[lkml]   [2003]   [Jan]   [18]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: 48GB NUMA-Q boots, with major IO-APIC hassles
On Sat, Jan 18, 2003 at 09:32:22PM -0500, Zwane Mwaikambo wrote:
>> You'll drop irqs when you have collisions with devices
>> attached to other busses/ioapics

On Sat, Jan 18, 2003 at 06:55:14PM -0800, William Lee Irwin III wrote:
> Those aren't reachable anyway. Any given IO-APIC can only reach
> devices within its own node. The only possible issue is the priority
> class bounded-depth queueing issue (max of 2 or 3 pending) which I've
> decided to ignore until something closer to working materializes.

Clarification: each node has its own APIC bus, and only logical DESTMOD
RTE's can interrupt cpus on remote nodes. The invariant of "any given
IO-APIC can only interrupt cpus on the same node" comes from the fact
that the RTE destinations s are programmed for physical broadcast, which
by definition cannot reach any further than the local node / APIC bus.

Essentially, because only interrupts with logical destinations are
routed by the cluster controllers, the guarantee of local-only
interruption is provided by using physical destinations in RTE's.

The net result is that so long as there are no vector clashes within a
given node, the software interrupt number is uniquely determined by the
vector and the node the interrupt was received on. And it is always
possible to assign unique vectors within a node as there are 190 vectors
and only 48 IO-APIC RTE's/pins.

All good? No more IDT overwriting and/or cross-node interrupt number
sharing concerns?

-- wli
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

 \ /
  Last update: 2005-03-22 13:32    [W:0.033 / U:5.204 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site