This x2apic_send_ipi(self) panic is reproducible with just cpu online/offline without using pmconfig to enable/disable cpu deep idle.
Bill On 11/11/08 19:22, Bill Holler wrote: > On 11/11/08 18:37, Bill Holler wrote: >> This fixes the deadlock. The deadlock was easily reproducible >> before this fix without cpu offline/online. With this fix the >> deadlock is not reproducible. >> >> Aggressive stress testing with multiple cpu online/offline scripts >> and multiple pmconfig scripts now hits this panic after a while: >> now hits this panic: >> [13]> ffffff0274e513c0 ::findstack >> stack pointer for thread ffffff0274e513c0: ffffff000b961990 >> ffffff000b961a80 panic+0x9c() >> ffffff000b961b00 die+0xf4() >> ffffff000b961c30 trap+0x37e() >> ffffff000b961c40 0xfffffffffb8002c0() >> ffffff000b961d60 rdmsr+2() >> ffffff000b961da0 pcplusmp`x2apic_send_ipi+0xc4() >> ffffff000b961dc0 send_dirint+0x19() >> ffffff000b961df0 poke_cpu+0x39() >> ffffff000b961e20 cpu_online+0xd8() >> ffffff000b961e80 p_online_internal+0x166() >> ffffff000b961eb0 p_online+0x19() >> ffffff000b961f00 _sys_sysenter_post_swapgs+0x23c() >> >> This looks like an earlier x2apic bug. > > The panic happened when sending an IPI to itself. > This was a known OpenSolaris bug. I will test this is > fixed by the onnv_102 merge. > > Regards, > Bill > >
