Long boot times (> 1 hour at times) on Solaris x86 machines when using jumbo frames

Last updated on AUGUST 17, 2017

Applies to:

Symptoms

 A Solaris 11.1 or later system requires many minutes - 20 minutes to a couple of hours - to boot when jumbo frames have been configured. 

A crash dump (or threadlist) reveals threads in VM2 memory allocation routines.   Such as this example (in this case an aggregation with interfaces using jumbo frames was being brought up):

Process: /usr/sbin/dladm up-aggr

genunix:flr_find_freectg_range()
genunix:flr_contig_alloc_cb()
unix:tile_pfns_iterate_range()
genunix:flr_contig_alloc()
unix:page_create_io()
unix:contig_alloc()
unix:kalloca_constrain()
unix:i_ddi_mem_alloc()
genunix:ddi_dma_mem_alloc()
mac:i_mac_dma_alloc()
mac:mac_block_alloc()
mac:i_mac_packet_pool_populate()
mac:i_mac_packet_pool_create()
mac:i_mac_bm_ring_create_pools()
mac:i_mac_bm_group_create_pools())
mac:i_mac_bm_create_pools()
mac:mac_bm_create_pools()
mac:mac_start()
mac:mac_client_datapath_setup()
mac:i_mac_unicast_add()
mac:mac_client_unicast_add()
mac:mac_unicast_add()
aggr:aggr_grp_add_port()
aggr:aggr_grp_create()
aggr:aggr_ioc_create(),
dld:drv_ioctl()
genunix:cdev_ioctl()
specfs:spec_ioctl()
genunix:fop_ioctl()
genunix:ioctl+0x188()
unix:_sys_sysenter_post_swapgs()

Changes

Cause