On 2018-01-12 01:22, Gustavo Lima Chaves wrote:
> This is a follow-up on
> https://groups.google.com/d/topic/jailhouse-dev/xIrkxRn1cOQ/discussion
> 
> This will enable Linux inmates with CONFIG_PCIEAER=y (default *is* y)
> not to barf when PCI devices are partitioned.
> 
> Then, it will enable more PCI drivers to function well during device
> handover from root to inmates, when their .remove() functions may
> write to the PCIe capability (previously read-only).
> 
> Next, it will suppress AER reporting on all functions being
> partitioned from root, since those reports would escalate to the root
> and jeopardize the freedom from interference (FFI) between cells.
> 
> On top of that, we now adhere to a more spec-friendly function level
> reset (FLR) routine on functions being handed over, also checking if
> they support FLR altogether in the first place. The original code
> would attempt to reset functions with no checks, what could lead to
> unsupported requests on them. Those are PCI-level and will reach their
> root complex, again bringing down FFI.
> 

Thanks for the patches. I commented on some details, but the general
direction looks good.

> Finally, we noticed that FLR was being issued repeatedly, what seemed
> redundant, now fixed.

Except for that: We need to silence PCI devices (reset) when removing
them from the root cell and adding them to a different address space AND
we also need to reset them when a cell is restarted. That leads to a
duplicate reset during the first cell startup, but I doubt that this is
a major issue, is it?

Jan

-- 
Siemens AG, Corporate Technology, CT RDA IOT SES-DE
Corporate Competence Center Embedded Linux

-- 
You received this message because you are subscribed to the Google Groups 
"Jailhouse" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
For more options, visit https://groups.google.com/d/optout.

Reply via email to