Re: [PATCH not-for-merge 0/5] Instrumentation for "Fixes around device realization"

[Markus Armbruster <armbru@redhat.com>]

Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk> writes:

> On 18/05/2020 06:19, Markus Armbruster wrote:
>
>> This is the instrumentation mentioned in "[PATCH 00/24] Fixes around
>> device realization".
>> 
>> PATCH 2/5 might have value on its own.  You tell me.
>> 
>> Shell script to smoke-test all machines:
>> 
>> #!/bin/sh
>> success=0
>> fail=0
>> ulimit -c 0
>> git-describe --dirty --match v\*
>> git-log --oneline -1
>> for i in bld/*-softmmu
>> do
>>     t=${i%-softmmu}
>>     t=${t##*/}
>>     q=$i/qemu-system-$t
>>     echo "= $t ="
>> 
>>     for m in `$q -M help | sed -n '/(alias of/d;2,$s/ .*//p'`
>>     do
>> 	echo "== $m =="
>> 	echo -e 'info qom-tree\ninfo qtree\nq' | $q -S -accel qtest -display none -L smoke-mon-roms -M $m -monitor stdio
>> 	if [ $? -eq 0 ]
>> 	then echo "*** Success: $m ***"; let success++
>> 	else echo "*** Fail: $m"; let fail++
>> 	fi
>>     done
>> done
>> echo $success succeeded, $fail failed
>> 
>> 
>> Markus Armbruster (5):
>>   qom: Instrument to detect missed realize
>>   qom: Make "info qom-tree" show children sorted
>>   qdev: Make "info qtree" show child devices sorted by QOM path
>>   qdev: Instrument to detect missed QOM parenting
>>   qdev: Instrument to detect bus mismatch
>> 
>>  hw/core/qdev.c     | 17 ++++++++++++++++
>>  qdev-monitor.c     | 32 ++++++++++++++++++++++++++++-
>>  qom/qom-hmp-cmds.c | 51 +++++++++++++++++++++++++++++++++++++++++++++-
>>  3 files changed, 98 insertions(+), 2 deletions(-)
>
> Thanks for sharing these patches! I certainly think that they have value and after a
> quick read through I'm thinking:
>
> - Patch 1 I assume is no longer needed once you previous series is merged

Correct, "[PATCH 24/24] qdev: Assert onboard devices all get realized
properly" supersedes.

> - Patches 2 & 3 would be really useful at the start of your previous series (as
> someone who has gone crossed-eyed enough trying to spot these differences, this is
> really helpful)

It's where they sat while I developed my fixes, so I don't go
cross-eyed, too :)

> - Patches 4 and 5 are good sanity checks for developers but I'm wondering what is the
> extent of work that needs to be done? Could existing failures be whitelisted with the
> aim of removal which would then at least prevent new devices being added that aren't
> correct?

Since realization assigns a QOM parent to orphans, "[PATCH 24/24] qdev:
Assert onboard devices all get realized properly" supersedes PATCH 4,
too.  I needed both PATCH 1 and 4 to track down missing realizes,
because some were visible only in "info qtree" (PATCH 4), and some only
in "info qom-tree" (PATCH 1).

"[PATCH 22/24] qdev: Assert devices are plugged into a bus that can take
them" supersedes PATCH 5.

Thanks!