Re: [Qemu-devel] [PATCH v2] ahci: add -drive support

[Gleb Natapov]

On Mon, Jul 09, 2012 at 11:36:47AM +0200, Alexander Graf wrote:
> 
> On 09.07.2012, at 11:27, Kevin Wolf wrote:
> 
> > Am 09.07.2012 11:13, schrieb Alexander Graf:
> >> 
> >> On 09.07.2012, at 11:11, Kevin Wolf wrote:
> >> 
> >>> Am 09.07.2012 10:50, schrieb Markus Armbruster:
> >>>> Alexander Graf <address@hidden> writes:
> >>>> 
> >>>>> We've had support for creating AHCI devices using -device for a while 
> >>>>> now,
> >>>>> but it's cumbersome to users. We really should provide an easier way for
> >>>>> them to leverage the power of AHCI!
> >>>>> 
> >>>>> So let's introduce a new if= option to -drive, giving users the same
> >>>>> command line experience as for scsi or ide.
> >>>>> 
> >>>>> Signed-off-by: Alexander Graf <address@hidden>
> >>>>> 
> >>>>> ---
> >>>>> 
> >>>>> v1 -> v2:
> >>>>> 
> >>>>> - support more than a single drive per adapter
> >>>>> - support index= option
> >>>>> - treat IF_AHCI the same as IF_IDE
> >>>> 
> >>>> Inhowfar?  Not obvious to me from the patch, or the diff patch v1.
> >>>> 
> >>>>> - add is_ata() helper to match AHCI || IDE
> >>>> 
> >>>> Not addressed:
> >>>> 
> >>>> Once we switch to q35, if=ahci will become a redundant wart: to add
> >>>> drives to the board's AHCI controller, you'll have to use if=ide.
> >>>> if=ahci will create new controllers, which is generally not what you
> >>>> want.  Ugh.
> >>> 
> >>> Can we even make it the default with q35 as long as our AHCI controller
> >>> doesn't also expose a legacy interface for compatibility?
> >> 
> >> What legacy interface? The ICH-9 can be controlled by the _BIOS_ to switch 
> >> between 2 PCI IDs. One for IDE mode, one for AHCI mode. I don't think that 
> >> would help us here, would it?
> > 
> > I didn't actually look into this much. I just supposed that the
> > existence of an AHCI Enable bit in the GHC register implies that some
> > compatibility mechanism can be implemented that is transparent to older
> > OSes.
> 
> Yeah, I was hoping the same too when I read the spec back then. 
> Unfortunately, it's really a firmware configuration bit which doesn't help us 
> at all.
> 
> I think there is a hack in some driver somewhere that on bootup checks if the 
> AHCI bit is disabled and then forcefully enables it, so that the device 
> during OS boot magically changes its ID and semantics. But I don't think we 
> really want to rely on that. IIRC it never went upstream and I doubt Windows 
> does it.
> 
Doesn't enabling the bit change PCI bars size/number?

> > If it can't, then I guess we can't change the meaning of -hda as long as
> > we claim that the command line is a stable API (or at least not if q35
> > is meant to become the default machine type at some point)
> 
> If -hda has the semantics of "create an IDE device", then no, we can't change 
> it. It doesn't however. IIRC on -M pseries -hda creates SCSI devices. On s390 
> -hda creates virtio devices. So if on -M q35 -hda would create if=ahci, I 
> don't see how that breaks the CLI.
> 
> 
> Alex
> 

--
                        Gleb.