Re: nfe problem on dom0 kernel 5.99.24 [need help]

[Jean-Yves Migeon <jeanyves.migeon%free.fr@localhost>]

On 03/02/10 01:28, Jean-Yves Migeon wrote:
> On 03/01/10 08:52, Adam Hamsik wrote:
>  >>> Do you have any ideas how can it be debug more ?
>  >>>
> > _xen_alloc_contig -> uvm_pglistalloc == 0
> > xen_alloc_contig: XENMEM_decrease_reservation failed: err 0 (pa
> 0x208f000 mfn 0x209d70)
> > _xen_alloc_contig 12
> >[snip]
> > _xen_alloc_contig -> uvm_pglistalloc
> > _xen_alloc_contig -> uvm_pglistalloc == 0
> > xen_alloc_contig: XENMEM_decrease_reservation failed: err 0 (pa
> 0x2093000 mfn 0xce0f3)
> > _xen_alloc_contig 12
> > nfe1: could not create DMA map
> > nfe1: could not allocate Tx ring
>
> I'll have a look tomorrow at hypervisor's code.
>
> How much RAM do you have for this host?
>
> Could you post the result obtained with the attached patch please?

Hmm, the hypervisor handles the M2P table by itself; the translation 
would have failed (it returns a poison value) in case of an incorrect MFN.

Would that be ok for you to add a couple printk() inside Xen? If yes, 
I'll prepare some diff for Xen 3.3. Unfortunately, the multiple error 
paths return the same value, so they are quite hard to identify without 
getting some printk in.

Please update to rev 1.19 of xen_bus_dma.c before (a minor fix for my 
previous patch).

Thanks for your patience!

--
Jean-Yves Migeon
jeanyves.migeon%free.fr@localhost