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

[Markus Armbruster]

Alexander Graf <address@hidden> writes:

> On 14.06.2012, at 09:29, Markus Armbruster wrote:
>
>> 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, bumping it en par with 
>>> virtio.
>> 
>> If I understand your patch correctly, this makes if=ahci work like
>> if=virtio and unlike if=ide:
>> 
>> * if=virtio: configure a new PCI device.
>> 
>> * if=ide: instruct the board to add an IDE device to its IDE controller.
>> 
>> For -M pc, the board's IDE controller happens to be piix3-ide.
>> 
>> For -M q35, I'd expect the board's IDE controller to be an ich9-ahci.
>> 
>> Once we switch to q35, if=ahci will become a redundant wart: to add
>> drives to the existing AHCI controller, you'll have to use if=ide.
>> if=ahci will create a new controller, which is generally not what you
>> want.  Ugh.
>
> Yeah, I couldn't come up with anything else that's not completely ugly like 
> the IF_SCSI implementation.

Here's a non-ugly solution: finish the q35 job, and if=ide just works :)

Except for old VMs that still have piix3-ide.  And for those VMs an easy
way to add AHCI controllers and drives isn't exactly a priority.

>> A few questions inline.
>> 
>>> Signed-off-by: Alexander Graf <address@hidden>
>>> ---
>>> blockdev.c      |   25 +++++++++++++++++++++++--
>>> blockdev.h      |    1 +
>>> qemu-options.hx |    7 ++++++-
>>> 3 files changed, 30 insertions(+), 3 deletions(-)
>>> 
>>> diff --git a/blockdev.c b/blockdev.c
>>> index 622ecba..5405f6c 100644
>>> --- a/blockdev.c
>>> +++ b/blockdev.c
>>> @@ -32,6 +32,7 @@ static const char *const if_name[IF_COUNT] = {
>>>     [IF_SD] = "sd",
>>>     [IF_VIRTIO] = "virtio",
>>>     [IF_XEN] = "xen",
>>> +    [IF_AHCI] = "ahci",
>>> };
>>> 
>>> static const int if_max_devs[IF_COUNT] = {
>>> @@ -519,7 +520,7 @@ DriveInfo *drive_init(QemuOpts *opts, int 
>>> default_to_scsi)
>>>     } else {
>>>         /* no id supplied -> create one */
>>>         dinfo->id = g_malloc0(32);
>>> -        if (type == IF_IDE || type == IF_SCSI)
>>> +        if (type == IF_IDE || type == IF_SCSI || type == IF_AHCI)
>>>             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
>>>         if (max_devs)
>>>             snprintf(dinfo->id, 32, "%s%i%s%i",
>>> @@ -549,6 +550,7 @@ DriveInfo *drive_init(QemuOpts *opts, int 
>>> default_to_scsi)
>>>     case IF_IDE:
>>>     case IF_SCSI:
>>>     case IF_XEN:
>>> +    case IF_AHCI:
>>>     case IF_NONE:
>>>         switch(media) {
>>>     case MEDIA_DISK:
>>> @@ -582,6 +584,25 @@ DriveInfo *drive_init(QemuOpts *opts, int 
>>> default_to_scsi)
>>>     default:
>>>         abort();
>>>     }
>>> +
>>> +    if (type == IF_AHCI) {
>>> +        static int ahci_bus = 0;
>>> +        char devname[] = "ahciXXX";
>>> +        char busname[] = "ahciXXX.0";
>>> +        snprintf(devname, sizeof(devname), "ahci%d", ahci_bus);
>>> +        snprintf(busname, sizeof(busname), "ahci%d.0", ahci_bus++);
>>> +
>>> +        /* add ahci host controller */
>>> +        opts = qemu_opts_create(qemu_find_opts("device"), devname, 0, 
>>> NULL);
>>> +        qemu_opt_set(opts, "driver", "ich9-ahci");
>>> +
>>> +        /* and attach a single ata disk to its bus */
>>> +        opts = qemu_opts_create(qemu_find_opts("device"), NULL, 0, NULL);
>>> +        qemu_opt_set(opts, "driver", "ide-drive");
>>> +        qemu_opt_set(opts, "bus", busname);
>>> +        qemu_opt_set(opts, "drive", dinfo->id);
>>> +    }
>>> +
>> 
>> Doesn't this create a new ich9-ahci controller per -drive?
>> 
>> If yes, it's problematic in practice, as you'll run out of PCI slots
>> pretty darn fast.  That problem made us replace virtio-blk by
>> virtio-scsi.  Let's not re-create it.
>
> Hrm. If you have a great idea on how to implement it, I'm all open for it. 
> Talking about it from a high level perspective I had the same feelings at 
> first. Looking at how to implement index= and bus= for real, I quickly 
> withdrew myself from the approach.

I'm afraid the sane way to do this is going to be complicated, just like
if=scsi.  I understand your reluctance to do that; I feel the same.
That's why I'm suggesting to make the problem go away via -M q35.

> The good news is that the limitation here is only a -drive if=ahci 
> limitation. It does not apply to -device. There you can still plug up to 6 
> ahci devices onto a single HBA.

A good convenience option covers a useful subset of common cases neatly.

Your -drive if=ahci covers one common case: add the first AHCI drive
(automatically adding the controller for it as well).  It doesn't cover
the common case of adding a second AHCI drive (to the same controller,
of course).  If I have to resort to -device for something as simple as
that, then I wonder why it's worth having -drive if=ahci at all.

>> IF_VIRTIO device option creation is done in the preceeding switch.  You
>> don't do that for IF_AHCI because you already do something else there:
>> handling the media option.  I think one switch for media and a separate
>> one for device options would be clearer.
>
> We can do that, yeah :).
>
>> 
>>>     if (!file || !*file) {
>>>         return dinfo;
>>>     }
>>> @@ -604,7 +625,7 @@ DriveInfo *drive_init(QemuOpts *opts, int 
>>> default_to_scsi)
>>>         ro = 1;
>>>     } else if (ro == 1) {
>>>         if (type != IF_SCSI && type != IF_VIRTIO && type != IF_FLOPPY &&
>>> -            type != IF_NONE && type != IF_PFLASH) {
>>> +            type != IF_NONE && type != IF_PFLASH && type != IF_AHCI) {
>>>             error_report("readonly not supported by this bus type");
>>>             goto err;
>>>         }
>> 
>> Are you sure AHCI can handle read-only?
>
> Isn't read-only handled in generic ATA code?

I think it is.  But treating IF_AHCI and IF_IDE differently here is
confusing, isn't it?

>> [...]
>>> diff --git a/blockdev.h b/blockdev.h
>>> index 260e16b..e14c1d5 100644
>>> --- a/blockdev.h
>>> +++ b/blockdev.h
>>> @@ -23,6 +23,7 @@ typedef enum {
>>>     IF_DEFAULT = -1,            /* for use with drive_add() only */
>>>     IF_NONE,
>>>     IF_IDE, IF_SCSI, IF_FLOPPY, IF_PFLASH, IF_MTD, IF_SD, IF_VIRTIO, IF_XEN,
>>> +    IF_AHCI,
>>>     IF_COUNT
>>> } BlockInterfaceType;
>>> 
>>> diff --git a/qemu-options.hx b/qemu-options.hx
>>> index 8b66264..9527c51 100644
>>> --- a/qemu-options.hx
>>> +++ b/qemu-options.hx
>>> @@ -160,7 +160,7 @@ Special files such as iSCSI devices can be specified 
>>> using protocol
>>> specific URLs. See the section for "Device URL Syntax" for more information.
>>> @item address@hidden
>>> This option defines on which type on interface the drive is connected.
>>> -Available types are: ide, scsi, sd, mtd, floppy, pflash, virtio.
>>> +Available types are: ide, scsi, sd, mtd, floppy, pflash, virtio, ahci.
>>> @item address@hidden,address@hidden
>>> These options define where is connected the drive by defining the bus 
>>> number and
>>> the unit id.
>>> @@ -260,6 +260,11 @@ You can connect a SCSI disk with unit ID 6 on the bus 
>>> #0:
>>> qemu-system-i386 -drive file=file,if=scsi,bus=0,unit=6
>>> @end example
>>> 
>>> +You can attach a SATA disk using AHCI:
>>> address@hidden
>>> +qemu-system-i386 -drive file=file,if=ahci
>>> address@hidden example
>>> +
>> 
>> If I'm reading drive_init() correctly, if=ahci doesn't attach a disk, it
>> creates a controller with a disk.  So this is somewhat misleading.
>
> It creates an HBA with a disk, yes. Which for almost everyone is the same 
> thing as attaching a disk. People who want to create more sophisticated 
> setups are still free to use the complex -device based syntax :).

I think we'd owe users a fair warning that if=ahci does something rather
unexpected, namely adding a controller.