Re: FreeBSD kernel doesn't boot on FUJITSU PRIMERGY RX200 S5 server

On Thursday 22 April 2010 4:07:37 pm Maxim Sobolev wrote:
> John Baldwin wrote:
> > On Thursday 22 April 2010 2:28:26 pm Maxim Sobolev wrote:
> >> John Baldwin wrote:
> >>> On Thursday 22 April 2010 6:05:04 am Maxim Sobolev wrote:
> >>>> Maxim Sobolev wrote:
> >>>>> There is already a code to detect non-existing AT keyboard and avoid 
> >>>>> attaching atkbd to it. The code is i386-only at the moment, I am 
trying 
> >>>>> to figure out how to modify it so that it works on amd64 as well.
> >>>> Looks like this huge delay is caused by the inb() being astonishingly 
> >>>> slow, which is not factored by the timeout routines. Reading keyboard 
> >>>> status port once takes about 0.003s! I am not sure if it's common 
> >>>> behaviour of the platform, or something specific to this particular 
> >>>> model. Do you know by any chance?
> >>> Well, many BIOSes trigger an SMI# when doing inb/outb to the keyboard 
ports so 
> >>> they can emulate a PS/2 keyboard when a USB keyboard is inserted.  Do 
you have 
> >>> any BIOS options related to the USB legacy compat?  I know of the 
Nehalem 
> >>> systems I've seen they have a separate option for controlling port 60/64 
> >>> emulation which we leave disabled by default.
> >> That makes sense. Unfortunately I don't have access to the BIOS 
> >> settings. This is a hosted system, and the provider keeps BIOS password 
> >> for themselves.
> >>
> >> I have a patch that fixes that issue by measuring status register 
> >> reading time first and then factoring it in the calculations of the 
> >> number of retries:
> >>
> >> http://sobomax.sippysoft.com/atkbdc.diff
> >>
> >> It also applies the same logic to detect broken/non-existing keyboard 
> >> controller to amd64 as we do to the i386. I'd appreciate if you can do a 
> >> review.
> > 
> > Hmm, not all i386 CPUs that we support have a TSC.  Is the change to
> > atkbdc_isa.c sufficient to fix the hang?  If so, I'd rather just commit 
that
> > bit and leave out the read_delay changes.
> 
> No, it's not sufficient. The problem here is that for some reason that 
> test passes on that system (probably emulation works) and so that normal 
> keyboard attach routine is invoked early in boot, when we don't even 
> have clock initialized. What if I make TSC-related changes amd64? Will 
> that be OK?

Hmm, I think you should definitely commit the atkbdc_isa.c change first of 
all.  I'm still thinking about the other change.  I wonder if we can figure 
out that a keyboard isn't present sooner somehow?  Do you know if the keyboard 
appears to be present but just slow vs if the keyboard is eventually found to 
not be present?

-- 
John Baldwin