Data Link Layer State Changed event
> may occur after pciehp_power_off_slot(), and it is handled when the slot is
> already set to OFF_STATE. This results in re-enabling the device and makes
> it impossible to actually safely remove it.
> 
> Add a check for a Presence Detect Changed bit to filter out this interrupt.
> 
> It is still possible to re-enable the device if it remains in the slot
> after pressing the Attention Button by pressing it again.
> 
> Signed-off-by: Sergey Miroshnichenko <s.miroshnichenko@yadro.com>
> Cc: Lukas Wunner <lukas@wunner.de>
> ---
>  drivers/pci/hotplug/pciehp_ctrl.c | 8 ++++++++
>  1 file changed, 8 insertions(+)
> 
> diff --git a/drivers/pci/hotplug/pciehp_ctrl.c b/drivers/pci/hotplug/pciehp_ctrl.c
> index 3f3df4c29f6e..941f77b71df4 100644
> --- a/drivers/pci/hotplug/pciehp_ctrl.c
> +++ b/drivers/pci/hotplug/pciehp_ctrl.c
> @@ -251,6 +251,14 @@ void pciehp_handle_presence_or_link_change(struct controller *ctrl, u32 events)
>  		return;
>  	}
>  
> +	/* Handle the DLLSC from the slot which just has been turned off */
> +	if ((events & PCI_EXP_SLTSTA_DLLSC) &&
> +	    !(events & PCI_EXP_SLTSTA_PDC) &&
> +	    present && !link_active) {
> +		mutex_unlock(&ctrl->state_lock);
> +		return;
> +	}
> +
>  	switch (ctrl->state) {
>  	case BLINKINGON_STATE:
>  		cancel_delayed_work(&ctrl->button_work);

Hm, I'd instead prefer amending remove_board() like this:

 	if (POWER_CTRL(ctrl)) {
 		pciehp_power_off_slot(ctrl);
 
 		/*
 		 * After turning power off, we must wait for at least 1 second
 		 * before taking any action that relies on power having been
 		 * removed from the slot/adapter.
 		 */
 		msleep(1000);
 
+		/* ignore link or presence changes caused by power off */
+		atomic_and(~(PCI_EXP_SLTSTA_DLLSC | PCI_EXP_SLTSTA_PDC),
+			   &ctrl->pending_events);
 	}

Would this work for you?

Thanks,

Lukas
Sergey Miroshnichenko March 12, 2019, 12:03 p.m. UTC | #2
On 3/5/19 2:01 PM, Lukas Wunner wrote:
> On Fri, Mar 01, 2019 at 07:12:51PM +0300, Sergey Miroshnichenko wrote:
>> During the safe removal procedure, a Data Link Layer State Changed event
>> may occur after pciehp_power_off_slot(), and it is handled when the slot is
>> already set to OFF_STATE. This results in re-enabling the device and makes
>> it impossible to actually safely remove it.
>>
>> Add a check for a Presence Detect Changed bit to filter out this interrupt.
>>
>> It is still possible to re-enable the device if it remains in the slot
>> after pressing the Attention Button by pressing it again.
>>
>> Signed-off-by: Sergey Miroshnichenko <s.miroshnichenko@yadro.com>
>> Cc: Lukas Wunner <lukas@wunner.de>
>> ---
>>  drivers/pci/hotplug/pciehp_ctrl.c | 8 ++++++++
>>  1 file changed, 8 insertions(+)
>>
>> diff --git a/drivers/pci/hotplug/pciehp_ctrl.c b/drivers/pci/hotplug/pciehp_ctrl.c
>> index 3f3df4c29f6e..941f77b71df4 100644
>> --- a/drivers/pci/hotplug/pciehp_ctrl.c
>> +++ b/drivers/pci/hotplug/pciehp_ctrl.c
>> @@ -251,6 +251,14 @@ void pciehp_handle_presence_or_link_change(struct controller *ctrl, u32 events)
>>  		return;
>>  	}
>>  
>> +	/* Handle the DLLSC from the slot which just has been turned off */
>> +	if ((events & PCI_EXP_SLTSTA_DLLSC) &&
>> +	    !(events & PCI_EXP_SLTSTA_PDC) &&
>> +	    present && !link_active) {
>> +		mutex_unlock(&ctrl->state_lock);
>> +		return;
>> +	}
>> +
>>  	switch (ctrl->state) {
>>  	case BLINKINGON_STATE:
>>  		cancel_delayed_work(&ctrl->button_work);
> 
> Hm, I'd instead prefer amending remove_board() like this:
> 
>  	if (POWER_CTRL(ctrl)) {
>  		pciehp_power_off_slot(ctrl);
>  
>  		/*
>  		 * After turning power off, we must wait for at least 1 second
>  		 * before taking any action that relies on power having been
>  		 * removed from the slot/adapter.
>  		 */
>  		msleep(1000);
>  
> +		/* ignore link or presence changes caused by power off */
> +		atomic_and(~(PCI_EXP_SLTSTA_DLLSC | PCI_EXP_SLTSTA_PDC),
> +			   &ctrl->pending_events);
>  	}
> 
> Would this work for you?

Thank you for the review and for the proposed change! This discards
exactly the event which caused the issue - I've just checked it on our
hardware, and will resend it as a v2.

Best regards,
Serge

> 
> Thanks,
> 
> Lukas
>

Patch
diff mbox series

diff --git a/drivers/pci/hotplug/pciehp_ctrl.c b/drivers/pci/hotplug/pciehp_ctrl.c
index 3f3df4c29f6e..941f77b71df4 100644
--- a/drivers/pci/hotplug/pciehp_ctrl.c
+++ b/drivers/pci/hotplug/pciehp_ctrl.c
@@ -251,6 +251,14 @@  void pciehp_handle_presence_or_link_change(struct controller *ctrl, u32 events)
 		return;
 	}
 
+	/* Handle the DLLSC from the slot which just has been turned off */
+	if ((events & PCI_EXP_SLTSTA_DLLSC) &&
+	    !(events & PCI_EXP_SLTSTA_PDC) &&
+	    present && !link_active) {
+		mutex_unlock(&ctrl->state_lock);
+		return;
+	}
+
 	switch (ctrl->state) {
 	case BLINKINGON_STATE:
 		cancel_delayed_work(&ctrl->button_work);