Re: [PATCH 1/3] kho: allow scratch areas with zero size

[Pratyush Yadav <pratyush@kernel.org>]

On Wed, Aug 13 2025, Mike Rapoport wrote:

> On Wed, Aug 13, 2025 at 03:45:29PM +0200, Pratyush Yadav wrote:
>> On Mon, Aug 11 2025, Mike Rapoport wrote:
>> 
>> > From: "Mike Rapoport (Microsoft)" <rppt@kernel.org>
>> >
>> > Parsing of kho_scratch parameter treats zero size as an invalid value,
>> > although it should be fine for user to request zero sized scratch area
>> > for some types if scratch memory, when for example there is no need to
>> > create scratch area in the low memory.
>> 
>> Can the system boot with 0 per-node memory? If not, then perhaps we
>> should only allow lowmem scratch to be zero?
>
> In most cases yes because most of boot time allocations have fallback to
> "any node".
> And there's also an option to omit the "global" scratch and boot with only
> per-node scratch areas, so I'd keep the possibility of setting any of these
> to 0.

Makes sense. In that case,

Reviewed-by: Pratyush Yadav <pratyush@kernel.org>

>
>> > Treat zero as a valid value for a scratch area size but reject
>> > kho_scratch parameter that defines no scratch memory at all.
>> >
>> > Signed-off-by: Mike Rapoport (Microsoft) <rppt@kernel.org>
>> > ---
>> >  kernel/kexec_handover.c | 7 ++++++-
>> >  1 file changed, 6 insertions(+), 1 deletion(-)
>> >
>> > diff --git a/kernel/kexec_handover.c b/kernel/kexec_handover.c
>> > index e49743ae52c5..c6ac5a5e51cb 100644
>> > --- a/kernel/kexec_handover.c
>> > +++ b/kernel/kexec_handover.c
>> > @@ -385,6 +385,7 @@ static int __init kho_parse_scratch_size(char *p)
>> >  {
>> >  	size_t len;
>> >  	unsigned long sizes[3];
>> > +	size_t total_size = 0;
>> >  	int i;
>> >  
>> >  	if (!p)
>> > @@ -421,11 +422,15 @@ static int __init kho_parse_scratch_size(char *p)
>> >  		}
>> >  
>> >  		sizes[i] = memparse(p, &endp);
>> > -		if (!sizes[i] || endp == p)
>> > +		if (endp == p)
>> >  			return -EINVAL;
>> >  		p = endp;
>> > +		total_size += sizes[i];
>> >  	}
>> >  
>> > +	if (!total_size)
>> > +		return -EINVAL;
>> > +
>> 
>> Looks good. BTW, unrelated to this patch, but should we also check that
>> p == '\0' here to make sure the whole argument was consumed?
>
> Care to send a patch? ;-) 

Will do :-)

>  
>> >  	scratch_size_lowmem = sizes[0];
>> >  	scratch_size_global = sizes[1];
>> >  	scratch_size_pernode = sizes[2];
>> 
>> -- 
>> Regards,
>> Pratyush Yadav

-- 
Regards,
Pratyush Yadav