Re: [Qemu-devel] [Qemu-ppc] [PATCH] hw/intc: fix failure return for xics_alloc_block()

[David Gibson <david@gibson.dropbear.id.au>]

[-- Attachment #1: Type: text/plain, Size: 3944 bytes --]

On Tue, Feb 23, 2016 at 06:46:44PM +0100, Greg Kurz wrote:
> On Wed, 10 Feb 2016 10:41:16 +0100
> Greg Kurz <gkurz@linux.vnet.ibm.com> wrote:
> 
> > On Mon, 8 Feb 2016 09:31:49 +0100
> > Greg Kurz <gkurz@linux.vnet.ibm.com> wrote:
> > 
> > > On Mon, 8 Feb 2016 11:45:19 +1000
> > > David Gibson <david@gibson.dropbear.id.au> wrote:
> > >   
> > > > On Fri, Feb 05, 2016 at 09:43:40AM +0100, Greg Kurz wrote:    
> > > > > From: Brian W. Hart <hartb@linux.vnet.ibm.com>
> > > > > 
> > > > > xics_alloc_block() does not return a clear error code when it
> > > > > fails to allocate a block of interrupts. Instead it returns the
> > > > > base interrupt number minus 1. This change updates it to return a
> > > > > clear -1 in case of failure (following the example of xics_alloc()).
> > > > > 
> > > > > The two callers of xics_alloc_block() are updated to check for
> > > > > a negative return as an error. They had previously checked for
> > > > > a 0 return as an error, which wrongly treated most failures as
> > > > > successes.
> > > > > 
> > > > > Fixes: bee763dbfb8cfceea112131970da07f215f293a6
> > > > > Signed-off-by: Brian W. Hart <hartb@linux.vnet.ibm.com>
> > > > > [only pass src and num to trace_xics_alloc_block_failed_no_left,
> > > > >  added trace_xics_alloc_block_failed_no_left definition to trace-events]
> > > > > Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>      
> > > > 
> > > > Hrm, it would probably be better to give xics_alloc_block() an Error
> > > > ** argument so it can report errors using the new API.
> > > >     
> > > 
> > > Sure. I can rework the patch to do so.
> > >   
> > 
> > The trace_xics_alloc_block_failed_no_left trace is more a debugging thing
> > than an error to be reported to the user. Also, rtas_ibm_change_msi()
> > already has a meaningful error message:
> > 
> >         error_report("Cannot allocate MSIs for device %x", config_addr);
> > 
> > So in the end, I'm not sure about the benefit of passing an Error **
> > down to xics_alloc_block().
> > 
> 
> Hi David !
> 
> Given the remarks above, do you still think we should pass Error ** ?

I still think using the Error API would be preferable, but it doesn't
make a huge difference.

> > > > TBH the whole xics_alloc_block() interface is kind of dubious, or at
> > > > least the ics_find_free_block() part of it.  Dynamically allocating
> > > > irqs to devices is basically awful for migration, so it's better to
> > > > have fixed allocations of all interrupts at the machine level.
> > > >     
> > > 
> > > I agree about the extra complexity, but isn't it the purpose of
> > > the ibm,change-msi RTAS call ? I'm not sure to understand what you
> > > are suggesting...
> > >   
> > 
> > And anyway, even if the decision is made one day to have fixed
> > allocations, shouldn't we consider fixing this bug first ?
> > 
> 
> According to the following commit changelog, the dynamic allocation was
> introduced to support PCI hot(un)plug. Maybe Alexey may explain why it
> got coded this way.
> 
> commit bee763dbfb8cfceea112131970da07f215f293a6
> Author: Alexey Kardashevskiy <aik@ozlabs.ru>
> Date:   Fri May 30 19:34:15 2014 +1000
> 
>     spapr: Move interrupt allocator to xics
> 
> I'm not sure of the amount of reflexion and work needed to address your
> concern... Given the time frame, what about deferring xics rework to 2.7
> and fix the bug we currently have in 2.6 ?

Yeah, sorry, I realised after I sent that the allocation does
actually make sense in this case - these are guest triggered
allocations that we really do need an allocator for, not host setup
allocations which we have used in the past but turned out to be
dubious.

-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]