From: Stefano Stabellini <stefano.stabellini@eu.citrix.com> To: Don Dutile <ddutile@redhat.com> Cc: Stefano Stabellini <Stefano.Stabellini@eu.citrix.com>, "stefano@stabellini.net" <stefano@stabellini.net>, "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>, "xen-devel@lists.xensource.com" <xen-devel@lists.xensource.com>, "sheng@linux.intel.com" <sheng@linux.intel.com>, "jeremy@goop.org" <jeremy@goop.org> Subject: Re: [PATCH 12/13] Unplug emulated disks and nics Date: Thu, 8 Jul 2010 14:13:17 +0100 [thread overview] Message-ID: <alpine.DEB.2.00.1007081246440.21432@kaball-desktop> (raw) In-Reply-To: <4C34DD1B.3010601@redhat.com> On Wed, 7 Jul 2010, Don Dutile wrote: > Stefano Stabellini wrote: > > On Thu, 1 Jul 2010, Don Dutile wrote: > >> The problem with this check/enable is that if you run > >> this on an older qemu-xen that doesn't have unplug support, > >> it fails pv-hvm configuration. > >> > >> But, all that means is that you can't use an xvd as the boot device, > >> and you have to use an emulated IDE device as boot device. > >> There are a couple ways to configure the vnif correctly (in guest > >> or in xen guest config file). > >> > >> So, on an older (say, rhel5) xen, I don't have this check; > >> the boot device is required to be spec'd as hda, not vda, and > >> xen-blkfront is not allowed to configure blk major nums > >> for IDE (& SCSI) (to avoid 2 drivers twiddling w/same phys backend... not good!). > >> > > > > Who is requiring that the boot device is spec'd as hda and not xvda? > > I don't think there is such limitation in xend or libxl at the moment. > > > > If you take a previous xen HVM guest spec & just run it on a guest > that has pv-hvm added to it, then one has hda spec'd as boot device (by default, > by not editing the guest config spec/file). > > Ideally, both config specs should/would work. I wouldn't want to cause data corruptions by default to people that specified xvda in their HVM config files by mistake (for example because they copied and pasted from a PV guest config file). But I agree that we should be able to do the right thing in your case scenario too. I propose the appended patch (to be merge with "Unplug emulated disks and nics"): if the user specifies xen_emul_unplug=ignore and the unplug protocol is not supported (old xen installations like rhel5), we continue with the PV on HVM initialization and we make sure that blkfront doesn't hook any IDE or SCSI device. Don, Jeremy, what do you think about it? --- diff --git a/arch/x86/xen/platform-pci-unplug.c b/arch/x86/xen/platform-pci-unplug.c index 72a3da6..2f7f3fb 100644 --- a/arch/x86/xen/platform-pci-unplug.c +++ b/arch/x86/xen/platform-pci-unplug.c @@ -29,9 +29,9 @@ #define XEN_PLATFORM_ERR_PROTOCOL -2 #define XEN_PLATFORM_ERR_BLACKLIST -3 -/* boolean to signal that the platform pci device can be used */ -bool xen_platform_pci_enabled; -EXPORT_SYMBOL_GPL(xen_platform_pci_enabled); +/* store the value of xen_emul_unplug after the unplug is done */ +int xen_platform_pci_unplug; +EXPORT_SYMBOL_GPL(xen_platform_pci_unplug); static int xen_emul_unplug; static int __init check_platform_magic(void) @@ -76,13 +76,13 @@ void __init xen_unplug_emulated_devices(void) /* If the version matches enable the Xen platform PCI driver. * Also enable the Xen platform PCI driver if the version is really old * and the user told us to ignore it. */ - if (!r || (r == XEN_PLATFORM_ERR_MAGIC && - (xen_emul_unplug & XEN_UNPLUG_IGNORE))) - xen_platform_pci_enabled = 1; + if (r && !(r == XEN_PLATFORM_ERR_MAGIC && + (xen_emul_unplug & XEN_UNPLUG_IGNORE))) + return; /* Set the default value of xen_emul_unplug depending on whether or * not the Xen PV frontends and the Xen platform PCI driver have * been compiled for this kernel (modules or built-in are both OK). */ - if (xen_platform_pci_enabled && !xen_emul_unplug) { + if (!xen_emul_unplug) { if (xen_must_unplug_nics()) { printk(KERN_INFO "Netfront and the Xen platform PCI driver have " "been compiled for this kernel: unplug emulated NICs.\n"); @@ -98,8 +98,9 @@ void __init xen_unplug_emulated_devices(void) } } /* Now unplug the emulated devices */ - if (xen_platform_pci_enabled && !(xen_emul_unplug & XEN_UNPLUG_IGNORE)) + if (!(xen_emul_unplug & XEN_UNPLUG_IGNORE)) outw(xen_emul_unplug, XEN_IOPORT_UNPLUG); + xen_platform_pci_unplug = xen_emul_unplug; } static int __init parse_xen_emul_unplug(char *arg) diff --git a/drivers/block/xen-blkfront.c b/drivers/block/xen-blkfront.c index 82ed403..6eb2989 100644 --- a/drivers/block/xen-blkfront.c +++ b/drivers/block/xen-blkfront.c @@ -48,6 +48,7 @@ #include <xen/grant_table.h> #include <xen/events.h> #include <xen/page.h> +#include <xen/platform_pci.h> #include <xen/interface/grant_table.h> #include <xen/interface/io/blkif.h> @@ -737,6 +738,22 @@ static int blkfront_probe(struct xenbus_device *dev, } } + /* no unplug has been done: do not hook devices != xen vbds */ + if (xen_hvm_domain() && (xen_platform_pci_unplug & XEN_UNPLUG_IGNORE)) { + int major; + + if (!VDEV_IS_EXTENDED(vdevice)) + major = BLKIF_MAJOR(vdevice); + else + major = XENVBD_MAJOR; + + if (major != XENVBD_MAJOR) { + printk(KERN_INFO + "%s: HVM does not support vbd %d as xen block device\n", + __FUNCTION__, vdevice); + return -ENODEV; + } + } info = kzalloc(sizeof(*info), GFP_KERNEL); if (!info) { xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure"); diff --git a/drivers/xen/platform-pci.c b/drivers/xen/platform-pci.c index be8b4f3..c01b5dd 100644 --- a/drivers/xen/platform-pci.c +++ b/drivers/xen/platform-pci.c @@ -196,7 +196,9 @@ static struct pci_driver platform_driver = { static int __init platform_pci_module_init(void) { - if (!xen_platform_pci_enabled) + /* no unplug has been done, IGNORE hasn't been specified: just + * return now */ + if (!xen_platform_pci_unplug) return -ENODEV; return pci_register_driver(&platform_driver); diff --git a/drivers/xen/xenbus/xenbus_probe.c b/drivers/xen/xenbus/xenbus_probe.c index 243279a..34287da 100644 --- a/drivers/xen/xenbus/xenbus_probe.c +++ b/drivers/xen/xenbus/xenbus_probe.c @@ -978,7 +978,7 @@ static void wait_for_devices(struct xenbus_driver *xendrv) #ifndef MODULE static int __init boot_wait_for_devices(void) { - if (xen_hvm_domain() && !xen_platform_pci_enabled) + if (xen_hvm_domain() && !xen_platform_pci_unplug) return -ENODEV; ready_to_wait_for_devices = 1; diff --git a/include/xen/platform_pci.h b/include/xen/platform_pci.h index afa8855..ce9d671 100644 --- a/include/xen/platform_pci.h +++ b/include/xen/platform_pci.h @@ -44,6 +44,6 @@ static inline int xen_must_unplug_disks(void) { #endif } -extern bool xen_platform_pci_enabled; +extern int xen_platform_pci_unplug; #endif /* _XEN_PLATFORM_PCI_H */
WARNING: multiple messages have this Message-ID (diff)
From: Stefano Stabellini <stefano.stabellini@eu.citrix.com> To: Don Dutile <ddutile@redhat.com> Cc: "jeremy@goop.org" <jeremy@goop.org>, "xen-devel@lists.xensource.com" <xen-devel@lists.xensource.com>, Stefano Stabellini <Stefano.Stabellini@eu.citrix.com>, "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>, "stefano@stabellini.net" <stefano@stabellini.net>, "sheng@linux.intel.com" <sheng@linux.intel.com> Subject: Re: [PATCH 12/13] Unplug emulated disks and nics Date: Thu, 8 Jul 2010 14:13:17 +0100 [thread overview] Message-ID: <alpine.DEB.2.00.1007081246440.21432@kaball-desktop> (raw) In-Reply-To: <4C34DD1B.3010601@redhat.com> On Wed, 7 Jul 2010, Don Dutile wrote: > Stefano Stabellini wrote: > > On Thu, 1 Jul 2010, Don Dutile wrote: > >> The problem with this check/enable is that if you run > >> this on an older qemu-xen that doesn't have unplug support, > >> it fails pv-hvm configuration. > >> > >> But, all that means is that you can't use an xvd as the boot device, > >> and you have to use an emulated IDE device as boot device. > >> There are a couple ways to configure the vnif correctly (in guest > >> or in xen guest config file). > >> > >> So, on an older (say, rhel5) xen, I don't have this check; > >> the boot device is required to be spec'd as hda, not vda, and > >> xen-blkfront is not allowed to configure blk major nums > >> for IDE (& SCSI) (to avoid 2 drivers twiddling w/same phys backend... not good!). > >> > > > > Who is requiring that the boot device is spec'd as hda and not xvda? > > I don't think there is such limitation in xend or libxl at the moment. > > > > If you take a previous xen HVM guest spec & just run it on a guest > that has pv-hvm added to it, then one has hda spec'd as boot device (by default, > by not editing the guest config spec/file). > > Ideally, both config specs should/would work. I wouldn't want to cause data corruptions by default to people that specified xvda in their HVM config files by mistake (for example because they copied and pasted from a PV guest config file). But I agree that we should be able to do the right thing in your case scenario too. I propose the appended patch (to be merge with "Unplug emulated disks and nics"): if the user specifies xen_emul_unplug=ignore and the unplug protocol is not supported (old xen installations like rhel5), we continue with the PV on HVM initialization and we make sure that blkfront doesn't hook any IDE or SCSI device. Don, Jeremy, what do you think about it? --- diff --git a/arch/x86/xen/platform-pci-unplug.c b/arch/x86/xen/platform-pci-unplug.c index 72a3da6..2f7f3fb 100644 --- a/arch/x86/xen/platform-pci-unplug.c +++ b/arch/x86/xen/platform-pci-unplug.c @@ -29,9 +29,9 @@ #define XEN_PLATFORM_ERR_PROTOCOL -2 #define XEN_PLATFORM_ERR_BLACKLIST -3 -/* boolean to signal that the platform pci device can be used */ -bool xen_platform_pci_enabled; -EXPORT_SYMBOL_GPL(xen_platform_pci_enabled); +/* store the value of xen_emul_unplug after the unplug is done */ +int xen_platform_pci_unplug; +EXPORT_SYMBOL_GPL(xen_platform_pci_unplug); static int xen_emul_unplug; static int __init check_platform_magic(void) @@ -76,13 +76,13 @@ void __init xen_unplug_emulated_devices(void) /* If the version matches enable the Xen platform PCI driver. * Also enable the Xen platform PCI driver if the version is really old * and the user told us to ignore it. */ - if (!r || (r == XEN_PLATFORM_ERR_MAGIC && - (xen_emul_unplug & XEN_UNPLUG_IGNORE))) - xen_platform_pci_enabled = 1; + if (r && !(r == XEN_PLATFORM_ERR_MAGIC && + (xen_emul_unplug & XEN_UNPLUG_IGNORE))) + return; /* Set the default value of xen_emul_unplug depending on whether or * not the Xen PV frontends and the Xen platform PCI driver have * been compiled for this kernel (modules or built-in are both OK). */ - if (xen_platform_pci_enabled && !xen_emul_unplug) { + if (!xen_emul_unplug) { if (xen_must_unplug_nics()) { printk(KERN_INFO "Netfront and the Xen platform PCI driver have " "been compiled for this kernel: unplug emulated NICs.\n"); @@ -98,8 +98,9 @@ void __init xen_unplug_emulated_devices(void) } } /* Now unplug the emulated devices */ - if (xen_platform_pci_enabled && !(xen_emul_unplug & XEN_UNPLUG_IGNORE)) + if (!(xen_emul_unplug & XEN_UNPLUG_IGNORE)) outw(xen_emul_unplug, XEN_IOPORT_UNPLUG); + xen_platform_pci_unplug = xen_emul_unplug; } static int __init parse_xen_emul_unplug(char *arg) diff --git a/drivers/block/xen-blkfront.c b/drivers/block/xen-blkfront.c index 82ed403..6eb2989 100644 --- a/drivers/block/xen-blkfront.c +++ b/drivers/block/xen-blkfront.c @@ -48,6 +48,7 @@ #include <xen/grant_table.h> #include <xen/events.h> #include <xen/page.h> +#include <xen/platform_pci.h> #include <xen/interface/grant_table.h> #include <xen/interface/io/blkif.h> @@ -737,6 +738,22 @@ static int blkfront_probe(struct xenbus_device *dev, } } + /* no unplug has been done: do not hook devices != xen vbds */ + if (xen_hvm_domain() && (xen_platform_pci_unplug & XEN_UNPLUG_IGNORE)) { + int major; + + if (!VDEV_IS_EXTENDED(vdevice)) + major = BLKIF_MAJOR(vdevice); + else + major = XENVBD_MAJOR; + + if (major != XENVBD_MAJOR) { + printk(KERN_INFO + "%s: HVM does not support vbd %d as xen block device\n", + __FUNCTION__, vdevice); + return -ENODEV; + } + } info = kzalloc(sizeof(*info), GFP_KERNEL); if (!info) { xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure"); diff --git a/drivers/xen/platform-pci.c b/drivers/xen/platform-pci.c index be8b4f3..c01b5dd 100644 --- a/drivers/xen/platform-pci.c +++ b/drivers/xen/platform-pci.c @@ -196,7 +196,9 @@ static struct pci_driver platform_driver = { static int __init platform_pci_module_init(void) { - if (!xen_platform_pci_enabled) + /* no unplug has been done, IGNORE hasn't been specified: just + * return now */ + if (!xen_platform_pci_unplug) return -ENODEV; return pci_register_driver(&platform_driver); diff --git a/drivers/xen/xenbus/xenbus_probe.c b/drivers/xen/xenbus/xenbus_probe.c index 243279a..34287da 100644 --- a/drivers/xen/xenbus/xenbus_probe.c +++ b/drivers/xen/xenbus/xenbus_probe.c @@ -978,7 +978,7 @@ static void wait_for_devices(struct xenbus_driver *xendrv) #ifndef MODULE static int __init boot_wait_for_devices(void) { - if (xen_hvm_domain() && !xen_platform_pci_enabled) + if (xen_hvm_domain() && !xen_platform_pci_unplug) return -ENODEV; ready_to_wait_for_devices = 1; diff --git a/include/xen/platform_pci.h b/include/xen/platform_pci.h index afa8855..ce9d671 100644 --- a/include/xen/platform_pci.h +++ b/include/xen/platform_pci.h @@ -44,6 +44,6 @@ static inline int xen_must_unplug_disks(void) { #endif } -extern bool xen_platform_pci_enabled; +extern int xen_platform_pci_unplug; #endif /* _XEN_PLATFORM_PCI_H */
next prev parent reply other threads:[~2010-07-08 13:12 UTC|newest] Thread overview: 53+ messages / expand[flat|nested] mbox.gz Atom feed top 2010-06-21 16:12 [PATCH 0/12] PV on HVM Xen Stefano Stabellini 2010-06-21 16:12 ` Stefano Stabellini 2010-06-21 16:13 ` [PATCH 01/13] Add support for hvm_op stefano 2010-06-21 16:13 ` [PATCH 02/13] early PV on HVM stefano 2010-06-21 16:13 ` [PATCH 03/13] evtchn delivery " stefano 2010-06-21 16:13 ` [PATCH 04/13] Xen PCI platform device driver stefano 2010-06-30 17:55 ` Konrad Rzeszutek Wilk 2010-06-30 17:55 ` Konrad Rzeszutek Wilk 2010-07-01 11:38 ` Stefano Stabellini 2010-06-21 16:13 ` [PATCH 05/13] Add suspend\resume support for PV on HVM guests stefano 2010-06-21 16:14 ` [PATCH 06/13] Allow xen platform pci device to be compiled as a module stefano 2010-06-21 16:14 ` [PATCH 0/12] PV on HVM Xen Stefano Stabellini 2010-06-21 16:14 ` Stefano Stabellini 2010-06-21 16:14 ` [PATCH 07/13] Fix find_unbound_irq in presence of ioapic irqs stefano 2010-06-21 16:14 ` [PATCH 08/13] Fix possible NULL pointer dereference in print_IO_APIC stefano 2010-06-21 16:14 ` [PATCH 09/13] __setup_vector_irq: handle NULL chip_data stefano 2010-06-21 16:14 ` [PATCH 10/13] Do not try to disable hpet if it hasn't been initialized before stefano 2010-06-30 17:53 ` Konrad Rzeszutek Wilk 2010-06-30 17:53 ` Konrad Rzeszutek Wilk 2010-07-09 1:05 ` john stultz 2010-06-30 21:24 ` Venkatesh Pallipadi 2010-06-30 21:24 ` Venkatesh Pallipadi 2010-07-02 10:44 ` Paolo Bonzini 2010-07-02 10:44 ` Paolo Bonzini 2010-06-21 16:14 ` [PATCH 11/13] Use xen_vcpuop_clockevent, xen_clocksource and xen wallclock stefano 2010-07-01 19:41 ` Don Dutile 2010-07-02 17:17 ` Stefano Stabellini 2010-07-02 17:17 ` Stefano Stabellini 2010-06-21 16:14 ` [PATCH 12/13] Unplug emulated disks and nics stefano 2010-07-01 19:41 ` Don Dutile 2010-07-01 19:41 ` Don Dutile 2010-07-05 11:58 ` Stefano Stabellini 2010-07-07 20:01 ` Don Dutile 2010-07-08 13:13 ` Stefano Stabellini [this message] 2010-07-08 13:13 ` Stefano Stabellini 2010-07-08 19:57 ` [Xen-devel] " Don Dutile 2010-07-08 21:29 ` Ian Campbell 2010-07-08 21:29 ` Ian Campbell 2010-07-08 21:59 ` [Xen-devel] " Don Dutile 2010-07-09 8:02 ` Ian Campbell 2010-07-09 8:02 ` Ian Campbell 2010-07-09 10:54 ` Stefano Stabellini 2010-07-09 10:54 ` Stefano Stabellini 2010-07-09 13:42 ` [Xen-devel] " Don Dutile 2010-06-21 16:14 ` [PATCH 13/13] Call HVMOP_pagetable_dying on exit_mmap stefano 2010-06-30 17:56 ` [PATCH 0/12] PV on HVM Xen Konrad Rzeszutek Wilk 2010-07-01 11:38 ` Stefano Stabellini 2010-07-01 19:43 ` [Xen-devel] " Dan Magenheimer 2010-07-02 10:43 ` Stefano Stabellini 2010-07-02 16:08 ` Dan Magenheimer 2010-07-02 17:14 ` Stefano Stabellini 2010-07-02 19:49 ` Dan Magenheimer 2010-07-02 19:49 ` Dan Magenheimer
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to=alpine.DEB.2.00.1007081246440.21432@kaball-desktop \ --to=stefano.stabellini@eu.citrix.com \ --cc=ddutile@redhat.com \ --cc=jeremy@goop.org \ --cc=linux-kernel@vger.kernel.org \ --cc=sheng@linux.intel.com \ --cc=stefano@stabellini.net \ --cc=xen-devel@lists.xensource.com \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: linkBe sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.