Re: [PATCH] Device suspend/resume fixes

[Todd Poynor <tpoynor@mvista.com>]

Patrick Mochel wrote:
>>(3) Flush signals between resume handlers in case a resume function
>>causes, for example, an ECHILD from modprobe or hotplug, so
>>interruptible APIs for the next handler aren't affected.
> 
> 
> Are there really cases in which you see this as an issue?

Ugh, I meant SIGCHILD, not ECHILD.  We have seen this in the past on a 
2.4 backport, although perhaps in a different environment than is 
normally used (this was on a PowerPC embedded platform that powers off 
the I/O ring during suspend).  For certain hotpluggable devices such as 
USB we've found it best to completely de-init and re-init the controller 
and re-sense plugged cards at suspend/resume time, which is when we saw 
troublesome signals (I2C setup then "timed out" before the controller 
was ready because a SIGCHILD was waiting).  I admit I don't know that 
much about the hotplug handlers, maybe I just didn't have things setup 
properly.  So if this seems bizarre then I'll withdraw it and try things 
again later when I have such hardware running in 2.6.  Thanks,

-- 
Todd Poynor
MontaVista Software