After 3 days of test (latency + windows 10 with cpu-z and stress test) I reached rare peaks of 130 us. For me it's OK but I think there is a small "regression" against the 4.9 Do you need any other test? R.
Il lun 8 apr 2019, 12:13 Jan Kiszka <jan.kis...@siemens.com> ha scritto: > On 08.04.19 11:56, cagnulein wrote: > > Wow! It works flawless! Latency is just a little higher than 4.9.146 (72 > VS 67) > > but it's acceptable for me. > > Now I started a 3 days test. I will let you know. > > That's good news! > > Let's see if the latency difference is still relevant after measuring > longer... > > Jan > > > Thanks! > > Roberto > > > > Il ven 5 apr 2019, 20:31 cagnulein <cagnul...@gmail.com > > <mailto:cagnul...@gmail.com>> ha scritto: > > > > Thank you, on Monday morning I will fire it up! :) > > R. > > > > Il ven 5 apr 2019, 20:29 Jan Kiszka <jan.kis...@siemens.com > > <mailto:jan.kis...@siemens.com>> ha scritto: > > > > On 05.04.19 15:27, Jan Kiszka wrote: > > > On 05.04.19 11:42, Jan Kiszka wrote: > > >> On 05.04.19 10:44, cagnulein wrote: > > >>> Hi, yes I double checked the patch and it's applied > correctly. > > >>> Here you can find my config. > > >>> > > >>> CONFIG_DEBUG_STACKOVERFLOW is disabled. > > >>> > > >>> I can't get the uart on this system, is there any other way > to > > catch log in > > >>> this scenario? > > >> > > >> UART is mandatory for debugging such crashes. Use a PCI > extension > > card if your > > >> board has no UART connector anymore. > > >> > > >> Well, you may try to lift Xenomai with KVM inside a KVM VM > and catch > > the UART > > >> output of the first level VM. This is how I'm debugging. > > >> > > >>> > > >>> Why should I test the 4.4.y? 4.9.146 works fine on the same > system > > (with the > > >>> same guest). Did I miss something? > > >> > > >> Well, 4.9 would be as broken as 4.4 currently is. But 4.4 is > > supported, 4.9 is > > >> discontinued by now. In any case, this indicates we have > another > > difference > > >> caused by changed in 4.14 (and before). > > >> > > >> I'll see if I can reproduce by booting Win10 in my nested > setup. > > >> > > > > > > Already running the UEFI BIOS triggers something, at least on > 4.4. > > Let's see... > > > > > > > As you will be able to see from the commits, there were a number > of > > further issues: > > > > https://gitlab.denx.de/Xenomai/ipipe-x86/commits/wip/kvm-4.14 > > > > With that, I'm able to start Win10 on the same core as the > latency test. > > But > > only in my virtual setup (nested kvm) which may mean that real > hw could > > expose > > further issue or latency problems. Please test. > > > > Jan > > > > -- > > Siemens AG, Corporate Technology, CT RDA IOT SES-DE > > Corporate Competence Center Embedded Linux > > > > -- > Siemens AG, Corporate Technology, CT RDA IOT SES-DE > Corporate Competence Center Embedded Linux >