Re: [PATCH 1/1] numa: check mem or memdev in numa configuration

[Igor Mammedov]

On Thu, 17 Feb 2022 13:24:08 +0100
Li Zhang <lizhang@suse.de> wrote:

> On 2/17/22 11:25 AM, Igor Mammedov wrote:
> > On Thu, 17 Feb 2022 10:38:32 +0100
> > Li Zhang <lizhang@suse.de> wrote:
> >   
> >> On 2/17/22 10:10 AM, Igor Mammedov wrote:  
> >>> On Wed, 16 Feb 2022 17:36:13 +0100
> >>> Li Zhang <lizhang@suse.de> wrote:
> >>>      
> >>>> If there is no mem or memdev in numa configuration, it always
> >>>> reports the error as the following:
> >>>>
> >>>> total memory for NUMA nodes (0x0) should equal RAM size (0x100000000)
> >>>>
> >>>> This error is confusing and the reason is that total memory of numa nodes
> >>>> is always 0 if there is no mem or memdev in numa configuration.
> >>>> So it's better to check mem or memdev in numa configuration.
> >>>>
> >>>> Signed-off-by: Li Zhang <lizhang@suse.de>
> >>>> ---
> >>>>    hw/core/numa.c | 5 +++++
> >>>>    1 file changed, 5 insertions(+)
> >>>>
> >>>> diff --git a/hw/core/numa.c b/hw/core/numa.c
> >>>> index 1aa05dcf42..11cbec51eb 100644
> >>>> --- a/hw/core/numa.c
> >>>> +++ b/hw/core/numa.c
> >>>> @@ -132,6 +132,11 @@ static void parse_numa_node(MachineState *ms, 
> >>>> NumaNodeOptions *node,
> >>>>    
> >>>>        have_memdevs = have_memdevs ? : node->has_memdev;
> >>>>        have_mem = have_mem ? : node->has_mem;
> >>>> +    if (!node->has_memdev && !node->has_mem) {
> >>>> +        error_setg(errp, "numa configuration should use mem= or memdev= 
> >>>> ");
> >>>> +        return;
> >>>> +    }  
> >>>
> >>> Wouldn't this breaks memory less numa nodes?  
> >>
> >> Yes, you are right. It will break it if there more numa nodes
> >> than memory, and the numa nodes have no memory backends specified.
> >>
> >> Is it allowed for users to specify more numa nodes than memory?  
> > yep, I think we support it at least for one of the targets
> > (but I don't remember which one(s))
> >   
> 
> Is it okay if I put a warning here, instead of an error and return?
> It won't break the special case. I wonder if it is annoying to get
> the warning.
issuing warning in perfectly valid case (memory-less node)
doesn't look like a good thing to do.

there is already a error message,

 "total memory for NUMA nodes (0x0) should equal RAM size (0x100000000)"

I'd suggest to just fix this error message to be less confusing
instead of adding dubious warning elsewhere.

> 
> Thanks
> Li
> 
> >>  
> >>>
> >>> I'd rather add/rephrase to original error message that memory
> >>> should be specified explicitly for desired numa nodes.
> >>> And I'd not mention 'mem=' since
> >>>     docs/about/removed-features.rst:``-numa node,mem=...`` (removed in 
> >>> 5.1)  
> >>
> >> Thanks for your suggestions, I will rephrase it.
> >>  
> >>>      
> >>>> +
> >>>>        if ((node->has_mem && have_memdevs) || (node->has_memdev && 
> >>>> have_mem)) {
> >>>>            error_setg(errp, "numa configuration should use either mem= 
> >>>> or memdev=,"
> >>>>                       "mixing both is not allowed");  
> >>>      
> >>  
> >   
>