Blkback cannot work properly on auto-translated guests if Xen doesn't
update the IOMMU when performing grant maps/unmaps, so only attach if
the newly introduced XENFEAT_hvm_gntmap_supports_iommu is found.
Signed-off-by: Roger Pau Monné <[email protected]>
Cc: Konrad Rzeszutek Wilk <[email protected]>
Cc: David Vrabel <[email protected]>
Cc: Boris Ostrovsky <[email protected]>
---
drivers/block/xen-blkback/blkback.c | 12 ++++++++++++
include/xen/interface/features.h | 6 ++++++
2 files changed, 18 insertions(+), 0 deletions(-)
diff --git a/drivers/block/xen-blkback/blkback.c b/drivers/block/xen-blkback/blkback.c
index 64c60ed..4c23ddd 100644
--- a/drivers/block/xen-blkback/blkback.c
+++ b/drivers/block/xen-blkback/blkback.c
@@ -1381,6 +1381,18 @@ static int __init xen_blkif_init(void)
if (!xen_domain())
return -ENODEV;
+#ifdef CONFIG_X86
+ if (xen_feature(XENFEAT_auto_translated_physmap) &&
+ !xen_feature(XENFEAT_hvm_gntmap_supports_iommu)) {
+ /*
+ * blkback cannot work properly on auto-translated guests
+ * if grant table mappings don't update the IOMMU entries.
+ */
+ pr_debug(DRV_PFX "Disabling blkback because Xen is missing feature XENFEAT_hvm_gntmap_supports_iommu\n");
+ return -ENODEV;
+ }
+#endif /* CONFIG_X86 */
+
rc = xen_blkif_interface_init();
if (rc)
goto failed_init;
diff --git a/include/xen/interface/features.h b/include/xen/interface/features.h
index 131a6cc..d1e0f23 100644
--- a/include/xen/interface/features.h
+++ b/include/xen/interface/features.h
@@ -53,6 +53,12 @@
/* operation as Dom0 is supported */
#define XENFEAT_dom0 11
+/*
+ * If set, GNTTABOP_map_grant_ref sets the proper IOMMU mappings so the
+ * resulting mapped page can be used for IO with hardware devices.
+ */
+#define XENFEAT_hvm_gntmap_supports_iommu 12
+
#define XENFEAT_NR_SUBMAPS 1
#endif /* __XEN_PUBLIC_FEATURES_H__ */
--
1.7.7.5 (Apple Git-26)
On 04/04/14 15:41, Roger Pau Monne wrote:
> Blkback cannot work properly on auto-translated guests if Xen doesn't
> update the IOMMU when performing grant maps/unmaps, so only attach if
> the newly introduced XENFEAT_hvm_gntmap_supports_iommu is found.
Can you explain the problem in more detail and which guest
configurations are affected?
This isn't a problem that is specific to blkback, but any backend that
grant maps and passes the foreign pages to a device so there needs to be
a generic solution.
Why can't this be fixed by having the swiotlb bounce foreign pages?
> --- a/drivers/block/xen-blkback/blkback.c
> +++ b/drivers/block/xen-blkback/blkback.c
> @@ -1381,6 +1381,18 @@ static int __init xen_blkif_init(void)
> if (!xen_domain())
> return -ENODEV;
>
> +#ifdef CONFIG_X86
> + if (xen_feature(XENFEAT_auto_translated_physmap) &&
> + !xen_feature(XENFEAT_hvm_gntmap_supports_iommu)) {
> + /*
> + * blkback cannot work properly on auto-translated guests
> + * if grant table mappings don't update the IOMMU entries.
> + */
> + pr_debug(DRV_PFX "Disabling blkback because Xen is missing feature XENFEAT_hvm_gntmap_supports_iommu\n");
> + return -ENODEV;
> + }
> +#endif /* CONFIG_X86 */
Why CONFIG_X86?
David
On 04/04/14 17:01, David Vrabel wrote:
> On 04/04/14 15:41, Roger Pau Monne wrote:
>> Blkback cannot work properly on auto-translated guests if Xen doesn't
>> update the IOMMU when performing grant maps/unmaps, so only attach if
>> the newly introduced XENFEAT_hvm_gntmap_supports_iommu is found.
>
> Can you explain the problem in more detail and which guest
> configurations are affected?
>
> This isn't a problem that is specific to blkback, but any backend that
> grant maps and passes the foreign pages to a device so there needs to be
> a generic solution.
>
> Why can't this be fixed by having the swiotlb bounce foreign pages?
Yes, it could be fixed with that, FreeBSD has been doing it for a long time.
But since nobody has complained about this, and the issue has been there
for a long time, I'm not sure it's worth adding all the bounce buffer
code for what seems to be an unused configuration.
Also, since Linux also supports PV, I would just say that if you want to
use driver domains you are forced to use PV guests (or have a recent Xen
version that properly updates the IOMMU on grant map/unmap).
>> --- a/drivers/block/xen-blkback/blkback.c
>> +++ b/drivers/block/xen-blkback/blkback.c
>> @@ -1381,6 +1381,18 @@ static int __init xen_blkif_init(void)
>> if (!xen_domain())
>> return -ENODEV;
>>
>> +#ifdef CONFIG_X86
>> + if (xen_feature(XENFEAT_auto_translated_physmap) &&
>> + !xen_feature(XENFEAT_hvm_gntmap_supports_iommu)) {
>> + /*
>> + * blkback cannot work properly on auto-translated guests
>> + * if grant table mappings don't update the IOMMU entries.
>> + */
>> + pr_debug(DRV_PFX "Disabling blkback because Xen is missing feature XENFEAT_hvm_gntmap_supports_iommu\n");
>> + return -ENODEV;
>> + }
>> +#endif /* CONFIG_X86 */
>
> Why CONFIG_X86?
ARM has a bounce buffer for foreign pages, so this is not needed.
>
> David
>
On Fri, Apr 04, 2014 at 04:41:11PM +0200, Roger Pau Monne wrote:
> Blkback cannot work properly on auto-translated guests if Xen doesn't
> update the IOMMU when performing grant maps/unmaps, so only attach if
> the newly introduced XENFEAT_hvm_gntmap_supports_iommu is found.
Could you point to the git commit that introduces it in the patch please?
>
> Signed-off-by: Roger Pau Monn? <[email protected]>
> Cc: Konrad Rzeszutek Wilk <[email protected]>
> Cc: David Vrabel <[email protected]>
> Cc: Boris Ostrovsky <[email protected]>
> ---
> drivers/block/xen-blkback/blkback.c | 12 ++++++++++++
> include/xen/interface/features.h | 6 ++++++
> 2 files changed, 18 insertions(+), 0 deletions(-)
>
> diff --git a/drivers/block/xen-blkback/blkback.c b/drivers/block/xen-blkback/blkback.c
> index 64c60ed..4c23ddd 100644
> --- a/drivers/block/xen-blkback/blkback.c
> +++ b/drivers/block/xen-blkback/blkback.c
> @@ -1381,6 +1381,18 @@ static int __init xen_blkif_init(void)
> if (!xen_domain())
> return -ENODEV;
>
> +#ifdef CONFIG_X86
> + if (xen_feature(XENFEAT_auto_translated_physmap) &&
> + !xen_feature(XENFEAT_hvm_gntmap_supports_iommu)) {
> + /*
> + * blkback cannot work properly on auto-translated guests
> + * if grant table mappings don't update the IOMMU entries.
> + */
> + pr_debug(DRV_PFX "Disabling blkback because Xen is missing feature XENFEAT_hvm_gntmap_supports_iommu\n");
> + return -ENODEV;
> + }
> +#endif /* CONFIG_X86 */
> +
> rc = xen_blkif_interface_init();
> if (rc)
> goto failed_init;
> diff --git a/include/xen/interface/features.h b/include/xen/interface/features.h
> index 131a6cc..d1e0f23 100644
> --- a/include/xen/interface/features.h
> +++ b/include/xen/interface/features.h
> @@ -53,6 +53,12 @@
> /* operation as Dom0 is supported */
> #define XENFEAT_dom0 11
>
> +/*
> + * If set, GNTTABOP_map_grant_ref sets the proper IOMMU mappings so the
> + * resulting mapped page can be used for IO with hardware devices.
> + */
> +#define XENFEAT_hvm_gntmap_supports_iommu 12
> +
> #define XENFEAT_NR_SUBMAPS 1
>
> #endif /* __XEN_PUBLIC_FEATURES_H__ */
> --
> 1.7.7.5 (Apple Git-26)
>
On 04/04/14 16:13, Roger Pau Monné wrote:
> On 04/04/14 17:01, David Vrabel wrote:
>> On 04/04/14 15:41, Roger Pau Monne wrote:
>>> Blkback cannot work properly on auto-translated guests if Xen doesn't
>>> update the IOMMU when performing grant maps/unmaps, so only attach if
>>> the newly introduced XENFEAT_hvm_gntmap_supports_iommu is found.
>>
>> Can you explain the problem in more detail and which guest
>> configurations are affected?
>>
>> This isn't a problem that is specific to blkback, but any backend that
>> grant maps and passes the foreign pages to a device so there needs to be
>> a generic solution.
>>
>> Why can't this be fixed by having the swiotlb bounce foreign pages?
>
> Yes, it could be fixed with that, FreeBSD has been doing it for a long time.
>
> But since nobody has complained about this, and the issue has been there
> for a long time, I'm not sure it's worth adding all the bounce buffer
> code for what seems to be an unused configuration.
Ok. Now that netback also grant maps. Can you make this a generic
function and call from netback and blkback?
David