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

[Dave Jiang <dave.jiang@intel.com>]

On 9/11/25 9:47 AM, Bowman, Terry wrote:
> 
> 
> On 9/11/2025 10:41 AM, Dave Jiang wrote:
>>
>> On 9/11/25 7:33 AM, Bowman, Terry wrote:
>>>
>>> On 8/28/2025 7:43 PM, Dave Jiang wrote:
>>>> On 8/26/25 6:35 PM, Terry Bowman wrote:
>>>>> The AER driver is now designed to forward CXL protocol errors to the CXL
>>>> I would rephrase it to:
>>>> The AER driver enqueues the CXL protocol error info to the created kfifo for the CXL driver to consume.
>>>>  
>>>>> driver. Update the CXL driver with functionality to dequeue the forwarded
>>>>> CXL error from the kfifo. Also, update the CXL driver to begin the protocol
>>>>> error handling processing using the work received from the FIFO.
>>>>>
>>>>> Update function cxl_proto_err_work_fn() to dequeue work forwarded by the
>>>>> AER service driver. This will begin the CXL protocol error processing with
>>>>> a call to cxl_handle_proto_error().
>>>>>
>>>>> Introduce logic to take the SBDF values from 'struct cxl_proto_error_info'
>>>>> and use in discovering the erring PCI device. The call to pci_get_domain_bus_and_slot()
>>>>> will return a reference counted 'struct pci_dev *'. This will serve as
>>>>> reference count to prevent releasing the CXL Endpoint's mapped RAS while
>>>>> handling the error. Use scope base __free() to put the reference count.
>>>>> This will change when adding support for CXL port devices in the future.
>>>>>
>>>>> Implement cxl_handle_proto_error() to differentiate between Restricted CXL
>>>>> Host (RCH) protocol errors and CXL virtual host (VH) protocol errors.
>>>>> Maintain the existing RCH handling. Export the AER driver's pcie_walk_rcec()
>>>>> allowing the CXL driver to walk the RCEC's secondary bus.
>>>>>
>>>>> VH correctable error (CE) processing will call the CXL CE handler. VH
>>>>> uncorrectable errors (UCE) will call cxl_do_recovery(), implemented as a
>>>>> stub for now and to be updated in future patch. Export pci_aer_clean_fatal_status()
>>>>> and pci_clean_device_status() used to clean up AER status after handling.
>>>>>
>>>>> Signed-off-by: Terry Bowman <terry.bowman@amd.com>
>>>>> Reviewed-by: Kuppuswamy Sathyanarayanan <sathyanarayanan.kuppuswamy@linux.intel.com>
>>>>>
>>>>> ---
>>>>> Changes in v10->v11:
>>>>> - Reword patch commit message to remove RCiEP details (Jonathan)
>>>>> - Add #include <linux/bitfield.h> (Terry)
>>>>> - is_cxl_rcd() - Fix short comment message wrap  (Jonathan)
>>>>> - is_cxl_rcd() - Combine return calls into 1  (Jonathan)
>>>>> - cxl_handle_proto_error() - Move comment earlier  (Jonathan)
>>>>> - Usse FIELD_GET() in discovering class code (Jonathan)
>>>>> - Remove BDF from cxl_proto_err_work_data. Use 'struct pci_dev *' (Dan)
>>>>> ---
>>>>>  drivers/cxl/core/ras.c  | 68 ++++++++++++++++++++++++++++++++++-------
>>>>>  drivers/pci/pci.c       |  1 +
>>>>>  drivers/pci/pci.h       |  7 -----
>>>>>  drivers/pci/pcie/aer.c  |  1 +
>>>>>  drivers/pci/pcie/rcec.c |  1 +
>>>>>  include/linux/aer.h     |  2 ++
>>>>>  include/linux/pci.h     | 10 ++++++
>>>>>  7 files changed, 72 insertions(+), 18 deletions(-)
>>>>>
>>>>> diff --git a/drivers/cxl/core/ras.c b/drivers/cxl/core/ras.c
>>>>> index b285448c2d9c..a2e95c49f965 100644
>>>>> --- a/drivers/cxl/core/ras.c
>>>>> +++ b/drivers/cxl/core/ras.c
>>>>> @@ -118,17 +118,6 @@ static void cxl_cper_prot_err_work_fn(struct work_struct *work)
>>>>>  }
>>>>>  static DECLARE_WORK(cxl_cper_prot_err_work, cxl_cper_prot_err_work_fn);
>>>>>  
>>>>> -int cxl_ras_init(void)
>>>>> -{
>>>>> -	return cxl_cper_register_prot_err_work(&cxl_cper_prot_err_work);
>>>>> -}
>>>>> -
>>>>> -void cxl_ras_exit(void)
>>>>> -{
>>>>> -	cxl_cper_unregister_prot_err_work(&cxl_cper_prot_err_work);
>>>>> -	cancel_work_sync(&cxl_cper_prot_err_work);
>>>>> -}
>>>>> -
>>>>>  static pci_ers_result_t cxl_handle_ras(struct device *dev, u64 serial, void __iomem *ras_base);
>>>>>  static void cxl_handle_cor_ras(struct device *dev, u64 serial, void __iomem *ras_base);
>>>>>  
>>>>> @@ -331,6 +320,10 @@ void cxl_endpoint_port_init_ras(struct cxl_port *ep)
>>>>>  }
>>>>>  EXPORT_SYMBOL_NS_GPL(cxl_endpoint_port_init_ras, "CXL");
>>>>>  
>>>>> +static void cxl_do_recovery(struct device *dev)
>>>>> +{
>>>>> +}
>>>>> +
>>>>>  static void cxl_handle_cor_ras(struct device *dev, u64 serial, void __iomem *ras_base)
>>>>>  {
>>>>>  	void __iomem *addr;
>>>>> @@ -472,3 +465,56 @@ pci_ers_result_t pci_error_detected(struct pci_dev *pdev,
>>>>>  	return rc;
>>>>>  }
>>>>>  EXPORT_SYMBOL_NS_GPL(pci_error_detected, "CXL");
>>>>> +
>>>>> +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.
>>>>
>>>> DJ
>>> Yes, this is called in the worker thread context. I added the pdev device locks
>>> later in cxl_report_error_detected() for the UCE case. I found it necessary to 
>>> put in this function and not in cxl_handle_proto_error() (here) because of the 
>>> traversing logic in the UCE handling flow where it needs to be locked but only
>>> exactly once. I didn't add for the CE because I wasn't certain a CE error was 
>>> enough reason to add a device lock. 
>>>
>>> The UCE flow is:
>>>
>>> cxl_handle_proto_error()
>>> --> cxl_do_recovery()
>>> ----> cxl_handle_proto_error() <--- Added device lock here because of topo traversing/iteration
>>>
>>>
>>> I tried adding a function checking for cxl_pci driver but ran into circular dependency 
>>> because the driver is defined in cxl_pci but called from cxl_core. I will revisit
>>> this again but need some ideas how to make that work as I expect it will require 
>>> some code moving. 
>> Is there a chance that the endpoint errors can just be handled via the standard AER flow via AER callback? Otherwise we may need to move the cxl_pci driver definition to core/pci.c in order to do this....
>>
>> DJ
> 
> I understand AER callbacks in your comment are the PCIe CE and UCE callback handlers. Calling 
> the AER handlers for the EPs would technically work because they are initialized to CXL 'PCIe' 
> callbacks already. The AER callbacks are used for handling EP fatal UCE errors.
> 
> One problem is this approach isn't consistent with the plan to use CXL device based error 
> handling because it will now be using PCIe error handling for EP CE and UCE errors. 
> Also, CXL device handling would be used for for CXL port devices which would be implemented
> differently.

Maybe Dan has an opinion on this. To me it seems we are at the point where we need to move some of the PCI driver code to cxl/core in order to allow the pci driver checking in the error handling. Of course the region driver is in cxl/core already so there's precedent.

DJ

> 
> Terry 
> 
>>> Terry
>>>
>>>>> +	struct cxl_memdev *cxlmd = cxlds->cxlmd;
>>>>> +	struct device *host_dev __free(put_device) = get_device(&cxlmd->dev);
>>>>> +> +	if (err_info->severity == AER_CORRECTABLE) {
>>>>> +		int aer = pdev->aer_cap;
>>>>> +
>>>>> +		if (aer)
>>>>> +			pci_clear_and_set_config_dword(pdev,
>>>>> +						       aer + PCI_ERR_COR_STATUS,
>>>>> +						       0, PCI_ERR_COR_INTERNAL);
>>>>> +
>>>>> +		cxl_cor_error_detected(&cxlmd->dev);
>>>>> +
>>>>> +		pcie_clear_device_status(pdev);
>>>>> +	} else {
>>>>> +		cxl_do_recovery(&cxlmd->dev);
>>>>> +	}
>>>>> +}
>>>>> +
>>>>> +static void cxl_proto_err_work_fn(struct work_struct *work)
>>>>> +{
>>>>> +	struct cxl_proto_err_work_data wd;
>>>>> +
>>>>> +	while (cxl_proto_err_kfifo_get(&wd))
>>>>> +		cxl_handle_proto_error(&wd);
>>>>> +}
>>>>> +
>>>>> +static struct work_struct cxl_proto_err_work;
>>>>> +static DECLARE_WORK(cxl_proto_err_work, cxl_proto_err_work_fn);
>>>>> +
>>>>> +int cxl_ras_init(void)
>>>>> +{
>>>>> +	if (cxl_cper_register_prot_err_work(&cxl_cper_prot_err_work))
>>>>> +		pr_err("Failed to initialize CXL RAS CPER\n");
>>>>> +
>>>>> +	cxl_register_proto_err_work(&cxl_proto_err_work);
>>>>> +
>>>>> +	return 0;
>>>>> +}
>>>>> +
>>>>> +void cxl_ras_exit(void)
>>>>> +{
>>>>> +	cxl_cper_unregister_prot_err_work(&cxl_cper_prot_err_work);
>>>>> +	cancel_work_sync(&cxl_cper_prot_err_work);
>>>>> +
>>>>> +	cxl_unregister_proto_err_work();
>>>>> +	cancel_work_sync(&cxl_proto_err_work);
>>>>> +}
>>>>> diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
>>>>> index d775ed37a79b..2c9827690cb3 100644
>>>>> --- a/drivers/pci/pci.c
>>>>> +++ b/drivers/pci/pci.c
>>>>> @@ -2328,6 +2328,7 @@ void pcie_clear_device_status(struct pci_dev *dev)
>>>>>  	pcie_capability_read_word(dev, PCI_EXP_DEVSTA, &sta);
>>>>>  	pcie_capability_write_word(dev, PCI_EXP_DEVSTA, sta);
>>>>>  }
>>>>> +EXPORT_SYMBOL_NS_GPL(pcie_clear_device_status, "CXL");
>>>>>  #endif
>>>>>  
>>>>>  /**
>>>>> diff --git a/drivers/pci/pci.h b/drivers/pci/pci.h
>>>>> index cfa75903dd3f..69ff7c2d214f 100644
>>>>> --- a/drivers/pci/pci.h
>>>>> +++ b/drivers/pci/pci.h
>>>>> @@ -671,16 +671,10 @@ static inline bool pci_dpc_recovered(struct pci_dev *pdev) { return false; }
>>>>>  void pci_rcec_init(struct pci_dev *dev);
>>>>>  void pci_rcec_exit(struct pci_dev *dev);
>>>>>  void pcie_link_rcec(struct pci_dev *rcec);
>>>>> -void pcie_walk_rcec(struct pci_dev *rcec,
>>>>> -		    int (*cb)(struct pci_dev *, void *),
>>>>> -		    void *userdata);
>>>>>  #else
>>>>>  static inline void pci_rcec_init(struct pci_dev *dev) { }
>>>>>  static inline void pci_rcec_exit(struct pci_dev *dev) { }
>>>>>  static inline void pcie_link_rcec(struct pci_dev *rcec) { }
>>>>> -static inline void pcie_walk_rcec(struct pci_dev *rcec,
>>>>> -				  int (*cb)(struct pci_dev *, void *),
>>>>> -				  void *userdata) { }
>>>>>  #endif
>>>>>  
>>>>>  #ifdef CONFIG_PCI_ATS
>>>>> @@ -1022,7 +1016,6 @@ void pci_restore_aer_state(struct pci_dev *dev);
>>>>>  static inline void pci_no_aer(void) { }
>>>>>  static inline void pci_aer_init(struct pci_dev *d) { }
>>>>>  static inline void pci_aer_exit(struct pci_dev *d) { }
>>>>> -static inline void pci_aer_clear_fatal_status(struct pci_dev *dev) { }
>>>>>  static inline int pci_aer_clear_status(struct pci_dev *dev) { return -EINVAL; }
>>>>>  static inline int pci_aer_raw_clear_status(struct pci_dev *dev) { return -EINVAL; }
>>>>>  static inline void pci_save_aer_state(struct pci_dev *dev) { }
>>>>> diff --git a/drivers/pci/pcie/aer.c b/drivers/pci/pcie/aer.c
>>>>> index 627d89ccea9c..45abe1622316 100644
>>>>> --- a/drivers/pci/pcie/aer.c
>>>>> +++ b/drivers/pci/pcie/aer.c
>>>>> @@ -288,6 +288,7 @@ void pci_aer_clear_fatal_status(struct pci_dev *dev)
>>>>>  	if (status)
>>>>>  		pci_write_config_dword(dev, aer + PCI_ERR_UNCOR_STATUS, status);
>>>>>  }
>>>>> +EXPORT_SYMBOL_GPL(pci_aer_clear_fatal_status);
>>>>>  
>>>>>  /**
>>>>>   * pci_aer_raw_clear_status - Clear AER error registers.
>>>>> diff --git a/drivers/pci/pcie/rcec.c b/drivers/pci/pcie/rcec.c
>>>>> index d0bcd141ac9c..fb6cf6449a1d 100644
>>>>> --- a/drivers/pci/pcie/rcec.c
>>>>> +++ b/drivers/pci/pcie/rcec.c
>>>>> @@ -145,6 +145,7 @@ void pcie_walk_rcec(struct pci_dev *rcec, int (*cb)(struct pci_dev *, void *),
>>>>>  
>>>>>  	walk_rcec(walk_rcec_helper, &rcec_data);
>>>>>  }
>>>>> +EXPORT_SYMBOL_NS_GPL(pcie_walk_rcec, "CXL");
>>>>>  
>>>>>  void pci_rcec_init(struct pci_dev *dev)
>>>>>  {
>>>>> diff --git a/include/linux/aer.h b/include/linux/aer.h
>>>>> index f8eb32805957..1f79f0be4bf7 100644
>>>>> --- a/include/linux/aer.h
>>>>> +++ b/include/linux/aer.h
>>>>> @@ -66,12 +66,14 @@ struct cxl_proto_err_work_data {
>>>>>  
>>>>>  #if defined(CONFIG_PCIEAER)
>>>>>  int pci_aer_clear_nonfatal_status(struct pci_dev *dev);
>>>>> +void pci_aer_clear_fatal_status(struct pci_dev *dev);
>>>>>  int pcie_aer_is_native(struct pci_dev *dev);
>>>>>  #else
>>>>>  static inline int pci_aer_clear_nonfatal_status(struct pci_dev *dev)
>>>>>  {
>>>>>  	return -EINVAL;
>>>>>  }
>>>>> +static inline void pci_aer_clear_fatal_status(struct pci_dev *dev) { }
>>>>>  static inline int pcie_aer_is_native(struct pci_dev *dev) { return 0; }
>>>>>  #endif
>>>>>  
>>>>> diff --git a/include/linux/pci.h b/include/linux/pci.h
>>>>> index 3dcab36c437f..3407d687459d 100644
>>>>> --- a/include/linux/pci.h
>>>>> +++ b/include/linux/pci.h
>>>>> @@ -1804,6 +1804,9 @@ extern bool pcie_ports_native;
>>>>>  
>>>>>  int pcie_set_target_speed(struct pci_dev *port, enum pci_bus_speed speed_req,
>>>>>  			  bool use_lt);
>>>>> +void pcie_walk_rcec(struct pci_dev *rcec,
>>>>> +		    int (*cb)(struct pci_dev *, void *),
>>>>> +		    void *userdata);
>>>>>  #else
>>>>>  #define pcie_ports_disabled	true
>>>>>  #define pcie_ports_native	false
>>>>> @@ -1814,8 +1817,15 @@ static inline int pcie_set_target_speed(struct pci_dev *port,
>>>>>  {
>>>>>  	return -EOPNOTSUPP;
>>>>>  }
>>>>> +
>>>>> +static inline void pcie_walk_rcec(struct pci_dev *rcec,
>>>>> +				  int (*cb)(struct pci_dev *, void *),
>>>>> +				  void *userdata) { }
>>>>> +
>>>>>  #endif
>>>>>  
>>>>> +void pcie_clear_device_status(struct pci_dev *dev);
>>>>> +
>>>>>  #define PCIE_LINK_STATE_L0S		(BIT(0) | BIT(1)) /* Upstr/dwnstr L0s */
>>>>>  #define PCIE_LINK_STATE_L1		BIT(2)	/* L1 state */
>>>>>  #define PCIE_LINK_STATE_L1_1		BIT(3)	/* ASPM L1.1 state */
>