Date:   Mon, 23 Nov 2020 23:57:35 +0000
From:   "Kelley, Sean V" <sean.v.kelley@...el.com>
To:     Bjorn Helgaas <helgaas@...nel.org>
CC:     "bhelgaas@...gle.com" <bhelgaas@...gle.com>,
        Jonathan Cameron <Jonathan.Cameron@...wei.com>,
        "xerces.zhao@...il.com" <xerces.zhao@...il.com>,
        "Wysocki, Rafael J" <rafael.j.wysocki@...el.com>,
        "Raj, Ashok" <ashok.raj@...el.com>,
        "Luck, Tony" <tony.luck@...el.com>,
        "Kuppuswamy, Sathyanarayanan" <sathyanarayanan.kuppuswamy@...el.com>,
        "Zhuo, Qiuxu" <qiuxu.zhuo@...el.com>,
        Linux PCI <linux-pci@...r.kernel.org>,
        "Linux Kernel Mailing List" <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH v12 10/15] PCI/ERR: Limit AER resets in pcie_do_recovery()

Hi Bjorn,

> On Nov 23, 2020, at 3:28 PM, Bjorn Helgaas <helgaas@...nel.org> wrote:
> 
> On Fri, Nov 20, 2020 at 04:10:31PM -0800, Sean V Kelley wrote:
>> In some cases a bridge may not exist as the hardware controlling may be
>> handled only by firmware and so is not visible to the OS. This scenario is
>> also possible in future use cases involving non-native use of RCECs by
>> firmware.
>> 
>> Explicitly apply conditional logic around these resets by limiting them to
>> Root Ports and Downstream Ports.
> 
> Can you help me understand this?  The subject says "Limit AER resets"
> and here you say "limit them to RPs and DPs", but it's not completely
> obvious how the resets are being limited, i.e., the patch doesn't add
> anything like:
> 
> +  if (type == PCI_EXP_TYPE_ROOT_PORT ||
> +      type == PCI_EXP_TYPE_DOWNSTREAM)
>      reset_subordinates(bridge);
> 
> It *does* add checks around pcie_clear_device_status(), but that also
> includes RC_EC.  And that's not a reset, so I don't think that's
> explicitly mentioned in the commit log.

The subject should have referred to the clearing of the device status rather than resets.
It originally came from this simpler patch in which I made use of reset instead of clear:

https://lore.kernel.org/linux-pci/20201002184735.1229220-8-seanvk.dev@oregontracks.org/

So a rephrase of clearing in place of resets would be more appropriate.

Then we added the notion of bridges…below

> 
> Also see the question below.
> 
>> Link: https://lore.kernel.org/r/20201002184735.1229220-8-seanvk.dev@oregontracks.org
>> Signed-off-by: Sean V Kelley <sean.v.kelley@...el.com>
>> Signed-off-by: Bjorn Helgaas <bhelgaas@...gle.com>
>> Acked-by: Jonathan Cameron <Jonathan.Cameron@...wei.com>
>> ---
>> drivers/pci/pcie/err.c | 31 +++++++++++++++++++++++++------
>> 1 file changed, 25 insertions(+), 6 deletions(-)
>> 
>> diff --git a/drivers/pci/pcie/err.c b/drivers/pci/pcie/err.c
>> index 8b53aecdb43d..7883c9791562 100644
>> --- a/drivers/pci/pcie/err.c
>> +++ b/drivers/pci/pcie/err.c
>> @@ -148,13 +148,17 @@ 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
>> + * @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
>>  *
>>  * If the device provided is a bridge, walk the subordinate bus, including
>>  * any bridged devices on buses under this bus.  Call the provided callback
>>  * on each device found.
>> + *
>> + * If the device provided has no subordinate bus, call the callback on the
>> + * device itself.
>>  */
>> static void pci_walk_bridge(struct pci_dev *bridge,
>> 			    int (*cb)(struct pci_dev *, void *),
>> @@ -162,6 +166,8 @@ static void pci_walk_bridge(struct pci_dev *bridge,
>> {
>> 	if (bridge->subordinate)
>> 		pci_walk_bus(bridge->subordinate, cb, userdata);
>> +	else
>> +		cb(bridge, userdata);
>> }
>> 
>> pci_ers_result_t pcie_do_recovery(struct pci_dev *dev,
>> @@ -174,10 +180,13 @@ pci_ers_result_t pcie_do_recovery(struct pci_dev *dev,
>> 
>> 	/*
>> 	 * Error recovery runs on all subordinates of the bridge.  If the
>> -	 * bridge detected the error, it is cleared at the end.
>> +	 * bridge detected the error, it is cleared at the end.  For RCiEPs
>> +	 * we should reset just the RCiEP itself.
>> 	 */
>> 	if (type == PCI_EXP_TYPE_ROOT_PORT ||
>> -	    type == PCI_EXP_TYPE_DOWNSTREAM)
>> +	    type == PCI_EXP_TYPE_DOWNSTREAM ||
>> +	    type == PCI_EXP_TYPE_RC_EC ||
>> +	    type == PCI_EXP_TYPE_RC_END)
>> 		bridge = dev;
>> 	else
>> 		bridge = pci_upstream_bridge(dev);
>> @@ -185,6 +194,12 @@ 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");
>> @@ -217,9 +232,13 @@ pci_ers_result_t pcie_do_recovery(struct pci_dev *dev,
>> 	pci_dbg(bridge, "broadcast resume message\n");
>> 	pci_walk_bridge(bridge, report_resume, &status);
>> 
>> -	if (pcie_aer_is_native(bridge))
>> -		pcie_clear_device_status(bridge);
>> -	pci_aer_clear_nonfatal_status(bridge);
>> +	if (type == PCI_EXP_TYPE_ROOT_PORT ||
>> +	    type == PCI_EXP_TYPE_DOWNSTREAM ||
>> +	    type == PCI_EXP_TYPE_RC_EC) {
>> +		if (pcie_aer_is_native(bridge))
>> +			pcie_clear_device_status(bridge);
>> +		pci_aer_clear_nonfatal_status(bridge);
> 
> This is hard to understand because "type" is from "dev", but "bridge"
> is not necessarily the same device.  Should it be this?
> 
>  type = pci_pcie_type(bridge);
>  if (type == PCI_EXP_TYPE_ROOT_PORT ||
>      ...)

Correct, it would be better if the type was based on the ‘bridge’.

Thanks,

Sean

> 
>> +	}
>> 	pci_info(bridge, "device recovery successful\n");
>> 	return status;
>> 
>> -- 
>> 2.29.2

Powered by blists - more mailing lists