xen-bus/block: explicitly assign event channels to an AioContext
PATCH 1/1 xen-bus/block: explicitly assign event channels to an AioContext
It is not safe to close an event channel from the QEMU main thread when that channel's poller is running in IOThread context.
This patch adds a new xen_device_set_event_channel_context() function to explicitly assign the channel AioContext, and modifies xen_device_bind_event_channel() to initially assign the channel's poller to the QEMU main thread context. The code in xen-block's dataplane is then modified to assign the channel to IOThread context during xen_block_dataplane_start() and de-assign it during in xen_block_dataplane_stop(), such that the channel is always assigned back to main thread context before it is closed. aio_set_fd_handler() already deals with all the necessary synchronization when moving an fd between AioContext-s so no extra code is needed to manage this.
---
Cc: Stefano Stabellini <[email protected]>
Cc: Anthony Perard <[email protected]>
Cc: Stefan Hajnoczi <[email protected]>
Cc: Kevin Wolf <[email protected]>
Cc: Max Reitz <[email protected]>
Tested against an HVM debian guest with a QCOW2 image as system disk, and as a hot-plugged/unplgged secondary disk.
--- hw/block/dataplane/xen-block.c | 20 ++++++++++++++++++-- hw/xen/xen-bus.c | 27 +++++++++++++++++++++++---- include/hw/xen/xen-bus.h | 5 ++++- 3 files changed, 45 insertions(+), 7 deletions(-)
hw/block/dataplane/xen-block.c +18 -2
@@ -685,12 +685,24 @@ void xen_block_dataplane_stop(XenBlockDataPlane *dataplane)return ; } + xendev = dataplane->xendev; +aio_context_acquire (dataplane->ctx); +if (dataplane->event_channel) { +/* Only reason for failure is a NULL channel */ +xen_device_set_event_channel_context (xendev, dataplane->event_channel, +qemu_get_aio_context (), + &error_abort); + }/* Xen doesn't have multiple users for nodes, so this can't fail */ blk_set_aio_context (dataplane->blk,qemu_get_aio_context (), &error_abort);aio_context_release (dataplane->ctx); - xendev = dataplane->xendev; +/* +* Now that the context has been moved onto the main thread, cancel +* further processing. +*/ +qemu_bh_cancel (dataplane->bh);if (dataplane->event_channel) {Error *local_err =NULL ;
@@ -807,7 +819,7 @@ void xen_block_dataplane_start(XenBlockDataPlane *dataplane, } dataplane->event_channel = -xen_device_bind_event_channel (xendev, dataplane->ctx, event_channel, +xen_device_bind_event_channel (xendev, event_channel, xen_block_dataplane_event, dataplane, &local_err);if (local_err) {
@@ -818,7 +830,11 @@ void xen_block_dataplane_start(XenBlockDataPlane *dataplane,aio_context_acquire (dataplane->ctx);/* If other users keep the BlockBackend in the iothread, that's ok */ blk_set_aio_context (dataplane->blk, dataplane->ctx,NULL ); +/* Only reason for failure is a NULL channel */ +xen_device_set_event_channel_context (xendev, dataplane->event_channel, + dataplane->ctx, &error_abort);aio_context_release (dataplane->ctx); +return ; stop:
hw/xen/xen-bus.c +23 -4
@@ -1089,8 +1089,26 @@ static void xen_device_event(void *opaque) } } +void xen_device_set_event_channel_context (XenDevice *xendev, +XenEventChannel *channel, +AioContext *ctx, +Error **errp) +{ +if (!channel) { +error_setg (errp,"bad channel" ); +return ; + } + +if (channel->ctx) +aio_set_fd_handler (channel->ctx,xenevtchn_fd (channel->xeh), true, +NULL ,NULL ,NULL ,NULL ); + + channel->ctx = ctx; +aio_set_fd_handler (channel->ctx,xenevtchn_fd (channel->xeh), true, + xen_device_event,NULL , xen_device_poll, channel); +} +XenEventChannel *xen_device_bind_event_channel (XenDevice *xendev, -AioContext *ctx,unsigned int port,XenEventHandler handler,void *opaque,Error **errp)
@@ -1116,9 +1134,10 @@ XenEventChannel *xen_device_bind_event_channel(XenDevice *xendev, channel->handler = handler; channel->opaque = opaque; - channel->ctx = ctx; -aio_set_fd_handler (channel->ctx,xenevtchn_fd (channel->xeh), true, - xen_device_event,NULL , xen_device_poll, channel); +/* Only reason for failure is a NULL channel */ +xen_device_set_event_channel_context (xendev, channel, +qemu_get_aio_context (), + &error_abort);QLIST_INSERT_HEAD (&xendev->event_channels, channel, list);
include/hw/xen/xen-bus.h +4 -1
@@ -128,10 +128,13 @@ void xen_device_copy_grant_refs(XenDevice *xendev, bool to_domain,typedef bool (*XenEventHandler )(void *opaque);XenEventChannel *xen_device_bind_event_channel (XenDevice *xendev, -AioContext *ctx,unsigned int port,XenEventHandler handler,void *opaque,Error **errp); +void xen_device_set_event_channel_context (XenDevice *xendev, +XenEventChannel *channel, +AioContext *ctx, +Error **errp);void xen_device_notify_event_channel (XenDevice *xendev,XenEventChannel *channel,Error **errp);
Signature · 7 lines
2.20.1 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Quoted 14 lines · Paul Durrant
It is not safe to close an event channel from the QEMU main thread when that channel's poller is running in IOThread context.
This patch adds a new
xen_device_set_event_channel_context()function to explicitly assign the channel AioContext, and modifiesxen_device_bind_event_channel()to initially assign the channel's poller to the QEMU main thread context. The code in xen-block's dataplane is then modified to assign the channel to IOThread context duringxen_block_dataplane_start()and de-assign it during inxen_block_dataplane_stop(), such that the channel is always assigned back to main thread context before it is closed.aio_set_fd_handler()already deals with all the necessary synchronization when moving an fd between AioContext-s so no extra code is needed to manage this.
Quoted 2 lines
Reported-by Julien Grall <[email protected]>Signed-off-by Paul Durrant <[email protected]>
Quoted 2 lines
Tested against an HVM debian guest with a QCOW2 image as system disk, and as a hot-plugged/unplgged secondary disk.
And I've run an osstest flight with the patch.
Thanks,
Signature · 6 lines
Anthony PERARD _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Hi Anthony,
Quoted 15 lines · Anthony PERARD
Quoted 14 lines · Paul Durrant
It is not safe to close an event channel from the QEMU main thread when that channel's poller is running in IOThread context.
This patch adds a new
xen_device_set_event_channel_context()function to explicitly assign the channel AioContext, and modifiesxen_device_bind_event_channel()to initially assign the channel's poller to the QEMU main thread context. The code in xen-block's dataplane is then modified to assign the channel to IOThread context duringxen_block_dataplane_start()and de-assign it during inxen_block_dataplane_stop(), such that the channel is always assigned back to main thread context before it is closed.aio_set_fd_handler()already deals with all the necessary synchronization when moving an fd between AioContext-s so no extra code is needed to manage this.
Quoted 4 lines
Quoted 2 lines
Reported-by Julien Grall <[email protected]>Signed-off-by Paul Durrant <[email protected]>Reviewed-by Anthony PERARD <[email protected]>
I can't find the patch in QEMU upstream. Are we missing any ack/review for this patch?
Quoted 8 lines
Quoted 3 lines
Tested against an HVM debian guest with a QCOW2 image as system disk, and as a hot-plugged/unplgged secondary disk.
And I've run an osstest flight with the patch.
Thanks,
Cheers,
Signature · 6 lines
Julien Grall _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Quoted 18 lines · Julien Grall
Hi Anthony,
Quoted 15 lines · Anthony PERARD
Quoted 14 lines · Paul Durrant
It is not safe to close an event channel from the QEMU main thread when that channel's poller is running in IOThread context.
This patch adds a new
xen_device_set_event_channel_context()function to explicitly assign the channel AioContext, and modifiesxen_device_bind_event_channel()to initially assign the channel's poller to the QEMU main thread context. The code in xen-block's dataplane is then modified to assign the channel to IOThread context duringxen_block_dataplane_start()and de-assign it during inxen_block_dataplane_stop(), such that the channel is always assigned back to main thread context before it is closed.aio_set_fd_handler()already deals with all the necessary synchronization when moving an fd between AioContext-s so no extra code is needed to manage this.
Quoted 7 lines
Quoted 4 lines
Quoted 2 lines
Reported-by Julien Grall <[email protected]>Signed-off-by Paul Durrant <[email protected]>Reviewed-by Anthony PERARD <[email protected]>I can't find the patch in QEMU upstream. Are we missing any ack/review for this patch?
No, I just need to prepare a pull request. It's in my list of patch for upstream, so there will be a pull request at some point before the next QEMU release.
Cheers,
Signature · 6 lines
Anthony PERARD _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel