From: Pierre Morel <pmorel@linux.ibm.com>
To: "Daniel P. Berrangé" <berrange@redhat.com>,
"Cornelia Huck" <cohuck@redhat.com>
Cc: thuth@redhat.com, ehabkost@redhat.com, david@redhat.com,
richard.henderson@linaro.org, qemu-devel@nongnu.org,
armbru@redhat.com, pasic@linux.ibm.com, borntraeger@de.ibm.com,
pbonzini@redhat.com, eblake@redhat.com
Subject: Re: [PATCH v1 1/9] s390x: smp: s390x dedicated smp parsing
Date: Fri, 16 Jul 2021 12:59:28 +0200 [thread overview]
Message-ID: <2582afdf-4071-a12c-bf83-0fdb837bb5f1@linux.ibm.com> (raw)
In-Reply-To: <YPFN83pKBt7F97kW@redhat.com>
On 7/16/21 11:14 AM, Daniel P. Berrangé wrote:
> On Fri, Jul 16, 2021 at 10:54:08AM +0200, Cornelia Huck wrote:
>> On Wed, Jul 14 2021, Pierre Morel <pmorel@linux.ibm.com> wrote:
>>
>>> We need a s390x dedicated SMP parsing to handle s390x specificities.
>>>
>>> In this patch we only handle threads, cores and sockets for
>>> s390x:
>>> - do not support threads, we always have 1 single thread per core
>>> - the sockets are filled one after the other with the cores
>>>
>>> Both these handlings are different from the standard smp_parse
>>> functionement and reflect the CPU topology in the simple case
>>> where all CPU belong to the same book.
>>>
>>> Topology levels above sockets, i.e. books, drawers, are not
>>> considered at this stage and will be introduced in a later patch.
>>>
>>> Signed-off-by: Pierre Morel <pmorel@linux.ibm.com>
>>> ---
>>> hw/s390x/s390-virtio-ccw.c | 42 ++++++++++++++++++++++++++++++++++++++
>>> 1 file changed, 42 insertions(+)
>>>
>>> diff --git a/hw/s390x/s390-virtio-ccw.c b/hw/s390x/s390-virtio-ccw.c
>>> index e4b18aef49..899d3a4137 100644
>>> --- a/hw/s390x/s390-virtio-ccw.c
>>> +++ b/hw/s390x/s390-virtio-ccw.c
>>> @@ -582,6 +582,47 @@ static ram_addr_t s390_fixup_ram_size(ram_addr_t sz)
>>> return newsz;
>>> }
>>>
>>> +/*
>>> + * In S390CCW machine we do not support threads for now,
>>> + * only sockets and cores.
>>> + */
>>> +static void s390_smp_parse(MachineState *ms, QemuOpts *opts)
>>
>> It seems you based this on an older version of the code? The current
>> signature of this function since 1e63fe685804 ("machine: pass QAPI
>> struct to mc->smp_parse") is
>>
>> void (*smp_parse)(MachineState *ms, SMPConfiguration *config, Error **errp);
>>
>> That affects your parsing, and also lets you get rid of the ugly exit(1)
>> statements.
>>
>>> +{
>>> + unsigned cpus = qemu_opt_get_number(opts, "cpus", 1);
>>> + unsigned sockets = qemu_opt_get_number(opts, "sockets", 1);
>>> + unsigned cores = qemu_opt_get_number(opts, "cores", 1);
>>> +
>>> + if (opts) {
>>> + if (cpus == 0 || sockets == 0 || cores == 0) {
>>
>> This behaviour looks different from what we do for other targets: if you
>> specify the value as 0, a value is calculated from the other values;
>> here, you error out. It's probably not a good idea to differ.
>
> I increasingly worry that we're making a mistake by going down the
> route of having custom smp_parse implementations per target, as this
> is showing signs of inconsistent behaviour and error reportings. I
> think the differences / restrictions have granularity at a different
> level that is being tested in many cases too.
>
> Whether threads != 1 is valid will likely vary depending on what
> CPU model is chosen, rather than what architecture is chosen.
> The same is true for dies != 1. We're not really checking this
> closely even in x86 - for example I can request nonsense such
> as a 25 year old i486 CPU model with hyperthreading and multiple
> dies
>
> qemu-system-x86_64 -cpu 486 -smp 16,cores=4,dies=2,threads=2
>
> In this patch, there is no error reporting if the user specifies
> dies != 1 or threads != 1 - it just silently ignores the request
> which is not good.
yes, I should change this
>
> Some machine types may have constraints on CPU sockets.
>
> This can of course all be handled by custom smp_parse impls, but
> this is ultimately going to lead to alot of duplicated and
> inconsistent logic I fear.
>
> I wonder if we would be better off having machine class callback
> that can report topology constraints for the current configuration,
> along lines of
>
> smp_constraints(MachineState *ms,
> int *max_sockets,
> int *max_dies,
> int *max_cores,
> int *max_threads)
I find the idee good, but what about making it really machine agnostic
by removing names and using a generic
smp_constraints(MachineState *ms,
int *nb_levels,
int *levels[]
);
Level can be replaced by another name like container.
The machine could also provide the level/container names according to
its internal documentation.
Regards,
Pierre
>
> And then have only a single smp_parse impl that takes into account
> these constraints, to report errors / fill in missing fields / etc ?
>
> Regards,
> Daniel
>
--
Pierre Morel
IBM Lab Boeblingen
next prev parent reply other threads:[~2021-07-16 11:02 UTC|newest]
Thread overview: 39+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-07-14 16:53 [PATCH v1 0/9] s390x: CPU Topology Pierre Morel
2021-07-14 16:53 ` [PATCH v1 1/9] s390x: smp: s390x dedicated smp parsing Pierre Morel
2021-07-16 8:54 ` Cornelia Huck
2021-07-16 9:14 ` Daniel P. Berrangé
2021-07-16 10:59 ` Pierre Morel [this message]
[not found] ` <e4865ad6-f8ec-e7ba-66ef-9c95334ba9b3@linux.ibm.com>
2021-07-19 15:43 ` Cornelia Huck
2021-07-19 15:52 ` Daniel P. Berrangé
2021-07-20 7:37 ` Pierre Morel
2021-07-20 8:33 ` Pierre Morel
2021-07-16 10:47 ` Pierre Morel
2021-07-14 16:53 ` [PATCH v1 2/9] s390x: toplogy: adding drawers and books to " Pierre Morel
2021-07-15 6:16 ` Markus Armbruster
2021-07-15 8:19 ` Pierre Morel
2021-07-15 10:48 ` Markus Armbruster
2021-07-16 9:10 ` Cornelia Huck
2021-07-16 9:18 ` Daniel P. Berrangé
2021-07-16 10:44 ` Cornelia Huck
2021-07-16 10:49 ` Daniel P. Berrangé
2021-07-19 15:50 ` Cornelia Huck
2021-07-20 7:52 ` Pierre Morel
2021-07-20 8:20 ` Cornelia Huck
2021-07-20 8:46 ` Pierre Morel
2021-07-20 9:00 ` Cornelia Huck
2021-07-20 9:19 ` Daniel P. Berrangé
2021-07-20 12:29 ` Pierre Morel
2021-07-16 9:23 ` Daniel P. Berrangé
2021-07-16 11:08 ` Pierre Morel
2021-07-14 16:53 ` [PATCH v1 3/9] s390x: cpu topology: CPU topology objects and structures Pierre Morel
2021-07-14 16:53 ` [PATCH v1 4/9] s390x: Topology list entries and SYSIB 15.x.x Pierre Morel
2021-07-14 16:53 ` [PATCH v1 5/9] s390x: topology: implementating Store Topology System Information Pierre Morel
2021-07-14 16:53 ` [PATCH v1 6/9] s390x: kvm: topology: interception of PTF instruction Pierre Morel
2021-07-16 9:22 ` Cornelia Huck
2021-07-16 11:23 ` Pierre Morel
2021-07-16 11:56 ` Cornelia Huck
2021-07-14 16:53 ` [PATCH v1 7/9] s390x: SCLP: reporting the maximum nested topology entries Pierre Morel
2021-07-16 9:24 ` Cornelia Huck
2021-07-16 11:12 ` Pierre Morel
2021-07-14 16:53 ` [PATCH v1 8/9] s390x: numa: define drawers and books for NUMA Pierre Morel
2021-07-14 16:53 ` [PATCH v1 9/9] s390x: numa: implement NUMA for S390x Pierre Morel
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=2582afdf-4071-a12c-bf83-0fdb837bb5f1@linux.ibm.com \
--to=pmorel@linux.ibm.com \
--cc=armbru@redhat.com \
--cc=berrange@redhat.com \
--cc=borntraeger@de.ibm.com \
--cc=cohuck@redhat.com \
--cc=david@redhat.com \
--cc=eblake@redhat.com \
--cc=ehabkost@redhat.com \
--cc=pasic@linux.ibm.com \
--cc=pbonzini@redhat.com \
--cc=qemu-devel@nongnu.org \
--cc=richard.henderson@linaro.org \
--cc=thuth@redhat.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).