Why is psr.ic off for ia64_pal_cache_flush?

Why is psr.ic off for ia64_pal_cache_flush?

[Tian, Kevin]

I know that currently there's no such invocation to ia64_pal_cache_flush
within ia64 linux kernel, but like to know why it's the only instance to
call PAL_CALL_IC_OFF under virtual mode? Since there's no DTR setup for
pal area, it can only work if DTC entry exists just before off psr.ic,
or pal procedure for this type doesn't issue any data access. Or else
nested dtlb miss may happen and it seems not to be handled by current
linux code.

Currently this stub is invoked by XEN (an open source VMM with code
derived from linux) to sync I/D cache after management tool loads kernel
image and initrd of another domain (guest OS). Though we may take place
of it by fc.i, I'm still curious to know historical background for this
design.

Thanks in advance,
Kevin

Re: Why is psr.ic off for ia64_pal_cache_flush?

[Dean Roe]

On Mon, Dec 26, 2005 at 01:44:55PM +0800, Tian, Kevin wrote:
> I know that currently there's no such invocation to ia64_pal_cache_flush
> within ia64 linux kernel, but like to know why it's the only instance to
> call PAL_CALL_IC_OFF under virtual mode? Since there's no DTR setup for
> pal area, it can only work if DTC entry exists just before off psr.ic,
> or pal procedure for this type doesn't issue any data access. Or else
> nested dtlb miss may happen and it seems not to be handled by current
> linux code.
> 
> Currently this stub is invoked by XEN (an open source VMM with code
> derived from linux) to sync I/D cache after management tool loads kernel
> image and initrd of another domain (guest OS). Though we may take place
> of it by fc.i, I'm still curious to know historical background for this
> design.
> 
> Thanks in advance,
> Kevin

Did you get any responses to this?  I played with ia64_pal_cache_flush
on an Altix (over a year ago) and couldn't get it to work without
converting it to use PAL_CALL.  Are you having a similar problem when
you use it for XEN?

Dean

-- 
Dean Roe
Silicon Graphics, Inc.
roe@sgi.com

RE: Why is psr.ic off for ia64_pal_cache_flush?

[Magenheimer, Dan (HP Labs Fort Collins)]

> Did you get any responses to this?  I played with ia64_pal_cache_flush
> on an Altix (over a year ago) and couldn't get it to work without
> converting it to use PAL_CALL.  Are you having a similar problem when
> you use it for XEN?

The PAL spec has changed so that PAL_CALL is correct and PAL_CALL_IC_OFF
is incorrect.  So this is a bug in Linux/ia64, though currently
harmless since ia64_pal_cache_flush is unused.

I think Kevin was going to submit a patch to the linux-ia64 list
so that others don't run into this problem in the future.

Dan