Re: [PATCH v3 4/4] x86/folio_zero_user: multi-page clearing

[Ankur Arora <ankur.a.arora@oracle.com>]

Raghavendra K T <raghavendra.kt@amd.com> writes:

> On 4/14/2025 12:35 PM, Ingo Molnar wrote:
>> * Ankur Arora <ankur.a.arora@oracle.com> wrote:
>
> [...]
>
>>> Performance for preempt=none|voluntary remains unchanged.
>> CONFIG_PREEMPT_VOLUNTARY=y is the default on a number of major
>> distributions, such as Ubuntu, and a lot of enterprise distro kernels -
>> and this patch does nothing for them, for no good reason.
>> So could you please provide a sensible size granularity cutoff of 16MB
>> or so on non-preemptive kernels, instead of this weird build-time
>> all-or-nothing binary cutoff based on preemption modes?
>>
>
> Agree with Ingo here. We are too harsh on none and voluntary with this
> limit. At least 512KB size?

512KB seems much too small. I was thinking of something larger: 8MB. With
a zeroing BW of 10GBps (typical value I've seen with relatively new
Intel/AMD hardware), that would mean a preemption delay of 0.8ms
for preempt=none/voluntary.

Even with relatively old hardware -- say skylake -- we should see ~5
GBps/1.6ms delay.

> I will try to see if I can get numbers for different sizes that does not
> interfere much with other workloads' perf. (in none/voluntary mode).

Thanks that'll be useful to make the case.

--
ankur