Patchwork [2/2] ata_piix: defer disks to the Hyper-V drivers by default

login
register
mail settings
Submitter Andy Whitcroft
Date April 14, 2012, 3:53 p.m.
Message ID <1334418796-11473-3-git-send-email-apw@canonical.com>
Download mbox | patch
Permalink /patch/152518/
State Not Applicable
Delegated to: David Miller
Headers show

Comments

Andy Whitcroft - April 14, 2012, 3:53 p.m.
When we are hosted on a Microsoft Hyper-V hypervisor the guest disks
are exposed both via the Hyper-V paravirtualised drivers and via an
emulated SATA disk drive.  In this case we want to use the paravirtualised
drivers if we can as they are much more efficient.  Note that the Hyper-V
paravirtualised drivers only expose the virtual hard disk devices, the
CDROM/DVD devices must still be enumerated.

Mark the host controller ATA_HOST_IGNORE_ATA to prevent enumeration
of disk devices.

BugLink: http://bugs.launchpad.net/bugs/929545
BugLink: http://bugs.launchpad.net/bugs/942316
Signed-off-by: Andy Whitcroft <apw@canonical.com>
---
 drivers/ata/ata_piix.c |   20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)
Alan Cox - April 14, 2012, 4:02 p.m.
O> +#ifdef CONFIG_X86
> +	/* On Hyper-V hypervisors the disks are exposed on both the emulated
> +	 * SATA controller and on the paravirtualised drivers.  The CD/DVD
> +	 * devices are only exposed on the emulated controller.  Request we
> +	 * ignore ATA devices on this host.
> +	 */
> +	if (prefer_ms_hyperv && x86_hyper == &x86_hyper_ms_hyperv) {
> +		host->flags |= ATA_HOST_IGNORE_ATA;
> +		dev_info(&pdev->dev, "Hyper-V hypervisor detected, ignoring ATA disks\n");
> +	}
> +#endif

Does hyper-v not set its own unique and identifable pci
subvendor/subdevice ?

Alan
--
To unsubscribe from this list: send the line "unsubscribe linux-ide" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
KY Srinivasan - April 15, 2012, 12:10 a.m.
> -----Original Message-----
> From: Alan Cox [mailto:alan@lxorguk.ukuu.org.uk]
> Sent: Saturday, April 14, 2012 12:02 PM
> To: Andy Whitcroft
> Cc: Jeff Garzik; linux-ide@vger.kernel.org; linux-kernel@vger.kernel.org; KY
> Srinivasan; Mike Sterling
> Subject: Re: [PATCH 2/2] ata_piix: defer disks to the Hyper-V drivers by default
> 
> O> +#ifdef CONFIG_X86
> > +	/* On Hyper-V hypervisors the disks are exposed on both the emulated
> > +	 * SATA controller and on the paravirtualised drivers.  The CD/DVD
> > +	 * devices are only exposed on the emulated controller.  Request we
> > +	 * ignore ATA devices on this host.
> > +	 */
> > +	if (prefer_ms_hyperv && x86_hyper == &x86_hyper_ms_hyperv) {
> > +		host->flags |= ATA_HOST_IGNORE_ATA;
> > +		dev_info(&pdev->dev, "Hyper-V hypervisor detected, ignoring
> ATA disks\n");
> > +	}
> > +#endif
> 
> Does hyper-v not set its own unique and identifable pci
> subvendor/subdevice ?

Number of elements of the virtual BIOS we present the guest can be used
to detect that we are running on Hyper-V. Additionally, we are exporting state based
on CPU detection which is what Andy is using here. As far as I know these are the only
ways to detect that we are running on Hyper-V. 

Regards,

K. Y
> 
> Alan
> 
> 



--
To unsubscribe from this list: send the line "unsubscribe linux-ide" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Andy Whitcroft - April 16, 2012, 11:20 a.m.
Add support for ignoring disks devices on specific host controllers to
libata.  Use this support to trigger us to ignore Hyper-V disk devices
on the emulated SATA device based on the system DMI information.

Jeff, I believe the first patch is what you had in mind in your comments.
Alan, the piix sadly has no subvendor/subdevice identifiers at all
though we are able to detect the platform from the DMI information,
this any better?

-apw

V3
  - switch detection method to DMI information

V2 
  - create a generic libata quirk
  - switch detection to piix_init_one

Andy Whitcroft (2):
  libata: add a host flag to ignore detected ATA devices
  ata_piix: defer disks to the Hyper-V drivers by default

 drivers/ata/ata_piix.c    |   20 ++++++++++++++++++++
 drivers/ata/libata-core.c |    7 +++++++
 include/linux/libata.h    |    1 +
 3 files changed, 28 insertions(+)
Alan Cox - April 16, 2012, 11:26 a.m.
On Mon, 16 Apr 2012 12:20:27 +0100
Andy Whitcroft <apw@canonical.com> wrote:

> Add support for ignoring disks devices on specific host controllers to
> libata.  Use this support to trigger us to ignore Hyper-V disk devices
> on the emulated SATA device based on the system DMI information.
> 
> Jeff, I believe the first patch is what you had in mind in your comments.
> Alan, the piix sadly has no subvendor/subdevice identifiers at all

So its 0x0000,0x0000 ?

> though we are able to detect the platform from the DMI information,
> this any better?

What happens if you have a PIIX device which is being passed through from
the underlying hardware or will hyper-v never support pass through ?

Alan
--
To unsubscribe from this list: send the line "unsubscribe linux-ide" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Jeff Garzik - April 16, 2012, 3:32 p.m.
On 04/16/2012 07:20 AM, Andy Whitcroft wrote:
> Jeff, I believe the first patch is what you had in mind in your comments.
> Alan, the piix sadly has no subvendor/subdevice identifiers at all
> though we are able to detect the platform from the DMI information,
> this any better?

Please post "lspci -vvv" for this device...

--
To unsubscribe from this list: send the line "unsubscribe linux-ide" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Andy Whitcroft - April 16, 2012, 6:28 p.m.
On Mon, Apr 16, 2012 at 11:32:22AM -0400, Jeff Garzik wrote:
> On 04/16/2012 07:20 AM, Andy Whitcroft wrote:
> >Jeff, I believe the first patch is what you had in mind in your comments.
> >Alan, the piix sadly has no subvendor/subdevice identifiers at all
> >though we are able to detect the platform from the DMI information,
> >this any better?
> 
> Please post "lspci -vvv" for this device...

lspci -nnvvv:

00:07.1 IDE interface [0101]: Intel Corporation 82371AB/EB/MB PIIX4 IDE [8086:7111] (rev 01) (prog-if 80 [Master])
	Control: I/O+ Mem- BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx-
	Status: Cap- 66MHz- UDF- FastB2B+ ParErr- DEVSEL=medium >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
	Latency: 0
	Region 0: [virtual] Memory at 000001f0 (32-bit, non-prefetchable) [size=8]
	Region 1: [virtual] Memory at 000003f0 (type 3, non-prefetchable) [size=1]
	Region 2: [virtual] Memory at 00000170 (32-bit, non-prefetchable) [size=8]
	Region 3: [virtual] Memory at 00000370 (type 3, non-prefetchable) [size=1]
	Region 4: I/O ports at ffa0 [size=16]
	Kernel driver in use: ata_piix

-apw
--
To unsubscribe from this list: send the line "unsubscribe linux-ide" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Andy Whitcroft - April 16, 2012, 6:29 p.m.
On Mon, Apr 16, 2012 at 12:26:34PM +0100, Alan Cox wrote:
> On Mon, 16 Apr 2012 12:20:27 +0100
> Andy Whitcroft <apw@canonical.com> wrote:
> 
> > Add support for ignoring disks devices on specific host controllers to
> > libata.  Use this support to trigger us to ignore Hyper-V disk devices
> > on the emulated SATA device based on the system DMI information.
> > 
> > Jeff, I believe the first patch is what you had in mind in your comments.
> > Alan, the piix sadly has no subvendor/subdevice identifiers at all
> 
> So its 0x0000,0x0000 ?
> 
> > though we are able to detect the platform from the DMI information,
> > this any better?
> 
> What happens if you have a PIIX device which is being passed through from
> the underlying hardware or will hyper-v never support pass through ?

A good question indeed.  KY perhaps you can comment.

-apw
--
To unsubscribe from this list: send the line "unsubscribe linux-ide" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
KY Srinivasan - April 16, 2012, 6:41 p.m.
> -----Original Message-----
> From: Andy Whitcroft [mailto:apw@canonical.com]
> Sent: Monday, April 16, 2012 2:29 PM
> To: Alan Cox
> Cc: Jeff Garzik; linux-ide@vger.kernel.org; linux-kernel@vger.kernel.org; KY
> Srinivasan; Mike Sterling
> Subject: Re: [PATCH 0/2] Hyper-V disk support V3
> 
> On Mon, Apr 16, 2012 at 12:26:34PM +0100, Alan Cox wrote:
> > On Mon, 16 Apr 2012 12:20:27 +0100
> > Andy Whitcroft <apw@canonical.com> wrote:
> >
> > > Add support for ignoring disks devices on specific host controllers to
> > > libata.  Use this support to trigger us to ignore Hyper-V disk devices
> > > on the emulated SATA device based on the system DMI information.
> > >
> > > Jeff, I believe the first patch is what you had in mind in your comments.
> > > Alan, the piix sadly has no subvendor/subdevice identifiers at all
> >
> > So its 0x0000,0x0000 ?
> >
> > > though we are able to detect the platform from the DMI information,
> > > this any better?
> >
> > What happens if you have a PIIX device which is being passed through from
> > the underlying hardware or will hyper-v never support pass through ?

The notion of "pass through" in Hyper-V  is a little different. IDE devices can be configured
under either one of the supported controllers and these devices can either be virtual disks
(VHDs) or physical disks. In either case these will be presented to the guest as IDE devices.

Regards,

K. Y



--
To unsubscribe from this list: send the line "unsubscribe linux-ide" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Alan Cox - April 18, 2012, 8:12 p.m.
> The notion of "pass through" in Hyper-V  is a little different. IDE devices can be configured
> under either one of the supported controllers and these devices can either be virtual disks
> (VHDs) or physical disks. In either case these will be presented to the guest as IDE devices.

So what ensures that by skipping it at the ATA device level we will
always find it as a VHD ?

Alan
--
To unsubscribe from this list: send the line "unsubscribe linux-ide" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
KY Srinivasan - April 18, 2012, 8:21 p.m.
> -----Original Message-----
> From: Alan Cox [mailto:alan@lxorguk.ukuu.org.uk]
> Sent: Wednesday, April 18, 2012 4:13 PM
> To: KY Srinivasan
> Cc: Andy Whitcroft; Jeff Garzik; linux-ide@vger.kernel.org; linux-
> kernel@vger.kernel.org; Mike Sterling
> Subject: Re: [PATCH 0/2] Hyper-V disk support V3
> 
> > The notion of "pass through" in Hyper-V  is a little different. IDE devices can be
> configured
> > under either one of the supported controllers and these devices can either be
> virtual disks
> > (VHDs) or physical disks. In either case these will be presented to the guest as
> IDE devices.
> 
> So what ensures that by skipping it at the ATA device level we will
> always find it as a VHD ?

If the Hyper-V storage driver is  present, then the Hyper-V storage driver will
manage the root device (and other devices skipped by the ata driver). 
So, the check if we are running on Hyper-V is not enough
to skip the device in the ata_piix driver; we need to also make sure that the Hyper-V
block driver is available to handle the skipped devices. I think the correct approach here is to
have a kernel (or an ata_piix module parameter) that can be set to override the skipping
of the block device by the ata driver. Distros that are including the Hyper-V storage driver
can set this parameter so that when running on Hyper-V, ata can skip the block device.

Regards,

K. Y
> 
> Alan
> 
> 



--
To unsubscribe from this list: send the line "unsubscribe linux-ide" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Andy Whitcroft - April 19, 2012, 7:40 a.m.
On Wed, Apr 18, 2012 at 09:12:35PM +0100, Alan Cox wrote:
> > The notion of "pass through" in Hyper-V  is a little different. IDE devices can be configured
> > under either one of the supported controllers and these devices can either be virtual disks
> > (VHDs) or physical disks. In either case these will be presented to the guest as IDE devices.
> 
> So what ensures that by skipping it at the ATA device level we will
> always find it as a VHD ?

My understanding of things is that the if you have disks assigned to the
guest that they will always appear both on the virtualised SATA contoller
and on the paravirtualised driver channel.  So that assuming a valid
configuration in which both drivers initialise they will be picked up by
one or the other.

KY for clarity are we saying the only way a disk can appear in the guest
is either on the emulated controller or paravirtualised, we will never
expose real devices into the guest.

Reading my own statement that ought to imply we really should only default
preferring paravirtualised to 'on' when the hyperv paravirtualised drivers
is enabled too; easy enough to do.  Howeever in a perfect world the
override would also have the inverse effect even if the paravirtualised
driver initialises first; not so easy but I assume we with some fiddling
we could make both drivers handle the same parameter.

/me considers this further.

-apw
--
To unsubscribe from this list: send the line "unsubscribe linux-ide" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Victor Miasnikov - April 19, 2012, 8:15 a.m.
Hi!

 Shortly:

1) 
The flag to disable the Hyper-V IDE drivers is:
 ata_piix.prefer_ms_hyperv=02)

#ifdef CONFIG_X86 
#if defined(CONFIG_HYPERV_STORAGE) || defined(CONFIG_HYPERV_STORAGE_MODULE)
#define ATA_PIIX_IGNORE_ATA_DISKS_ON_HYPERV 1
#endif
#endif


---

>>Alan Cox:
>> So what ensures that by 
>> skipping it at the ATA device level we will always find it as a VHD ?

> KY Srinivasan:
> So, the check if we are running on Hyper-V is not enough
> to skip the device in the ata_piix driver; 
> we need to also make sure that the Hyper-V block driver is available to handle the skipped devices. 

>  I think the correct approach here is to have a kernel (or an ata_piix module parameter) 
> that can be set to override the skipping of the block device by the ata driver. 

 Already implemented, see:
module_param(prefer_ms_hyperv, int, 0);

 and work as need in real life:

http://www.mail-archive.com/ubuntu-bugs@lists.ubuntu.com/msg3482442.html
==
Tim Miller Dyck
Wed, 14 Mar 2012 07:15:52 -0700
 . . .
The boot flags can be viewed by pressing F5? F6? (one of those) on the
setup boot screen and then navigating to the boot line with cursor keys
and adding the boot parameter below before the double dash at the end of
the line.
The flag to disable the Hyper-V IDE drivers is:

 ata_piix.prefer_ms_hyperv=0  . . .
==
> Distros that are including the Hyper-V storage driver can set this parameter so 
> that when running on Hyper-V, ata can skip the block device.


 Best:
Distros that are not-including the Hyper-V storage driver don't worry about "ata can skip" etc.:

See re-incarnatin Olaf Hering idea:

#ifdef CONFIG_X86 
#if defined(CONFIG_HYPERV_STORAGE) || defined(CONFIG_HYPERV_STORAGE_MODULE)
#define ATA_PIIX_IGNORE_ATA_DISKS_ON_HYPERV 1
#endif
#endif

 and

#ifdef CONFIG_X86 

 replace to

#ifdef ATA_PIIX_IGNORE_ATA_DISKS_ON_HYPERV 


Best regards, Victor Miasnikov
Blog:  http://vvm.blog.tut.by/


--
To unsubscribe from this list: send the line "unsubscribe linux-ide" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
KY Srinivasan - April 19, 2012, 1:17 p.m.
> -----Original Message-----
> From: Andy Whitcroft [mailto:apw@canonical.com]
> Sent: Thursday, April 19, 2012 3:40 AM
> To: Alan Cox
> Cc: KY Srinivasan; Jeff Garzik; linux-ide@vger.kernel.org; linux-
> kernel@vger.kernel.org; Mike Sterling
> Subject: Re: [PATCH 0/2] Hyper-V disk support V3
> 
> On Wed, Apr 18, 2012 at 09:12:35PM +0100, Alan Cox wrote:
> > > The notion of "pass through" in Hyper-V  is a little different. IDE devices can
> be configured
> > > under either one of the supported controllers and these devices can either
> be virtual disks
> > > (VHDs) or physical disks. In either case these will be presented to the guest as
> IDE devices.
> >
> > So what ensures that by skipping it at the ATA device level we will
> > always find it as a VHD ?
> 
> My understanding of things is that the if you have disks assigned to the
> guest that they will always appear both on the virtualised SATA contoller
> and on the paravirtualised driver channel.  So that assuming a valid
> configuration in which both drivers initialise they will be picked up by
> one or the other.
> 
> KY for clarity are we saying the only way a disk can appear in the guest
> is either on the emulated controller or paravirtualised, we will never
> expose real devices into the guest.

You are right. Depending on how the device is configured for the guest, even
in the pass through case, the device will show up as either an emulated device
(if configured under the IDE controllers) or as a device under the synthetic scsi
controller (as seen by the guest). The storvsc driver can handle both these configurations
while the native ata driver can only handle emulated devices. Since, presently we can 
only boot off of a device configured under the IDE controller, we have the issue which your
patch addresses.

K. Y 



--
To unsubscribe from this list: send the line "unsubscribe linux-ide" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Andy Whitcroft - April 19, 2012, 2:33 p.m.
Add support for ignoring disks devices on specific host controllers to
libata.  Use this support to trigger us to ignore Hyper-V disk devices
on the emulated SATA device based on the system DMI information.

I have cleaned up the Hyper-V detection so it is only included when the
paravirtualised drivers are enabled.  I have also reintroduced the
override which had mysteriously gone missing in V3.

Comments?

-apw

V4
  - only enable the quirk when the paravirtualised drivers are compiled
  - fix up the command line flag to disable this support

V3
  - switch detection method to DMI information

V2 
  - create a generic libata quirk
  - switch detection to piix_init_one

Andy Whitcroft (2):
  libata: add a host flag to ignore detected ATA devices
  ata_piix: defer disks to the Hyper-V drivers by default

 drivers/ata/ata_piix.c    |   20 ++++++++++++++++++++
 drivers/ata/libata-core.c |    7 +++++++
 include/linux/libata.h    |    1 +
 3 files changed, 28 insertions(+)

Patch

diff --git a/drivers/ata/ata_piix.c b/drivers/ata/ata_piix.c
index 68013f9..b52d8ab 100644
--- a/drivers/ata/ata_piix.c
+++ b/drivers/ata/ata_piix.c
@@ -94,6 +94,9 @@ 
 #include <scsi/scsi_host.h>
 #include <linux/libata.h>
 #include <linux/dmi.h>
+#ifdef CONFIG_X86
+#include <asm/hypervisor.h>
+#endif
 
 #define DRV_NAME	"ata_piix"
 #define DRV_VERSION	"2.13"
@@ -1567,6 +1570,11 @@  static bool piix_broken_system_poweroff(struct pci_dev *pdev)
  *	Zero on success, or -ERRNO value.
  */
 
+#ifdef CONFIG_X86
+static int prefer_ms_hyperv = 1;
+module_param(prefer_ms_hyperv, int, 0);
+#endif
+
 static int __devinit piix_init_one(struct pci_dev *pdev,
 				   const struct pci_device_id *ent)
 {
@@ -1667,6 +1675,18 @@  static int __devinit piix_init_one(struct pci_dev *pdev,
 	}
 	host->flags |= ATA_HOST_PARALLEL_SCAN;
 
+#ifdef CONFIG_X86
+	/* On Hyper-V hypervisors the disks are exposed on both the emulated
+	 * SATA controller and on the paravirtualised drivers.  The CD/DVD
+	 * devices are only exposed on the emulated controller.  Request we
+	 * ignore ATA devices on this host.
+	 */
+	if (prefer_ms_hyperv && x86_hyper == &x86_hyper_ms_hyperv) {
+		host->flags |= ATA_HOST_IGNORE_ATA;
+		dev_info(&pdev->dev, "Hyper-V hypervisor detected, ignoring ATA disks\n");
+	}
+#endif
+
 	pci_set_master(pdev);
 	return ata_pci_sff_activate_host(host, ata_bmdma_interrupt, sht);
 }