Jon Noack wrote: > Yeah, it is most likely bugs uncovered by native preemption. For now > you can remask then by commenting out the "#define PREEMPTION" in > src/sys/<ARCH>/include/param.h and rebuilding your kernel. Robert > Watson's comments on this: > "This results in a quite usable kernel, although quite probably more > poor interrupt handling latency, etc." I just tried a new kernel (2004.07.16.01.00.00) with preemption turned on, and my problems with interrupt storms and SATA still remains. With preemption turned off the machine seems stable, but somewhat sluggish. With preemption enabled I get interrupt storms and DMA problems with the SATA devices like below when trying to access all discs at once (running fsck on all 11 file systems simultaneously). ad20 is one of two SATA devices in the system. Interrupt storm detected on "irq20: atapci4"; throttling interrupt source ad20: FAILURE - already active DMA on this device ad20: setting up DMA failed Attached is the dmesg.boot from the old kernel (2004.07.12.22.00.00) with preemption turned off that seems to be stable. I'm using HZ=2000, DEVICE_POLLING and ZERO_COPY_SOCKETS, plus device apic. I also have net.isr enabled. 'vmstat -i' gives the following output: interrupt total rate irq1: atkbd0 248 0 irq0: clk 8176853 1993 irq4: sio0 12 0 irq6: fdc0 6 0 irq8: rtc 524902 127 irq13: npx0 1 0 irq14: ata0 33148 8 irq15: ata1 24641 6 irq16: atapci0+ 867398 211 irq17: atapci2+ 296877 72 irq18: re0 2 0 irq19: ahc0 94035 22 irq20: atapci4 184381 44 irq23: vr0 2 0 Total 10202506 2487 /Daniel Eriksson
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:38:01 UTC