Re: [Qemu-devel] [ANNOUNCE] QEMU 2.12.0-rc4 is now available

[Laszlo Ersek]

On 04/24/18 15:39, Eric Blake wrote:
> On 04/24/2018 07:46 AM, Laszlo Ersek wrote:
> 
>> I just noticed two QAPI bugs introduced in 2.12, in the following two
>> commits:
>>
>> (1) ca230ff33f89 ("qmp: add architecture specific cpu data for
>> query-cpus-fast", 2018-02-26)
>>
>> This commit added @arch to @CpuInfoFast, but it failed to set the new
>> field to CPU_INFO_ARCH_OTHER in qmp_query_cpus_fast(), when TARGET_S390X
>> is not defined.
> 
> We zero-initialize info->value, and info->value->arch == 0 happens to be
> CPU_INFO_ARCH_X86 which is associated with the empty type CpuInfoOther,
> which is no different than what we would have had with
> CPU_INFO_ARCH_OTHER.  So the drawback is that on x86 architectures,
> things look correct; while on sparc, ppc, mips, and tricore (which have
> a specific type) and on all other platforms (which should use "other"),
> the output wrongly reports "arch":"x86" but does not crash.

I agree.

> 
>>
>> (2) 25fa194b7b11 ("RISC-V Build Infrastructure", 2018-03-07)
>>
>> This commit added the @riscv enum constant to @CpuInfoArch (used in both
>> @CpuInfo and @CpuInfoFast -- the return types of the @query-cpus and
>> @query-cpus-fast commands, respectively), and assigned, in both return
>> structures, the @CpuInfoRISCV sub-structure to the new enum value.
> 
> Ouch - we SHOULD have used 'CpuInfoOther' rather than 'CpuInfoRISCV' for
> @CpuInfoFast.  We do NOT want to report the 'pc' field in the fast query.

Right.

>>
>> However, qmp_query_cpus_fast() does not populate the sub-structure, when
>> TARGET_RISCV is defined; only qmp_query_cpus() does.
> 
> Indeed, the output wrongly reports "arch":"x86" instead of
> "arch":"riscv" - but at least things do not crash.
> 
>>
>> The fixes don't look complicated. Do you want to hold the release while
>> I post the fixes?
>>
>> I don't think these bugs are "show stoppers". On the other hand, QAPI is
>> an external interface.
> 
> It's too late for the 2.12 release.  We'll definitely fix things in
> 2.12.1 (so cc the patches to qemu-stable),

OK, will do.

> and should document in the
> release notes that query-cpus-fast is known to report bogus "arch" data
> for all architectures except x86 and s390, so applications should be
> careful to not rely on the bogus information.

<https://wiki.qemu.org/ChangeLog/2.12#Known_issues> refers to
<https://wiki.qemu.org/Planning/2.12>, so I've added the above sentence
to <https://wiki.qemu.org/Planning/2.12#Issues_that_will_not_be_fixed>.

Thanks, Eric!
Laszlo