Re: [PATCH] octeontx: add linux 5.10 testing kernel support
On Sat, 2021-09-04 at 15:18 +0200, Daniel Danzberger wrote: > > > > > > > +CONFIG_TRANSPARENT_HUGEPAGE_ALWAYS=y > > > > > > CONFIG_TRANSPARENT_HUGEPAGE_ALWAYS=y might not be a performance > > > win, > > > especially with only 1 GiB of RAM. Should be measured, otherwise > > > MADVISE should be selected instead. > > Same here, it was selected in the 5.4 kernel config and I have only > > run > > tested with this option so far. > > Let me do some benchmarking with MADVISE before we continue here > > ... > Looks like none of the processes running even gets a huge page: > -- > root@OpenWrt:~# grep AnonHugePages /proc/meminfo > AnonHugePages: 0 kB > root@OpenWrt:~# > -- > Even my test tool that just allocs one 8MB block doesn't get one. > Looks like CONFIG_TRANSPARENT_HUGEPAGE isn't working on this > platform. After further testing, huge pages only get used when memory is allocated >= 2MB with mmap() on arm64. Unlike on x86 where processes get a huge page when using malloc() or mmap() >= 2MB. So I think we can safely keep CONFIG_TRANSPARENT_HUGEPAGE_ALWAYS=y like in 5.4 -- Regards Daniel Danzberger embeDD GmbH, Alter Postplatz 2, CH-6370 Stans ___ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
Re: [PATCH] octeontx: add linux 5.10 testing kernel support
On Fri, 2021-09-03 at 15:08 +0200, Daniel Danzberger wrote: > > > > > > +CONFIG_HZ=250 > > > > Why 250 Hz? Does the 100 Hz all other targets use cause any > > measurable > > increase in latency? > > [snipped] > I did not notice this changed. The config I made for 5.10 was before > your commit 3326b5e75c277b4fac21bffd2085df4aa40d2775 which changed > the > default frequency to 100 Hz. > There should be no issue with 100 Hz on octecontx, but I will confirm > this with new builds on my test devices. Confirmed. No issues here with 100 Hz. > > > > > > +CONFIG_TRANSPARENT_HUGEPAGE_ALWAYS=y > > > > CONFIG_TRANSPARENT_HUGEPAGE_ALWAYS=y might not be a performance > > win, > > especially with only 1 GiB of RAM. Should be measured, otherwise > > MADVISE should be selected instead. > Same here, it was selected in the 5.4 kernel config and I have only > run > tested with this option so far. > Let me do some benchmarking with MADVISE before we continue here ... Looks like none of the processes running even gets a huge page: -- root@OpenWrt:~# grep AnonHugePages /proc/meminfo AnonHugePages: 0 kB root@OpenWrt:~# -- Even my test tool that just allocs one 8MB block doesn't get one. Looks like CONFIG_TRANSPARENT_HUGEPAGE isn't working on this platform. > > > > Best regards, > > Rui > > > -- Regards Daniel Danzberger embeDD GmbH, Alter Postplatz 2, CH-6370 Stans ___ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
Re: [PATCH] octeontx: add linux 5.10 testing kernel support
> > [snipped] > > > +CONFIG_EXT4_FS=y > > +CONFIG_EXT4_FS_POSIX_ACL=y > > +CONFIG_F2FS_FS=y > > +CONFIG_FANOTIFY=y > > +CONFIG_FAT_FS=y > > Do all of these filesystems have to be built-in? We have config > options for them. I don't think so. I took the kernel 5.4 config as a template, because I didn't want to break things for others when upgrading to 5.10. But I agree, there is no reason for a default buil-in. > > [snipped] > > > +CONFIG_HZ=250 > > Why 250 Hz? Does the 100 Hz all other targets use cause any > measurable > increase in latency? > [snipped] I did not notice this changed. The config I made for 5.10 was before your commit 3326b5e75c277b4fac21bffd2085df4aa40d2775 which changed the default frequency to 100 Hz. There should be no issue with 100 Hz on octecontx, but I will confirm this with new builds on my test devices. > > > +CONFIG_TRANSPARENT_HUGEPAGE_ALWAYS=y > > CONFIG_TRANSPARENT_HUGEPAGE_ALWAYS=y might not be a performance win, > especially with only 1 GiB of RAM. Should be measured, otherwise > MADVISE should be selected instead. Same here, it was selected in the 5.4 kernel config and I have only run tested with this option so far. Let me do some benchmarking with MADVISE before we continue here ... > > Best regards, > Rui > -- Regards Daniel Danzberger embeDD GmbH, Alter Postplatz 2, CH-6370 Stans ___ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
Re: [PATCH] octeontx: add linux 5.10 testing kernel support
Hi again, Daniel, On Fri, 3 Sept 2021 at 12:46, Daniel Danzberger wrote: > [snipped] > +CONFIG_EXT4_FS=y > +CONFIG_EXT4_FS_POSIX_ACL=y > +CONFIG_F2FS_FS=y > +CONFIG_FANOTIFY=y > +CONFIG_FAT_FS=y Do all of these filesystems have to be built-in? We have config options for them. [snipped] > +CONFIG_HZ=250 Why 250 Hz? Does the 100 Hz all other targets use cause any measurable increase in latency? [snipped] > +CONFIG_TRANSPARENT_HUGEPAGE_ALWAYS=y CONFIG_TRANSPARENT_HUGEPAGE_ALWAYS=y might not be a performance win, especially with only 1 GiB of RAM. Should be measured, otherwise MADVISE should be selected instead. Best regards, Rui ___ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
Re: [PATCH] octeontx: add linux 5.10 testing kernel support
Hi, forgot that. There were some more redundand symbols. New patch is on the way ... On Fri, 2021-09-03 at 10:26 +0100, Rui Salvaterra wrote: > Hi, Daniel, > > On Fri, 3 Sept 2021 at 10:18, Daniel Danzberger > wrote: > > > > [snipped for context] > > > +CONFIG_RCU_NEED_SEGCBLIST=y > > +CONFIG_RCU_STALL_COMMON=y > > Please refresh your kernel configuration. At least these two symbols > are part of the generic kconfigs. > > Thanks, > Rui > -- Regards Daniel Danzberger embeDD GmbH, Alter Postplatz 2, CH-6370 Stans ___ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
Re: [PATCH] octeontx: add linux 5.10 testing kernel support
Hi, Daniel, On Fri, 3 Sept 2021 at 10:18, Daniel Danzberger wrote: > [snipped for context] > +CONFIG_RCU_NEED_SEGCBLIST=y > +CONFIG_RCU_STALL_COMMON=y Please refresh your kernel configuration. At least these two symbols are part of the generic kconfigs. Thanks, Rui ___ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel