RE: [RFC PATCH v2 2/3] ACPI: extlog: Trace CPER PCI Express Error Section

[Dan Williams]

Fabio M. De Francesco wrote:
> Currently, extlog_print() (ELOG) only reports CPER PCIe section (UEFI
> v2.10, Appendix N.2.7) to the kernel log via print_extlog_rcd(). Instead,
> the similar ghes_do_proc() (GHES) prints to kernel log and calls
> pci_print_aer() to report via the ftrace infrastructure.
> 
> Add support to report the CPER PCIe Error section also via the ftrace
> infrastructure by calling pci_print_aer() to make ELOG act consistently
> with GHES.
> 
> Cc: Dan Williams <dan.j.williams@xxxxxxxxx>
> Signed-off-by: Fabio M. De Francesco <fabio.m.de.francesco@xxxxxxxxxxxxxxx>
> ---
>  drivers/acpi/acpi_extlog.c | 35 +++++++++++++++++++++++++++++++++++
>  1 file changed, 35 insertions(+)
> 
> diff --git a/drivers/acpi/acpi_extlog.c b/drivers/acpi/acpi_extlog.c
> index 4e62d7235d33..fb7b0c73f86a 100644
> --- a/drivers/acpi/acpi_extlog.c
> +++ b/drivers/acpi/acpi_extlog.c
> @@ -131,6 +131,36 @@ static int print_extlog_rcd(const char *pfx,
>  	return 1;
>  }
>  
> +static void extlog_print_pcie(struct cper_sec_pcie *pcie_err,
> +			      int severity)
> +{
> +#ifdef CONFIG_ACPI_APEI_PCIEAER

Whenever possible do not use ifdef in C files, see "21) Conditional Compilation"
in coding style.

I suspect the issue here is that pci_print_aer() is not always defined, but the
solution there is something like:

diff --git a/include/linux/aer.h b/include/linux/aer.h
index 4b97f38f3fcf..6ff54197480d 100644
--- a/include/linux/aer.h
+++ b/include/linux/aer.h
@@ -42,16 +42,20 @@ int pcie_read_tlp_log(struct pci_dev *dev, int where, struct pcie_tlp_log *log);
 #if defined(CONFIG_PCIEAER)
 int pci_aer_clear_nonfatal_status(struct pci_dev *dev);
 int pcie_aer_is_native(struct pci_dev *dev);
+void pci_print_aer(struct pci_dev *dev, int aer_severity,
+                   struct aer_capability_regs *aer);
 #else
 static inline int pci_aer_clear_nonfatal_status(struct pci_dev *dev)
 {
        return -EINVAL;
 }
 static inline int pcie_aer_is_native(struct pci_dev *dev) { return 0; }
+static inline void pci_print_aer(struct pci_dev *dev, int aer_severity,
+                                struct aer_capability_regs *aer)
+{
+}
 #endif
 
-void pci_print_aer(struct pci_dev *dev, int aer_severity,
-                   struct aer_capability_regs *aer);
 int cper_severity_to_aer(int cper_severity);
 void aer_recover_queue(int domain, unsigned int bus, unsigned int devfn,
                       int severity, struct aer_capability_regs *aer_regs);


> +	struct aer_capability_regs *aer;
> +	struct pci_dev *pdev;
> +	unsigned int devfn;
> +	unsigned int bus;
> +	int aer_severity;
> +	int domain;
> +
> +	if (pcie_err->validation_bits & CPER_PCIE_VALID_DEVICE_ID &&
> +	    pcie_err->validation_bits & CPER_PCIE_VALID_AER_INFO) {
> +		aer_severity = cper_severity_to_aer(severity);
> +		aer = (struct aer_capability_regs *)pcie_err->aer_info;
> +		domain = pcie_err->device_id.segment;
> +		bus = pcie_err->device_id.bus;
> +		devfn = PCI_DEVFN(pcie_err->device_id.device,
> +				  pcie_err->device_id.function);
> +		pdev = pci_get_domain_bus_and_slot(domain, bus, devfn);

This pci_get_domain_bus_and_slot() takes a refernce on @pdev that reference
needs to be dropped after operations on @pdev are complete.

> +		if (!pdev) {
> +			pr_err("no pci_dev for %04x:%02x:%02x.%x\n",
> +			       domain, bus, PCI_SLOT(devfn), PCI_FUNC(devfn));

I am not sure this error message is useful. I see this was copied from
aer_recover_work_func(). I wonder if a better approach is to just teach
pci_print_aer() to be tolerant of a NULL @pdev argument.

Most of what pci_print_aer() is just a device name which can be recreated as:

sprintf(name, "%04x:%02x:%02x.%d", domain, bus, PCI_SLOT(dev->devfn),
        PCI_FUNC(dev->devfn));

For now, I would just return if a device cannot be found, and skip the print.

> +	}
> +#endif
> +}
> +
>  static int extlog_print(struct notifier_block *nb, unsigned long val,
>  			void *data)
>  {
> @@ -182,6 +212,10 @@ static int extlog_print(struct notifier_block *nb, unsigned long val,
>  			if (gdata->error_data_length >= sizeof(*mem))
>  				trace_extlog_mem_event(mem, err_seq, fru_id, fru_text,
>  						       (u8)gdata->error_severity);
> +		} else if (guid_equal(sec_type, &CPER_SEC_PCIE)) {
> +			struct cper_sec_pcie *pcie_err = acpi_hest_get_payload(gdata);
> +
> +			extlog_print_pcie(pcie_err, gdata->error_severity);
>  		} else {
>  			void *err = acpi_hest_get_payload(gdata);
>  
> @@ -331,3 +365,4 @@ module_exit(extlog_exit);
>  MODULE_AUTHOR("Chen, Gong <gong.chen@xxxxxxxxx>");
>  MODULE_DESCRIPTION("Extended MCA Error Log Driver");
>  MODULE_LICENSE("GPL");
> +MODULE_IMPORT_NS(CXL);

This is awkward now that pci_print_aer() is consumed by a module that is not
"CXL", so perhaps we should either drop the namespacing of the pci_print_aer()
symbol or put it in its own namespace. I think the former is the way to go.