hw/vfio-user/container.c | 10 +--------- 1 file changed, 1 insertion(+), 9 deletions(-)
As the full cpr implementation is yet to be merged upstream, do not register
the vfio-user container with cpr. Full vfio-user support for cpr can be
merged later as a follow-up series.
Signed-off-by: Mark Cave-Ayland <mark.caveayland@nutanix.com>
---
hw/vfio-user/container.c | 10 +---------
1 file changed, 1 insertion(+), 9 deletions(-)
diff --git a/hw/vfio-user/container.c b/hw/vfio-user/container.c
index 3133fef177..4ee99fc2cc 100644
--- a/hw/vfio-user/container.c
+++ b/hw/vfio-user/container.c
@@ -225,14 +225,10 @@ vfio_user_container_connect(AddressSpace *as, VFIODevice *vbasedev,
bcontainer = &container->bcontainer;
- if (!vfio_cpr_register_container(bcontainer, errp)) {
- goto free_container_exit;
- }
-
ret = ram_block_uncoordinated_discard_disable(true);
if (ret) {
error_setg_errno(errp, -ret, "Cannot set discarding of RAM broken");
- goto unregister_container_exit;
+ goto free_container_exit;
}
vioc = VFIO_IOMMU_GET_CLASS(bcontainer);
@@ -261,9 +257,6 @@ listener_release_exit:
enable_discards_exit:
ram_block_uncoordinated_discard_disable(false);
-unregister_container_exit:
- vfio_cpr_unregister_container(bcontainer);
-
free_container_exit:
object_unref(container);
@@ -286,7 +279,6 @@ static void vfio_user_container_disconnect(VFIOUserContainer *container)
vioc->release(bcontainer);
}
- vfio_cpr_unregister_container(bcontainer);
object_unref(container);
vfio_address_space_put(space);
--
2.43.0On 7/2/25 13:59, Mark Cave-Ayland wrote:
> As the full cpr implementation is yet to be merged upstream, do not register
> the vfio-user container with cpr. Full vfio-user support for cpr can be
> merged later as a follow-up series.
>
> Signed-off-by: Mark Cave-Ayland <mark.caveayland@nutanix.com>
> ---
> hw/vfio-user/container.c | 10 +---------
> 1 file changed, 1 insertion(+), 9 deletions(-)
>
> diff --git a/hw/vfio-user/container.c b/hw/vfio-user/container.c
> index 3133fef177..4ee99fc2cc 100644
> --- a/hw/vfio-user/container.c
> +++ b/hw/vfio-user/container.c
> @@ -225,14 +225,10 @@ vfio_user_container_connect(AddressSpace *as, VFIODevice *vbasedev,
>
> bcontainer = &container->bcontainer;
>
> - if (!vfio_cpr_register_container(bcontainer, errp)) {
> - goto free_container_exit;
> - }
> -
> ret = ram_block_uncoordinated_discard_disable(true);
> if (ret) {
> error_setg_errno(errp, -ret, "Cannot set discarding of RAM broken");
> - goto unregister_container_exit;
> + goto free_container_exit;
> }
>
> vioc = VFIO_IOMMU_GET_CLASS(bcontainer);
> @@ -261,9 +257,6 @@ listener_release_exit:
> enable_discards_exit:
> ram_block_uncoordinated_discard_disable(false);
>
> -unregister_container_exit:
> - vfio_cpr_unregister_container(bcontainer);
> -
> free_container_exit:
> object_unref(container);
>
> @@ -286,7 +279,6 @@ static void vfio_user_container_disconnect(VFIOUserContainer *container)
> vioc->release(bcontainer);
> }
>
> - vfio_cpr_unregister_container(bcontainer);
> object_unref(container);
>
> vfio_address_space_put(space);
Applied to vfio-next.
Thanks,
C.
On 7/2/2025 7:59 AM, Mark Cave-Ayland wrote:
> As the full cpr implementation is yet to be merged upstream, do not register
> the vfio-user container with cpr. Full vfio-user support for cpr can be
> merged later as a follow-up series.
>
> Signed-off-by: Mark Cave-Ayland <mark.caveayland@nutanix.com>
This is fine, or I could drop my pending patch "vfio/container: delete old cpr register".
Thus the vfio-user container would be registered for cpr-reboot mode, which is fine
at this time.
- Steve
> ---
> hw/vfio-user/container.c | 10 +---------
> 1 file changed, 1 insertion(+), 9 deletions(-)
>
> diff --git a/hw/vfio-user/container.c b/hw/vfio-user/container.c
> index 3133fef177..4ee99fc2cc 100644
> --- a/hw/vfio-user/container.c
> +++ b/hw/vfio-user/container.c
> @@ -225,14 +225,10 @@ vfio_user_container_connect(AddressSpace *as, VFIODevice *vbasedev,
>
> bcontainer = &container->bcontainer;
>
> - if (!vfio_cpr_register_container(bcontainer, errp)) {
> - goto free_container_exit;
> - }
> -
> ret = ram_block_uncoordinated_discard_disable(true);
> if (ret) {
> error_setg_errno(errp, -ret, "Cannot set discarding of RAM broken");
> - goto unregister_container_exit;
> + goto free_container_exit;
> }
>
> vioc = VFIO_IOMMU_GET_CLASS(bcontainer);
> @@ -261,9 +257,6 @@ listener_release_exit:
> enable_discards_exit:
> ram_block_uncoordinated_discard_disable(false);
>
> -unregister_container_exit:
> - vfio_cpr_unregister_container(bcontainer);
> -
> free_container_exit:
> object_unref(container);
>
> @@ -286,7 +279,6 @@ static void vfio_user_container_disconnect(VFIOUserContainer *container)
> vioc->release(bcontainer);
> }
>
> - vfio_cpr_unregister_container(bcontainer);
> object_unref(container);
>
> vfio_address_space_put(space);
On 02/07/2025 13:07, Steven Sistare wrote: > On 7/2/2025 7:59 AM, Mark Cave-Ayland wrote: >> As the full cpr implementation is yet to be merged upstream, do not >> register >> the vfio-user container with cpr. Full vfio-user support for cpr can be >> merged later as a follow-up series. >> >> Signed-off-by: Mark Cave-Ayland <mark.caveayland@nutanix.com> > > This is fine, or I could drop my pending patch "vfio/container: delete > old cpr register". > Thus the vfio-user container would be registered for cpr-reboot mode, > which is fine > at this time. Have you got a pointer to a commit showing how the registration should be done without using the old register/unregister functions? We are certainly interested in looking at cpr here at Nutanix, but since any work is in its early stages I don't think it matters too much if the old functions were to disappear. ATB, Mark.
On 7/2/2025 9:21 AM, Mark Cave-Ayland wrote: > On 02/07/2025 13:07, Steven Sistare wrote: > >> On 7/2/2025 7:59 AM, Mark Cave-Ayland wrote: >>> As the full cpr implementation is yet to be merged upstream, do not register >>> the vfio-user container with cpr. Full vfio-user support for cpr can be >>> merged later as a follow-up series. >>> >>> Signed-off-by: Mark Cave-Ayland <mark.caveayland@nutanix.com> >> >> This is fine, or I could drop my pending patch "vfio/container: delete old cpr register". >> Thus the vfio-user container would be registered for cpr-reboot mode, which is fine >> at this time. > > Have you got a pointer to a commit showing how the registration should be done without using the old register/unregister functions? We are certainly interested in looking at cpr here at Nutanix, but since any work is in its early stages I don't think it matters too much if the old functions were to disappear. No. cpr-transfer for vfio-user is going to be a bit tricky because of the external process, and the existing registrations will not cover it. - Steve
On 02/07/2025 14:24, Steven Sistare wrote: > On 7/2/2025 9:21 AM, Mark Cave-Ayland wrote: >> On 02/07/2025 13:07, Steven Sistare wrote: >> >>> On 7/2/2025 7:59 AM, Mark Cave-Ayland wrote: >>>> As the full cpr implementation is yet to be merged upstream, do not >>>> register >>>> the vfio-user container with cpr. Full vfio-user support for cpr can be >>>> merged later as a follow-up series. >>>> >>>> Signed-off-by: Mark Cave-Ayland <mark.caveayland@nutanix.com> >>> >>> This is fine, or I could drop my pending patch "vfio/container: >>> delete old cpr register". >>> Thus the vfio-user container would be registered for cpr-reboot mode, >>> which is fine >>> at this time. >> >> Have you got a pointer to a commit showing how the registration should >> be done without using the old register/unregister functions? We are >> certainly interested in looking at cpr here at Nutanix, but since any >> work is in its early stages I don't think it matters too much if the >> old functions were to disappear. > > No. cpr-transfer for vfio-user is going to be a bit tricky because of > the external process, > and the existing registrations will not cover it. Got it. In that case I'm happy for Cédric to take my earlier patch to remove the old register/unregister functions from vfio-user to avoid blocking your work. ATB, Mark.
On 7/2/2025 9:30 AM, Mark Cave-Ayland wrote: > On 02/07/2025 14:24, Steven Sistare wrote: > >> On 7/2/2025 9:21 AM, Mark Cave-Ayland wrote: >>> On 02/07/2025 13:07, Steven Sistare wrote: >>> >>>> On 7/2/2025 7:59 AM, Mark Cave-Ayland wrote: >>>>> As the full cpr implementation is yet to be merged upstream, do not register >>>>> the vfio-user container with cpr. Full vfio-user support for cpr can be >>>>> merged later as a follow-up series. >>>>> >>>>> Signed-off-by: Mark Cave-Ayland <mark.caveayland@nutanix.com> >>>> >>>> This is fine, or I could drop my pending patch "vfio/container: delete old cpr register". >>>> Thus the vfio-user container would be registered for cpr-reboot mode, which is fine >>>> at this time. >>> >>> Have you got a pointer to a commit showing how the registration should be done without using the old register/unregister functions? We are certainly interested in looking at cpr here at Nutanix, but since any work is in its early stages I don't think it matters too much if the old functions were to disappear. >> >> No. cpr-transfer for vfio-user is going to be a bit tricky because of the external process, >> and the existing registrations will not cover it. > > Got it. In that case I'm happy for Cédric to take my earlier patch to remove the old register/unregister functions from vfio-user to avoid blocking your work. It will not block my work. Leaving the old registrations in place in your patch and mine allows cpr-reboot to work for vfio-user. I would simply drop my patch that deletes it, and rework the registration to support all cpr modes in the future. - Steve
On 02/07/2025 14:39, Steven Sistare wrote: > On 7/2/2025 9:30 AM, Mark Cave-Ayland wrote: >> On 02/07/2025 14:24, Steven Sistare wrote: >> >>> On 7/2/2025 9:21 AM, Mark Cave-Ayland wrote: >>>> On 02/07/2025 13:07, Steven Sistare wrote: >>>> >>>>> On 7/2/2025 7:59 AM, Mark Cave-Ayland wrote: >>>>>> As the full cpr implementation is yet to be merged upstream, do >>>>>> not register >>>>>> the vfio-user container with cpr. Full vfio-user support for cpr >>>>>> can be >>>>>> merged later as a follow-up series. >>>>>> >>>>>> Signed-off-by: Mark Cave-Ayland <mark.caveayland@nutanix.com> >>>>> >>>>> This is fine, or I could drop my pending patch "vfio/container: >>>>> delete old cpr register". >>>>> Thus the vfio-user container would be registered for cpr-reboot >>>>> mode, which is fine >>>>> at this time. >>>> >>>> Have you got a pointer to a commit showing how the registration >>>> should be done without using the old register/unregister functions? >>>> We are certainly interested in looking at cpr here at Nutanix, but >>>> since any work is in its early stages I don't think it matters too >>>> much if the old functions were to disappear. >>> >>> No. cpr-transfer for vfio-user is going to be a bit tricky because >>> of the external process, >>> and the existing registrations will not cover it. >> >> Got it. In that case I'm happy for Cédric to take my earlier patch to >> remove the old register/unregister functions from vfio-user to avoid >> blocking your work. > > It will not block my work. Leaving the old registrations in place in > your patch and > mine allows cpr-reboot to work for vfio-user. I would simply drop my > patch that > deletes it, and rework the registration to support all cpr modes in the > future. Okay, great! I'm happy for you to CC me on your future work to help review patches in this area. ATB, Mark.
On 7/2/25 13:59, Mark Cave-Ayland wrote:
> As the full cpr implementation is yet to be merged upstream, do not register
> the vfio-user container with cpr. Full vfio-user support for cpr can be
> merged later as a follow-up series.
>
> Signed-off-by: Mark Cave-Ayland <mark.caveayland@nutanix.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
Thanks,
C.
> ---
> hw/vfio-user/container.c | 10 +---------
> 1 file changed, 1 insertion(+), 9 deletions(-)
>
> diff --git a/hw/vfio-user/container.c b/hw/vfio-user/container.c
> index 3133fef177..4ee99fc2cc 100644
> --- a/hw/vfio-user/container.c
> +++ b/hw/vfio-user/container.c
> @@ -225,14 +225,10 @@ vfio_user_container_connect(AddressSpace *as, VFIODevice *vbasedev,
>
> bcontainer = &container->bcontainer;
>
> - if (!vfio_cpr_register_container(bcontainer, errp)) {
> - goto free_container_exit;
> - }
> -
> ret = ram_block_uncoordinated_discard_disable(true);
> if (ret) {
> error_setg_errno(errp, -ret, "Cannot set discarding of RAM broken");
> - goto unregister_container_exit;
> + goto free_container_exit;
> }
>
> vioc = VFIO_IOMMU_GET_CLASS(bcontainer);
> @@ -261,9 +257,6 @@ listener_release_exit:
> enable_discards_exit:
> ram_block_uncoordinated_discard_disable(false);
>
> -unregister_container_exit:
> - vfio_cpr_unregister_container(bcontainer);
> -
> free_container_exit:
> object_unref(container);
>
> @@ -286,7 +279,6 @@ static void vfio_user_container_disconnect(VFIOUserContainer *container)
> vioc->release(bcontainer);
> }
>
> - vfio_cpr_unregister_container(bcontainer);
> object_unref(container);
>
> vfio_address_space_put(space);
© 2016 - 2026 Red Hat, Inc.