On 8/24/21 6:51 AM, wangyanan (Y) wrote:
> On 2021/8/23 21:17, Philippe Mathieu-Daudé wrote:
>> On 8/23/21 2:27 PM, Yanan Wang wrote:
>>> We have two requirements for a valid SMP configuration:
>>> the product of "sockets * cores * threads" must represent all the
>>> possible cpus, i.e., max_cpus, and then must include the initially
>>> present cpus, i.e., smp_cpus.
>>>
>>> So we only need to ensure 1) "sockets * cores * threads == maxcpus"
>>> at first and then ensure 2) "maxcpus >= cpus". With a reasonable
>>> order of the sanity check, we can simplify the error reporting code.
>>> When reporting an error message we also report the exact value of
>>> each topology member to make users easily see what's going on.
>>>
>>> Signed-off-by: Yanan Wang <wangyana...@huawei.com>
>>> Reviewed-by: Andrew Jones <drjo...@redhat.com>
>>> Reviewed-by: Pankaj Gupta <pankaj.gu...@ionos.com>
>>> ---
>>> hw/core/machine.c | 22 +++++++++-------------
>>> hw/i386/pc.c | 24 ++++++++++--------------
>>> 2 files changed, 19 insertions(+), 27 deletions(-)
>>>
>>> diff --git a/hw/core/machine.c b/hw/core/machine.c
>>> index 85908abc77..093c0d382d 100644
>>> --- a/hw/core/machine.c
>>> +++ b/hw/core/machine.c
>>> @@ -779,25 +779,21 @@ static void smp_parse(MachineState *ms,
>>> SMPConfiguration *config, Error **errp)
>>> maxcpus = maxcpus > 0 ? maxcpus : sockets * cores * threads;
>>> cpus = cpus > 0 ? cpus : maxcpus;
>>> - if (sockets * cores * threads < cpus) {
>>> - error_setg(errp, "cpu topology: "
>>> - "sockets (%u) * cores (%u) * threads (%u) < "
>>> - "smp_cpus (%u)",
>>> - sockets, cores, threads, cpus);
>>> + if (sockets * cores * threads != maxcpus) {
>>> + error_setg(errp, "Invalid CPU topology: "
>>> + "product of the hierarchy must match maxcpus: "
>>> + "sockets (%u) * cores (%u) * threads (%u) "
>>> + "!= maxcpus (%u)",
>>> + sockets, cores, threads, maxcpus);
>>> return;
>>> }
>> Thinking about scalability, MachineClass could have a
>> parse_cpu_topology() handler, and this would be the
>> generic one. Principally because architectures don't
>> use the same terms, and die/socket/core/thread arrangement
>> is machine specific (besides being arch-spec).
>> Not a problem as of today, but the way we try to handle
>> this generically seems over-engineered to me.
> Hi Philippe,
>
> The reason for introducing a generic implementation and avoiding
> specific ones is that we thought there is little difference in parsing
> logic between the specific parsers. Most part of the parsing is the
> automatic calculation of missing values and the related error reporting,
> in which the only difference between parsers is the handling of specific
> (no matter of arch-specific or machine-specifc) parameters.
>
> So it may be better to keep the parsing logic unified if we can easily
> realize that. And actually we can use compat stuff to handle specific
> topology parameters well. See implementation in patch #10.
>
> There have been patches on list introducing new specific members
> (s390 related in [1] and ARM related in [2]), and in each of them there
> is a specific parser needed. However, based on generic one we can
> extend without the increasing code duplication.
>
> There is also some discussion about generic/specific parser in [1],
> which can be a reference.
>
> [1]
> https://lore.kernel.org/qemu-devel/1626281596-31061-2-git-send-email-pmo...@linux.ibm.com/
>
> [2]
> https://lore.kernel.org/qemu-devel/20210516103228.37792-1-wangyana...@huawei.com/
OK I read Daniel's rationale here:
https://lore.kernel.org/qemu-devel/ypfn83pkbt7f9...@redhat.com/
Thanks,
Phil.
