Re: [PATCH] x86/cpu: use smp_call_function_many() in arch_freq_prepare_all()

[Thomas Gleixner]

On Wed, Mar 30 2022 at 10:05, Eric Dumazet wrote:
> On Wed, Mar 30, 2022 at 10:02 AM Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:
>> On Wed, Mar 30 2022 at 09:51, Eric Dumazet wrote:
>> > On Wed, Mar 30, 2022 at 8:58 AM Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:
>> >> which I hate with a passion because that allows *unpriviledged* user
>> >> space to inject systemwide IPIs every 10ms just to read these counters
>> >> which are providing not more than some estimate and are of no value for
>> >> the only sane use case of /proc/cpuinfo, i.e. #1 above.
>> >
>> > You do realize that before my patch, this is already happening ?
>> >
>> > My "optimization" simply replace an open loop of individual IPI with
>> > use of the broadcast IPI capability.
>> >
>> > Are you saying we should remove IPI broadcast and use loops
>> > of IPI, one cpu at a time ?
>>
>> I rather have no IPIs at all...
>
> Can you send an actual patch, with a changelog then ?

I can polish up the patch I sent, split it up and add changelogs. Sure.

Thanks,

        tglx