Re: [Qemu-devel] [PATCH RFC] vfio error recovery: kernel support

[Michael S. Tsirkin]

On Tue, Jan 24, 2017 at 09:00:07PM +0800, Cao jin wrote:
> It take me for a long while to find what I missed.  I missed much
> details in pci-error-recovery.txt, and I was not conscious of a
> multi-function device which could have different functions(driver) on it.
> 
> Thank you both.

Pls note the actual code isn't always in sync with the documentation.
You want to look at that too.

-- 
MST

Re: [Qemu-devel] [PATCH RFC] vfio error recovery: kernel support

[Cao jin]

On 01/21/2017 01:01 AM, Michael S. Tsirkin wrote:
> On Fri, Jan 20, 2017 at 06:13:22PM +0800, Cao jin wrote:
>>
>>
>> On 01/20/2017 04:16 AM, Michael S. Tsirkin wrote:
>>> This is a design and an initial patch for kernel side for AER
>>> support in VFIO.
>>>
>>> 0. What happens now (PCIE AER only)
>>>Fatal errors cause a link reset.
>>>Non fatal errors don't.
>>>All errors stop the VM eventually, but not immediately
>>>because it's detected and reported asynchronously.
>>>Interrupts are forwarded as usual.
>>>Correctable errors are not reported to guest at all.
>>>Note: PPC EEH is different. This focuses on AER.
>>>
>>> 1. Correctable errors
>>>I don't see a need to report these to guest. So let's not.
>>>
>>> 2. Fatal errors
>>>It's not easy to handle them gracefully since link reset
>>>is needed. As a first step, let's use the existing mechanism
>>>in that case.
>>>
>>> 2. Non-fatal errors
>>>Here we could make progress by reporting them to guest
>>>and have guest handle them.
>>>Issues:
>>> a. this behaviour should only be enabled with new userspace
>>>old userspace should work without changes
>>> Suggestion: One way to address this would be to add a new eventfd
>>> non_fatal_err_trigger. If not set, invoke err_trigger.
>>>
>>> b. drivers are supposed to stop MMIO when error is reported
>>> if vm keeps going, we will keep doing MMIO/config
>>> Suggestion 1: ignore this. vm stop happens much later when userspace 
>>> runs anyway,
>>> so we are not making things much worse
>>> Suggestion 2: try to stop MMIO/config, resume on resume call
>>>
>>> Patch below implements Suggestion 1.
>>>
>>> c. PF driver might detect that function is completely broken,
>>> if vm keeps going, we will keep doing MMIO/config
>>> Suggestion 1: ignore this. vm stop happens much later when userspace 
>>> runs anyway,
>>> so we are not making things much worse
>>> Suggestion 2: detect this and invoke err_trigger to stop VM
>>>
>>> Patch below implements Suggestion 2.
>>>
>>> Aside: we currently return PCI_ERS_RESULT_DISCONNECT when device
>>> is not attached. This seems bogus, likely based on the confusing name.
>>> We probably should return PCI_ERS_RESULT_CAN_RECOVER.
>>>
>>> The following patch does not change that.
>>>
>>> Signed-off-by: Michael S. Tsirkin 
>>>
>>> ---
>>>
>>> The patch is completely untested. Let's discuss the design first.
>>> Cao jin, if this is deemed acceptable please take it from here.
>>>
>>
>> Ok, thanks very much.
>>
>>> diff --git a/drivers/vfio/pci/vfio_pci.c b/drivers/vfio/pci/vfio_pci.c
>>> index dce511f..fdca683 100644
>>> --- a/drivers/vfio/pci/vfio_pci.c
>>> +++ b/drivers/vfio/pci/vfio_pci.c
>>> @@ -1292,7 +1292,9 @@ static pci_ers_result_t 
>>> vfio_pci_aer_err_detected(struct pci_dev *pdev,
>>>  
>>> mutex_lock(>igate);
>>>  
>>> -   if (vdev->err_trigger)
>>> +   if (state == pci_channel_io_normal && vdev->non_fatal_err_trigger)
>>> +   eventfd_signal(vdev->err_trigger, 1);
>>> +   else if (vdev->err_trigger)
>>> eventfd_signal(vdev->err_trigger, 1);
>>>  
>>> mutex_unlock(>igate);
>>> @@ -1302,8 +1304,38 @@ static pci_ers_result_t 
>>> vfio_pci_aer_err_detected(struct pci_dev *pdev,
>>> return PCI_ERS_RESULT_CAN_RECOVER;
>>>  }
>>>  
>>> +static pci_ers_result_t vfio_pci_aer_slot_reset(struct pci_dev *pdev,
>>> +   pci_channel_state_t state)
>>> +{
>>> +   struct vfio_pci_device *vdev;
>>> +   struct vfio_device *device;
>>> +
>>> +   device = vfio_device_get_from_dev(>dev);
>>> +   if (!device)
>>> +   goto err_dev;
>>> +
>>> +   vdev = vfio_device_data(device);
>>> +   if (!vdev)
>>> +   goto err_dev;
>>> +
>>> +   mutex_lock(>igate);
>>> +
>>> +   if (vdev->err_trigger)
>>> +   eventfd_signal(vdev->err_trigger, 1);
>>> +
>>> +   mutex_unlock(>igate);
>>> +
>>> +   vfio_device_put(device);
>>> +
>>> +err_data:
>>> +   vfio_device_put(device);
>>> +err_dev:
>>> +   return PCI_ERS_RESULT_RECOVERED;
>>> +}
>>> +
>>>  static const struct pci_error_handlers vfio_err_handlers = {
>>> .error_detected = vfio_pci_aer_err_detected,
>>> +   .slot_reset = vfio_pci_aer_slot_reset,
>>>  };
>>>  
>>
>> if .slot_reset wants to be called, .error_detected should return
>> PCI_ERS_RESULT_NEED_RESET, as pci-error-recovery.txt said, so does code.
>>
>> Is .slot_reset now just a copy of .error_detected and we are going do
>> some tricks here? or else don't get why .slot_reset signal user again.
> 
> 
> No. We do not want a slot reset. But it might be triggered by
> another (PF) driver on the slot. If that happens some driver did something
> to make devices in the slot go to reset and our driver must recover.
> We can't recover however, so let's stop the VM.
> 
> Basically the design is simple
> - if you can keep going - do
> - if you can't - ask qemu to stop guest
> 
> 

Re: [Qemu-devel] [PATCH RFC] vfio error recovery: kernel support

[Michael S. Tsirkin]

On Fri, Jan 20, 2017 at 06:13:22PM +0800, Cao jin wrote:
> 
> 
> On 01/20/2017 04:16 AM, Michael S. Tsirkin wrote:
> > This is a design and an initial patch for kernel side for AER
> > support in VFIO.
> > 
> > 0. What happens now (PCIE AER only)
> >Fatal errors cause a link reset.
> >Non fatal errors don't.
> >All errors stop the VM eventually, but not immediately
> >because it's detected and reported asynchronously.
> >Interrupts are forwarded as usual.
> >Correctable errors are not reported to guest at all.
> >Note: PPC EEH is different. This focuses on AER.
> > 
> > 1. Correctable errors
> >I don't see a need to report these to guest. So let's not.
> > 
> > 2. Fatal errors
> >It's not easy to handle them gracefully since link reset
> >is needed. As a first step, let's use the existing mechanism
> >in that case.
> >
> > 2. Non-fatal errors
> >Here we could make progress by reporting them to guest
> >and have guest handle them.
> >Issues:
> > a. this behaviour should only be enabled with new userspace
> >old userspace should work without changes
> > Suggestion: One way to address this would be to add a new eventfd
> > non_fatal_err_trigger. If not set, invoke err_trigger.
> > 
> > b. drivers are supposed to stop MMIO when error is reported
> > if vm keeps going, we will keep doing MMIO/config
> > Suggestion 1: ignore this. vm stop happens much later when userspace 
> > runs anyway,
> > so we are not making things much worse
> > Suggestion 2: try to stop MMIO/config, resume on resume call
> > 
> > Patch below implements Suggestion 1.
> > 
> > c. PF driver might detect that function is completely broken,
> > if vm keeps going, we will keep doing MMIO/config
> > Suggestion 1: ignore this. vm stop happens much later when userspace 
> > runs anyway,
> > so we are not making things much worse
> > Suggestion 2: detect this and invoke err_trigger to stop VM
> > 
> > Patch below implements Suggestion 2.
> > 
> > Aside: we currently return PCI_ERS_RESULT_DISCONNECT when device
> > is not attached. This seems bogus, likely based on the confusing name.
> > We probably should return PCI_ERS_RESULT_CAN_RECOVER.
> > 
> > The following patch does not change that.
> > 
> > Signed-off-by: Michael S. Tsirkin 
> > 
> > ---
> > 
> > The patch is completely untested. Let's discuss the design first.
> > Cao jin, if this is deemed acceptable please take it from here.
> > 
> 
> Ok, thanks very much.
> 
> > diff --git a/drivers/vfio/pci/vfio_pci.c b/drivers/vfio/pci/vfio_pci.c
> > index dce511f..fdca683 100644
> > --- a/drivers/vfio/pci/vfio_pci.c
> > +++ b/drivers/vfio/pci/vfio_pci.c
> > @@ -1292,7 +1292,9 @@ static pci_ers_result_t 
> > vfio_pci_aer_err_detected(struct pci_dev *pdev,
> >  
> > mutex_lock(>igate);
> >  
> > -   if (vdev->err_trigger)
> > +   if (state == pci_channel_io_normal && vdev->non_fatal_err_trigger)
> > +   eventfd_signal(vdev->err_trigger, 1);
> > +   else if (vdev->err_trigger)
> > eventfd_signal(vdev->err_trigger, 1);
> >  
> > mutex_unlock(>igate);
> > @@ -1302,8 +1304,38 @@ static pci_ers_result_t 
> > vfio_pci_aer_err_detected(struct pci_dev *pdev,
> > return PCI_ERS_RESULT_CAN_RECOVER;
> >  }
> >  
> > +static pci_ers_result_t vfio_pci_aer_slot_reset(struct pci_dev *pdev,
> > +   pci_channel_state_t state)
> > +{
> > +   struct vfio_pci_device *vdev;
> > +   struct vfio_device *device;
> > +
> > +   device = vfio_device_get_from_dev(>dev);
> > +   if (!device)
> > +   goto err_dev;
> > +
> > +   vdev = vfio_device_data(device);
> > +   if (!vdev)
> > +   goto err_dev;
> > +
> > +   mutex_lock(>igate);
> > +
> > +   if (vdev->err_trigger)
> > +   eventfd_signal(vdev->err_trigger, 1);
> > +
> > +   mutex_unlock(>igate);
> > +
> > +   vfio_device_put(device);
> > +
> > +err_data:
> > +   vfio_device_put(device);
> > +err_dev:
> > +   return PCI_ERS_RESULT_RECOVERED;
> > +}
> > +
> >  static const struct pci_error_handlers vfio_err_handlers = {
> > .error_detected = vfio_pci_aer_err_detected,
> > +   .slot_reset = vfio_pci_aer_slot_reset,
> >  };
> >  
> 
> if .slot_reset wants to be called, .error_detected should return
> PCI_ERS_RESULT_NEED_RESET, as pci-error-recovery.txt said, so does code.
> 
> Is .slot_reset now just a copy of .error_detected and we are going do
> some tricks here? or else don't get why .slot_reset signal user again.


No. We do not want a slot reset. But it might be triggered by
another (PF) driver on the slot. If that happens some driver did something
to make devices in the slot go to reset and our driver must recover.
We can't recover however, so let's stop the VM.

Basically the design is simple
- if you can keep going - do
- if you can't - ask qemu to stop guest

Don't worry about adding more conditions to trigger reset etc
at this 

Re: [Qemu-devel] [PATCH RFC] vfio error recovery: kernel support

[Alex Williamson]

On Fri, 20 Jan 2017 18:13:22 +0800
Cao jin  wrote:

> On 01/20/2017 04:16 AM, Michael S. Tsirkin wrote:
> > This is a design and an initial patch for kernel side for AER
> > support in VFIO.
> > 
> > 0. What happens now (PCIE AER only)
> >Fatal errors cause a link reset.
> >Non fatal errors don't.
> >All errors stop the VM eventually, but not immediately
> >because it's detected and reported asynchronously.
> >Interrupts are forwarded as usual.
> >Correctable errors are not reported to guest at all.
> >Note: PPC EEH is different. This focuses on AER.
> > 
> > 1. Correctable errors
> >I don't see a need to report these to guest. So let's not.
> > 
> > 2. Fatal errors
> >It's not easy to handle them gracefully since link reset
> >is needed. As a first step, let's use the existing mechanism
> >in that case.
> >
> > 2. Non-fatal errors
> >Here we could make progress by reporting them to guest
> >and have guest handle them.
> >Issues:
> > a. this behaviour should only be enabled with new userspace
> >old userspace should work without changes
> > Suggestion: One way to address this would be to add a new eventfd
> > non_fatal_err_trigger. If not set, invoke err_trigger.
> > 
> > b. drivers are supposed to stop MMIO when error is reported
> > if vm keeps going, we will keep doing MMIO/config
> > Suggestion 1: ignore this. vm stop happens much later when userspace 
> > runs anyway,
> > so we are not making things much worse
> > Suggestion 2: try to stop MMIO/config, resume on resume call
> > 
> > Patch below implements Suggestion 1.
> > 
> > c. PF driver might detect that function is completely broken,
> > if vm keeps going, we will keep doing MMIO/config
> > Suggestion 1: ignore this. vm stop happens much later when userspace 
> > runs anyway,
> > so we are not making things much worse
> > Suggestion 2: detect this and invoke err_trigger to stop VM
> > 
> > Patch below implements Suggestion 2.
> > 
> > Aside: we currently return PCI_ERS_RESULT_DISCONNECT when device
> > is not attached. This seems bogus, likely based on the confusing name.
> > We probably should return PCI_ERS_RESULT_CAN_RECOVER.
> > 
> > The following patch does not change that.
> > 
> > Signed-off-by: Michael S. Tsirkin 
> > 
> > ---
> > 
> > The patch is completely untested. Let's discuss the design first.
> > Cao jin, if this is deemed acceptable please take it from here.
> >   
> 
> Ok, thanks very much.
> 
> > diff --git a/drivers/vfio/pci/vfio_pci.c b/drivers/vfio/pci/vfio_pci.c
> > index dce511f..fdca683 100644
> > --- a/drivers/vfio/pci/vfio_pci.c
> > +++ b/drivers/vfio/pci/vfio_pci.c
> > @@ -1292,7 +1292,9 @@ static pci_ers_result_t 
> > vfio_pci_aer_err_detected(struct pci_dev *pdev,
> >  
> > mutex_lock(>igate);
> >  
> > -   if (vdev->err_trigger)
> > +   if (state == pci_channel_io_normal && vdev->non_fatal_err_trigger)
> > +   eventfd_signal(vdev->err_trigger, 1);
> > +   else if (vdev->err_trigger)
> > eventfd_signal(vdev->err_trigger, 1);
> >  
> > mutex_unlock(>igate);
> > @@ -1302,8 +1304,38 @@ static pci_ers_result_t 
> > vfio_pci_aer_err_detected(struct pci_dev *pdev,
> > return PCI_ERS_RESULT_CAN_RECOVER;
> >  }
> >  
> > +static pci_ers_result_t vfio_pci_aer_slot_reset(struct pci_dev *pdev,
> > +   pci_channel_state_t state)
> > +{
> > +   struct vfio_pci_device *vdev;
> > +   struct vfio_device *device;
> > +
> > +   device = vfio_device_get_from_dev(>dev);
> > +   if (!device)
> > +   goto err_dev;
> > +
> > +   vdev = vfio_device_data(device);
> > +   if (!vdev)
> > +   goto err_dev;
> > +
> > +   mutex_lock(>igate);
> > +
> > +   if (vdev->err_trigger)
> > +   eventfd_signal(vdev->err_trigger, 1);
> > +
> > +   mutex_unlock(>igate);
> > +
> > +   vfio_device_put(device);
> > +
> > +err_data:
> > +   vfio_device_put(device);
> > +err_dev:
> > +   return PCI_ERS_RESULT_RECOVERED;
> > +}
> > +
> >  static const struct pci_error_handlers vfio_err_handlers = {
> > .error_detected = vfio_pci_aer_err_detected,
> > +   .slot_reset = vfio_pci_aer_slot_reset,
> >  };
> >
> 
> if .slot_reset wants to be called, .error_detected should return
> PCI_ERS_RESULT_NEED_RESET, as pci-error-recovery.txt said, so does code.
> 
> Is .slot_reset now just a copy of .error_detected and we are going do
> some tricks here? or else don't get why .slot_reset signal user again.

If error_detected returns NEED_RESET, then slot_reset will always be
called and every error escalated to fatal and we've made no progress.
The point of having slot_reset is to test whether any other driver
escalated to a NEEDS_RESET; we don't want it to be called.
 
> >  static struct pci_driver vfio_pci_driver = {
> > diff --git a/drivers/vfio/pci/vfio_pci_intrs.c 
> > b/drivers/vfio/pci/vfio_pci_intrs.c

Re: [Qemu-devel] [PATCH RFC] vfio error recovery: kernel support

[Cao jin]

On 01/20/2017 04:16 AM, Michael S. Tsirkin wrote:
> This is a design and an initial patch for kernel side for AER
> support in VFIO.
> 
> 0. What happens now (PCIE AER only)
>Fatal errors cause a link reset.
>Non fatal errors don't.
>All errors stop the VM eventually, but not immediately
>because it's detected and reported asynchronously.
>Interrupts are forwarded as usual.
>Correctable errors are not reported to guest at all.
>Note: PPC EEH is different. This focuses on AER.
> 
> 1. Correctable errors
>I don't see a need to report these to guest. So let's not.
> 
> 2. Fatal errors
>It's not easy to handle them gracefully since link reset
>is needed. As a first step, let's use the existing mechanism
>in that case.
>
> 2. Non-fatal errors
>Here we could make progress by reporting them to guest
>and have guest handle them.
>Issues:
> a. this behaviour should only be enabled with new userspace
>old userspace should work without changes
> Suggestion: One way to address this would be to add a new eventfd
> non_fatal_err_trigger. If not set, invoke err_trigger.
> 
> b. drivers are supposed to stop MMIO when error is reported
> if vm keeps going, we will keep doing MMIO/config
> Suggestion 1: ignore this. vm stop happens much later when userspace runs 
> anyway,
> so we are not making things much worse
> Suggestion 2: try to stop MMIO/config, resume on resume call
> 
> Patch below implements Suggestion 1.
> 
> c. PF driver might detect that function is completely broken,
> if vm keeps going, we will keep doing MMIO/config
> Suggestion 1: ignore this. vm stop happens much later when userspace runs 
> anyway,
> so we are not making things much worse
> Suggestion 2: detect this and invoke err_trigger to stop VM
> 
> Patch below implements Suggestion 2.
> 
> Aside: we currently return PCI_ERS_RESULT_DISCONNECT when device
> is not attached. This seems bogus, likely based on the confusing name.
> We probably should return PCI_ERS_RESULT_CAN_RECOVER.
> 
> The following patch does not change that.
> 
> Signed-off-by: Michael S. Tsirkin 
> 
> ---
> 
> The patch is completely untested. Let's discuss the design first.
> Cao jin, if this is deemed acceptable please take it from here.
> 

Ok, thanks very much.

> diff --git a/drivers/vfio/pci/vfio_pci.c b/drivers/vfio/pci/vfio_pci.c
> index dce511f..fdca683 100644
> --- a/drivers/vfio/pci/vfio_pci.c
> +++ b/drivers/vfio/pci/vfio_pci.c
> @@ -1292,7 +1292,9 @@ static pci_ers_result_t 
> vfio_pci_aer_err_detected(struct pci_dev *pdev,
>  
>   mutex_lock(>igate);
>  
> - if (vdev->err_trigger)
> + if (state == pci_channel_io_normal && vdev->non_fatal_err_trigger)
> + eventfd_signal(vdev->err_trigger, 1);
> + else if (vdev->err_trigger)
>   eventfd_signal(vdev->err_trigger, 1);
>  
>   mutex_unlock(>igate);
> @@ -1302,8 +1304,38 @@ static pci_ers_result_t 
> vfio_pci_aer_err_detected(struct pci_dev *pdev,
>   return PCI_ERS_RESULT_CAN_RECOVER;
>  }
>  
> +static pci_ers_result_t vfio_pci_aer_slot_reset(struct pci_dev *pdev,
> + pci_channel_state_t state)
> +{
> + struct vfio_pci_device *vdev;
> + struct vfio_device *device;
> +
> + device = vfio_device_get_from_dev(>dev);
> + if (!device)
> + goto err_dev;
> +
> + vdev = vfio_device_data(device);
> + if (!vdev)
> + goto err_dev;
> +
> + mutex_lock(>igate);
> +
> + if (vdev->err_trigger)
> + eventfd_signal(vdev->err_trigger, 1);
> +
> + mutex_unlock(>igate);
> +
> + vfio_device_put(device);
> +
> +err_data:
> + vfio_device_put(device);
> +err_dev:
> + return PCI_ERS_RESULT_RECOVERED;
> +}
> +
>  static const struct pci_error_handlers vfio_err_handlers = {
>   .error_detected = vfio_pci_aer_err_detected,
> + .slot_reset = vfio_pci_aer_slot_reset,
>  };
>  

if .slot_reset wants to be called, .error_detected should return
PCI_ERS_RESULT_NEED_RESET, as pci-error-recovery.txt said, so does code.

Is .slot_reset now just a copy of .error_detected and we are going do
some tricks here? or else don't get why .slot_reset signal user again.

>  static struct pci_driver vfio_pci_driver = {
> diff --git a/drivers/vfio/pci/vfio_pci_intrs.c 
> b/drivers/vfio/pci/vfio_pci_intrs.c
> index 1c46045..e883db5 100644
> --- a/drivers/vfio/pci/vfio_pci_intrs.c
> +++ b/drivers/vfio/pci/vfio_pci_intrs.c
> @@ -611,6 +611,17 @@ static int vfio_pci_set_err_trigger(struct 
> vfio_pci_device *vdev,
>  count, flags, data);
>  }
>  
> +static int vfio_pci_set_non_fatal_err_trigger(struct vfio_pci_device *vdev,
> + unsigned index, unsigned start,
> + unsigned count, uint32_t flags, void *data)
> +{
> + if (index != 

Re: [Qemu-devel] [PATCH RFC] vfio error recovery: kernel support

[Michael S. Tsirkin]

On Thu, Jan 19, 2017 at 03:57:34PM -0700, Alex Williamson wrote:
> On Fri, 20 Jan 2017 00:21:02 +0200
> "Michael S. Tsirkin"  wrote:
> 
> > On Thu, Jan 19, 2017 at 03:10:56PM -0700, Alex Williamson wrote:
> > > On Thu, 19 Jan 2017 22:16:03 +0200
> > > "Michael S. Tsirkin"  wrote:
> > >   
> > > > This is a design and an initial patch for kernel side for AER
> > > > support in VFIO.
> > > > 
> > > > 0. What happens now (PCIE AER only)
> > > >Fatal errors cause a link reset.
> > > >Non fatal errors don't.
> > > >All errors stop the VM eventually, but not immediately
> > > >because it's detected and reported asynchronously.
> > > >Interrupts are forwarded as usual.
> > > >Correctable errors are not reported to guest at all.
> > > >Note: PPC EEH is different. This focuses on AER.
> > > > 
> > > > 1. Correctable errors
> > > >I don't see a need to report these to guest. So let's not.
> > > > 
> > > > 2. Fatal errors
> > > >It's not easy to handle them gracefully since link reset
> > > >is needed. As a first step, let's use the existing mechanism
> > > >in that case.
> > > >
> > > > 2. Non-fatal errors
> > > >Here we could make progress by reporting them to guest
> > > >and have guest handle them.
> > > >Issues:
> > > > a. this behaviour should only be enabled with new userspace
> > > >old userspace should work without changes
> > > > Suggestion: One way to address this would be to add a new eventfd
> > > > non_fatal_err_trigger. If not set, invoke err_trigger.
> > > > 
> > > > b. drivers are supposed to stop MMIO when error is reported
> > > > if vm keeps going, we will keep doing MMIO/config
> > > > Suggestion 1: ignore this. vm stop happens much later when 
> > > > userspace runs anyway,
> > > > so we are not making things much worse
> > > > Suggestion 2: try to stop MMIO/config, resume on resume call
> > > > 
> > > > Patch below implements Suggestion 1.  
> > > 
> > > Although this is really against the documentation,  
> > 
> > documentation is out of sync with code unfortunately.
> > I have a todo to rewrite it to match reality, for now
> > you will have to read the recovery function code.
> > Fortunately it is rather short.
> > 
> > > which states
> > > error_detected() is the point at which the driver should quiesce the
> > > device and not touch it further (until diagnostic poking at
> > > mmio_enabled or full access at resume callback).  
> > 
> > Right. But note it's not a regression.
> 
> Agreed.
>  
> > > > c. PF driver might detect that function is completely broken,
> > > > if vm keeps going, we will keep doing MMIO/config
> > > > Suggestion 1: ignore this. vm stop happens much later when 
> > > > userspace runs anyway,
> > > > so we are not making things much worse
> > > > Suggestion 2: detect this and invoke err_trigger to stop VM
> > > > 
> > > > Patch below implements Suggestion 2.
> > > > 
> > > > Aside: we currently return PCI_ERS_RESULT_DISCONNECT when device
> > > > is not attached. This seems bogus, likely based on the confusing name.
> > > > We probably should return PCI_ERS_RESULT_CAN_RECOVER.  
> > > 
> > > Not sure I agree here, if we get called for the error_detected callback
> > > and we can't find a handle for the device, we certainly don't want to
> > > see any of the other callbacks for this device and we can't do anything
> > > about recovering it.  
> > 
> > But we aren't actually driving it from any VMs so it's in the same
> > state it was and not doing any DMA or MMIO.
> 
> If either of the two cases where we return DISCONNECT occur, then we're
> probably getting an error_detected callback as the device is being
> unbound from vfio-pci, or we're just in a completely inconsistent state
> where we're getting aer callbacks for devices not even bound to
> vfio-pci.  We really just want to not be involved with recover of such
> devices. Maybe we should vote PCI_ERS_RESULT_NONE.

I guest I'll tweak slot_reset to behave same as error_detected.
Further enhacements can happen later.


> > >  What's wrong with putting the device into a
> > > failed state in that case?  
> > 
> > That you will wedge the PF too for no good reason.
> > 
> > > I actually question whether CAN_RECOVER is really the right return for
> > > the existing path.  If we consider this to be a fatal error, should we
> > > be voting NEED_RESET?  We're certainly not doing anything to return the
> > > device to a working state.  
> > 
> > 
> > Yes we do - we stop VM and reset device on VM shutdown.
> > At least for VFs this is likely enough as by design they
> > must not wedge each other on driver bugs.
> 
> Right, if the device supports FLR, which VFs are required to do, then
> we can always reset the device.  I think that we have a lot of use
> cases of PF assignment though, I don't think that's just something we
> can push to the side as not recommended.
>

Re: [Qemu-devel] [PATCH RFC] vfio error recovery: kernel support

[Alex Williamson]

On Fri, 20 Jan 2017 00:21:02 +0200
"Michael S. Tsirkin"  wrote:

> On Thu, Jan 19, 2017 at 03:10:56PM -0700, Alex Williamson wrote:
> > On Thu, 19 Jan 2017 22:16:03 +0200
> > "Michael S. Tsirkin"  wrote:
> >   
> > > This is a design and an initial patch for kernel side for AER
> > > support in VFIO.
> > > 
> > > 0. What happens now (PCIE AER only)
> > >Fatal errors cause a link reset.
> > >Non fatal errors don't.
> > >All errors stop the VM eventually, but not immediately
> > >because it's detected and reported asynchronously.
> > >Interrupts are forwarded as usual.
> > >Correctable errors are not reported to guest at all.
> > >Note: PPC EEH is different. This focuses on AER.
> > > 
> > > 1. Correctable errors
> > >I don't see a need to report these to guest. So let's not.
> > > 
> > > 2. Fatal errors
> > >It's not easy to handle them gracefully since link reset
> > >is needed. As a first step, let's use the existing mechanism
> > >in that case.
> > >
> > > 2. Non-fatal errors
> > >Here we could make progress by reporting them to guest
> > >and have guest handle them.
> > >Issues:
> > > a. this behaviour should only be enabled with new userspace
> > >old userspace should work without changes
> > > Suggestion: One way to address this would be to add a new eventfd
> > > non_fatal_err_trigger. If not set, invoke err_trigger.
> > > 
> > > b. drivers are supposed to stop MMIO when error is reported
> > > if vm keeps going, we will keep doing MMIO/config
> > > Suggestion 1: ignore this. vm stop happens much later when userspace 
> > > runs anyway,
> > > so we are not making things much worse
> > > Suggestion 2: try to stop MMIO/config, resume on resume call
> > > 
> > > Patch below implements Suggestion 1.  
> > 
> > Although this is really against the documentation,  
> 
> documentation is out of sync with code unfortunately.
> I have a todo to rewrite it to match reality, for now
> you will have to read the recovery function code.
> Fortunately it is rather short.
> 
> > which states
> > error_detected() is the point at which the driver should quiesce the
> > device and not touch it further (until diagnostic poking at
> > mmio_enabled or full access at resume callback).  
> 
> Right. But note it's not a regression.

Agreed.
 
> > > c. PF driver might detect that function is completely broken,
> > > if vm keeps going, we will keep doing MMIO/config
> > > Suggestion 1: ignore this. vm stop happens much later when userspace 
> > > runs anyway,
> > > so we are not making things much worse
> > > Suggestion 2: detect this and invoke err_trigger to stop VM
> > > 
> > > Patch below implements Suggestion 2.
> > > 
> > > Aside: we currently return PCI_ERS_RESULT_DISCONNECT when device
> > > is not attached. This seems bogus, likely based on the confusing name.
> > > We probably should return PCI_ERS_RESULT_CAN_RECOVER.  
> > 
> > Not sure I agree here, if we get called for the error_detected callback
> > and we can't find a handle for the device, we certainly don't want to
> > see any of the other callbacks for this device and we can't do anything
> > about recovering it.  
> 
> But we aren't actually driving it from any VMs so it's in the same
> state it was and not doing any DMA or MMIO.

If either of the two cases where we return DISCONNECT occur, then we're
probably getting an error_detected callback as the device is being
unbound from vfio-pci, or we're just in a completely inconsistent state
where we're getting aer callbacks for devices not even bound to
vfio-pci.  We really just want to not be involved with recover of such
devices. Maybe we should vote PCI_ERS_RESULT_NONE.

> >  What's wrong with putting the device into a
> > failed state in that case?  
> 
> That you will wedge the PF too for no good reason.
> 
> > I actually question whether CAN_RECOVER is really the right return for
> > the existing path.  If we consider this to be a fatal error, should we
> > be voting NEED_RESET?  We're certainly not doing anything to return the
> > device to a working state.  
> 
> 
> Yes we do - we stop VM and reset device on VM shutdown.
> At least for VFs this is likely enough as by design they
> must not wedge each other on driver bugs.

Right, if the device supports FLR, which VFs are required to do, then
we can always reset the device.  I think that we have a lot of use
cases of PF assignment though, I don't think that's just something we
can push to the side as not recommended.

> >  Should we be more harsh if err_trigger is
> > not registered, putting the device into DISCONNECT?  Should only the new
> > path you've added below for non-fatal errors return CAN_RECOVER?  
> 
> So anyone assigning PFs deserves the resulting pain. I don't
> want to speculate about the best strategy there.
> But for VFs I think CAN_RECOVER is reasonable because
> they should be 

Re: [Qemu-devel] [PATCH RFC] vfio error recovery: kernel support

[Michael S. Tsirkin]

On Thu, Jan 19, 2017 at 03:10:56PM -0700, Alex Williamson wrote:
> On Thu, 19 Jan 2017 22:16:03 +0200
> "Michael S. Tsirkin"  wrote:
> 
> > This is a design and an initial patch for kernel side for AER
> > support in VFIO.
> > 
> > 0. What happens now (PCIE AER only)
> >Fatal errors cause a link reset.
> >Non fatal errors don't.
> >All errors stop the VM eventually, but not immediately
> >because it's detected and reported asynchronously.
> >Interrupts are forwarded as usual.
> >Correctable errors are not reported to guest at all.
> >Note: PPC EEH is different. This focuses on AER.
> > 
> > 1. Correctable errors
> >I don't see a need to report these to guest. So let's not.
> > 
> > 2. Fatal errors
> >It's not easy to handle them gracefully since link reset
> >is needed. As a first step, let's use the existing mechanism
> >in that case.
> >
> > 2. Non-fatal errors
> >Here we could make progress by reporting them to guest
> >and have guest handle them.
> >Issues:
> > a. this behaviour should only be enabled with new userspace
> >old userspace should work without changes
> > Suggestion: One way to address this would be to add a new eventfd
> > non_fatal_err_trigger. If not set, invoke err_trigger.
> > 
> > b. drivers are supposed to stop MMIO when error is reported
> > if vm keeps going, we will keep doing MMIO/config
> > Suggestion 1: ignore this. vm stop happens much later when userspace 
> > runs anyway,
> > so we are not making things much worse
> > Suggestion 2: try to stop MMIO/config, resume on resume call
> > 
> > Patch below implements Suggestion 1.
> 
> Although this is really against the documentation,

documentation is out of sync with code unfortunately.
I have a todo to rewrite it to match reality, for now
you will have to read the recovery function code.
Fortunately it is rather short.

> which states
> error_detected() is the point at which the driver should quiesce the
> device and not touch it further (until diagnostic poking at
> mmio_enabled or full access at resume callback).

Right. But note it's not a regression.

> > c. PF driver might detect that function is completely broken,
> > if vm keeps going, we will keep doing MMIO/config
> > Suggestion 1: ignore this. vm stop happens much later when userspace 
> > runs anyway,
> > so we are not making things much worse
> > Suggestion 2: detect this and invoke err_trigger to stop VM
> > 
> > Patch below implements Suggestion 2.
> > 
> > Aside: we currently return PCI_ERS_RESULT_DISCONNECT when device
> > is not attached. This seems bogus, likely based on the confusing name.
> > We probably should return PCI_ERS_RESULT_CAN_RECOVER.
> 
> Not sure I agree here, if we get called for the error_detected callback
> and we can't find a handle for the device, we certainly don't want to
> see any of the other callbacks for this device and we can't do anything
> about recovering it.

But we aren't actually driving it from any VMs so it's in the same
state it was and not doing any DMA or MMIO.

>  What's wrong with putting the device into a
> failed state in that case?

That you will wedge the PF too for no good reason.

> I actually question whether CAN_RECOVER is really the right return for
> the existing path.  If we consider this to be a fatal error, should we
> be voting NEED_RESET?  We're certainly not doing anything to return the
> device to a working state.


Yes we do - we stop VM and reset device on VM shutdown.
At least for VFs this is likely enough as by design they
must not wedge each other on driver bugs.

>  Should we be more harsh if err_trigger is
> not registered, putting the device into DISCONNECT?  Should only the new
> path you've added below for non-fatal errors return CAN_RECOVER?

So anyone assigning PFs deserves the resulting pain. I don't
want to speculate about the best strategy there.
But for VFs I think CAN_RECOVER is reasonable because
they should be independent of each other.

Also pls note any status except CAN_RECOVER mostly just wedges hardware
ATM. Maybe AER should do link resets more aggressively but it does not.

> > The following patch does not change that.
> > 
> > Signed-off-by: Michael S. Tsirkin 
> > 
> > ---
> > 
> > The patch is completely untested. Let's discuss the design first.
> > Cao jin, if this is deemed acceptable please take it from here.
> > 
> > diff --git a/drivers/vfio/pci/vfio_pci.c b/drivers/vfio/pci/vfio_pci.c
> > index dce511f..fdca683 100644
> > --- a/drivers/vfio/pci/vfio_pci.c
> > +++ b/drivers/vfio/pci/vfio_pci.c
> > @@ -1292,7 +1292,9 @@ static pci_ers_result_t 
> > vfio_pci_aer_err_detected(struct pci_dev *pdev,
> >  
> > mutex_lock(>igate);
> >  
> > -   if (vdev->err_trigger)
> > +   if (state == pci_channel_io_normal && vdev->non_fatal_err_trigger)
> > +   eventfd_signal(vdev->err_trigger, 1);
> 
> 

Re: [Qemu-devel] [PATCH RFC] vfio error recovery: kernel support

[Alex Williamson]

On Thu, 19 Jan 2017 22:16:03 +0200
"Michael S. Tsirkin"  wrote:

> This is a design and an initial patch for kernel side for AER
> support in VFIO.
> 
> 0. What happens now (PCIE AER only)
>Fatal errors cause a link reset.
>Non fatal errors don't.
>All errors stop the VM eventually, but not immediately
>because it's detected and reported asynchronously.
>Interrupts are forwarded as usual.
>Correctable errors are not reported to guest at all.
>Note: PPC EEH is different. This focuses on AER.
> 
> 1. Correctable errors
>I don't see a need to report these to guest. So let's not.
> 
> 2. Fatal errors
>It's not easy to handle them gracefully since link reset
>is needed. As a first step, let's use the existing mechanism
>in that case.
>
> 2. Non-fatal errors
>Here we could make progress by reporting them to guest
>and have guest handle them.
>Issues:
> a. this behaviour should only be enabled with new userspace
>old userspace should work without changes
> Suggestion: One way to address this would be to add a new eventfd
> non_fatal_err_trigger. If not set, invoke err_trigger.
> 
> b. drivers are supposed to stop MMIO when error is reported
> if vm keeps going, we will keep doing MMIO/config
> Suggestion 1: ignore this. vm stop happens much later when userspace runs 
> anyway,
> so we are not making things much worse
> Suggestion 2: try to stop MMIO/config, resume on resume call
> 
> Patch below implements Suggestion 1.

Although this is really against the documentation, which states
error_detected() is the point at which the driver should quiesce the
device and not touch it further (until diagnostic poking at
mmio_enabled or full access at resume callback).
 
> c. PF driver might detect that function is completely broken,
> if vm keeps going, we will keep doing MMIO/config
> Suggestion 1: ignore this. vm stop happens much later when userspace runs 
> anyway,
> so we are not making things much worse
> Suggestion 2: detect this and invoke err_trigger to stop VM
> 
> Patch below implements Suggestion 2.
> 
> Aside: we currently return PCI_ERS_RESULT_DISCONNECT when device
> is not attached. This seems bogus, likely based on the confusing name.
> We probably should return PCI_ERS_RESULT_CAN_RECOVER.

Not sure I agree here, if we get called for the error_detected callback
and we can't find a handle for the device, we certainly don't want to
see any of the other callbacks for this device and we can't do anything
about recovering it.  What's wrong with putting the device into a
failed state in that case?

I actually question whether CAN_RECOVER is really the right return for
the existing path.  If we consider this to be a fatal error, should we
be voting NEED_RESET?  We're certainly not doing anything to return the
device to a working state.  Should we be more harsh if err_trigger is
not registered, putting the device into DISCONNECT?  Should only the new
path you've added below for non-fatal errors return CAN_RECOVER?

> The following patch does not change that.
> 
> Signed-off-by: Michael S. Tsirkin 
> 
> ---
> 
> The patch is completely untested. Let's discuss the design first.
> Cao jin, if this is deemed acceptable please take it from here.
> 
> diff --git a/drivers/vfio/pci/vfio_pci.c b/drivers/vfio/pci/vfio_pci.c
> index dce511f..fdca683 100644
> --- a/drivers/vfio/pci/vfio_pci.c
> +++ b/drivers/vfio/pci/vfio_pci.c
> @@ -1292,7 +1292,9 @@ static pci_ers_result_t 
> vfio_pci_aer_err_detected(struct pci_dev *pdev,
>  
>   mutex_lock(>igate);
>  
> - if (vdev->err_trigger)
> + if (state == pci_channel_io_normal && vdev->non_fatal_err_trigger)
> + eventfd_signal(vdev->err_trigger, 1);

s/err_trigger/non_fatal_err_trigger/

> + else if (vdev->err_trigger)
>   eventfd_signal(vdev->err_trigger, 1);
>  
>   mutex_unlock(>igate);
> @@ -1302,8 +1304,38 @@ static pci_ers_result_t 
> vfio_pci_aer_err_detected(struct pci_dev *pdev,
>   return PCI_ERS_RESULT_CAN_RECOVER;
>  }
>  
> +static pci_ers_result_t vfio_pci_aer_slot_reset(struct pci_dev *pdev,
> + pci_channel_state_t state)
> +{
> + struct vfio_pci_device *vdev;
> + struct vfio_device *device;
> +
> + device = vfio_device_get_from_dev(>dev);
> + if (!device)
> + goto err_dev;
> +
> + vdev = vfio_device_data(device);
> + if (!vdev)
> + goto err_dev;

s/err_dev/err_data/

> +
> + mutex_lock(>igate);
> +
> + if (vdev->err_trigger)
> + eventfd_signal(vdev->err_trigger, 1);
> +
> + mutex_unlock(>igate);
> +
> + vfio_device_put(device);
> +
> +err_data:
> + vfio_device_put(device);
> +err_dev:
> + return PCI_ERS_RESULT_RECOVERED;
> +}
> +
>  static const struct pci_error_handlers vfio_err_handlers = {
>   .error_detected =