RE: [Xen-ia64-devel] Problem with xen-unstable cset 7937 (grant tabletransfer fix)

["Magenheimer, Dan (HP Labs Fort Collins)" <dan.magenheimer@xxxxxx>]

Thanks very much!

There is a typo in gnttab_shared_mfn in the asm-ia64/grant_table.h
you created ((t)shared should be (t)->shared) and I found a
misspelling in include/asm-ia64/mm.h (PGT_writeable should
be PGT_writable... this is never actually used on ia64 but
needs to be defined syntactically to compile).  With these
fixes, it compiles.

However, it doesn't run... booting domU gets "Cannot open
root device".  I will repeat the experiment with xen-ia64 tip
(to verify somethin else isn't screwed up) and then with
xen-ia64 tip with just the grant_table changes to see if
that works.

Or did I miss something that you had left for me to fix
semantically in the new ia64-specific header file?

Thanks,
Dan


> -----Original Message-----
> From: Magenheimer, Dan (HP Labs Fort Collins) 
> Sent: Tuesday, November 22, 2005 9:39 AM
> To: 'Keir Fraser'
> Cc: xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
> Subject: RE: [Xen-ia64-devel] Problem with xen-unstable cset 
> 7937 (grant tabletransfer fix)
> 
> > However, I see no need to ifdef the code that uses pfn_valid, 
> > get_page[_type] and put_page[_type]. They are easily 
> noop'ed if they 
> > mean nothing on ia64 (in fact I notice that you already 
> > define them). I 
> > would imagine that at least pfn_valid() makes sense though?
> 
> True, I think I can get around these easily enough.  One could
> reasonably argue that the whole concept of page type is very 
> x86-specific
> but since it can be easily overridden, we can skip dealing 
> with it now.
> 
> > Apart from that, I also didn't understand the changes in the block 
> > commented 'fundamental difference here'.
> 
> "Fundamental" is overstated I guess.  What I meant is that the control
> flow and the fact that the code block is completely different 
> depending
> on whether dom0 or domU would make it hard to hide the ia64-specific
> implementation behind the existing code, so some separate
> arch_grant_setup_table may be necessary.
> 
> > But looks to me like we ought to be able to move ia64 onto common 
> > grant_table.c implementation fairly easily. We should shoot 
> > down a few 
> > of the easy ifdefs first though.
> 
> OK, if you could:
> - create an asm/grant_table.h and put ORDER_GRANT_FRAMES in it
> - move the big block in _gnttab_map_grant_ref (doing things
>   with l1_pgentry etc) into arch_gnttab_map_grant_ref (or some
>   other way out of grant_table.c) so that I can no-op it
> - replace the "fundamental" block in grant_setup_table with
>   a function call -- arch_grant_setup_table(d,op,uop)?? --
>   this probably becomes a static inline in asm/grant_table.h
> - replace the asm statement with a function call that I can
>   define differently for ia64
> then I think I can handle the rest and all the ifdef's will
> be gone too.
> 
> Thanks,
> Dan
> 

_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel