On Sat, Apr 22, 2017 at 05:23:49PM +0800, Hailiang Zhang wrote: > Hi, Hi Hailiang,
> I think the bellow patch can fix your problme. > [PATCH 2/4] qmp-cont: invalidate on RUN_STATE_PRELAUNCH > https://patchwork.kernel.org/patch/9591885/ Hmm, the above patch ("qmp-cont: invalidate on RUN_STATE_PRELAUNCH") is not merged in Git, as it's stalled on design discussion between Kevin Wolf and Vladimir. And the below patch, from you, seems to be not submitted upstream (2.8 stable tree, perhaps). Do you intend to do so? > Actually, we encounter the same problem in our test, we fix it with the > follow patch: > > From 0e4d6d706afd9909b5fd71536b45c58af60892f8 Mon Sep 17 00:00:00 2001 > From: zhanghailiang<zhang.zhanghaili...@huawei.com> > Date: Tue, 21 Mar 2017 09:44:36 +0800 > Subject: [PATCH] migration: Re-activate blocks whenever migration been > cancelled > > In commit 1d2acc3162d9c7772510c973f446353fbdd1f9a8, we try to fix the bug > 'bdrv_co_do_pwritev: Assertion `!(bs->open_flags & 0x0800)' failed' > which occured in migration cancelling process. > > But it seems that we didn't cover all the cases, we caught such a case > which > slipped from the old fixup in our test: if libvirtd cancelled the > migration > process for a shutting down VM, it will send 'system_reset' command > first, > and then 'cont' command behind, after VM resumes to run, it will trigger > the above > error reports, because we didn't regain the control of blocks for VM. > > Signed-off-by: zhanghailiang<zhang.zhanghaili...@huawei.com> > Signed-off-by: Hongyang Yang<yanghongy...@huawei.com> > --- > block.c | 12 +++++++++++- > include/block/block.h | 1 + > include/migration/migration.h | 3 --- > migration/migration.c | 7 +------ > qmp.c | 4 +--- > 5 files changed, 14 insertions(+), 13 deletions(-) [...] -- /kashyap
