From: Alexander Bluhm Subject: Re: igc(4) tso To: Moritz Buhl Cc: tech@openbsd.org, bket@openbsd.org, jan@openbsd.org Date: Fri, 3 May 2024 14:13:44 +0200 On Thu, May 02, 2024 at 11:58:45PM +0200, Alexander Bluhm wrote: > On Wed, May 01, 2024 at 06:24:36PM +0200, Moritz Buhl wrote: > > The VLAN_HWTAGGING diff is in snapshots for a while, I saw no > > additional feedback. How about adding TSO to igc(4) next? > > > > I ran thorough tests on bluhms' setup: > > http://bluhm.genua.de/netlink/results/2024-04-30T00%3A37%3A41Z/netlink.html > > and additional tests on my home router using pppoe(4) and vlan(4). > > I did test it with more options. With jumbo frames I see this crash. > Next I will test igc with jumbos, but without diff. Looks like igc(4) with jumbo frames is broken regardless of the TSO diff. panic: pool_cache_item_magic_check: mcl2k2 cpu free list modified: item addr 0xfffffd80bc08b480+16 0x33c144d5e0383f2e!=0x4fa073e50b458552 Stopped at db_enter+0x14: popq %rbp TID PID UID PRFLAGS PFLAGS CPU COMMAND 393473 76617 1000 0x18100002 0 2 tcpbench 117181 23314 0 0x14000 0x200 1 softnet3 379304 26108 0 0x14000 0x200 3 softnet0 db_enter() at db_enter+0x14 panic(ffffffff82108f78) at panic+0xc3 pool_cache_get(ffffffff8261ca18) at pool_cache_get+0x24a pool_get(ffffffff8261ca18,2) at pool_get+0x62 m_clget(0,2,802) at m_clget+0x1b4 igc_get_buf(ffff800000304688,381) at igc_get_buf+0x9f igc_rxfill(ffff800000304688) at igc_rxfill+0xce igc_intr_queue(ffff8000002fd6d8) at igc_intr_queue+0x55 intr_handler(ffff80005c0417e0,ffff80000009f280) at intr_handler+0x72 Xintr_ioapic_edge6_untramp() at Xintr_ioapic_edge6_untramp+0x18f acpicpu_idle() at acpicpu_idle+0x11f sched_idle(ffff80005a60dff0) at sched_idle+0x272 end trace frame: 0x0, count: 3 https://www.openbsd.org/ddb.html describes the minimum info required in bug reports. Insufficient info makes it difficult to find and fix bugs. ddb{5}> client_loop: send disconnect: Broken pipe