Does probe filter (BLDCFG_USE_HT_ASSIST) work in any AMD family?
I have server Supermicro H8QGI board and 4 AMD Opteron f15 processors. Linux won't boot up with HT-ASSIST feature enabled. No errors... Linux just stuck in some place like this
[ 0.369877] CPU0: AM Opteron(TM) Processor 6234 tepping 02 [ 0.380023] Performance Events: AMD Family 15h PMU driver. [ 0.380026] ... version: 0 [ 0.384024] ... bit width: 48 [ 0.388024] ... generic registers: 6 [ 0.392025] ... value mask: 0000ffffffffffff [ 0.396025] ... max period: 00007fffffffffff [ 0.400025] ... fixed-purpose events: 0 [ 0.404025] ... event mask: 000000000000003f [ 0.408286] NMI watchdog enabled, takes one hw-pmu counter. [ 0.412251] Booting Node 0, Processors #1 [ 0.016000] do_IRQ: 1.55 No irq handler for vector (irq -1) [ 0.440077] NMI watchdog enabled, takes one hw-pmu counter. [ 0.444274] #2 [ 0.016000] do_IRQ: 2.55 No irq handler for vector (irq -1) [ 0.468056] NMI watchdog enabled, takes one hw-pmu counter. [ 0.472212] #3 [ 0.016000] do_IRQ: 3.55 No irq handler for vector (irq -1) [ 0.496055] NMI watchdog enabled, takes one hw-pmu counter."do_IRQ" errors present also in log without probe filter... i don't know are these things connected... anyway i opened another topic about do_IRQ (http://www.coreboot.org/pipermail/coreboot/2012-December/072998.html) Also h8qgi buildOpts.c have weird comment about BLDCFG_USE_HT_ASSIST. It says that this feature don't work in f10 family. So... does this feature work in any AMD family?
It's very important problem... because Linux can't really deal with AMD ACPI SLIT table without probe filter. Kernel code think that SLIT table is wrong, because it have non-diagonal elements equal to diagonal (drivers\acpi\numa.c -> "slit_valid" function). So it loads standard table that produces performance 5 times lower! (in STREAM)
-
Константин Аладышев