Hi Herbert, On Wed, Jun 10, 2020 at 11:28 AM Mauricio Faria de Oliveira <mfo@xxxxxxxxxxxxx> wrote: > > On Tue, Jun 9, 2020 at 9:21 PM Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx> wrote: > > > > On Tue, Jun 09, 2020 at 12:17:32PM -0300, Mauricio Faria de Oliveira wrote: > > > > > > Per your knowledge/experience with the crypto subsystem, the changed code > > > paths are not hot enough to suffer from such implications? > > > > I don't think replacing a spin-lock with a pair of atomic ops is > > going to be too much of an issue. We can always look at this again > > if someone comes up with real numbers of course. > > Right; I meant the other places as well, where atomic ops were added > (in addition to the existing spinlocks.) > > But indeed, real numbers would be great and tell whether or not > there's performance differences. > > We're working on that -- Brian (bug reporter) has access to detailed > metrics/stats from the workload, and kindly agreed to set up two > identical instances to compare the numbers. I'll keep you posted. > Just wanted to let you know that the performance is really close for the two patches, both on the CPU utilization profile and also on workload results (#requests over time). And the tests ran for two days, so it looks stable. Thanks! > Thank you, > Mauricio > > > > > > > Cheers, > > -- > > Email: Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx> > > Home Page: http://gondor.apana.org.au/~herbert/ > > PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt > > > > -- > Mauricio Faria de Oliveira -- Mauricio Faria de Oliveira
