[v9,12/15] PCI/RCEC: Add RCiEP's linked RCEC to AER/ERR

Commit Message

Sean V Kelley Oct. 16, 2020, 12:11 a.m. UTC

From: Qiuxu Zhuo <[e-mail]>

When attempting error recovery for an RCiEP associated with an RCEC device,
there needs to be a way to update the Root Error Status, the Uncorrectable
Error Status and the Uncorrectable Error Severity of the parent RCEC.  In
some non-native cases in which there is no OS-visible device associated
with the RCiEP, there is nothing to act upon as the firmware is acting
before the OS.

Add handling for the linked RCEC in AER/ERR while taking into account
non-native cases.

Co-developed-by: Sean V Kelley <[e-mail]>
Link: https://lore.kernel.org/r/[e-mail]
Signed-off-by: Sean V Kelley <[e-mail]>
Signed-off-by: Qiuxu Zhuo <[e-mail]>
Signed-off-by: Bjorn Helgaas <[e-mail]>
Reviewed-by: Jonathan Cameron <[e-mail]>
---
 drivers/pci/pcie/aer.c | 53 ++++++++++++++++++++++++++++++------------
 drivers/pci/pcie/err.c | 20 ++++++++--------
 2 files changed, 48 insertions(+), 25 deletions(-)

Comments

Bjorn Helgaas Oct. 16, 2020, 8:30 p.m. UTC | #1

[+to Jonathan]

On Thu, Oct 15, 2020 at 05:11:10PM -0700, Sean V Kelley wrote:
> From: Qiuxu Zhuo <[e-mail]>
> 
> When attempting error recovery for an RCiEP associated with an RCEC device,
> there needs to be a way to update the Root Error Status, the Uncorrectable
> Error Status and the Uncorrectable Error Severity of the parent RCEC.  In
> some non-native cases in which there is no OS-visible device associated
> with the RCiEP, there is nothing to act upon as the firmware is acting
> before the OS.
> 
> Add handling for the linked RCEC in AER/ERR while taking into account
> non-native cases.
> 
> Co-developed-by: Sean V Kelley <[e-mail]>
> Link: https://lore.kernel.org/r/[e-mail]
> Signed-off-by: Sean V Kelley <[e-mail]>
> Signed-off-by: Qiuxu Zhuo <[e-mail]>
> Signed-off-by: Bjorn Helgaas <[e-mail]>
> Reviewed-by: Jonathan Cameron <[e-mail]>
> ---
>  drivers/pci/pcie/aer.c | 53 ++++++++++++++++++++++++++++++------------
>  drivers/pci/pcie/err.c | 20 ++++++++--------
>  2 files changed, 48 insertions(+), 25 deletions(-)
> 
> diff --git a/drivers/pci/pcie/aer.c b/drivers/pci/pcie/aer.c
> index 65dff5f3457a..083f69b67bfd 100644
> --- a/drivers/pci/pcie/aer.c
> +++ b/drivers/pci/pcie/aer.c
> @@ -1357,27 +1357,50 @@ static int aer_probe(struct pcie_device *dev)
>   */
>  static pci_ers_result_t aer_root_reset(struct pci_dev *dev)
>  {
> -	int aer = dev->aer_cap;
> +	int type = pci_pcie_type(dev);
> +	struct pci_dev *root;
> +	int aer = 0;
> +	int rc = 0;
>  	u32 reg32;
> -	int rc;
>  
> +	if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)

"type == PCI_EXP_TYPE_RC_END"

> +		/*
> +		 * The reset should only clear the Root Error Status
> +		 * of the RCEC. Only perform this for the
> +		 * native case, i.e., an RCEC is present.
> +		 */
> +		root = dev->rcec;
> +	else
> +		root = dev;
>  
> -	/* Disable Root's interrupt in response to error messages */
> -	pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, &reg32);
> -	reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
> -	pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, reg32);
> +	if (root)
> +		aer = dev->aer_cap;
>  
> -	rc = pci_bus_error_reset(dev);
> -	pci_info(dev, "Root Port link has been reset\n");
> +	if (aer) {
> +		/* Disable Root's interrupt in response to error messages */
> +		pci_read_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, &reg32);
> +		reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
> +		pci_write_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, reg32);

Not directly related to *this* patch, but my assumption was that in
the APEI case, the firmware should retain ownership of the AER
Capability, so the OS should not touch PCI_ERR_ROOT_COMMAND and
PCI_ERR_ROOT_STATUS.

But this code appears to ignore that ownership.  Jonathan, you must
have looked at this recently for 068c29a248b6 ("PCI/ERR: Clear PCIe
Device Status errors only if OS owns AER").  Do you have any insight
about this?

> -	/* Clear Root Error Status */
> -	pci_read_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, &reg32);
> -	pci_write_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, reg32);
> +		/* Clear Root Error Status */
> +		pci_read_config_dword(root, aer + PCI_ERR_ROOT_STATUS, &reg32);
> +		pci_write_config_dword(root, aer + PCI_ERR_ROOT_STATUS, reg32);
>  
> -	/* Enable Root Port's interrupt in response to error messages */
> -	pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, &reg32);
> -	reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
> -	pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, reg32);
> +		/* Enable Root Port's interrupt in response to error messages */
> +		pci_read_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, &reg32);
> +		reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
> +		pci_write_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, reg32);
> +	}
> +
> +	if ((type == PCI_EXP_TYPE_RC_EC) || (type == PCI_EXP_TYPE_RC_END)) {
> +		if (pcie_has_flr(root)) {
> +			rc = pcie_flr(root);
> +			pci_info(dev, "has been reset (%d)\n", rc);
> +		}
> +	} else {
> +		rc = pci_bus_error_reset(root);

Don't we want "dev" for both the FLR and pci_bus_error_reset()?  I
think "root == dev" except when dev is an RCiEP.  When dev is an
RCiEP, "root" is the RCEC (if present), and we want to reset the
RCiEP, not the RCEC.

> +		pci_info(dev, "Root Port link has been reset (%d)\n", rc);
> +	}

There are a couple changes here that I think should be split out.

Based on my theory that when firmware retains control of AER, the OS
should not touch PCI_ERR_ROOT_COMMAND and PCI_ERR_ROOT_STATUS, and any
updates to them would have to be done by firmware before we get here,
I suggested reordering this:

  - clear PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
  - do reset
  - clear PCI_ERR_ROOT_STATUS (for APEI, presumably done by firmware?)
  - enable PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK

to this:

  - clear PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
  - clear PCI_ERR_ROOT_STATUS
  - enable PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
  - do reset

If my theory is correct, I think we should still reorder this, but:

  - It's a significant behavior change that deserves its own patch so
    we can document/bisect/revert.

  - I'm not sure why we clear the PCI_ERR_ROOT_COMMAND error reporting
    bits.  In the new "clear COMMAND, clear STATUS, enable COMMAND"
    order, it looks superfluous.  There's no reason to disable error
    reporting while clearing the status bits.

    The current "clear, reset, enable" order suggests that the reset
    might cause errors that we should ignore.  I don't know whether
    that's the case or not.  It dates from 6c2b374d7485 ("PCI-Express
    AER implemetation: AER core and aerdriver"), which doesn't
    elaborate.

  - Should we also test for OS ownership of AER before touching
    PCI_ERR_ROOT_STATUS?

  - If we remove the PCI_ERR_ROOT_COMMAND fiddling (and I tentatively
    think we *should* unless we can justify it), that would also
    deserve its own patch.  Possibly (1) remove PCI_ERR_ROOT_COMMAND
    fiddling, (2) reorder PCI_ERR_ROOT_STATUS clearing and reset, (3)
    test for OS ownership of AER (?), (4) the rest of this patch.

>  	return rc ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
>  }
> diff --git a/drivers/pci/pcie/err.c b/drivers/pci/pcie/err.c
> index 7883c9791562..cbc5abfe767b 100644
> --- a/drivers/pci/pcie/err.c
> +++ b/drivers/pci/pcie/err.c
> @@ -148,10 +148,10 @@ static int report_resume(struct pci_dev *dev, void *data)
>  
>  /**
>   * pci_walk_bridge - walk bridges potentially AER affected
> - * @bridge:	bridge which may be a Port, an RCEC with associated RCiEPs,
> - *		or an RCiEP associated with an RCEC
> - * @cb:		callback to be called for each device found
> - * @userdata:	arbitrary pointer to be passed to callback
> + * @bridge   bridge which may be an RCEC with associated RCiEPs,
> + *           or a Port.
> + * @cb       callback to be called for each device found
> + * @userdata arbitrary pointer to be passed to callback.
>   *
>   * If the device provided is a bridge, walk the subordinate bus, including
>   * any bridged devices on buses under this bus.  Call the provided callback
> @@ -164,8 +164,14 @@ static void pci_walk_bridge(struct pci_dev *bridge,
>  			    int (*cb)(struct pci_dev *, void *),
>  			    void *userdata)
>  {
> +	/*
> +	 * In a non-native case where there is no OS-visible reporting
> +	 * device the bridge will be NULL, i.e., no RCEC, no Downstream Port.
> +	 */
>  	if (bridge->subordinate)
>  		pci_walk_bus(bridge->subordinate, cb, userdata);
> +	else if (bridge->rcec)
> +		cb(bridge->rcec, userdata);
>  	else
>  		cb(bridge, userdata);
>  }
> @@ -194,12 +200,6 @@ pci_ers_result_t pcie_do_recovery(struct pci_dev *dev,
>  	pci_dbg(bridge, "broadcast error_detected message\n");
>  	if (state == pci_channel_io_frozen) {
>  		pci_walk_bridge(bridge, report_frozen_detected, &status);
> -		if (type == PCI_EXP_TYPE_RC_END) {
> -			pci_warn(dev, "subordinate device reset not possible for RCiEP\n");
> -			status = PCI_ERS_RESULT_NONE;
> -			goto failed;
> -		}
> -
>  		status = reset_subordinates(bridge);
>  		if (status != PCI_ERS_RESULT_RECOVERED) {
>  			pci_warn(bridge, "subordinate device reset failed\n");
> -- 
> 2.28.0
>

Bjorn Helgaas Oct. 16, 2020, 10:29 p.m. UTC | #2

[+cc Christoph, Ethan, Sinan, Keith; sorry should have cc'd you to
begin with since you're looking at this code too.  Particularly
interested in your thoughts about whether we should be touching
PCI_ERR_ROOT_COMMAND and PCI_ERR_ROOT_STATUS when we don't own AER.]

On Fri, Oct 16, 2020 at 03:30:37PM -0500, Bjorn Helgaas wrote:
> [+to Jonathan]
> 
> On Thu, Oct 15, 2020 at 05:11:10PM -0700, Sean V Kelley wrote:
> > From: Qiuxu Zhuo <[e-mail]>
> > 
> > When attempting error recovery for an RCiEP associated with an RCEC device,
> > there needs to be a way to update the Root Error Status, the Uncorrectable
> > Error Status and the Uncorrectable Error Severity of the parent RCEC.  In
> > some non-native cases in which there is no OS-visible device associated
> > with the RCiEP, there is nothing to act upon as the firmware is acting
> > before the OS.
> > 
> > Add handling for the linked RCEC in AER/ERR while taking into account
> > non-native cases.
> > 
> > Co-developed-by: Sean V Kelley <[e-mail]>
> > Link: https://lore.kernel.org/r/[e-mail]
> > Signed-off-by: Sean V Kelley <[e-mail]>
> > Signed-off-by: Qiuxu Zhuo <[e-mail]>
> > Signed-off-by: Bjorn Helgaas <[e-mail]>
> > Reviewed-by: Jonathan Cameron <[e-mail]>
> > ---
> >  drivers/pci/pcie/aer.c | 53 ++++++++++++++++++++++++++++++------------
> >  drivers/pci/pcie/err.c | 20 ++++++++--------
> >  2 files changed, 48 insertions(+), 25 deletions(-)
> > 
> > diff --git a/drivers/pci/pcie/aer.c b/drivers/pci/pcie/aer.c
> > index 65dff5f3457a..083f69b67bfd 100644
> > --- a/drivers/pci/pcie/aer.c
> > +++ b/drivers/pci/pcie/aer.c
> > @@ -1357,27 +1357,50 @@ static int aer_probe(struct pcie_device *dev)
> >   */
> >  static pci_ers_result_t aer_root_reset(struct pci_dev *dev)
> >  {
> > -	int aer = dev->aer_cap;
> > +	int type = pci_pcie_type(dev);
> > +	struct pci_dev *root;
> > +	int aer = 0;
> > +	int rc = 0;
> >  	u32 reg32;
> > -	int rc;
> >  
> > +	if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
> 
> "type == PCI_EXP_TYPE_RC_END"
> 
> > +		/*
> > +		 * The reset should only clear the Root Error Status
> > +		 * of the RCEC. Only perform this for the
> > +		 * native case, i.e., an RCEC is present.
> > +		 */
> > +		root = dev->rcec;
> > +	else
> > +		root = dev;
> >  
> > -	/* Disable Root's interrupt in response to error messages */
> > -	pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, &reg32);
> > -	reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
> > -	pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, reg32);
> > +	if (root)
> > +		aer = dev->aer_cap;
> >  
> > -	rc = pci_bus_error_reset(dev);
> > -	pci_info(dev, "Root Port link has been reset\n");
> > +	if (aer) {
> > +		/* Disable Root's interrupt in response to error messages */
> > +		pci_read_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, &reg32);
> > +		reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
> > +		pci_write_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, reg32);
> 
> Not directly related to *this* patch, but my assumption was that in
> the APEI case, the firmware should retain ownership of the AER
> Capability, so the OS should not touch PCI_ERR_ROOT_COMMAND and
> PCI_ERR_ROOT_STATUS.
> 
> But this code appears to ignore that ownership.  Jonathan, you must
> have looked at this recently for 068c29a248b6 ("PCI/ERR: Clear PCIe
> Device Status errors only if OS owns AER").  Do you have any insight
> about this?
> 
> > -	/* Clear Root Error Status */
> > -	pci_read_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, &reg32);
> > -	pci_write_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, reg32);
> > +		/* Clear Root Error Status */
> > +		pci_read_config_dword(root, aer + PCI_ERR_ROOT_STATUS, &reg32);
> > +		pci_write_config_dword(root, aer + PCI_ERR_ROOT_STATUS, reg32);
> >  
> > -	/* Enable Root Port's interrupt in response to error messages */
> > -	pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, &reg32);
> > -	reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
> > -	pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, reg32);
> > +		/* Enable Root Port's interrupt in response to error messages */
> > +		pci_read_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, &reg32);
> > +		reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
> > +		pci_write_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, reg32);
> > +	}
> > +
> > +	if ((type == PCI_EXP_TYPE_RC_EC) || (type == PCI_EXP_TYPE_RC_END)) {
> > +		if (pcie_has_flr(root)) {
> > +			rc = pcie_flr(root);
> > +			pci_info(dev, "has been reset (%d)\n", rc);
> > +		}
> > +	} else {
> > +		rc = pci_bus_error_reset(root);
> 
> Don't we want "dev" for both the FLR and pci_bus_error_reset()?  I
> think "root == dev" except when dev is an RCiEP.  When dev is an
> RCiEP, "root" is the RCEC (if present), and we want to reset the
> RCiEP, not the RCEC.
> 
> > +		pci_info(dev, "Root Port link has been reset (%d)\n", rc);
> > +	}
> 
> There are a couple changes here that I think should be split out.
> 
> Based on my theory that when firmware retains control of AER, the OS
> should not touch PCI_ERR_ROOT_COMMAND and PCI_ERR_ROOT_STATUS, and any
> updates to them would have to be done by firmware before we get here,
> I suggested reordering this:
> 
>   - clear PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>   - do reset
>   - clear PCI_ERR_ROOT_STATUS (for APEI, presumably done by firmware?)
>   - enable PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
> 
> to this:
> 
>   - clear PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>   - clear PCI_ERR_ROOT_STATUS
>   - enable PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>   - do reset
> 
> If my theory is correct, I think we should still reorder this, but:
> 
>   - It's a significant behavior change that deserves its own patch so
>     we can document/bisect/revert.
> 
>   - I'm not sure why we clear the PCI_ERR_ROOT_COMMAND error reporting
>     bits.  In the new "clear COMMAND, clear STATUS, enable COMMAND"
>     order, it looks superfluous.  There's no reason to disable error
>     reporting while clearing the status bits.
> 
>     The current "clear, reset, enable" order suggests that the reset
>     might cause errors that we should ignore.  I don't know whether
>     that's the case or not.  It dates from 6c2b374d7485 ("PCI-Express
>     AER implemetation: AER core and aerdriver"), which doesn't
>     elaborate.
> 
>   - Should we also test for OS ownership of AER before touching
>     PCI_ERR_ROOT_STATUS?
> 
>   - If we remove the PCI_ERR_ROOT_COMMAND fiddling (and I tentatively
>     think we *should* unless we can justify it), that would also
>     deserve its own patch.  Possibly (1) remove PCI_ERR_ROOT_COMMAND
>     fiddling, (2) reorder PCI_ERR_ROOT_STATUS clearing and reset, (3)
>     test for OS ownership of AER (?), (4) the rest of this patch.
> 
> >  	return rc ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
> >  }
> > diff --git a/drivers/pci/pcie/err.c b/drivers/pci/pcie/err.c
> > index 7883c9791562..cbc5abfe767b 100644
> > --- a/drivers/pci/pcie/err.c
> > +++ b/drivers/pci/pcie/err.c
> > @@ -148,10 +148,10 @@ static int report_resume(struct pci_dev *dev, void *data)
> >  
> >  /**
> >   * pci_walk_bridge - walk bridges potentially AER affected
> > - * @bridge:	bridge which may be a Port, an RCEC with associated RCiEPs,
> > - *		or an RCiEP associated with an RCEC
> > - * @cb:		callback to be called for each device found
> > - * @userdata:	arbitrary pointer to be passed to callback
> > + * @bridge   bridge which may be an RCEC with associated RCiEPs,
> > + *           or a Port.
> > + * @cb       callback to be called for each device found
> > + * @userdata arbitrary pointer to be passed to callback.
> >   *
> >   * If the device provided is a bridge, walk the subordinate bus, including
> >   * any bridged devices on buses under this bus.  Call the provided callback
> > @@ -164,8 +164,14 @@ static void pci_walk_bridge(struct pci_dev *bridge,
> >  			    int (*cb)(struct pci_dev *, void *),
> >  			    void *userdata)
> >  {
> > +	/*
> > +	 * In a non-native case where there is no OS-visible reporting
> > +	 * device the bridge will be NULL, i.e., no RCEC, no Downstream Port.
> > +	 */
> >  	if (bridge->subordinate)
> >  		pci_walk_bus(bridge->subordinate, cb, userdata);
> > +	else if (bridge->rcec)
> > +		cb(bridge->rcec, userdata);
> >  	else
> >  		cb(bridge, userdata);
> >  }
> > @@ -194,12 +200,6 @@ pci_ers_result_t pcie_do_recovery(struct pci_dev *dev,
> >  	pci_dbg(bridge, "broadcast error_detected message\n");
> >  	if (state == pci_channel_io_frozen) {
> >  		pci_walk_bridge(bridge, report_frozen_detected, &status);
> > -		if (type == PCI_EXP_TYPE_RC_END) {
> > -			pci_warn(dev, "subordinate device reset not possible for RCiEP\n");
> > -			status = PCI_ERS_RESULT_NONE;
> > -			goto failed;
> > -		}
> > -
> >  		status = reset_subordinates(bridge);
> >  		if (status != PCI_ERS_RESULT_RECOVERED) {
> >  			pci_warn(bridge, "subordinate device reset failed\n");
> > -- 
> > 2.28.0
> >

Kuppuswamy, Sathyanarayanan Oct. 17, 2020, 12:28 a.m. UTC | #3

On 10/16/20 3:29 PM, Bjorn Helgaas wrote:
> [+cc Christoph, Ethan, Sinan, Keith; sorry should have cc'd you to
> begin with since you're looking at this code too.  Particularly
> interested in your thoughts about whether we should be touching
> PCI_ERR_ROOT_COMMAND and PCI_ERR_ROOT_STATUS when we don't own AER.]
This part is not very clear in ACPI spec or PCI firmware spec.

IMO, since AEPI notifies the OS about the error, then I guess
we are allowed to clear the PCI_ERR_ROOT_STATUS register
after handling the error.

>
> On Fri, Oct 16, 2020 at 03:30:37PM -0500, Bjorn Helgaas wrote:
>> [+to Jonathan]
>>
>> On Thu, Oct 15, 2020 at 05:11:10PM -0700, Sean V Kelley wrote:
>>> From: Qiuxu Zhuo <[e-mail]>
>>>
>>> When attempting error recovery for an RCiEP associated with an RCEC device,
>>> there needs to be a way to update the Root Error Status, the Uncorrectable
>>> Error Status and the Uncorrectable Error Severity of the parent RCEC.  In
>>> some non-native cases in which there is no OS-visible device associated
>>> with the RCiEP, there is nothing to act upon as the firmware is acting
>>> before the OS.
>>>
>>> Add handling for the linked RCEC in AER/ERR while taking into account
>>> non-native cases.
>>>
>>> Co-developed-by: Sean V Kelley <[e-mail]>
>>> Link: https://lore.kernel.org/r/[e-mail]
>>> Signed-off-by: Sean V Kelley <[e-mail]>
>>> Signed-off-by: Qiuxu Zhuo <[e-mail]>
>>> Signed-off-by: Bjorn Helgaas <[e-mail]>
>>> Reviewed-by: Jonathan Cameron <[e-mail]>
>>> ---
>>>   drivers/pci/pcie/aer.c | 53 ++++++++++++++++++++++++++++++------------
>>>   drivers/pci/pcie/err.c | 20 ++++++++--------
>>>   2 files changed, 48 insertions(+), 25 deletions(-)
>>>
>>> diff --git a/drivers/pci/pcie/aer.c b/drivers/pci/pcie/aer.c
>>> index 65dff5f3457a..083f69b67bfd 100644
>>> --- a/drivers/pci/pcie/aer.c
>>> +++ b/drivers/pci/pcie/aer.c
>>> @@ -1357,27 +1357,50 @@ static int aer_probe(struct pcie_device *dev)
>>>    */
>>>   static pci_ers_result_t aer_root_reset(struct pci_dev *dev)
>>>   {
>>> -	int aer = dev->aer_cap;
>>> +	int type = pci_pcie_type(dev);
>>> +	struct pci_dev *root;
>>> +	int aer = 0;
>>> +	int rc = 0;
>>>   	u32 reg32;
>>> -	int rc;
>>>   
>>> +	if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
>> "type == PCI_EXP_TYPE_RC_END"
>>
>>> +		/*
>>> +		 * The reset should only clear the Root Error Status
>>> +		 * of the RCEC. Only perform this for the
>>> +		 * native case, i.e., an RCEC is present.
>>> +		 */
>>> +		root = dev->rcec;
>>> +	else
>>> +		root = dev;
>>>   
>>> -	/* Disable Root's interrupt in response to error messages */
>>> -	pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, &reg32);
>>> -	reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
>>> -	pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, reg32);
>>> +	if (root)
>>> +		aer = dev->aer_cap;
>>>   
>>> -	rc = pci_bus_error_reset(dev);
>>> -	pci_info(dev, "Root Port link has been reset\n");
>>> +	if (aer) {
>>> +		/* Disable Root's interrupt in response to error messages */
>>> +		pci_read_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, &reg32);
>>> +		reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
>>> +		pci_write_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, reg32);
>> Not directly related to *this* patch, but my assumption was that in
>> the APEI case, the firmware should retain ownership of the AER
>> Capability, so the OS should not touch PCI_ERR_ROOT_COMMAND and
>> PCI_ERR_ROOT_STATUS.
>>
>> But this code appears to ignore that ownership.  Jonathan, you must
>> have looked at this recently for 068c29a248b6 ("PCI/ERR: Clear PCIe
>> Device Status errors only if OS owns AER").  Do you have any insight
>> about this?
>>
>>> -	/* Clear Root Error Status */
>>> -	pci_read_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, &reg32);
>>> -	pci_write_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, reg32);
>>> +		/* Clear Root Error Status */
>>> +		pci_read_config_dword(root, aer + PCI_ERR_ROOT_STATUS, &reg32);
>>> +		pci_write_config_dword(root, aer + PCI_ERR_ROOT_STATUS, reg32);
>>>   
>>> -	/* Enable Root Port's interrupt in response to error messages */
>>> -	pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, &reg32);
>>> -	reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
>>> -	pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, reg32);
>>> +		/* Enable Root Port's interrupt in response to error messages */
>>> +		pci_read_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, &reg32);
>>> +		reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
>>> +		pci_write_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, reg32);
>>> +	}
>>> +
>>> +	if ((type == PCI_EXP_TYPE_RC_EC) || (type == PCI_EXP_TYPE_RC_END)) {
>>> +		if (pcie_has_flr(root)) {
>>> +			rc = pcie_flr(root);
>>> +			pci_info(dev, "has been reset (%d)\n", rc);
>>> +		}
>>> +	} else {
>>> +		rc = pci_bus_error_reset(root);
>> Don't we want "dev" for both the FLR and pci_bus_error_reset()?  I
>> think "root == dev" except when dev is an RCiEP.  When dev is an
>> RCiEP, "root" is the RCEC (if present), and we want to reset the
>> RCiEP, not the RCEC.
>>
>>> +		pci_info(dev, "Root Port link has been reset (%d)\n", rc);
>>> +	}
>> There are a couple changes here that I think should be split out.
>>
>> Based on my theory that when firmware retains control of AER, the OS
>> should not touch PCI_ERR_ROOT_COMMAND and PCI_ERR_ROOT_STATUS, and any
>> updates to them would have to be done by firmware before we get here,
>> I suggested reordering this:
>>
>>    - clear PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>>    - do reset
>>    - clear PCI_ERR_ROOT_STATUS (for APEI, presumably done by firmware?)
>>    - enable PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>>
>> to this:
>>
>>    - clear PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>>    - clear PCI_ERR_ROOT_STATUS
>>    - enable PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>>    - do reset
>>
>> If my theory is correct, I think we should still reorder this, but:
>>
>>    - It's a significant behavior change that deserves its own patch so
>>      we can document/bisect/revert.
>>
>>    - I'm not sure why we clear the PCI_ERR_ROOT_COMMAND error reporting
>>      bits.  In the new "clear COMMAND, clear STATUS, enable COMMAND"
>>      order, it looks superfluous.  There's no reason to disable error
>>      reporting while clearing the status bits.
>>
>>      The current "clear, reset, enable" order suggests that the reset
>>      might cause errors that we should ignore.  I don't know whether
>>      that's the case or not.  It dates from 6c2b374d7485 ("PCI-Express
>>      AER implemetation: AER core and aerdriver"), which doesn't
>>      elaborate.
>>
>>    - Should we also test for OS ownership of AER before touching
>>      PCI_ERR_ROOT_STATUS?
>>
>>    - If we remove the PCI_ERR_ROOT_COMMAND fiddling (and I tentatively
>>      think we *should* unless we can justify it), that would also
>>      deserve its own patch.  Possibly (1) remove PCI_ERR_ROOT_COMMAND
>>      fiddling, (2) reorder PCI_ERR_ROOT_STATUS clearing and reset, (3)
>>      test for OS ownership of AER (?), (4) the rest of this patch.
>>
>>>   	return rc ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
>>>   }
>>> diff --git a/drivers/pci/pcie/err.c b/drivers/pci/pcie/err.c
>>> index 7883c9791562..cbc5abfe767b 100644
>>> --- a/drivers/pci/pcie/err.c
>>> +++ b/drivers/pci/pcie/err.c
>>> @@ -148,10 +148,10 @@ static int report_resume(struct pci_dev *dev, void *data)
>>>   
>>>   /**
>>>    * pci_walk_bridge - walk bridges potentially AER affected
>>> - * @bridge:	bridge which may be a Port, an RCEC with associated RCiEPs,
>>> - *		or an RCiEP associated with an RCEC
>>> - * @cb:		callback to be called for each device found
>>> - * @userdata:	arbitrary pointer to be passed to callback
>>> + * @bridge   bridge which may be an RCEC with associated RCiEPs,
>>> + *           or a Port.
>>> + * @cb       callback to be called for each device found
>>> + * @userdata arbitrary pointer to be passed to callback.
>>>    *
>>>    * If the device provided is a bridge, walk the subordinate bus, including
>>>    * any bridged devices on buses under this bus.  Call the provided callback
>>> @@ -164,8 +164,14 @@ static void pci_walk_bridge(struct pci_dev *bridge,
>>>   			    int (*cb)(struct pci_dev *, void *),
>>>   			    void *userdata)
>>>   {
>>> +	/*
>>> +	 * In a non-native case where there is no OS-visible reporting
>>> +	 * device the bridge will be NULL, i.e., no RCEC, no Downstream Port.
>>> +	 */
>>>   	if (bridge->subordinate)
>>>   		pci_walk_bus(bridge->subordinate, cb, userdata);
>>> +	else if (bridge->rcec)
>>> +		cb(bridge->rcec, userdata);
>>>   	else
>>>   		cb(bridge, userdata);
>>>   }
>>> @@ -194,12 +200,6 @@ pci_ers_result_t pcie_do_recovery(struct pci_dev *dev,
>>>   	pci_dbg(bridge, "broadcast error_detected message\n");
>>>   	if (state == pci_channel_io_frozen) {
>>>   		pci_walk_bridge(bridge, report_frozen_detected, &status);
>>> -		if (type == PCI_EXP_TYPE_RC_END) {
>>> -			pci_warn(dev, "subordinate device reset not possible for RCiEP\n");
>>> -			status = PCI_ERS_RESULT_NONE;
>>> -			goto failed;
>>> -		}
>>> -
>>>   		status = reset_subordinates(bridge);
>>>   		if (status != PCI_ERS_RESULT_RECOVERED) {
>>>   			pci_warn(bridge, "subordinate device reset failed\n");
>>> -- 
>>> 2.28.0
>>>

Kuppuswamy Sathyanarayanan Oct. 17, 2020, 1:45 a.m. UTC | #4

On 10/16/20 3:29 PM, Bjorn Helgaas wrote:
> [+cc Christoph, Ethan, Sinan, Keith; sorry should have cc'd you to
> begin with since you're looking at this code too.  Particularly
> interested in your thoughts about whether we should be touching
> PCI_ERR_ROOT_COMMAND and PCI_ERR_ROOT_STATUS when we don't own AER.]
This part is not very clear in ACPI spec or PCI firmware spec.

IMO, since AEPI notifies the OS about the error, then I guess
we are allowed to clear the PCI_ERR_ROOT_STATUS register
after handling the error (similar to EDR case).
>
> On Fri, Oct 16, 2020 at 03:30:37PM -0500, Bjorn Helgaas wrote:
>> [+to Jonathan]
>>
>> On Thu, Oct 15, 2020 at 05:11:10PM -0700, Sean V Kelley wrote:
>>> From: Qiuxu Zhuo <[e-mail]>
>>>
>>> When attempting error recovery for an RCiEP associated with an RCEC device,
>>> there needs to be a way to update the Root Error Status, the Uncorrectable
>>> Error Status and the Uncorrectable Error Severity of the parent RCEC.  In
>>> some non-native cases in which there is no OS-visible device associated
>>> with the RCiEP, there is nothing to act upon as the firmware is acting
>>> before the OS.
>>>
>>> Add handling for the linked RCEC in AER/ERR while taking into account
>>> non-native cases.
>>>
>>> Co-developed-by: Sean V Kelley <[e-mail]>
>>> Link: https://lore.kernel.org/r/[e-mail]
>>> Signed-off-by: Sean V Kelley <[e-mail]>
>>> Signed-off-by: Qiuxu Zhuo <[e-mail]>
>>> Signed-off-by: Bjorn Helgaas <[e-mail]>
>>> Reviewed-by: Jonathan Cameron <[e-mail]>
>>> ---
>>>   drivers/pci/pcie/aer.c | 53 ++++++++++++++++++++++++++++++------------
>>>   drivers/pci/pcie/err.c | 20 ++++++++--------
>>>   2 files changed, 48 insertions(+), 25 deletions(-)
>>>
>>> diff --git a/drivers/pci/pcie/aer.c b/drivers/pci/pcie/aer.c
>>> index 65dff5f3457a..083f69b67bfd 100644
>>> --- a/drivers/pci/pcie/aer.c
>>> +++ b/drivers/pci/pcie/aer.c
>>> @@ -1357,27 +1357,50 @@ static int aer_probe(struct pcie_device *dev)
>>>    */
>>>   static pci_ers_result_t aer_root_reset(struct pci_dev *dev)
>>>   {
>>> -	int aer = dev->aer_cap;
>>> +	int type = pci_pcie_type(dev);
>>> +	struct pci_dev *root;
>>> +	int aer = 0;
>>> +	int rc = 0;
>>>   	u32 reg32;
>>> -	int rc;
>>>   
>>> +	if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
>> "type == PCI_EXP_TYPE_RC_END"
>>
>>> +		/*
>>> +		 * The reset should only clear the Root Error Status
>>> +		 * of the RCEC. Only perform this for the
>>> +		 * native case, i.e., an RCEC is present.
>>> +		 */
>>> +		root = dev->rcec;
>>> +	else
>>> +		root = dev;
>>>   
>>> -	/* Disable Root's interrupt in response to error messages */
>>> -	pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, &reg32);
>>> -	reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
>>> -	pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, reg32);
>>> +	if (root)
>>> +		aer = dev->aer_cap;
>>>   
>>> -	rc = pci_bus_error_reset(dev);
>>> -	pci_info(dev, "Root Port link has been reset\n");
>>> +	if (aer) {
>>> +		/* Disable Root's interrupt in response to error messages */
>>> +		pci_read_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, &reg32);
>>> +		reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
>>> +		pci_write_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, reg32);
>> Not directly related to *this* patch, but my assumption was that in
>> the APEI case, the firmware should retain ownership of the AER
>> Capability, so the OS should not touch PCI_ERR_ROOT_COMMAND and
>> PCI_ERR_ROOT_STATUS.
>>
>> But this code appears to ignore that ownership.  Jonathan, you must
>> have looked at this recently for 068c29a248b6 ("PCI/ERR: Clear PCIe
>> Device Status errors only if OS owns AER").  Do you have any insight
>> about this?
>>
>>> -	/* Clear Root Error Status */
>>> -	pci_read_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, &reg32);
>>> -	pci_write_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, reg32);
>>> +		/* Clear Root Error Status */
>>> +		pci_read_config_dword(root, aer + PCI_ERR_ROOT_STATUS, &reg32);
>>> +		pci_write_config_dword(root, aer + PCI_ERR_ROOT_STATUS, reg32);
>>>   
>>> -	/* Enable Root Port's interrupt in response to error messages */
>>> -	pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, &reg32);
>>> -	reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
>>> -	pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, reg32);
>>> +		/* Enable Root Port's interrupt in response to error messages */
>>> +		pci_read_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, &reg32);
>>> +		reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
>>> +		pci_write_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, reg32);
>>> +	}
>>> +
>>> +	if ((type == PCI_EXP_TYPE_RC_EC) || (type == PCI_EXP_TYPE_RC_END)) {
>>> +		if (pcie_has_flr(root)) {
>>> +			rc = pcie_flr(root);
>>> +			pci_info(dev, "has been reset (%d)\n", rc);
>>> +		}
>>> +	} else {
>>> +		rc = pci_bus_error_reset(root);
>> Don't we want "dev" for both the FLR and pci_bus_error_reset()?  I
>> think "root == dev" except when dev is an RCiEP.  When dev is an
>> RCiEP, "root" is the RCEC (if present), and we want to reset the
>> RCiEP, not the RCEC.
>>
>>> +		pci_info(dev, "Root Port link has been reset (%d)\n", rc);
>>> +	}
>> There are a couple changes here that I think should be split out.
>>
>> Based on my theory that when firmware retains control of AER, the OS
>> should not touch PCI_ERR_ROOT_COMMAND and PCI_ERR_ROOT_STATUS, and any
>> updates to them would have to be done by firmware before we get here,
>> I suggested reordering this:
>>
>>    - clear PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>>    - do reset
>>    - clear PCI_ERR_ROOT_STATUS (for APEI, presumably done by firmware?)
>>    - enable PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>>
>> to this:
>>
>>    - clear PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>>    - clear PCI_ERR_ROOT_STATUS
>>    - enable PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>>    - do reset
>>
>> If my theory is correct, I think we should still reorder this, but:
>>
>>    - It's a significant behavior change that deserves its own patch so
>>      we can document/bisect/revert.
>>
>>    - I'm not sure why we clear the PCI_ERR_ROOT_COMMAND error reporting
>>      bits.  In the new "clear COMMAND, clear STATUS, enable COMMAND"
>>      order, it looks superfluous.  There's no reason to disable error
>>      reporting while clearing the status bits.
>>
>>      The current "clear, reset, enable" order suggests that the reset
>>      might cause errors that we should ignore.  I don't know whether
>>      that's the case or not.  It dates from 6c2b374d7485 ("PCI-Express
>>      AER implemetation: AER core and aerdriver"), which doesn't
>>      elaborate.
>>
>>    - Should we also test for OS ownership of AER before touching
>>      PCI_ERR_ROOT_STATUS?
>>
>>    - If we remove the PCI_ERR_ROOT_COMMAND fiddling (and I tentatively
>>      think we *should* unless we can justify it), that would also
>>      deserve its own patch.  Possibly (1) remove PCI_ERR_ROOT_COMMAND
>>      fiddling, (2) reorder PCI_ERR_ROOT_STATUS clearing and reset, (3)
>>      test for OS ownership of AER (?), (4) the rest of this patch.
>>
>>>   	return rc ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
>>>   }
>>> diff --git a/drivers/pci/pcie/err.c b/drivers/pci/pcie/err.c
>>> index 7883c9791562..cbc5abfe767b 100644
>>> --- a/drivers/pci/pcie/err.c
>>> +++ b/drivers/pci/pcie/err.c
>>> @@ -148,10 +148,10 @@ static int report_resume(struct pci_dev *dev, void *data)
>>>   
>>>   /**
>>>    * pci_walk_bridge - walk bridges potentially AER affected
>>> - * @bridge:	bridge which may be a Port, an RCEC with associated RCiEPs,
>>> - *		or an RCiEP associated with an RCEC
>>> - * @cb:		callback to be called for each device found
>>> - * @userdata:	arbitrary pointer to be passed to callback
>>> + * @bridge   bridge which may be an RCEC with associated RCiEPs,
>>> + *           or a Port.
>>> + * @cb       callback to be called for each device found
>>> + * @userdata arbitrary pointer to be passed to callback.
>>>    *
>>>    * If the device provided is a bridge, walk the subordinate bus, including
>>>    * any bridged devices on buses under this bus.  Call the provided callback
>>> @@ -164,8 +164,14 @@ static void pci_walk_bridge(struct pci_dev *bridge,
>>>   			    int (*cb)(struct pci_dev *, void *),
>>>   			    void *userdata)
>>>   {
>>> +	/*
>>> +	 * In a non-native case where there is no OS-visible reporting
>>> +	 * device the bridge will be NULL, i.e., no RCEC, no Downstream Port.
>>> +	 */
>>>   	if (bridge->subordinate)
>>>   		pci_walk_bus(bridge->subordinate, cb, userdata);
>>> +	else if (bridge->rcec)
>>> +		cb(bridge->rcec, userdata);
>>>   	else
>>>   		cb(bridge, userdata);
>>>   }
>>> @@ -194,12 +200,6 @@ pci_ers_result_t pcie_do_recovery(struct pci_dev *dev,
>>>   	pci_dbg(bridge, "broadcast error_detected message\n");
>>>   	if (state == pci_channel_io_frozen) {
>>>   		pci_walk_bridge(bridge, report_frozen_detected, &status);
>>> -		if (type == PCI_EXP_TYPE_RC_END) {
>>> -			pci_warn(dev, "subordinate device reset not possible for RCiEP\n");
>>> -			status = PCI_ERS_RESULT_NONE;
>>> -			goto failed;
>>> -		}
>>> -
>>>   		status = reset_subordinates(bridge);
>>>   		if (status != PCI_ERS_RESULT_RECOVERED) {
>>>   			pci_warn(bridge, "subordinate device reset failed\n");
>>> -- 
>>> 2.28.0
>>>

Kelley, Sean V Oct. 17, 2020, 4:14 p.m. UTC | #5

On 16 Oct 2020, at 13:30, Bjorn Helgaas wrote:

> [+to Jonathan]
>
> On Thu, Oct 15, 2020 at 05:11:10PM -0700, Sean V Kelley wrote:
>> From: Qiuxu Zhuo <[e-mail]>
>>
>> When attempting error recovery for an RCiEP associated with an RCEC 
>> device,
>> there needs to be a way to update the Root Error Status, the 
>> Uncorrectable
>> Error Status and the Uncorrectable Error Severity of the parent RCEC. 
>>  In
>> some non-native cases in which there is no OS-visible device 
>> associated
>> with the RCiEP, there is nothing to act upon as the firmware is 
>> acting
>> before the OS.
>>
>> Add handling for the linked RCEC in AER/ERR while taking into account
>> non-native cases.
>>
>> Co-developed-by: Sean V Kelley <[e-mail]>
>> Link: 
>> https://lore.kernel.org/r/[e-mail]
>> Signed-off-by: Sean V Kelley <[e-mail]>
>> Signed-off-by: Qiuxu Zhuo <[e-mail]>
>> Signed-off-by: Bjorn Helgaas <[e-mail]>
>> Reviewed-by: Jonathan Cameron <[e-mail]>
>> ---
>>  drivers/pci/pcie/aer.c | 53 
>> ++++++++++++++++++++++++++++++------------
>>  drivers/pci/pcie/err.c | 20 ++++++++--------
>>  2 files changed, 48 insertions(+), 25 deletions(-)
>>
>> diff --git a/drivers/pci/pcie/aer.c b/drivers/pci/pcie/aer.c
>> index 65dff5f3457a..083f69b67bfd 100644
>> --- a/drivers/pci/pcie/aer.c
>> +++ b/drivers/pci/pcie/aer.c
>> @@ -1357,27 +1357,50 @@ static int aer_probe(struct pcie_device *dev)
>>   */
>>  static pci_ers_result_t aer_root_reset(struct pci_dev *dev)
>>  {
>> -	int aer = dev->aer_cap;
>> +	int type = pci_pcie_type(dev);
>> +	struct pci_dev *root;
>> +	int aer = 0;
>> +	int rc = 0;
>>  	u32 reg32;
>> -	int rc;
>>
>> +	if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
>
> "type == PCI_EXP_TYPE_RC_END"

Right, I merged your suggested changes which added the type. Will 
correct.

>
>> +		/*
>> +		 * The reset should only clear the Root Error Status
>> +		 * of the RCEC. Only perform this for the
>> +		 * native case, i.e., an RCEC is present.
>> +		 */
>> +		root = dev->rcec;
>> +	else
>> +		root = dev;
>>
>> -	/* Disable Root's interrupt in response to error messages */
>> -	pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, &reg32);
>> -	reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
>> -	pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, reg32);
>> +	if (root)
>> +		aer = dev->aer_cap;
>>
>> -	rc = pci_bus_error_reset(dev);
>> -	pci_info(dev, "Root Port link has been reset\n");
>> +	if (aer) {
>> +		/* Disable Root's interrupt in response to error messages */
>> +		pci_read_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, &reg32);
>> +		reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
>> +		pci_write_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, reg32);
>
> Not directly related to *this* patch, but my assumption was that in
> the APEI case, the firmware should retain ownership of the AER
> Capability, so the OS should not touch PCI_ERR_ROOT_COMMAND and
> PCI_ERR_ROOT_STATUS.
>
> But this code appears to ignore that ownership.  Jonathan, you must
> have looked at this recently for 068c29a248b6 ("PCI/ERR: Clear PCIe
> Device Status errors only if OS owns AER").  Do you have any insight
> about this?
>
>> -	/* Clear Root Error Status */
>> -	pci_read_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, &reg32);
>> -	pci_write_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, reg32);
>> +		/* Clear Root Error Status */
>> +		pci_read_config_dword(root, aer + PCI_ERR_ROOT_STATUS, &reg32);
>> +		pci_write_config_dword(root, aer + PCI_ERR_ROOT_STATUS, reg32);
>>
>> -	/* Enable Root Port's interrupt in response to error messages */
>> -	pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, &reg32);
>> -	reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
>> -	pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, reg32);
>> +		/* Enable Root Port's interrupt in response to error messages */
>> +		pci_read_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, &reg32);
>> +		reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
>> +		pci_write_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, reg32);
>> +	}
>> +
>> +	if ((type == PCI_EXP_TYPE_RC_EC) || (type == PCI_EXP_TYPE_RC_END)) 
>> {
>> +		if (pcie_has_flr(root)) {
>> +			rc = pcie_flr(root);
>> +			pci_info(dev, "has been reset (%d)\n", rc);
>> +		}
>> +	} else {
>> +		rc = pci_bus_error_reset(root);
>
> Don't we want "dev" for both the FLR and pci_bus_error_reset()?  I
> think "root == dev" except when dev is an RCiEP.  When dev is an
> RCiEP, "root" is the RCEC (if present), and we want to reset the
> RCiEP, not the RCEC.

Right, when I did the goto in the earlier incarnation, I always set root 
to dev at the start and in the merge it needs to be dev always except 
for the RC_END where RCEC exists. Will change without bringing back the 
goto…

+       struct pci_dev *root = dev;

…

+non_native:
+       if ((type == PCI_EXP_TYPE_RC_EC) || (type ==
PCI_EXP_TYPE_RC_END)) {
+               rc = flr_on_rc(root);
+               pci_info(dev, "has been reset (%d)\n", rc);
+       } else {
+               rc = pci_bus_error_reset(root);
+               pci_info(dev, "Root Port link has been reset (%d)\n",
rc);
+       }


>
>> +		pci_info(dev, "Root Port link has been reset (%d)\n", rc);
>> +	}
>
> There are a couple changes here that I think should be split out.
>
> Based on my theory that when firmware retains control of AER, the OS
> should not touch PCI_ERR_ROOT_COMMAND and PCI_ERR_ROOT_STATUS, and any
> updates to them would have to be done by firmware before we get here,
> I suggested reordering this:
>
>   - clear PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>   - do reset
>   - clear PCI_ERR_ROOT_STATUS (for APEI, presumably done by firmware?)
>   - enable PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>
> to this:
>
>   - clear PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>   - clear PCI_ERR_ROOT_STATUS
>   - enable PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>   - do reset
>
> If my theory is correct, I think we should still reorder this, but:
>
>   - It's a significant behavior change that deserves its own patch so
>     we can document/bisect/revert.
>
>   - I'm not sure why we clear the PCI_ERR_ROOT_COMMAND error reporting
>     bits.  In the new "clear COMMAND, clear STATUS, enable COMMAND"
>     order, it looks superfluous.  There's no reason to disable error
>     reporting while clearing the status bits.
>
>     The current "clear, reset, enable" order suggests that the reset
>     might cause errors that we should ignore.  I don't know whether
>     that's the case or not.  It dates from 6c2b374d7485 ("PCI-Express
>     AER implemetation: AER core and aerdriver"), which doesn't
>     elaborate.
>
>   - Should we also test for OS ownership of AER before touching
>     PCI_ERR_ROOT_STATUS?
>
>   - If we remove the PCI_ERR_ROOT_COMMAND fiddling (and I tentatively
>     think we *should* unless we can justify it), that would also
>     deserve its own patch.  Possibly (1) remove PCI_ERR_ROOT_COMMAND
>     fiddling, (2) reorder PCI_ERR_ROOT_STATUS clearing and reset, (3)
>     test for OS ownership of AER (?), (4) the rest of this patch.

You’ve highlighted some good questions.

I think we should remove the fiddling until we have a clearer picture 
and put that into its own patch.

Sean
>
>>  	return rc ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
>>  }
>> diff --git a/drivers/pci/pcie/err.c b/drivers/pci/pcie/err.c
>> index 7883c9791562..cbc5abfe767b 100644
>> --- a/drivers/pci/pcie/err.c
>> +++ b/drivers/pci/pcie/err.c
>> @@ -148,10 +148,10 @@ static int report_resume(struct pci_dev *dev, 
>> void *data)
>>
>>  /**
>>   * pci_walk_bridge - walk bridges potentially AER affected
>> - * @bridge:	bridge which may be a Port, an RCEC with associated 
>> RCiEPs,
>> - *		or an RCiEP associated with an RCEC
>> - * @cb:		callback to be called for each device found
>> - * @userdata:	arbitrary pointer to be passed to callback
>> + * @bridge   bridge which may be an RCEC with associated RCiEPs,
>> + *           or a Port.
>> + * @cb       callback to be called for each device found
>> + * @userdata arbitrary pointer to be passed to callback.
>>   *
>>   * If the device provided is a bridge, walk the subordinate bus, 
>> including
>>   * any bridged devices on buses under this bus.  Call the provided 
>> callback
>> @@ -164,8 +164,14 @@ static void pci_walk_bridge(struct pci_dev 
>> *bridge,
>>  			    int (*cb)(struct pci_dev *, void *),
>>  			    void *userdata)
>>  {
>> +	/*
>> +	 * In a non-native case where there is no OS-visible reporting
>> +	 * device the bridge will be NULL, i.e., no RCEC, no Downstream 
>> Port.
>> +	 */
>>  	if (bridge->subordinate)
>>  		pci_walk_bus(bridge->subordinate, cb, userdata);
>> +	else if (bridge->rcec)
>> +		cb(bridge->rcec, userdata);
>>  	else
>>  		cb(bridge, userdata);
>>  }
>> @@ -194,12 +200,6 @@ pci_ers_result_t pcie_do_recovery(struct pci_dev 
>> *dev,
>>  	pci_dbg(bridge, "broadcast error_detected message\n");
>>  	if (state == pci_channel_io_frozen) {
>>  		pci_walk_bridge(bridge, report_frozen_detected, &status);
>> -		if (type == PCI_EXP_TYPE_RC_END) {
>> -			pci_warn(dev, "subordinate device reset not possible for 
>> RCiEP\n");
>> -			status = PCI_ERS_RESULT_NONE;
>> -			goto failed;
>> -		}
>> -
>>  		status = reset_subordinates(bridge);
>>  		if (status != PCI_ERS_RESULT_RECOVERED) {
>>  			pci_warn(bridge, "subordinate device reset failed\n");
>> -- 
>> 2.28.0
>>

Ethan Zhao Oct. 19, 2020, 10:49 a.m. UTC | #6

On Sat, Oct 17, 2020 at 6:29 AM Bjorn Helgaas <[e-mail]> wrote:
>
> [+cc Christoph, Ethan, Sinan, Keith; sorry should have cc'd you to
> begin with since you're looking at this code too.  Particularly
> interested in your thoughts about whether we should be touching
> PCI_ERR_ROOT_COMMAND and PCI_ERR_ROOT_STATUS when we don't own AER.]

aer_root_reset() function has a prefix  'aer_', looks like it's a
function of aer driver, will
only be called by aer driver at runtime. if so it's up to the
owner/aer to know if OSPM is
granted to init. while actually some of the functions and runtime service of
aer driver is also shared by GHES driver (running time) and DPC driver
(compiling time ?)
etc. then it is confused now.

Shall we move some of the shared functions and running time service to
pci/err.c ?
if so , just like pcie_do_recovery(), it's share by firmware_first  mode GHES
ghes_probe()
->ghes_irq_func
  ->ghes_proc
    ->ghes_do_proc()
      ->ghes_handle_aer()
        ->aer_recover_work_func()
          ->pcie_do_recovery()
            ->aer_root_reset()

and aer driver etc.  if aer wants to do some access might conflict
with firmware(or
firmware in embedded controller) should check _OSC_ etc first.  blindly issue
PCI_ERR_ROOT_COMMAND  or clear PCI_ERR_ROOT_STATUS *likely*
cause errors by error handling itself.

Thanks,
Ethan

>
> On Fri, Oct 16, 2020 at 03:30:37PM -0500, Bjorn Helgaas wrote:
> > [+to Jonathan]
> >
> > On Thu, Oct 15, 2020 at 05:11:10PM -0700, Sean V Kelley wrote:
> > > From: Qiuxu Zhuo <[e-mail]>
> > >
> > > When attempting error recovery for an RCiEP associated with an RCEC device,
> > > there needs to be a way to update the Root Error Status, the Uncorrectable
> > > Error Status and the Uncorrectable Error Severity of the parent RCEC.  In
> > > some non-native cases in which there is no OS-visible device associated
> > > with the RCiEP, there is nothing to act upon as the firmware is acting
> > > before the OS.
> > >
> > > Add handling for the linked RCEC in AER/ERR while taking into account
> > > non-native cases.
> > >
> > > Co-developed-by: Sean V Kelley <[e-mail]>
> > > Link: https://lore.kernel.org/r/[e-mail]
> > > Signed-off-by: Sean V Kelley <[e-mail]>
> > > Signed-off-by: Qiuxu Zhuo <[e-mail]>
> > > Signed-off-by: Bjorn Helgaas <[e-mail]>
> > > Reviewed-by: Jonathan Cameron <[e-mail]>
> > > ---
> > >  drivers/pci/pcie/aer.c | 53 ++++++++++++++++++++++++++++++------------
> > >  drivers/pci/pcie/err.c | 20 ++++++++--------
> > >  2 files changed, 48 insertions(+), 25 deletions(-)
> > >
> > > diff --git a/drivers/pci/pcie/aer.c b/drivers/pci/pcie/aer.c
> > > index 65dff5f3457a..083f69b67bfd 100644
> > > --- a/drivers/pci/pcie/aer.c
> > > +++ b/drivers/pci/pcie/aer.c
> > > @@ -1357,27 +1357,50 @@ static int aer_probe(struct pcie_device *dev)
> > >   */
> > >  static pci_ers_result_t aer_root_reset(struct pci_dev *dev)
> > >  {
> > > -   int aer = dev->aer_cap;
> > > +   int type = pci_pcie_type(dev);
> > > +   struct pci_dev *root;
> > > +   int aer = 0;
> > > +   int rc = 0;
> > >     u32 reg32;
> > > -   int rc;
> > >
> > > +   if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
> >
> > "type == PCI_EXP_TYPE_RC_END"
> >
> > > +           /*
> > > +            * The reset should only clear the Root Error Status
> > > +            * of the RCEC. Only perform this for the
> > > +            * native case, i.e., an RCEC is present.
> > > +            */
> > > +           root = dev->rcec;
> > > +   else
> > > +           root = dev;
> > >
> > > -   /* Disable Root's interrupt in response to error messages */
> > > -   pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, &reg32);
> > > -   reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
> > > -   pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, reg32);
> > > +   if (root)
> > > +           aer = dev->aer_cap;
> > >
> > > -   rc = pci_bus_error_reset(dev);
> > > -   pci_info(dev, "Root Port link has been reset\n");
> > > +   if (aer) {
> > > +           /* Disable Root's interrupt in response to error messages */
> > > +           pci_read_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, &reg32);
> > > +           reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
> > > +           pci_write_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, reg32);
> >
> > Not directly related to *this* patch, but my assumption was that in
> > the APEI case, the firmware should retain ownership of the AER
> > Capability, so the OS should not touch PCI_ERR_ROOT_COMMAND and
> > PCI_ERR_ROOT_STATUS.
> >
> > But this code appears to ignore that ownership.  Jonathan, you must
> > have looked at this recently for 068c29a248b6 ("PCI/ERR: Clear PCIe
> > Device Status errors only if OS owns AER").  Do you have any insight
> > about this?
> >
> > > -   /* Clear Root Error Status */
> > > -   pci_read_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, &reg32);
> > > -   pci_write_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, reg32);
> > > +           /* Clear Root Error Status */
> > > +           pci_read_config_dword(root, aer + PCI_ERR_ROOT_STATUS, &reg32);
> > > +           pci_write_config_dword(root, aer + PCI_ERR_ROOT_STATUS, reg32);
> > >
> > > -   /* Enable Root Port's interrupt in response to error messages */
> > > -   pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, &reg32);
> > > -   reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
> > > -   pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, reg32);
> > > +           /* Enable Root Port's interrupt in response to error messages */
> > > +           pci_read_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, &reg32);
> > > +           reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
> > > +           pci_write_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, reg32);
> > > +   }
> > > +
> > > +   if ((type == PCI_EXP_TYPE_RC_EC) || (type == PCI_EXP_TYPE_RC_END)) {
> > > +           if (pcie_has_flr(root)) {
> > > +                   rc = pcie_flr(root);
> > > +                   pci_info(dev, "has been reset (%d)\n", rc);
> > > +           }
> > > +   } else {
> > > +           rc = pci_bus_error_reset(root);
> >
> > Don't we want "dev" for both the FLR and pci_bus_error_reset()?  I
> > think "root == dev" except when dev is an RCiEP.  When dev is an
> > RCiEP, "root" is the RCEC (if present), and we want to reset the
> > RCiEP, not the RCEC.
> >
> > > +           pci_info(dev, "Root Port link has been reset (%d)\n", rc);
> > > +   }
> >
> > There are a couple changes here that I think should be split out.
> >
> > Based on my theory that when firmware retains control of AER, the OS
> > should not touch PCI_ERR_ROOT_COMMAND and PCI_ERR_ROOT_STATUS, and any
> > updates to them would have to be done by firmware before we get here,
> > I suggested reordering this:
> >
> >   - clear PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
> >   - do reset
> >   - clear PCI_ERR_ROOT_STATUS (for APEI, presumably done by firmware?)
> >   - enable PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
> >
> > to this:
> >
> >   - clear PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
> >   - clear PCI_ERR_ROOT_STATUS
> >   - enable PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
> >   - do reset
> >
> > If my theory is correct, I think we should still reorder this, but:
> >
> >   - It's a significant behavior change that deserves its own patch so
> >     we can document/bisect/revert.
> >
> >   - I'm not sure why we clear the PCI_ERR_ROOT_COMMAND error reporting
> >     bits.  In the new "clear COMMAND, clear STATUS, enable COMMAND"
> >     order, it looks superfluous.  There's no reason to disable error
> >     reporting while clearing the status bits.
> >
> >     The current "clear, reset, enable" order suggests that the reset
> >     might cause errors that we should ignore.  I don't know whether
> >     that's the case or not.  It dates from 6c2b374d7485 ("PCI-Express
> >     AER implemetation: AER core and aerdriver"), which doesn't
> >     elaborate.
> >
> >   - Should we also test for OS ownership of AER before touching
> >     PCI_ERR_ROOT_STATUS?
> >
> >   - If we remove the PCI_ERR_ROOT_COMMAND fiddling (and I tentatively
> >     think we *should* unless we can justify it), that would also
> >     deserve its own patch.  Possibly (1) remove PCI_ERR_ROOT_COMMAND
> >     fiddling, (2) reorder PCI_ERR_ROOT_STATUS clearing and reset, (3)
> >     test for OS ownership of AER (?), (4) the rest of this patch.
> >
> > >     return rc ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
> > >  }
> > > diff --git a/drivers/pci/pcie/err.c b/drivers/pci/pcie/err.c
> > > index 7883c9791562..cbc5abfe767b 100644
> > > --- a/drivers/pci/pcie/err.c
> > > +++ b/drivers/pci/pcie/err.c
> > > @@ -148,10 +148,10 @@ static int report_resume(struct pci_dev *dev, void *data)
> > >
> > >  /**
> > >   * pci_walk_bridge - walk bridges potentially AER affected
> > > - * @bridge:        bridge which may be a Port, an RCEC with associated RCiEPs,
> > > - *         or an RCiEP associated with an RCEC
> > > - * @cb:            callback to be called for each device found
> > > - * @userdata:      arbitrary pointer to be passed to callback
> > > + * @bridge   bridge which may be an RCEC with associated RCiEPs,
> > > + *           or a Port.
> > > + * @cb       callback to be called for each device found
> > > + * @userdata arbitrary pointer to be passed to callback.
> > >   *
> > >   * If the device provided is a bridge, walk the subordinate bus, including
> > >   * any bridged devices on buses under this bus.  Call the provided callback
> > > @@ -164,8 +164,14 @@ static void pci_walk_bridge(struct pci_dev *bridge,
> > >                         int (*cb)(struct pci_dev *, void *),
> > >                         void *userdata)
> > >  {
> > > +   /*
> > > +    * In a non-native case where there is no OS-visible reporting
> > > +    * device the bridge will be NULL, i.e., no RCEC, no Downstream Port.
> > > +    */
> > >     if (bridge->subordinate)
> > >             pci_walk_bus(bridge->subordinate, cb, userdata);
> > > +   else if (bridge->rcec)
> > > +           cb(bridge->rcec, userdata);
> > >     else
> > >             cb(bridge, userdata);
> > >  }
> > > @@ -194,12 +200,6 @@ pci_ers_result_t pcie_do_recovery(struct pci_dev *dev,
> > >     pci_dbg(bridge, "broadcast error_detected message\n");
> > >     if (state == pci_channel_io_frozen) {
> > >             pci_walk_bridge(bridge, report_frozen_detected, &status);
> > > -           if (type == PCI_EXP_TYPE_RC_END) {
> > > -                   pci_warn(dev, "subordinate device reset not possible for RCiEP\n");
> > > -                   status = PCI_ERS_RESULT_NONE;
> > > -                   goto failed;
> > > -           }
> > > -
> > >             status = reset_subordinates(bridge);
> > >             if (status != PCI_ERS_RESULT_RECOVERED) {
> > >                     pci_warn(bridge, "subordinate device reset failed\n");
> > > --
> > > 2.28.0
> > >

Kelley, Sean V Oct. 19, 2020, 6:31 p.m. UTC | #7

On 19 Oct 2020, at 3:49, Ethan Zhao wrote:

> On Sat, Oct 17, 2020 at 6:29 AM Bjorn Helgaas <[e-mail]> 
> wrote:
>>
>> [+cc Christoph, Ethan, Sinan, Keith; sorry should have cc'd you to
>> begin with since you're looking at this code too.  Particularly
>> interested in your thoughts about whether we should be touching
>> PCI_ERR_ROOT_COMMAND and PCI_ERR_ROOT_STATUS when we don't own AER.]
>
> aer_root_reset() function has a prefix  'aer_', looks like it's a
> function of aer driver, will
> only be called by aer driver at runtime. if so it's up to the
> owner/aer to know if OSPM is
> granted to init. while actually some of the functions and runtime 
> service of
> aer driver is also shared by GHES driver (running time) and DPC driver
> (compiling time ?)
> etc. then it is confused now.
>
> Shall we move some of the shared functions and running time service to
> pci/err.c ?
> if so , just like pcie_do_recovery(), it's share by firmware_first  
> mode GHES
> ghes_probe()
> ->ghes_irq_func
>   ->ghes_proc
>     ->ghes_do_proc()
>       ->ghes_handle_aer()
>         ->aer_recover_work_func()
>           ->pcie_do_recovery()
>             ->aer_root_reset()
>
> and aer driver etc.  if aer wants to do some access might conflict
> with firmware(or
> firmware in embedded controller) should check _OSC_ etc first.  
> blindly issue
> PCI_ERR_ROOT_COMMAND  or clear PCI_ERR_ROOT_STATUS *likely*
> cause errors by error handling itself.

If _OSC negotiation ends up with FW being in control of AER, that means 
OS is not in charge and should not be messing with AER I guess. That 
seems appropriate to me then.

Thanks,

Sean



>
> Thanks,
> Ethan
>
>>
>> On Fri, Oct 16, 2020 at 03:30:37PM -0500, Bjorn Helgaas wrote:
>>> [+to Jonathan]
>>>
>>> On Thu, Oct 15, 2020 at 05:11:10PM -0700, Sean V Kelley wrote:
>>>> From: Qiuxu Zhuo <[e-mail]>
>>>>
>>>> When attempting error recovery for an RCiEP associated with an RCEC 
>>>> device,
>>>> there needs to be a way to update the Root Error Status, the 
>>>> Uncorrectable
>>>> Error Status and the Uncorrectable Error Severity of the parent 
>>>> RCEC.  In
>>>> some non-native cases in which there is no OS-visible device 
>>>> associated
>>>> with the RCiEP, there is nothing to act upon as the firmware is 
>>>> acting
>>>> before the OS.
>>>>
>>>> Add handling for the linked RCEC in AER/ERR while taking into 
>>>> account
>>>> non-native cases.
>>>>
>>>> Co-developed-by: Sean V Kelley <[e-mail]>
>>>> Link: 
>>>> https://lore.kernel.org/r/[e-mail]
>>>> Signed-off-by: Sean V Kelley <[e-mail]>
>>>> Signed-off-by: Qiuxu Zhuo <[e-mail]>
>>>> Signed-off-by: Bjorn Helgaas <[e-mail]>
>>>> Reviewed-by: Jonathan Cameron <[e-mail]>
>>>> ---
>>>>  drivers/pci/pcie/aer.c | 53 
>>>> ++++++++++++++++++++++++++++++------------
>>>>  drivers/pci/pcie/err.c | 20 ++++++++--------
>>>>  2 files changed, 48 insertions(+), 25 deletions(-)
>>>>
>>>> diff --git a/drivers/pci/pcie/aer.c b/drivers/pci/pcie/aer.c
>>>> index 65dff5f3457a..083f69b67bfd 100644
>>>> --- a/drivers/pci/pcie/aer.c
>>>> +++ b/drivers/pci/pcie/aer.c
>>>> @@ -1357,27 +1357,50 @@ static int aer_probe(struct pcie_device 
>>>> *dev)
>>>>   */
>>>>  static pci_ers_result_t aer_root_reset(struct pci_dev *dev)
>>>>  {
>>>> -   int aer = dev->aer_cap;
>>>> +   int type = pci_pcie_type(dev);
>>>> +   struct pci_dev *root;
>>>> +   int aer = 0;
>>>> +   int rc = 0;
>>>>     u32 reg32;
>>>> -   int rc;
>>>>
>>>> +   if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
>>>
>>> "type == PCI_EXP_TYPE_RC_END"
>>>
>>>> +           /*
>>>> +            * The reset should only clear the Root Error Status
>>>> +            * of the RCEC. Only perform this for the
>>>> +            * native case, i.e., an RCEC is present.
>>>> +            */
>>>> +           root = dev->rcec;
>>>> +   else
>>>> +           root = dev;
>>>>
>>>> -   /* Disable Root's interrupt in response to error messages */
>>>> -   pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, &reg32);
>>>> -   reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
>>>> -   pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, reg32);
>>>> +   if (root)
>>>> +           aer = dev->aer_cap;
>>>>
>>>> -   rc = pci_bus_error_reset(dev);
>>>> -   pci_info(dev, "Root Port link has been reset\n");
>>>> +   if (aer) {
>>>> +           /* Disable Root's interrupt in response to error 
>>>> messages */
>>>> +           pci_read_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, 
>>>> &reg32);
>>>> +           reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
>>>> +           pci_write_config_dword(root, aer + 
>>>> PCI_ERR_ROOT_COMMAND, reg32);
>>>
>>> Not directly related to *this* patch, but my assumption was that in
>>> the APEI case, the firmware should retain ownership of the AER
>>> Capability, so the OS should not touch PCI_ERR_ROOT_COMMAND and
>>> PCI_ERR_ROOT_STATUS.
>>>
>>> But this code appears to ignore that ownership.  Jonathan, you must
>>> have looked at this recently for 068c29a248b6 ("PCI/ERR: Clear PCIe
>>> Device Status errors only if OS owns AER").  Do you have any insight
>>> about this?
>>>
>>>> -   /* Clear Root Error Status */
>>>> -   pci_read_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, &reg32);
>>>> -   pci_write_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, reg32);
>>>> +           /* Clear Root Error Status */
>>>> +           pci_read_config_dword(root, aer + PCI_ERR_ROOT_STATUS, 
>>>> &reg32);
>>>> +           pci_write_config_dword(root, aer + PCI_ERR_ROOT_STATUS, 
>>>> reg32);
>>>>
>>>> -   /* Enable Root Port's interrupt in response to error messages 
>>>> */
>>>> -   pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, &reg32);
>>>> -   reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
>>>> -   pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, reg32);
>>>> +           /* Enable Root Port's interrupt in response to error 
>>>> messages */
>>>> +           pci_read_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, 
>>>> &reg32);
>>>> +           reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
>>>> +           pci_write_config_dword(root, aer + 
>>>> PCI_ERR_ROOT_COMMAND, reg32);
>>>> +   }
>>>> +
>>>> +   if ((type == PCI_EXP_TYPE_RC_EC) || (type == 
>>>> PCI_EXP_TYPE_RC_END)) {
>>>> +           if (pcie_has_flr(root)) {
>>>> +                   rc = pcie_flr(root);
>>>> +                   pci_info(dev, "has been reset (%d)\n", rc);
>>>> +           }
>>>> +   } else {
>>>> +           rc = pci_bus_error_reset(root);
>>>
>>> Don't we want "dev" for both the FLR and pci_bus_error_reset()?  I
>>> think "root == dev" except when dev is an RCiEP.  When dev is an
>>> RCiEP, "root" is the RCEC (if present), and we want to reset the
>>> RCiEP, not the RCEC.
>>>
>>>> +           pci_info(dev, "Root Port link has been reset (%d)\n", 
>>>> rc);
>>>> +   }
>>>
>>> There are a couple changes here that I think should be split out.
>>>
>>> Based on my theory that when firmware retains control of AER, the OS
>>> should not touch PCI_ERR_ROOT_COMMAND and PCI_ERR_ROOT_STATUS, and 
>>> any
>>> updates to them would have to be done by firmware before we get 
>>> here,
>>> I suggested reordering this:
>>>
>>>   - clear PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>>>   - do reset
>>>   - clear PCI_ERR_ROOT_STATUS (for APEI, presumably done by 
>>> firmware?)
>>>   - enable PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>>>
>>> to this:
>>>
>>>   - clear PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>>>   - clear PCI_ERR_ROOT_STATUS
>>>   - enable PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>>>   - do reset
>>>
>>> If my theory is correct, I think we should still reorder this, but:
>>>
>>>   - It's a significant behavior change that deserves its own patch 
>>> so
>>>     we can document/bisect/revert.
>>>
>>>   - I'm not sure why we clear the PCI_ERR_ROOT_COMMAND error 
>>> reporting
>>>     bits.  In the new "clear COMMAND, clear STATUS, enable COMMAND"
>>>     order, it looks superfluous.  There's no reason to disable error
>>>     reporting while clearing the status bits.
>>>
>>>     The current "clear, reset, enable" order suggests that the reset
>>>     might cause errors that we should ignore.  I don't know whether
>>>     that's the case or not.  It dates from 6c2b374d7485 
>>> ("PCI-Express
>>>     AER implemetation: AER core and aerdriver"), which doesn't
>>>     elaborate.
>>>
>>>   - Should we also test for OS ownership of AER before touching
>>>     PCI_ERR_ROOT_STATUS?
>>>
>>>   - If we remove the PCI_ERR_ROOT_COMMAND fiddling (and I 
>>> tentatively
>>>     think we *should* unless we can justify it), that would also
>>>     deserve its own patch.  Possibly (1) remove PCI_ERR_ROOT_COMMAND
>>>     fiddling, (2) reorder PCI_ERR_ROOT_STATUS clearing and reset, 
>>> (3)
>>>     test for OS ownership of AER (?), (4) the rest of this patch.
>>>
>>>>     return rc ? PCI_ERS_RESULT_DISCONNECT : 
>>>> PCI_ERS_RESULT_RECOVERED;
>>>>  }
>>>> diff --git a/drivers/pci/pcie/err.c b/drivers/pci/pcie/err.c
>>>> index 7883c9791562..cbc5abfe767b 100644
>>>> --- a/drivers/pci/pcie/err.c
>>>> +++ b/drivers/pci/pcie/err.c
>>>> @@ -148,10 +148,10 @@ static int report_resume(struct pci_dev *dev, 
>>>> void *data)
>>>>
>>>>  /**
>>>>   * pci_walk_bridge - walk bridges potentially AER affected
>>>> - * @bridge:        bridge which may be a Port, an RCEC with 
>>>> associated RCiEPs,
>>>> - *         or an RCiEP associated with an RCEC
>>>> - * @cb:            callback to be called for each device found
>>>> - * @userdata:      arbitrary pointer to be passed to callback
>>>> + * @bridge   bridge which may be an RCEC with associated RCiEPs,
>>>> + *           or a Port.
>>>> + * @cb       callback to be called for each device found
>>>> + * @userdata arbitrary pointer to be passed to callback.
>>>>   *
>>>>   * If the device provided is a bridge, walk the subordinate bus, 
>>>> including
>>>>   * any bridged devices on buses under this bus.  Call the provided 
>>>> callback
>>>> @@ -164,8 +164,14 @@ static void pci_walk_bridge(struct pci_dev 
>>>> *bridge,
>>>>                         int (*cb)(struct pci_dev *, void *),
>>>>                         void *userdata)
>>>>  {
>>>> +   /*
>>>> +    * In a non-native case where there is no OS-visible reporting
>>>> +    * device the bridge will be NULL, i.e., no RCEC, no Downstream 
>>>> Port.
>>>> +    */
>>>>     if (bridge->subordinate)
>>>>             pci_walk_bus(bridge->subordinate, cb, userdata);
>>>> +   else if (bridge->rcec)
>>>> +           cb(bridge->rcec, userdata);
>>>>     else
>>>>             cb(bridge, userdata);
>>>>  }
>>>> @@ -194,12 +200,6 @@ pci_ers_result_t pcie_do_recovery(struct 
>>>> pci_dev *dev,
>>>>     pci_dbg(bridge, "broadcast error_detected message\n");
>>>>     if (state == pci_channel_io_frozen) {
>>>>             pci_walk_bridge(bridge, report_frozen_detected, 
>>>> &status);
>>>> -           if (type == PCI_EXP_TYPE_RC_END) {
>>>> -                   pci_warn(dev, "subordinate device reset not 
>>>> possible for RCiEP\n");
>>>> -                   status = PCI_ERS_RESULT_NONE;
>>>> -                   goto failed;
>>>> -           }
>>>> -
>>>>             status = reset_subordinates(bridge);
>>>>             if (status != PCI_ERS_RESULT_RECOVERED) {
>>>>                     pci_warn(bridge, "subordinate device reset 
>>>> failed\n");
>>>> --
>>>> 2.28.0
>>>>

Kuppuswamy Sathyanarayanan Oct. 19, 2020, 6:59 p.m. UTC | #8

On 10/19/20 11:31 AM, Sean V Kelley wrote:
> On 19 Oct 2020, at 3:49, Ethan Zhao wrote:
>
>> On Sat, Oct 17, 2020 at 6:29 AM Bjorn Helgaas <[e-mail]> wrote:
>>>
>>> [+cc Christoph, Ethan, Sinan, Keith; sorry should have cc'd you to
>>> begin with since you're looking at this code too. Particularly
>>> interested in your thoughts about whether we should be touching
>>> PCI_ERR_ROOT_COMMAND and PCI_ERR_ROOT_STATUS when we don't own AER.]
>>
>> aer_root_reset() function has a prefix  'aer_', looks like it's a
>> function of aer driver, will
>> only be called by aer driver at runtime. if so it's up to the
>> owner/aer to know if OSPM is
>> granted to init. while actually some of the functions and runtime service of
>> aer driver is also shared by GHES driver (running time) and DPC driver
>> (compiling time ?)
>> etc. then it is confused now.
>>
>> Shall we move some of the shared functions and running time service to
>> pci/err.c ?
>> if so , just like pcie_do_recovery(), it's share by firmware_first  mode GHES
>> ghes_probe()
>> ->ghes_irq_func
>>   ->ghes_proc
>>     ->ghes_do_proc()
>>       ->ghes_handle_aer()
>>         ->aer_recover_work_func()
>>           ->pcie_do_recovery()
>>             ->aer_root_reset()
>>
>> and aer driver etc.  if aer wants to do some access might conflict
>> with firmware(or
>> firmware in embedded controller) should check _OSC_ etc first. blindly issue
>> PCI_ERR_ROOT_COMMAND  or clear PCI_ERR_ROOT_STATUS *likely*
>> cause errors by error handling itself.
>
> If _OSC negotiation ends up with FW being in control of AER, that means OS is not in charge and 
> should not be messing with AER I guess. That seems appropriate to me then.
But APEI based notification is more like a hybrid approach (frimware first detects the
error and notifies OS). Since spec does not clarify what OS is allowed to do, its bit of a
gray area now. My point is, since firmware allows OS to process the error by sending
the notification, I think its OK to clear the status once the error is handled.
>
> Thanks,
>
> Sean
>
>
>
>>
>> Thanks,
>> Ethan
>>
>>>
>>> On Fri, Oct 16, 2020 at 03:30:37PM -0500, Bjorn Helgaas wrote:
>>>> [+to Jonathan]
>>>>
>>>> On Thu, Oct 15, 2020 at 05:11:10PM -0700, Sean V Kelley wrote:
>>>>> From: Qiuxu Zhuo <[e-mail]>
>>>>>
>>>>> When attempting error recovery for an RCiEP associated with an RCEC device,
>>>>> there needs to be a way to update the Root Error Status, the Uncorrectable
>>>>> Error Status and the Uncorrectable Error Severity of the parent RCEC.  In
>>>>> some non-native cases in which there is no OS-visible device associated
>>>>> with the RCiEP, there is nothing to act upon as the firmware is acting
>>>>> before the OS.
>>>>>
>>>>> Add handling for the linked RCEC in AER/ERR while taking into account
>>>>> non-native cases.
>>>>>
>>>>> Co-developed-by: Sean V Kelley <[e-mail]>
>>>>> Link: https://lore.kernel.org/r/[e-mail]
>>>>> Signed-off-by: Sean V Kelley <[e-mail]>
>>>>> Signed-off-by: Qiuxu Zhuo <[e-mail]>
>>>>> Signed-off-by: Bjorn Helgaas <[e-mail]>
>>>>> Reviewed-by: Jonathan Cameron <[e-mail]>
>>>>> ---
>>>>>  drivers/pci/pcie/aer.c | 53 ++++++++++++++++++++++++++++++------------
>>>>>  drivers/pci/pcie/err.c | 20 ++++++++--------
>>>>>  2 files changed, 48 insertions(+), 25 deletions(-)
>>>>>
>>>>> diff --git a/drivers/pci/pcie/aer.c b/drivers/pci/pcie/aer.c
>>>>> index 65dff5f3457a..083f69b67bfd 100644
>>>>> --- a/drivers/pci/pcie/aer.c
>>>>> +++ b/drivers/pci/pcie/aer.c
>>>>> @@ -1357,27 +1357,50 @@ static int aer_probe(struct pcie_device *dev)
>>>>>   */
>>>>>  static pci_ers_result_t aer_root_reset(struct pci_dev *dev)
>>>>>  {
>>>>> -   int aer = dev->aer_cap;
>>>>> +   int type = pci_pcie_type(dev);
>>>>> +   struct pci_dev *root;
>>>>> +   int aer = 0;
>>>>> +   int rc = 0;
>>>>>     u32 reg32;
>>>>> -   int rc;
>>>>>
>>>>> +   if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
>>>>
>>>> "type == PCI_EXP_TYPE_RC_END"
>>>>
>>>>> +           /*
>>>>> +            * The reset should only clear the Root Error Status
>>>>> +            * of the RCEC. Only perform this for the
>>>>> +            * native case, i.e., an RCEC is present.
>>>>> +            */
>>>>> +           root = dev->rcec;
>>>>> +   else
>>>>> +           root = dev;
>>>>>
>>>>> -   /* Disable Root's interrupt in response to error messages */
>>>>> -   pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, &reg32);
>>>>> -   reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
>>>>> -   pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, reg32);
>>>>> +   if (root)
>>>>> +           aer = dev->aer_cap;
>>>>>
>>>>> -   rc = pci_bus_error_reset(dev);
>>>>> -   pci_info(dev, "Root Port link has been reset\n");
>>>>> +   if (aer) {
>>>>> +           /* Disable Root's interrupt in response to error messages */
>>>>> +           pci_read_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, &reg32);
>>>>> +           reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
>>>>> +           pci_write_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, reg32);
>>>>
>>>> Not directly related to *this* patch, but my assumption was that in
>>>> the APEI case, the firmware should retain ownership of the AER
>>>> Capability, so the OS should not touch PCI_ERR_ROOT_COMMAND and
>>>> PCI_ERR_ROOT_STATUS.
>>>>
>>>> But this code appears to ignore that ownership.  Jonathan, you must
>>>> have looked at this recently for 068c29a248b6 ("PCI/ERR: Clear PCIe
>>>> Device Status errors only if OS owns AER").  Do you have any insight
>>>> about this?
>>>>
>>>>> -   /* Clear Root Error Status */
>>>>> -   pci_read_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, &reg32);
>>>>> -   pci_write_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, reg32);
>>>>> +           /* Clear Root Error Status */
>>>>> +           pci_read_config_dword(root, aer + PCI_ERR_ROOT_STATUS, &reg32);
>>>>> +           pci_write_config_dword(root, aer + PCI_ERR_ROOT_STATUS, reg32);
>>>>>
>>>>> -   /* Enable Root Port's interrupt in response to error messages */
>>>>> -   pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, &reg32);
>>>>> -   reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
>>>>> -   pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, reg32);
>>>>> +           /* Enable Root Port's interrupt in response to error messages */
>>>>> +           pci_read_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, &reg32);
>>>>> +           reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
>>>>> +           pci_write_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, reg32);
>>>>> +   }
>>>>> +
>>>>> +   if ((type == PCI_EXP_TYPE_RC_EC) || (type == PCI_EXP_TYPE_RC_END)) {
>>>>> +           if (pcie_has_flr(root)) {
>>>>> +                   rc = pcie_flr(root);
>>>>> +                   pci_info(dev, "has been reset (%d)\n", rc);
>>>>> +           }
>>>>> +   } else {
>>>>> +           rc = pci_bus_error_reset(root);
>>>>
>>>> Don't we want "dev" for both the FLR and pci_bus_error_reset()?  I
>>>> think "root == dev" except when dev is an RCiEP.  When dev is an
>>>> RCiEP, "root" is the RCEC (if present), and we want to reset the
>>>> RCiEP, not the RCEC.
>>>>
>>>>> +           pci_info(dev, "Root Port link has been reset (%d)\n", rc);
>>>>> +   }
>>>>
>>>> There are a couple changes here that I think should be split out.
>>>>
>>>> Based on my theory that when firmware retains control of AER, the OS
>>>> should not touch PCI_ERR_ROOT_COMMAND and PCI_ERR_ROOT_STATUS, and any
>>>> updates to them would have to be done by firmware before we get here,
>>>> I suggested reordering this:
>>>>
>>>>   - clear PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>>>>   - do reset
>>>>   - clear PCI_ERR_ROOT_STATUS (for APEI, presumably done by firmware?)
>>>>   - enable PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>>>>
>>>> to this:
>>>>
>>>>   - clear PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>>>>   - clear PCI_ERR_ROOT_STATUS
>>>>   - enable PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>>>>   - do reset
>>>>
>>>> If my theory is correct, I think we should still reorder this, but:
>>>>
>>>>   - It's a significant behavior change that deserves its own patch so
>>>>     we can document/bisect/revert.
>>>>
>>>>   - I'm not sure why we clear the PCI_ERR_ROOT_COMMAND error reporting
>>>>     bits.  In the new "clear COMMAND, clear STATUS, enable COMMAND"
>>>>     order, it looks superfluous.  There's no reason to disable error
>>>>     reporting while clearing the status bits.
>>>>
>>>>     The current "clear, reset, enable" order suggests that the reset
>>>>     might cause errors that we should ignore.  I don't know whether
>>>>     that's the case or not.  It dates from 6c2b374d7485 ("PCI-Express
>>>>     AER implemetation: AER core and aerdriver"), which doesn't
>>>>     elaborate.
>>>>
>>>>   - Should we also test for OS ownership of AER before touching
>>>>     PCI_ERR_ROOT_STATUS?
>>>>
>>>>   - If we remove the PCI_ERR_ROOT_COMMAND fiddling (and I tentatively
>>>>     think we *should* unless we can justify it), that would also
>>>>     deserve its own patch.  Possibly (1) remove PCI_ERR_ROOT_COMMAND
>>>>     fiddling, (2) reorder PCI_ERR_ROOT_STATUS clearing and reset, (3)
>>>>     test for OS ownership of AER (?), (4) the rest of this patch.
>>>>
>>>>>     return rc ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
>>>>>  }
>>>>> diff --git a/drivers/pci/pcie/err.c b/drivers/pci/pcie/err.c
>>>>> index 7883c9791562..cbc5abfe767b 100644
>>>>> --- a/drivers/pci/pcie/err.c
>>>>> +++ b/drivers/pci/pcie/err.c
>>>>> @@ -148,10 +148,10 @@ static int report_resume(struct pci_dev *dev, void *data)
>>>>>
>>>>>  /**
>>>>>   * pci_walk_bridge - walk bridges potentially AER affected
>>>>> - * @bridge:        bridge which may be a Port, an RCEC with associated RCiEPs,
>>>>> - *         or an RCiEP associated with an RCEC
>>>>> - * @cb:            callback to be called for each device found
>>>>> - * @userdata:      arbitrary pointer to be passed to callback
>>>>> + * @bridge   bridge which may be an RCEC with associated RCiEPs,
>>>>> + *           or a Port.
>>>>> + * @cb       callback to be called for each device found
>>>>> + * @userdata arbitrary pointer to be passed to callback.
>>>>>   *
>>>>>   * If the device provided is a bridge, walk the subordinate bus, including
>>>>>   * any bridged devices on buses under this bus.  Call the provided callback
>>>>> @@ -164,8 +164,14 @@ static void pci_walk_bridge(struct pci_dev *bridge,
>>>>>                         int (*cb)(struct pci_dev *, void *),
>>>>>                         void *userdata)
>>>>>  {
>>>>> +   /*
>>>>> +    * In a non-native case where there is no OS-visible reporting
>>>>> +    * device the bridge will be NULL, i.e., no RCEC, no Downstream Port.
>>>>> +    */
>>>>>     if (bridge->subordinate)
>>>>>             pci_walk_bus(bridge->subordinate, cb, userdata);
>>>>> +   else if (bridge->rcec)
>>>>> +           cb(bridge->rcec, userdata);
>>>>>     else
>>>>>             cb(bridge, userdata);
>>>>>  }
>>>>> @@ -194,12 +200,6 @@ pci_ers_result_t pcie_do_recovery(struct pci_dev *dev,
>>>>>     pci_dbg(bridge, "broadcast error_detected message\n");
>>>>>     if (state == pci_channel_io_frozen) {
>>>>>             pci_walk_bridge(bridge, report_frozen_detected, &status);
>>>>> -           if (type == PCI_EXP_TYPE_RC_END) {
>>>>> -                   pci_warn(dev, "subordinate device reset not possible for RCiEP\n");
>>>>> -                   status = PCI_ERS_RESULT_NONE;
>>>>> -                   goto failed;
>>>>> -           }
>>>>> -
>>>>>             status = reset_subordinates(bridge);
>>>>>             if (status != PCI_ERS_RESULT_RECOVERED) {
>>>>>                     pci_warn(bridge, "subordinate device reset failed\n");
>>>>> -- 
>>>>> 2.28.0
>>>>>

Kelley, Sean V Oct. 19, 2020, 7:07 p.m. UTC | #9

On Sat, 2020-10-17 at 09:14 -0700, Sean V Kelley wrote:
> On 16 Oct 2020, at 13:30, Bjorn Helgaas wrote:
> 
> > [+to Jonathan]
> > 
> > On Thu, Oct 15, 2020 at 05:11:10PM -0700, Sean V Kelley wrote:
> > > From: Qiuxu Zhuo <[e-mail]>
> > > 
> > > When attempting error recovery for an RCiEP associated with an
> > > RCEC 
> > > device,
> > > there needs to be a way to update the Root Error Status, the 
> > > Uncorrectable
> > > Error Status and the Uncorrectable Error Severity of the parent
> > > RCEC. 
> > >  In
> > > some non-native cases in which there is no OS-visible device 
> > > associated
> > > with the RCiEP, there is nothing to act upon as the firmware is 
> > > acting
> > > before the OS.
> > > 
> > > Add handling for the linked RCEC in AER/ERR while taking into
> > > account
> > > non-native cases.
> > > 
> > > Co-developed-by: Sean V Kelley <[e-mail]>
> > > Link: 
> > > https://lore.kernel.org/r/[e-mail]
> > > Signed-off-by: Sean V Kelley <[e-mail]>
> > > Signed-off-by: Qiuxu Zhuo <[e-mail]>
> > > Signed-off-by: Bjorn Helgaas <[e-mail]>
> > > Reviewed-by: Jonathan Cameron <[e-mail]>
> > > ---
> > >  drivers/pci/pcie/aer.c | 53 
> > > ++++++++++++++++++++++++++++++------------
> > >  drivers/pci/pcie/err.c | 20 ++++++++--------
> > >  2 files changed, 48 insertions(+), 25 deletions(-)
> > > 
> > > diff --git a/drivers/pci/pcie/aer.c b/drivers/pci/pcie/aer.c
> > > index 65dff5f3457a..083f69b67bfd 100644
> > > --- a/drivers/pci/pcie/aer.c
> > > +++ b/drivers/pci/pcie/aer.c
> > > @@ -1357,27 +1357,50 @@ static int aer_probe(struct pcie_device
> > > *dev)
> > >   */
> > >  static pci_ers_result_t aer_root_reset(struct pci_dev *dev)
> > >  {
> > > -       int aer = dev->aer_cap;
> > > +       int type = pci_pcie_type(dev);
> > > +       struct pci_dev *root;
> > > +       int aer = 0;
> > > +       int rc = 0;
> > >         u32 reg32;
> > > -       int rc;
> > > 
> > > +       if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
> > 
> > "type == PCI_EXP_TYPE_RC_END"
> 
> Right, I merged your suggested changes which added the type. Will 
> correct.
> 
> > 
> > > +               /*
> > > +                * The reset should only clear the Root Error
> > > Status
> > > +                * of the RCEC. Only perform this for the
> > > +                * native case, i.e., an RCEC is present.
> > > +                */
> > > +               root = dev->rcec;
> > > +       else
> > > +               root = dev;
> > > 
> > > -       /* Disable Root's interrupt in response to error messages
> > > */
> > > -       pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND,
> > > &reg32);
> > > -       reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
> > > -       pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND,
> > > reg32);
> > > +       if (root)
> > > +               aer = dev->aer_cap;
> > > 
> > > -       rc = pci_bus_error_reset(dev);
> > > -       pci_info(dev, "Root Port link has been reset\n");
> > > +       if (aer) {
> > > +               /* Disable Root's interrupt in response to error
> > > messages */
> > > +               pci_read_config_dword(root, aer +
> > > PCI_ERR_ROOT_COMMAND, &reg32);
> > > +               reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
> > > +               pci_write_config_dword(root, aer +
> > > PCI_ERR_ROOT_COMMAND, reg32);
> > 
> > Not directly related to *this* patch, but my assumption was that in
> > the APEI case, the firmware should retain ownership of the AER
> > Capability, so the OS should not touch PCI_ERR_ROOT_COMMAND and
> > PCI_ERR_ROOT_STATUS.
> > 
> > But this code appears to ignore that ownership.  Jonathan, you must
> > have looked at this recently for 068c29a248b6 ("PCI/ERR: Clear PCIe
> > Device Status errors only if OS owns AER").  Do you have any
> > insight
> > about this?
> > 
> > > -       /* Clear Root Error Status */
> > > -       pci_read_config_dword(dev, aer + PCI_ERR_ROOT_STATUS,
> > > &reg32);
> > > -       pci_write_config_dword(dev, aer + PCI_ERR_ROOT_STATUS,
> > > reg32);
> > > +               /* Clear Root Error Status */
> > > +               pci_read_config_dword(root, aer +
> > > PCI_ERR_ROOT_STATUS, &reg32);
> > > +               pci_write_config_dword(root, aer +
> > > PCI_ERR_ROOT_STATUS, reg32);
> > > 
> > > -       /* Enable Root Port's interrupt in response to error
> > > messages */
> > > -       pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND,
> > > &reg32);
> > > -       reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
> > > -       pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND,
> > > reg32);
> > > +               /* Enable Root Port's interrupt in response to
> > > error messages */
> > > +               pci_read_config_dword(root, aer +
> > > PCI_ERR_ROOT_COMMAND, &reg32);
> > > +               reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
> > > +               pci_write_config_dword(root, aer +
> > > PCI_ERR_ROOT_COMMAND, reg32);
> > > +       }
> > > +
> > > +       if ((type == PCI_EXP_TYPE_RC_EC) || (type ==
> > > PCI_EXP_TYPE_RC_END)) 
> > > {
> > > +               if (pcie_has_flr(root)) {
> > > +                       rc = pcie_flr(root);
> > > +                       pci_info(dev, "has been reset (%d)\n",
> > > rc);
> > > +               }
> > > +       } else {
> > > +               rc = pci_bus_error_reset(root);
> > 
> > Don't we want "dev" for both the FLR and pci_bus_error_reset()?  I
> > think "root == dev" except when dev is an RCiEP.  When dev is an
> > RCiEP, "root" is the RCEC (if present), and we want to reset the
> > RCiEP, not the RCEC.
> 
> Right, when I did the goto in the earlier incarnation, I always set
> root 
> to dev at the start and in the merge it needs to be dev always except
> for the RC_END where RCEC exists. Will change without bringing back
> the 
> goto…
> 
> +       struct pci_dev *root = dev;
> 
> …
> 
> +non_native:
> +       if ((type == PCI_EXP_TYPE_RC_EC) || (type ==
> PCI_EXP_TYPE_RC_END)) {
> +               rc = flr_on_rc(root);
> +               pci_info(dev, "has been reset (%d)\n", rc);
> +       } else {
> +               rc = pci_bus_error_reset(root);
> +               pci_info(dev, "Root Port link has been reset (%d)\n",
> rc);
> +       }
> 
> 
> > 
> > > +               pci_info(dev, "Root Port link has been reset
> > > (%d)\n", rc);
> > > +       }
> > 
> > There are a couple changes here that I think should be split out.
> > 
> > Based on my theory that when firmware retains control of AER, the
> > OS
> > should not touch PCI_ERR_ROOT_COMMAND and PCI_ERR_ROOT_STATUS, and
> > any
> > updates to them would have to be done by firmware before we get
> > here,
> > I suggested reordering this:
> > 
> >   - clear PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
> >   - do reset
> >   - clear PCI_ERR_ROOT_STATUS (for APEI, presumably done by
> > firmware?)
> >   - enable PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
> > 
> > to this:
> > 
> >   - clear PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
> >   - clear PCI_ERR_ROOT_STATUS
> >   - enable PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
> >   - do reset
> > 
> > If my theory is correct, I think we should still reorder this, but:
> > 
> >   - It's a significant behavior change that deserves its own patch
> > so
> >     we can document/bisect/revert.
> > 
> >   - I'm not sure why we clear the PCI_ERR_ROOT_COMMAND error
> > reporting
> >     bits.  In the new "clear COMMAND, clear STATUS, enable COMMAND"
> >     order, it looks superfluous.  There's no reason to disable
> > error
> >     reporting while clearing the status bits.
> > 
> >     The current "clear, reset, enable" order suggests that the
> > reset
> >     might cause errors that we should ignore.  I don't know whether
> >     that's the case or not.  It dates from 6c2b374d7485 ("PCI-
> > Express
> >     AER implemetation: AER core and aerdriver"), which doesn't
> >     elaborate.
> > 
> >   - Should we also test for OS ownership of AER before touching
> >     PCI_ERR_ROOT_STATUS?
> > 
> >   - If we remove the PCI_ERR_ROOT_COMMAND fiddling (and I
> > tentatively
> >     think we *should* unless we can justify it), that would also
> >     deserve its own patch.  Possibly (1) remove
> > PCI_ERR_ROOT_COMMAND
> >     fiddling, (2) reorder PCI_ERR_ROOT_STATUS clearing and reset,
> > (3)
> >     test for OS ownership of AER (?), (4) the rest of this patch.
> 
> You’ve highlighted some good questions.



Reading Ethan's reply and also thinking about separation from an _OSC
perspective perhaps something like this could be done with a check in
aer_root_reset().


diff --git a/drivers/pci/pcie/aer.c b/drivers/pci/pcie/aer.c
index 65dff5f3457a..70bf637042ff 100644
--- a/drivers/pci/pcie/aer.c
+++ b/drivers/pci/pcie/aer.c
@@ -1357,27 +1357,46 @@ static int aer_probe(struct pcie_device *dev)
  */
 static pci_ers_result_t aer_root_reset(struct pci_dev *dev)
 {
+       int type = pci_pcie_type(dev);
+       struct pci_dev *root = dev;
        int aer = dev->aer_cap;
+       int rc = 0;
        u32 reg32;
-       int rc;
 
+       if (!pcie_aer_is_native(dev))
+               return PCI_ERS_RESULT_RECOVERD;
+
+       if (type == PCI_EXP_TYPE_RC_END)
+               /*
+                * The reset should only clear the Root Error Status
+                * of the RCEC. Only perform this for the
+                * native case, i.e., an RCEC is present.
+                */
+               root = dev->rcec;
 
        /* Disable Root's interrupt in response to error messages */
-       pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, &reg32);
+       pci_read_config_dword(root, aer + PCI_ERR_ROOT_COMMAND,
&reg32);
        reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
-       pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, reg32);
-
-       rc = pci_bus_error_reset(dev);
-       pci_info(dev, "Root Port link has been reset\n");
+       pci_write_config_dword(root, aer + PCI_ERR_ROOT_COMMAND,
reg32);
 
        /* Clear Root Error Status */
-       pci_read_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, &reg32);
-       pci_write_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, reg32);
+       pci_read_config_dword(root, aer + PCI_ERR_ROOT_STATUS, &reg32);
+       pci_write_config_dword(root, aer + PCI_ERR_ROOT_STATUS, reg32);
 
        /* Enable Root Port's interrupt in response to error messages
*/
-       pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, &reg32);
+       pci_read_config_dword(root, aer + PCI_ERR_ROOT_COMMAND,
&reg32);
        reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
-       pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, reg32);
+       pci_write_config_dword(root, aer + PCI_ERR_ROOT_COMMAND,
reg32);
+
+       if (type == PCI_EXP_TYPE_RC_EC || type == PCI_EXP_TYPE_RC_END)
{
+               if (pcie_has_flr(dev)) {
+                       rc = pcie_flr(dev);
+                       pci_info(dev, "has been reset (%d)\n", rc);
+               }
+       } else {
+               rc = pci_bus_error_reset(dev);
+               pci_info(dev, "Root Port link has been reset (%d)\n",
rc);
+       }


> 
> I think we should remove the fiddling until we have a clearer picture
> and put that into its own patch.
> 
> Sean
> > 
> > >         return rc ? PCI_ERS_RESULT_DISCONNECT :
> > > PCI_ERS_RESULT_RECOVERED;
> > >  }
> > > diff --git a/drivers/pci/pcie/err.c b/drivers/pci/pcie/err.c
> > > index 7883c9791562..cbc5abfe767b 100644
> > > --- a/drivers/pci/pcie/err.c
> > > +++ b/drivers/pci/pcie/err.c
> > > @@ -148,10 +148,10 @@ static int report_resume(struct pci_dev
> > > *dev, 
> > > void *data)
> > > 
> > >  /**
> > >   * pci_walk_bridge - walk bridges potentially AER affected
> > > - * @bridge:    bridge which may be a Port, an RCEC with
> > > associated 
> > > RCiEPs,
> > > - *             or an RCiEP associated with an RCEC
> > > - * @cb:                callback to be called for each device
> > > found
> > > - * @userdata:  arbitrary pointer to be passed to callback
> > > + * @bridge   bridge which may be an RCEC with associated RCiEPs,
> > > + *           or a Port.
> > > + * @cb       callback to be called for each device found
> > > + * @userdata arbitrary pointer to be passed to callback.
> > >   *
> > >   * If the device provided is a bridge, walk the subordinate bus,
> > > including
> > >   * any bridged devices on buses under this bus.  Call the
> > > provided 
> > > callback
> > > @@ -164,8 +164,14 @@ static void pci_walk_bridge(struct pci_dev 
> > > *bridge,
> > >                             int (*cb)(struct pci_dev *, void *),
> > >                             void *userdata)
> > >  {
> > > +       /*
> > > +        * In a non-native case where there is no OS-visible
> > > reporting
> > > +        * device the bridge will be NULL, i.e., no RCEC, no
> > > Downstream 
> > > Port.
> > > +        */
> > >         if (bridge->subordinate)
> > >                 pci_walk_bus(bridge->subordinate, cb, userdata);
> > > +       else if (bridge->rcec)
> > > +               cb(bridge->rcec, userdata);
> > >         else
> > >                 cb(bridge, userdata);
> > >  }
> > > @@ -194,12 +200,6 @@ pci_ers_result_t pcie_do_recovery(struct
> > > pci_dev 
> > > *dev,
> > >         pci_dbg(bridge, "broadcast error_detected message\n");
> > >         if (state == pci_channel_io_frozen) {
> > >                 pci_walk_bridge(bridge, report_frozen_detected,
> > > &status);
> > > -               if (type == PCI_EXP_TYPE_RC_END) {
> > > -                       pci_warn(dev, "subordinate device reset
> > > not possible for 
> > > RCiEP\n");
> > > -                       status = PCI_ERS_RESULT_NONE;
> > > -                       goto failed;
> > > -               }
> > > -
> > >                 status = reset_subordinates(bridge);
> > >                 if (status != PCI_ERS_RESULT_RECOVERED) {
> > >                         pci_warn(bridge, "subordinate device
> > > reset failed\n");
> > > -- 
> > > 2.28.0
> > >

Kelley, Sean V Oct. 19, 2020, 8:50 p.m. UTC | #10

On 19 Oct 2020, at 11:59, Kuppuswamy, Sathyanarayanan wrote:

> On 10/19/20 11:31 AM, Sean V Kelley wrote:
>> On 19 Oct 2020, at 3:49, Ethan Zhao wrote:
>>
>>> On Sat, Oct 17, 2020 at 6:29 AM Bjorn Helgaas <[e-mail]> 
>>> wrote:
>>>>
>>>> [+cc Christoph, Ethan, Sinan, Keith; sorry should have cc'd you to
>>>> begin with since you're looking at this code too. Particularly
>>>> interested in your thoughts about whether we should be touching
>>>> PCI_ERR_ROOT_COMMAND and PCI_ERR_ROOT_STATUS when we don't own 
>>>> AER.]
>>>
>>> aer_root_reset() function has a prefix  'aer_', looks like it's a
>>> function of aer driver, will
>>> only be called by aer driver at runtime. if so it's up to the
>>> owner/aer to know if OSPM is
>>> granted to init. while actually some of the functions and runtime 
>>> service of
>>> aer driver is also shared by GHES driver (running time) and DPC 
>>> driver
>>> (compiling time ?)
>>> etc. then it is confused now.
>>>
>>> Shall we move some of the shared functions and running time service 
>>> to
>>> pci/err.c ?
>>> if so , just like pcie_do_recovery(), it's share by firmware_first  
>>> mode GHES
>>> ghes_probe()
>>> ->ghes_irq_func
>>>   ->ghes_proc
>>>     ->ghes_do_proc()
>>>       ->ghes_handle_aer()
>>>         ->aer_recover_work_func()
>>>           ->pcie_do_recovery()
>>>             ->aer_root_reset()
>>>
>>> and aer driver etc.  if aer wants to do some access might conflict
>>> with firmware(or
>>> firmware in embedded controller) should check _OSC_ etc first. 
>>> blindly issue
>>> PCI_ERR_ROOT_COMMAND  or clear PCI_ERR_ROOT_STATUS *likely*
>>> cause errors by error handling itself.
>>
>> If _OSC negotiation ends up with FW being in control of AER, that 
>> means OS is not in charge and should not be messing with AER I guess. 
>> That seems appropriate to me then.
> But APEI based notification is more like a hybrid approach (frimware 
> first detects the
> error and notifies OS). Since spec does not clarify what OS is allowed 
> to do, its bit of a
> gray area now. My point is, since firmware allows OS to process the 
> error by sending
> the notification, I think its OK to clear the status once the error is 
> handled.

I don’t disagree as long as AER is granted to the OS via _OSC. But if 
it’s not granted explicitly via _OSC even in the APEI case where 
it’s either an SCI or NMI and not an MSI, I’m unsure whether the OS 
should be touching those registers.

Sean

>>
>> Thanks,
>>
>> Sean
>>
>>
>>
>>>
>>> Thanks,
>>> Ethan
>>>
>>>>
>>>> On Fri, Oct 16, 2020 at 03:30:37PM -0500, Bjorn Helgaas wrote:
>>>>> [+to Jonathan]
>>>>>
>>>>> On Thu, Oct 15, 2020 at 05:11:10PM -0700, Sean V Kelley wrote:
>>>>>> From: Qiuxu Zhuo <[e-mail]>
>>>>>>
>>>>>> When attempting error recovery for an RCiEP associated with an 
>>>>>> RCEC device,
>>>>>> there needs to be a way to update the Root Error Status, the 
>>>>>> Uncorrectable
>>>>>> Error Status and the Uncorrectable Error Severity of the parent 
>>>>>> RCEC.  In
>>>>>> some non-native cases in which there is no OS-visible device 
>>>>>> associated
>>>>>> with the RCiEP, there is nothing to act upon as the firmware is 
>>>>>> acting
>>>>>> before the OS.
>>>>>>
>>>>>> Add handling for the linked RCEC in AER/ERR while taking into 
>>>>>> account
>>>>>> non-native cases.
>>>>>>
>>>>>> Co-developed-by: Sean V Kelley <[e-mail]>
>>>>>> Link: 
>>>>>> https://lore.kernel.org/r/[e-mail]
>>>>>> Signed-off-by: Sean V Kelley <[e-mail]>
>>>>>> Signed-off-by: Qiuxu Zhuo <[e-mail]>
>>>>>> Signed-off-by: Bjorn Helgaas <[e-mail]>
>>>>>> Reviewed-by: Jonathan Cameron <[e-mail]>
>>>>>> ---
>>>>>>  drivers/pci/pcie/aer.c | 53 
>>>>>> ++++++++++++++++++++++++++++++------------
>>>>>>  drivers/pci/pcie/err.c | 20 ++++++++--------
>>>>>>  2 files changed, 48 insertions(+), 25 deletions(-)
>>>>>>
>>>>>> diff --git a/drivers/pci/pcie/aer.c b/drivers/pci/pcie/aer.c
>>>>>> index 65dff5f3457a..083f69b67bfd 100644
>>>>>> --- a/drivers/pci/pcie/aer.c
>>>>>> +++ b/drivers/pci/pcie/aer.c
>>>>>> @@ -1357,27 +1357,50 @@ static int aer_probe(struct pcie_device 
>>>>>> *dev)
>>>>>>   */
>>>>>>  static pci_ers_result_t aer_root_reset(struct pci_dev *dev)
>>>>>>  {
>>>>>> -   int aer = dev->aer_cap;
>>>>>> +   int type = pci_pcie_type(dev);
>>>>>> +   struct pci_dev *root;
>>>>>> +   int aer = 0;
>>>>>> +   int rc = 0;
>>>>>>     u32 reg32;
>>>>>> -   int rc;
>>>>>>
>>>>>> +   if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
>>>>>
>>>>> "type == PCI_EXP_TYPE_RC_END"
>>>>>
>>>>>> +           /*
>>>>>> +            * The reset should only clear the Root 
>>>>>> Error Status
>>>>>> +            * of the RCEC. Only perform this for the
>>>>>> +            * native case, i.e., an RCEC is present.
>>>>>> +            */
>>>>>> +           root = dev->rcec;
>>>>>> +   else
>>>>>> +           root = dev;
>>>>>>
>>>>>> -   /* Disable Root's interrupt in response to error messages 
>>>>>> */
>>>>>> -   pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, 
>>>>>> &reg32);
>>>>>> -   reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
>>>>>> -   pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, 
>>>>>> reg32);
>>>>>> +   if (root)
>>>>>> +           aer = dev->aer_cap;
>>>>>>
>>>>>> -   rc = pci_bus_error_reset(dev);
>>>>>> -   pci_info(dev, "Root Port link has been reset\n");
>>>>>> +   if (aer) {
>>>>>> +           /* Disable Root's interrupt in response to 
>>>>>> error messages */
>>>>>> +           pci_read_config_dword(root, aer + 
>>>>>> PCI_ERR_ROOT_COMMAND, &reg32);
>>>>>> +           reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
>>>>>> +           pci_write_config_dword(root, aer + 
>>>>>> PCI_ERR_ROOT_COMMAND, reg32);
>>>>>
>>>>> Not directly related to *this* patch, but my assumption was that 
>>>>> in
>>>>> the APEI case, the firmware should retain ownership of the AER
>>>>> Capability, so the OS should not touch PCI_ERR_ROOT_COMMAND and
>>>>> PCI_ERR_ROOT_STATUS.
>>>>>
>>>>> But this code appears to ignore that ownership.  Jonathan, you 
>>>>> must
>>>>> have looked at this recently for 068c29a248b6 ("PCI/ERR: Clear 
>>>>> PCIe
>>>>> Device Status errors only if OS owns AER").  Do you have any 
>>>>> insight
>>>>> about this?
>>>>>
>>>>>> -   /* Clear Root Error Status */
>>>>>> -   pci_read_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, 
>>>>>> &reg32);
>>>>>> -   pci_write_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, 
>>>>>> reg32);
>>>>>> +           /* Clear Root Error Status */
>>>>>> +           pci_read_config_dword(root, aer + 
>>>>>> PCI_ERR_ROOT_STATUS, &reg32);
>>>>>> +           pci_write_config_dword(root, aer + 
>>>>>> PCI_ERR_ROOT_STATUS, reg32);
>>>>>>
>>>>>> -   /* Enable Root Port's interrupt in response to error 
>>>>>> messages */
>>>>>> -   pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, 
>>>>>> &reg32);
>>>>>> -   reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
>>>>>> -   pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, 
>>>>>> reg32);
>>>>>> +           /* Enable Root Port's interrupt in response 
>>>>>> to error messages */
>>>>>> +           pci_read_config_dword(root, aer + 
>>>>>> PCI_ERR_ROOT_COMMAND, &reg32);
>>>>>> +           reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
>>>>>> +           pci_write_config_dword(root, aer + 
>>>>>> PCI_ERR_ROOT_COMMAND, reg32);
>>>>>> +   }
>>>>>> +
>>>>>> +   if ((type == PCI_EXP_TYPE_RC_EC) || (type == 
>>>>>> PCI_EXP_TYPE_RC_END)) {
>>>>>> +           if (pcie_has_flr(root)) {
>>>>>> +                   rc = pcie_flr(root);
>>>>>> +                   pci_info(dev, "has been 
>>>>>> reset (%d)\n", rc);
>>>>>> +           }
>>>>>> +   } else {
>>>>>> +           rc = pci_bus_error_reset(root);
>>>>>
>>>>> Don't we want "dev" for both the FLR and pci_bus_error_reset()?  
>>>>> I
>>>>> think "root == dev" except when dev is an RCiEP.  When dev is an
>>>>> RCiEP, "root" is the RCEC (if present), and we want to reset the
>>>>> RCiEP, not the RCEC.
>>>>>
>>>>>> +           pci_info(dev, "Root Port link has been 
>>>>>> reset (%d)\n", rc);
>>>>>> +   }
>>>>>
>>>>> There are a couple changes here that I think should be split out.
>>>>>
>>>>> Based on my theory that when firmware retains control of AER, the 
>>>>> OS
>>>>> should not touch PCI_ERR_ROOT_COMMAND and PCI_ERR_ROOT_STATUS, and 
>>>>> any
>>>>> updates to them would have to be done by firmware before we get 
>>>>> here,
>>>>> I suggested reordering this:
>>>>>
>>>>>   - clear PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>>>>>   - do reset
>>>>>   - clear PCI_ERR_ROOT_STATUS (for APEI, presumably done by 
>>>>> firmware?)
>>>>>   - enable PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>>>>>
>>>>> to this:
>>>>>
>>>>>   - clear PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>>>>>   - clear PCI_ERR_ROOT_STATUS
>>>>>   - enable PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
>>>>>   - do reset
>>>>>
>>>>> If my theory is correct, I think we should still reorder this, 
>>>>> but:
>>>>>
>>>>>   - It's a significant behavior change that deserves its own 
>>>>> patch so
>>>>>     we can document/bisect/revert.
>>>>>
>>>>>   - I'm not sure why we clear the PCI_ERR_ROOT_COMMAND error 
>>>>> reporting
>>>>>     bits.  In the new "clear COMMAND, clear STATUS, enable 
>>>>> COMMAND"
>>>>>     order, it looks superfluous.  There's no reason to disable 
>>>>> error
>>>>>     reporting while clearing the status bits.
>>>>>
>>>>>     The current "clear, reset, enable" order suggests that the 
>>>>> reset
>>>>>     might cause errors that we should ignore.  I don't know 
>>>>> whether
>>>>>     that's the case or not.  It dates from 6c2b374d7485 
>>>>> ("PCI-Express
>>>>>     AER implemetation: AER core and aerdriver"), which doesn't
>>>>>     elaborate.
>>>>>
>>>>>   - Should we also test for OS ownership of AER before touching
>>>>>     PCI_ERR_ROOT_STATUS?
>>>>>
>>>>>   - If we remove the PCI_ERR_ROOT_COMMAND fiddling (and I 
>>>>> tentatively
>>>>>     think we *should* unless we can justify it), that would 
>>>>> also
>>>>>     deserve its own patch.  Possibly (1) remove 
>>>>> PCI_ERR_ROOT_COMMAND
>>>>>     fiddling, (2) reorder PCI_ERR_ROOT_STATUS clearing and 
>>>>> reset, (3)
>>>>>     test for OS ownership of AER (?), (4) the rest of this 
>>>>> patch.
>>>>>
>>>>>>     return rc ? PCI_ERS_RESULT_DISCONNECT : 
>>>>>> PCI_ERS_RESULT_RECOVERED;
>>>>>>  }
>>>>>> diff --git a/drivers/pci/pcie/err.c b/drivers/pci/pcie/err.c
>>>>>> index 7883c9791562..cbc5abfe767b 100644
>>>>>> --- a/drivers/pci/pcie/err.c
>>>>>> +++ b/drivers/pci/pcie/err.c
>>>>>> @@ -148,10 +148,10 @@ static int report_resume(struct pci_dev 
>>>>>> *dev, void *data)
>>>>>>
>>>>>>  /**
>>>>>>   * pci_walk_bridge - walk bridges potentially AER affected
>>>>>> - * @bridge:        bridge which may be a Port, an RCEC 
>>>>>> with associated RCiEPs,
>>>>>> - *         or an RCiEP associated with an RCEC
>>>>>> - * @cb:            callback to be called for each 
>>>>>> device found
>>>>>> - * @userdata:      arbitrary pointer to be passed to 
>>>>>> callback
>>>>>> + * @bridge   bridge which may be an RCEC with associated 
>>>>>> RCiEPs,
>>>>>> + *           or a Port.
>>>>>> + * @cb       callback to be called for each device found
>>>>>> + * @userdata arbitrary pointer to be passed to callback.
>>>>>>   *
>>>>>>   * If the device provided is a bridge, walk the subordinate 
>>>>>> bus, including
>>>>>>   * any bridged devices on buses under this bus.  Call the 
>>>>>> provided callback
>>>>>> @@ -164,8 +164,14 @@ static void pci_walk_bridge(struct pci_dev 
>>>>>> *bridge,
>>>>>>                         int (*cb)(struct 
>>>>>> pci_dev *, void *),
>>>>>>                         void *userdata)
>>>>>>  {
>>>>>> +   /*
>>>>>> +    * In a non-native case where there is no OS-visible 
>>>>>> reporting
>>>>>> +    * device the bridge will be NULL, i.e., no RCEC, no 
>>>>>> Downstream Port.
>>>>>> +    */
>>>>>>     if (bridge->subordinate)
>>>>>>             pci_walk_bus(bridge->subordinate, cb, 
>>>>>> userdata);
>>>>>> +   else if (bridge->rcec)
>>>>>> +           cb(bridge->rcec, userdata);
>>>>>>     else
>>>>>>             cb(bridge, userdata);
>>>>>>  }
>>>>>> @@ -194,12 +200,6 @@ pci_ers_result_t pcie_do_recovery(struct 
>>>>>> pci_dev *dev,
>>>>>>     pci_dbg(bridge, "broadcast error_detected message\n");
>>>>>>     if (state == pci_channel_io_frozen) {
>>>>>>             pci_walk_bridge(bridge, 
>>>>>> report_frozen_detected, &status);
>>>>>> -           if (type == PCI_EXP_TYPE_RC_END) {
>>>>>> -                   pci_warn(dev, "subordinate 
>>>>>> device reset not possible for RCiEP\n");
>>>>>> -                   status = 
>>>>>> PCI_ERS_RESULT_NONE;
>>>>>> -                   goto failed;
>>>>>> -           }
>>>>>> -
>>>>>>             status = reset_subordinates(bridge);
>>>>>>             if (status != PCI_ERS_RESULT_RECOVERED) {
>>>>>>                     pci_warn(bridge, 
>>>>>> "subordinate device reset failed\n");
>>>>>> -- 
>>>>>> 2.28.0
>>>>>>
> -- 
> Sathyanarayanan Kuppuswamy
> Linux Kernel Developer

Jonathan Cameron Oct. 20, 2020, 12:59 p.m. UTC | #11

On Mon, 19 Oct 2020 13:50:17 -0700
Sean V Kelley <[e-mail]> wrote:

> On 19 Oct 2020, at 11:59, Kuppuswamy, Sathyanarayanan wrote:
> 
> > On 10/19/20 11:31 AM, Sean V Kelley wrote:  
> >> On 19 Oct 2020, at 3:49, Ethan Zhao wrote:
> >>  
> >>> On Sat, Oct 17, 2020 at 6:29 AM Bjorn Helgaas <[e-mail]> 
> >>> wrote:  
> >>>>
> >>>> [+cc Christoph, Ethan, Sinan, Keith; sorry should have cc'd you to
> >>>> begin with since you're looking at this code too. Particularly
> >>>> interested in your thoughts about whether we should be touching
> >>>> PCI_ERR_ROOT_COMMAND and PCI_ERR_ROOT_STATUS when we don't own 
> >>>> AER.]  
> >>>
> >>> aer_root_reset() function has a prefix  'aer_', looks like it's a
> >>> function of aer driver, will
> >>> only be called by aer driver at runtime. if so it's up to the
> >>> owner/aer to know if OSPM is
> >>> granted to init. while actually some of the functions and runtime 
> >>> service of
> >>> aer driver is also shared by GHES driver (running time) and DPC 
> >>> driver
> >>> (compiling time ?)
> >>> etc. then it is confused now.
> >>>
> >>> Shall we move some of the shared functions and running time service 
> >>> to
> >>> pci/err.c ?
> >>> if so , just like pcie_do_recovery(), it's share by firmware_first  
> >>> mode GHES
> >>> ghes_probe()  
> >>> ->ghes_irq_func  
> >>>   ->ghes_proc
> >>>     ->ghes_do_proc()
> >>>       ->ghes_handle_aer()
> >>>         ->aer_recover_work_func()
> >>>           ->pcie_do_recovery()
> >>>             ->aer_root_reset()
> >>>
> >>> and aer driver etc.  if aer wants to do some access might conflict
> >>> with firmware(or
> >>> firmware in embedded controller) should check _OSC_ etc first. 
> >>> blindly issue
> >>> PCI_ERR_ROOT_COMMAND  or clear PCI_ERR_ROOT_STATUS *likely*
> >>> cause errors by error handling itself.  
> >>
> >> If _OSC negotiation ends up with FW being in control of AER, that 
> >> means OS is not in charge and should not be messing with AER I guess. 
> >> That seems appropriate to me then.  
> > But APEI based notification is more like a hybrid approach (frimware 
> > first detects the
> > error and notifies OS). Since spec does not clarify what OS is allowed 
> > to do, its bit of a
> > gray area now. My point is, since firmware allows OS to process the 
> > error by sending
> > the notification, I think its OK to clear the status once the error is 
> > handled.  
> 
> I don’t disagree as long as AER is granted to the OS via _OSC. But if 
> it’s not granted explicitly via _OSC even in the APEI case where 
> it’s either an SCI or NMI and not an MSI, I’m unsure whether the OS 
> should be touching those registers.

My assumption was indeed this.  If AER hasn't been granted to the OS,
it shouldn't be doing anything involving AER itself.  It should constrain
itself to dealing with the End Points etc due to the need there for
driver interaction.

I fully agree with the comment that the specifications aren't entirely
clear on these cases.

It is possible that no one is currently generating the particular
combination of severity bits in the APEI path to actually hit this.
It requires the outer record to be marked recoverable, but the inner
part to be marked fatal.  Kind of an odd mix.

In the GHES case, you get to this path by having a
Generic Error Status Block - recoverable (must not be fatal to avoid panic()
in APEI layer) containing one more more Generic Error Blocks, one of
which is fatal.

Response of our firmware team is that this particularly combination is
probably crazy.

So good to clean up this corner, but it is probably not a problem
anyone has actually hit so far.

Jonathan


> 
> Sean
> 
> >>
> >> Thanks,
> >>
> >> Sean
> >>
> >>
> >>  
> >>>
> >>> Thanks,
> >>> Ethan
> >>>  
> >>>>
> >>>> On Fri, Oct 16, 2020 at 03:30:37PM -0500, Bjorn Helgaas wrote:  
> >>>>> [+to Jonathan]
> >>>>>
> >>>>> On Thu, Oct 15, 2020 at 05:11:10PM -0700, Sean V Kelley wrote:  
> >>>>>> From: Qiuxu Zhuo <[e-mail]>
> >>>>>>
> >>>>>> When attempting error recovery for an RCiEP associated with an 
> >>>>>> RCEC device,
> >>>>>> there needs to be a way to update the Root Error Status, the 
> >>>>>> Uncorrectable
> >>>>>> Error Status and the Uncorrectable Error Severity of the parent 
> >>>>>> RCEC.  In
> >>>>>> some non-native cases in which there is no OS-visible device 
> >>>>>> associated
> >>>>>> with the RCiEP, there is nothing to act upon as the firmware is 
> >>>>>> acting
> >>>>>> before the OS.
> >>>>>>
> >>>>>> Add handling for the linked RCEC in AER/ERR while taking into 
> >>>>>> account
> >>>>>> non-native cases.
> >>>>>>
> >>>>>> Co-developed-by: Sean V Kelley <[e-mail]>
> >>>>>> Link: 
> >>>>>> https://lore.kernel.org/r/[e-mail]
> >>>>>> Signed-off-by: Sean V Kelley <[e-mail]>
> >>>>>> Signed-off-by: Qiuxu Zhuo <[e-mail]>
> >>>>>> Signed-off-by: Bjorn Helgaas <[e-mail]>
> >>>>>> Reviewed-by: Jonathan Cameron <[e-mail]>
> >>>>>> ---
> >>>>>>  drivers/pci/pcie/aer.c | 53 
> >>>>>> ++++++++++++++++++++++++++++++------------
> >>>>>>  drivers/pci/pcie/err.c | 20 ++++++++--------
> >>>>>>  2 files changed, 48 insertions(+), 25 deletions(-)
> >>>>>>
> >>>>>> diff --git a/drivers/pci/pcie/aer.c b/drivers/pci/pcie/aer.c
> >>>>>> index 65dff5f3457a..083f69b67bfd 100644
> >>>>>> --- a/drivers/pci/pcie/aer.c
> >>>>>> +++ b/drivers/pci/pcie/aer.c
> >>>>>> @@ -1357,27 +1357,50 @@ static int aer_probe(struct pcie_device 
> >>>>>> *dev)
> >>>>>>   */
> >>>>>>  static pci_ers_result_t aer_root_reset(struct pci_dev *dev)
> >>>>>>  {
> >>>>>> -   int aer = dev->aer_cap;
> >>>>>> +   int type = pci_pcie_type(dev);
> >>>>>> +   struct pci_dev *root;
> >>>>>> +   int aer = 0;
> >>>>>> +   int rc = 0;
> >>>>>>     u32 reg32;
> >>>>>> -   int rc;
> >>>>>>
> >>>>>> +   if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)  
> >>>>>
> >>>>> "type == PCI_EXP_TYPE_RC_END"
> >>>>>  
> >>>>>> +           /*
> >>>>>> +            * The reset should only clear the Root 
> >>>>>> Error Status
> >>>>>> +            * of the RCEC. Only perform this for the
> >>>>>> +            * native case, i.e., an RCEC is present.
> >>>>>> +            */
> >>>>>> +           root = dev->rcec;
> >>>>>> +   else
> >>>>>> +           root = dev;
> >>>>>>
> >>>>>> -   /* Disable Root's interrupt in response to error messages 
> >>>>>> */
> >>>>>> -   pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, 
> >>>>>> &reg32);
> >>>>>> -   reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
> >>>>>> -   pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, 
> >>>>>> reg32);
> >>>>>> +   if (root)
> >>>>>> +           aer = dev->aer_cap;
> >>>>>>
> >>>>>> -   rc = pci_bus_error_reset(dev);
> >>>>>> -   pci_info(dev, "Root Port link has been reset\n");
> >>>>>> +   if (aer) {
> >>>>>> +           /* Disable Root's interrupt in response to 
> >>>>>> error messages */
> >>>>>> +           pci_read_config_dword(root, aer + 
> >>>>>> PCI_ERR_ROOT_COMMAND, &reg32);
> >>>>>> +           reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
> >>>>>> +           pci_write_config_dword(root, aer + 
> >>>>>> PCI_ERR_ROOT_COMMAND, reg32);  
> >>>>>
> >>>>> Not directly related to *this* patch, but my assumption was that 
> >>>>> in
> >>>>> the APEI case, the firmware should retain ownership of the AER
> >>>>> Capability, so the OS should not touch PCI_ERR_ROOT_COMMAND and
> >>>>> PCI_ERR_ROOT_STATUS.
> >>>>>
> >>>>> But this code appears to ignore that ownership.  Jonathan, you 
> >>>>> must
> >>>>> have looked at this recently for 068c29a248b6 ("PCI/ERR: Clear 
> >>>>> PCIe
> >>>>> Device Status errors only if OS owns AER").  Do you have any 
> >>>>> insight
> >>>>> about this?
> >>>>>  
> >>>>>> -   /* Clear Root Error Status */
> >>>>>> -   pci_read_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, 
> >>>>>> &reg32);
> >>>>>> -   pci_write_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, 
> >>>>>> reg32);
> >>>>>> +           /* Clear Root Error Status */
> >>>>>> +           pci_read_config_dword(root, aer + 
> >>>>>> PCI_ERR_ROOT_STATUS, &reg32);
> >>>>>> +           pci_write_config_dword(root, aer + 
> >>>>>> PCI_ERR_ROOT_STATUS, reg32);
> >>>>>>
> >>>>>> -   /* Enable Root Port's interrupt in response to error 
> >>>>>> messages */
> >>>>>> -   pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, 
> >>>>>> &reg32);
> >>>>>> -   reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
> >>>>>> -   pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, 
> >>>>>> reg32);
> >>>>>> +           /* Enable Root Port's interrupt in response 
> >>>>>> to error messages */
> >>>>>> +           pci_read_config_dword(root, aer + 
> >>>>>> PCI_ERR_ROOT_COMMAND, &reg32);
> >>>>>> +           reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
> >>>>>> +           pci_write_config_dword(root, aer + 
> >>>>>> PCI_ERR_ROOT_COMMAND, reg32);
> >>>>>> +   }
> >>>>>> +
> >>>>>> +   if ((type == PCI_EXP_TYPE_RC_EC) || (type == 
> >>>>>> PCI_EXP_TYPE_RC_END)) {
> >>>>>> +           if (pcie_has_flr(root)) {
> >>>>>> +                   rc = pcie_flr(root);
> >>>>>> +                   pci_info(dev, "has been 
> >>>>>> reset (%d)\n", rc);
> >>>>>> +           }
> >>>>>> +   } else {
> >>>>>> +           rc = pci_bus_error_reset(root);  
> >>>>>
> >>>>> Don't we want "dev" for both the FLR and pci_bus_error_reset()?  
> >>>>> I
> >>>>> think "root == dev" except when dev is an RCiEP.  When dev is an
> >>>>> RCiEP, "root" is the RCEC (if present), and we want to reset the
> >>>>> RCiEP, not the RCEC.
> >>>>>  
> >>>>>> +           pci_info(dev, "Root Port link has been 
> >>>>>> reset (%d)\n", rc);
> >>>>>> +   }  
> >>>>>
> >>>>> There are a couple changes here that I think should be split out.
> >>>>>
> >>>>> Based on my theory that when firmware retains control of AER, the 
> >>>>> OS
> >>>>> should not touch PCI_ERR_ROOT_COMMAND and PCI_ERR_ROOT_STATUS, and 
> >>>>> any
> >>>>> updates to them would have to be done by firmware before we get 
> >>>>> here,
> >>>>> I suggested reordering this:
> >>>>>
> >>>>>   - clear PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
> >>>>>   - do reset
> >>>>>   - clear PCI_ERR_ROOT_STATUS (for APEI, presumably done by 
> >>>>> firmware?)
> >>>>>   - enable PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
> >>>>>
> >>>>> to this:
> >>>>>
> >>>>>   - clear PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
> >>>>>   - clear PCI_ERR_ROOT_STATUS
> >>>>>   - enable PCI_ERR_ROOT_COMMAND ROOT_PORT_INTR_ON_MESG_MASK
> >>>>>   - do reset
> >>>>>
> >>>>> If my theory is correct, I think we should still reorder this, 
> >>>>> but:
> >>>>>
> >>>>>   - It's a significant behavior change that deserves its own 
> >>>>> patch so
> >>>>>     we can document/bisect/revert.
> >>>>>
> >>>>>   - I'm not sure why we clear the PCI_ERR_ROOT_COMMAND error 
> >>>>> reporting
> >>>>>     bits.  In the new "clear COMMAND, clear STATUS, enable 
> >>>>> COMMAND"
> >>>>>     order, it looks superfluous.  There's no reason to disable 
> >>>>> error
> >>>>>     reporting while clearing the status bits.
> >>>>>
> >>>>>     The current "clear, reset, enable" order suggests that the 
> >>>>> reset
> >>>>>     might cause errors that we should ignore.  I don't know 
> >>>>> whether
> >>>>>     that's the case or not.  It dates from 6c2b374d7485 
> >>>>> ("PCI-Express
> >>>>>     AER implemetation: AER core and aerdriver"), which doesn't
> >>>>>     elaborate.
> >>>>>
> >>>>>   - Should we also test for OS ownership of AER before touching
> >>>>>     PCI_ERR_ROOT_STATUS?
> >>>>>
> >>>>>   - If we remove the PCI_ERR_ROOT_COMMAND fiddling (and I 
> >>>>> tentatively
> >>>>>     think we *should* unless we can justify it), that would 
> >>>>> also
> >>>>>     deserve its own patch.  Possibly (1) remove 
> >>>>> PCI_ERR_ROOT_COMMAND
> >>>>>     fiddling, (2) reorder PCI_ERR_ROOT_STATUS clearing and 
> >>>>> reset, (3)
> >>>>>     test for OS ownership of AER (?), (4) the rest of this 
> >>>>> patch.
> >>>>>  
> >>>>>>     return rc ? PCI_ERS_RESULT_DISCONNECT : 
> >>>>>> PCI_ERS_RESULT_RECOVERED;
> >>>>>>  }
> >>>>>> diff --git a/drivers/pci/pcie/err.c b/drivers/pci/pcie/err.c
> >>>>>> index 7883c9791562..cbc5abfe767b 100644
> >>>>>> --- a/drivers/pci/pcie/err.c
> >>>>>> +++ b/drivers/pci/pcie/err.c
> >>>>>> @@ -148,10 +148,10 @@ static int report_resume(struct pci_dev 
> >>>>>> *dev, void *data)
> >>>>>>
> >>>>>>  /**
> >>>>>>   * pci_walk_bridge - walk bridges potentially AER affected
> >>>>>> - * @bridge:        bridge which may be a Port, an RCEC 
> >>>>>> with associated RCiEPs,
> >>>>>> - *         or an RCiEP associated with an RCEC
> >>>>>> - * @cb:            callback to be called for each 
> >>>>>> device found
> >>>>>> - * @userdata:      arbitrary pointer to be passed to 
> >>>>>> callback
> >>>>>> + * @bridge   bridge which may be an RCEC with associated 
> >>>>>> RCiEPs,
> >>>>>> + *           or a Port.
> >>>>>> + * @cb       callback to be called for each device found
> >>>>>> + * @userdata arbitrary pointer to be passed to callback.
> >>>>>>   *
> >>>>>>   * If the device provided is a bridge, walk the subordinate 
> >>>>>> bus, including
> >>>>>>   * any bridged devices on buses under this bus.  Call the 
> >>>>>> provided callback
> >>>>>> @@ -164,8 +164,14 @@ static void pci_walk_bridge(struct pci_dev 
> >>>>>> *bridge,
> >>>>>>                         int (*cb)(struct 
> >>>>>> pci_dev *, void *),
> >>>>>>                         void *userdata)
> >>>>>>  {
> >>>>>> +   /*
> >>>>>> +    * In a non-native case where there is no OS-visible 
> >>>>>> reporting
> >>>>>> +    * device the bridge will be NULL, i.e., no RCEC, no 
> >>>>>> Downstream Port.
> >>>>>> +    */
> >>>>>>     if (bridge->subordinate)
> >>>>>>             pci_walk_bus(bridge->subordinate, cb, 
> >>>>>> userdata);
> >>>>>> +   else if (bridge->rcec)
> >>>>>> +           cb(bridge->rcec, userdata);
> >>>>>>     else
> >>>>>>             cb(bridge, userdata);
> >>>>>>  }
> >>>>>> @@ -194,12 +200,6 @@ pci_ers_result_t pcie_do_recovery(struct 
> >>>>>> pci_dev *dev,
> >>>>>>     pci_dbg(bridge, "broadcast error_detected message\n");
> >>>>>>     if (state == pci_channel_io_frozen) {
> >>>>>>             pci_walk_bridge(bridge, 
> >>>>>> report_frozen_detected, &status);
> >>>>>> -           if (type == PCI_EXP_TYPE_RC_END) {
> >>>>>> -                   pci_warn(dev, "subordinate 
> >>>>>> device reset not possible for RCiEP\n");
> >>>>>> -                   status = 
> >>>>>> PCI_ERS_RESULT_NONE;
> >>>>>> -                   goto failed;
> >>>>>> -           }
> >>>>>> -
> >>>>>>             status = reset_subordinates(bridge);
> >>>>>>             if (status != PCI_ERS_RESULT_RECOVERED) {
> >>>>>>                     pci_warn(bridge, 
> >>>>>> "subordinate device reset failed\n");
> >>>>>> -- 
> >>>>>> 2.28.0
> >>>>>>  
> > -- 
> > Sathyanarayanan Kuppuswamy
> > Linux Kernel Developer

Bjorn Helgaas Oct. 20, 2020, 4:28 p.m. UTC | #12

On Tue, Oct 20, 2020 at 01:59:20PM +0100, Jonathan Cameron wrote:
> On Mon, 19 Oct 2020 13:50:17 -0700
> Sean V Kelley <[e-mail]> wrote:
> > On 19 Oct 2020, at 11:59, Kuppuswamy, Sathyanarayanan wrote:
> > > On 10/19/20 11:31 AM, Sean V Kelley wrote:  
> > >> On 19 Oct 2020, at 3:49, Ethan Zhao wrote:
> > >>> On Sat, Oct 17, 2020 at 6:29 AM Bjorn Helgaas <[e-mail]> 
> > >>> wrote:  
> > >>>>
> > >>>> [+cc Christoph, Ethan, Sinan, Keith; sorry should have cc'd you to
> > >>>> begin with since you're looking at this code too. Particularly
> > >>>> interested in your thoughts about whether we should be touching
> > >>>> PCI_ERR_ROOT_COMMAND and PCI_ERR_ROOT_STATUS when we don't own 
> > >>>> AER.]  
> > >>>
> > >>> aer_root_reset() function has a prefix  'aer_', looks like it's a
> > >>> function of aer driver, will
> > >>> only be called by aer driver at runtime. if so it's up to the
> > >>> owner/aer to know if OSPM is
> > >>> granted to init. while actually some of the functions and runtime 
> > >>> service of
> > >>> aer driver is also shared by GHES driver (running time) and DPC 
> > >>> driver
> > >>> (compiling time ?)
> > >>> etc. then it is confused now.
> > >>>
> > >>> Shall we move some of the shared functions and running time service 
> > >>> to
> > >>> pci/err.c ?
> > >>> if so , just like pcie_do_recovery(), it's share by firmware_first  
> > >>> mode GHES
> > >>> ghes_probe()  
> > >>> ->ghes_irq_func  
> > >>>   ->ghes_proc
> > >>>     ->ghes_do_proc()
> > >>>       ->ghes_handle_aer()
> > >>>         ->aer_recover_work_func()
> > >>>           ->pcie_do_recovery()
> > >>>             ->aer_root_reset()
> > >>>
> > >>> and aer driver etc.  if aer wants to do some access might conflict
> > >>> with firmware(or
> > >>> firmware in embedded controller) should check _OSC_ etc first. 
> > >>> blindly issue
> > >>> PCI_ERR_ROOT_COMMAND  or clear PCI_ERR_ROOT_STATUS *likely*
> > >>> cause errors by error handling itself.  
> > >>
> > >> If _OSC negotiation ends up with FW being in control of AER, that 
> > >> means OS is not in charge and should not be messing with AER I guess. 
> > >> That seems appropriate to me then.  
> > >
> > > But APEI based notification is more like a hybrid approach (frimware 
> > > first detects the
> > > error and notifies OS). Since spec does not clarify what OS is allowed 
> > > to do, its bit of a
> > > gray area now. My point is, since firmware allows OS to process the 
> > > error by sending
> > > the notification, I think its OK to clear the status once the error is 
> > > handled.  
> > 
> > I don’t disagree as long as AER is granted to the OS via _OSC. But if 
> > it’s not granted explicitly via _OSC even in the APEI case where 
> > it’s either an SCI or NMI and not an MSI, I’m unsure whether the OS 
> > should be touching those registers.
> 
> My assumption was indeed this.  If AER hasn't been granted to the OS,
> it shouldn't be doing anything involving AER itself.  It should constrain
> itself to dealing with the End Points etc due to the need there for
> driver interaction.
> 
> I fully agree with the comment that the specifications aren't entirely
> clear on these cases.
> 
> It is possible that no one is currently generating the particular
> combination of severity bits in the APEI path to actually hit this.
> It requires the outer record to be marked recoverable, but the inner
> part to be marked fatal.  Kind of an odd mix.
> 
> In the GHES case, you get to this path by having a
> Generic Error Status Block - recoverable (must not be fatal to avoid panic()
> in APEI layer) containing one more more Generic Error Blocks, one of
> which is fatal.
> 
> Response of our firmware team is that this particularly combination is
> probably crazy.

Thanks a lot for researching this and outlining these details.  I
hadn't worked all that out.  It makes me a lot less worried about
breaking something if we tweak this.

> So good to clean up this corner, but it is probably not a problem
> anyone has actually hit so far.

IMO if the OS has not been granted AER control via _OSC, it shouldn't
be touching these registers.  I don't want to speculate based on what
the intent might have been with APEI.  If the intent was that the OS
should write those registers even if it doesn't own the AER
capability, that could easily have been put in the spec.

IIUC APEI enables cases where the device with Root Error Command and
Root Error Status registers (or device-specific equivalents) is not
even visible to the OS.  In those cases the OS *cannot* fiddle with
them.

I assume APEI tells us about an error with an Endpoint.  I do not
think we should be groping around for an an upstream device and poking
things in it.  I'm not even 100% comfortable with the fact that we
find an upstream device and reset all devices below it.  Maybe that's
OK as a "bigger hammer," but I don't know about it being the default
approach.

I was hoping to get this series merged for v5.10, but I don't think
it's really practical to merge it at this stage with four days left in
the merge window.  When we do merge this, I propose tightening up
aer_root_reset() along these lines as preliminary patches, since this
has nothing to do with RCEC/RCiEP, and if they *do* cause any issues,
I don't want these patches to be implicated.

Bjorn

Patch

diff --git a/drivers/pci/pcie/aer.c b/drivers/pci/pcie/aer.c
index 65dff5f3457a..083f69b67bfd 100644
--- a/drivers/pci/pcie/aer.c
+++ b/drivers/pci/pcie/aer.c
@@ -1357,27 +1357,50 @@  static int aer_probe(struct pcie_device *dev)
  */
 static pci_ers_result_t aer_root_reset(struct pci_dev *dev)
 {
-	int aer = dev->aer_cap;
+	int type = pci_pcie_type(dev);
+	struct pci_dev *root;
+	int aer = 0;
+	int rc = 0;
 	u32 reg32;
-	int rc;
 
+	if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
+		/*
+		 * The reset should only clear the Root Error Status
+		 * of the RCEC. Only perform this for the
+		 * native case, i.e., an RCEC is present.
+		 */
+		root = dev->rcec;
+	else
+		root = dev;
 
-	/* Disable Root's interrupt in response to error messages */
-	pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, &reg32);
-	reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
-	pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, reg32);
+	if (root)
+		aer = dev->aer_cap;
 
-	rc = pci_bus_error_reset(dev);
-	pci_info(dev, "Root Port link has been reset\n");
+	if (aer) {
+		/* Disable Root's interrupt in response to error messages */
+		pci_read_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, &reg32);
+		reg32 &= ~ROOT_PORT_INTR_ON_MESG_MASK;
+		pci_write_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, reg32);
 
-	/* Clear Root Error Status */
-	pci_read_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, &reg32);
-	pci_write_config_dword(dev, aer + PCI_ERR_ROOT_STATUS, reg32);
+		/* Clear Root Error Status */
+		pci_read_config_dword(root, aer + PCI_ERR_ROOT_STATUS, &reg32);
+		pci_write_config_dword(root, aer + PCI_ERR_ROOT_STATUS, reg32);
 
-	/* Enable Root Port's interrupt in response to error messages */
-	pci_read_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, &reg32);
-	reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
-	pci_write_config_dword(dev, aer + PCI_ERR_ROOT_COMMAND, reg32);
+		/* Enable Root Port's interrupt in response to error messages */
+		pci_read_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, &reg32);
+		reg32 |= ROOT_PORT_INTR_ON_MESG_MASK;
+		pci_write_config_dword(root, aer + PCI_ERR_ROOT_COMMAND, reg32);
+	}
+
+	if ((type == PCI_EXP_TYPE_RC_EC) || (type == PCI_EXP_TYPE_RC_END)) {
+		if (pcie_has_flr(root)) {
+			rc = pcie_flr(root);
+			pci_info(dev, "has been reset (%d)\n", rc);
+		}
+	} else {
+		rc = pci_bus_error_reset(root);
+		pci_info(dev, "Root Port link has been reset (%d)\n", rc);
+	}
 
 	return rc ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
 }
diff --git a/drivers/pci/pcie/err.c b/drivers/pci/pcie/err.c
index 7883c9791562..cbc5abfe767b 100644
--- a/drivers/pci/pcie/err.c
+++ b/drivers/pci/pcie/err.c
@@ -148,10 +148,10 @@  static int report_resume(struct pci_dev *dev, void *data)
 
 /**
  * pci_walk_bridge - walk bridges potentially AER affected
- * @bridge:	bridge which may be a Port, an RCEC with associated RCiEPs,
- *		or an RCiEP associated with an RCEC
- * @cb:		callback to be called for each device found
- * @userdata:	arbitrary pointer to be passed to callback
+ * @bridge   bridge which may be an RCEC with associated RCiEPs,
+ *           or a Port.
+ * @cb       callback to be called for each device found
+ * @userdata arbitrary pointer to be passed to callback.
  *
  * If the device provided is a bridge, walk the subordinate bus, including
  * any bridged devices on buses under this bus.  Call the provided callback
@@ -164,8 +164,14 @@  static void pci_walk_bridge(struct pci_dev *bridge,
 			    int (*cb)(struct pci_dev *, void *),
 			    void *userdata)
 {
+	/*
+	 * In a non-native case where there is no OS-visible reporting
+	 * device the bridge will be NULL, i.e., no RCEC, no Downstream Port.
+	 */
 	if (bridge->subordinate)
 		pci_walk_bus(bridge->subordinate, cb, userdata);
+	else if (bridge->rcec)
+		cb(bridge->rcec, userdata);
 	else
 		cb(bridge, userdata);
 }
@@ -194,12 +200,6 @@  pci_ers_result_t pcie_do_recovery(struct pci_dev *dev,
 	pci_dbg(bridge, "broadcast error_detected message\n");
 	if (state == pci_channel_io_frozen) {
 		pci_walk_bridge(bridge, report_frozen_detected, &status);
-		if (type == PCI_EXP_TYPE_RC_END) {
-			pci_warn(dev, "subordinate device reset not possible for RCiEP\n");
-			status = PCI_ERS_RESULT_NONE;
-			goto failed;
-		}
-
 		status = reset_subordinates(bridge);
 		if (status != PCI_ERS_RESULT_RECOVERED) {
 			pci_warn(bridge, "subordinate device reset failed\n");