On 06/25/2014 08:03 AM, Gavin Shan wrote:

On Tue, Jun 24, 2014 at 05:00:52PM +1000, Benjamin Herrenschmidt wrote:

quoted

On Tue, 2014-06-24 at 14:57 +0800, Mike Qiu wrote:

quoted

Is that mean *host* side error injection should base on
"CONFIG_IOMMU_API" ? If it is just host side(no guest, no pass through),
can't we do error inject?

Maybe I misunderstand :)

Ah no, make different patches, we don't want to use IOMMU group ID, just
PE numbers. Maybe we should expose in sysfs the PEs from the platform
code with the error injection files underneath ...

Yeah, "errinjct" needs grab PCI_domain_nr+PE number from sysfs. We
already had PE number sysfs file:

[root@ltcfbl8eb 0000:01:00.1]# pwd
/sys/bus/pci/devices/0000:01:00.1
[root@ltcfbl8eb 0000:01:00.1]# cat eeh_pe_config_addr
0x1

For guest support, we will rely on VFIO group ioctl command, which
naturally depends on pass-through.

---

We probably implement it like this. If there're anything wrong, please
correct me:

- Introduce EEH callback struct eeh_ops::err_inject(), which will be
   implemented for PowerNV (NULL for pSeries) by calling the PCI error
   injection dedicated OPAL API (opal_pci_err_inject()).
- Introduce global function eeh.c::eeh_err_inject(), which calls to
   eeh_ops::err_inject() and newly introduced VFIO EEH operation
   will be implemented based on this function.
- Introduce debugfs /sys/kernel/debug/powerpc/PCIxxxx/errinjct, which

Here maybe  "/sys/kernel/debug/powerpc/errinjct" is better, because it 
will supply "PCI_domain_nr" in parameters, so no need supply errinjct 
for each PCI domain.

Another reason is error inject not only for PCI(in future), so better 
not in PCI domain entry.

Also it simple for userland tools to has a fixed path.

Thanks
Mike

   receives PCI error injection parameters from "errinjct". It could
   have format: "ei_token:addr:mask:PCI_domain_nr:PE_num:function".
   Eventually, eeh_err_inject() is invoked to call the corresponding
   OPAL API.

Thanks,
Gavin