在 2017/3/7 14:37, Fam Zheng 写道:
On Tue, 03/07 03:53, QingFeng Hao wrote:
Hi All,
I am not sure if the fix is correct because I am not very clear about the
logic in vmstate.c. From my test, once size=0, the iotests case 68 failed
due to the assert. So just send this draft patch for your comments!
The patch is based on commit 17783ac828a "Merge remote-tracking branch
'remotes/dgibson/tags/ppc-for-2.9-20170303' into staging".
I cannot reproduce the failure on either 17783ac828a or current head
56b51708e9e. Both passes for me. I wonder where do you get the size=0.
The error happens when running "savevm 0" in case 068. It can be
manually reproduced
by "./check -qcow2 68" or "./s390x-softmmu/qemu-system-s390x -nodefaults \
-machine accel=qtest -no-shutdown -nographic -monitor stdio -serial none \
-hda /home/mc/gitcheck/work/qemu-master/tree/qemu/tests/t.img.bak", then
type
"savevm 0". t.img.bak is the backup image for t.img generated by 068.
I added the print in vmstate_save_state:
QJSON *vmdesc_loop = vmdesc;
+ error_report("haoqf:%s:opaque:%p, offset:%lx, size:%d, field name:%s,
vname:%s\n", __FUNCTION__, opaque, field->offset, size, field->name,
vmsd->name);
And here is the test log:
haoqf:vmstate_save_state:opaque:0x2aa5a5715c0, offset:122e1, size:1,
field name:env.sigp_order, vname:cpu
haoqf:vmstate_size: field size:4, offset:0
haoqf:vmstate_save_state:opaque:0x2aa5a5715c0, offset:12300, size:4,
field name:irqstate_saved_size, vname:cpu
haoqf:vmstate_size: field size:0, offset:74496
haoqf:vmstate_size: calculated size:0
haoqf:vmstate_save_state:opaque:0x2aa5a5715c0, offset:122f8, size:0,
field name:irqstate, vname:cpu
haoqf:vmstate_save_state:firstelem:(nil), elements: 1
qemu-system-s390x: ../migration/vmstate.c:336: vmstate_save_state:
Assertion `first_elem || !n_elems' failed.
Aborted (core dumped)
I also did the test for x86 with: "./x86_64-softmmu/qemu-system-x86_64
-nodefaults \
-machine accel=qtest -no-shutdown -nographic -monitor stdio -serial none \
-hda /home/mc/gitcheck/work/qemu-master/tree/qemu/tests/t.img.bak",
and then ran "savevm 0", but it didn't core and the size are all non-zero:
haoqf:vmstate_save calling vmstate_save_state:
haoqf:vmstate_size: field size:4, offset:0
haoqf:vmstate_save_state:opaque:0x2aa13325438, offset:4, size:4, field
name:size, vname:globalstate
haoqf:vmstate_size: field size:100, offset:0
haoqf:vmstate_save_state:opaque:0x2aa13325438, offset:8, size:100, field
name:runstate, vname:globalstate
haoqf:vmstate_save:called vmstate_save_state
So probably x86 doesn't have this problem.
Fam
--
Regards
QingFeng Hao
