Re: [PATCH v11 18/23] PCI/AER: Dequeue forwarded CXL error

["Bowman, Terry" <terry.bowman@amd.com>]

On 8/29/2025 2:10 AM, Lukas Wunner wrote:
> On Thu, Aug 28, 2025 at 05:43:31PM -0700, Dave Jiang wrote:
>> On 8/26/25 6:35 PM, Terry Bowman wrote:
>>> +static void cxl_handle_proto_error(struct cxl_proto_err_work_data *err_info)
>>> +{
>>> +	struct pci_dev *pdev = err_info->pdev;
>>> +	struct cxl_dev_state *cxlds = pci_get_drvdata(pdev);
>> So this function is called from the workqueue thread to consume data
>> from the kfifo right? Do we need to take the device lock of the pdev
>> to ensure that a driver is bound to the device before we attempt to
>> retrieve the data? And do we also need to verify that the driver bound
>> is the cxl_pci driver (and not something like vfio_pci)? Otherwise I
>> think assuming the drv data is cxl_dev_state may cause crash.
> In v10 of this series, there used to be a cxl_pci_drv_bound() function
> to verify that the cxl_pci_driver is bound and not some other driver.
> That function was called from cxl_rch_handle_error_iter().
>
> It seems this is gone in v11?
>
> Thanks,
>
> Lukas
Hi Lukas,

Yes, this was removed due to a build issue. I am adding back with the fix.

You mentioned cxl_rch_handle_error_iter() above and I want to clarify my 
understanding that this is not needed in the RCH case. RCH handling includes 
traversal to find the first downstream PCI EP and calls the EP's 
pci_driver::err_handlers callbacks. These are part of pci_driver and therefore 
don't need the driver check as they will work for any bound pci_driver. 

The check is needed for VH EP CE and non-fatal UCE because they are handled 
by CXL error callbacks defined in the cxl_core/cxl_pci modules and not in the 
pci_driver::err_handlers.

Terry