> On 25 Oct 2018, at 01:57, Stephen Hemminger <step...@networkplumber.org> > wrote: > > On Wed, 24 Oct 2018 23:09:15 +0200 > Damjan Marion <dmar...@me.com> wrote: > >> — >> Damjan >> >>> On 24 Oct 2018, at 23:04, Stephen Hemminger <step...@networkplumber.org> >>> wrote: >>> >>> On Wed, 24 Oct 2018 21:07:15 +0200 >>> Damjan Marion <dmar...@me.com> wrote: >>> >>>>> On 24 Oct 2018, at 20:41, Stephen Hemminger <step...@networkplumber.org> >>>>> wrote: >>>>> >>>>> On Wed, 24 Oct 2018 11:31:38 -0700 >>>>> "Stephen Hemminger" <step...@networkplumber.org >>>>> <mailto:step...@networkplumber.org>> wrote: >>>> >>>> [snip] >>>> >>>>> >>>>> The problem is that the setup code decides to make 1 huge page. And since >>>>> huge page size >>>>> on this system is 256M, DPDK can't even get stared. >>>> >>>> Already fixed in master... (hopefully works on Aarch64) >>>> >>> >>> Still broken, added some instrumentation to clib_sysfs_prealloc_hugepages >>> and got. >>> >>> clib_sysfs_prealloc_hugepages:261: found log2=21 page_size=2048 n=0 >>> clib_sysfs_prealloc_hugepages:264: pre-allocating 1 additional 2048K >>> hugepages on numa node 0 >>> clib_sysfs_set_nr_hugepages:158: set >>> /sys/devices/system/node/node0/hugepages/hugepages-2048kB/nr_hugepages to 1 >>> >>> DPDK won't start with on 2M of memory. >> >> Yeah, i already fixed that but it is not merged yet. This is just prealloc >> issue, if pages are preallocated everything should work ok. > > Got it, fixed for me. > Wonder why CI never caught this??
Likely pages are pre-allocated on boot time which is not bad thing to do. I added this sysfs prealloc functionality as a last attempt to survive if pages are not there...
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#10983): https://lists.fd.io/g/vpp-dev/message/10983 Mute This Topic: https://lists.fd.io/mt/27570325/21656 Group Owner: vpp-dev+ow...@lists.fd.io Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com] -=-=-=-=-=-=-=-=-=-=-=-
