Messages in this thread | | | Date | Tue, 29 Jun 1999 17:58:06 +0200 | From | Jamie Lokier <> | Subject | Re: Improving raw network performance [was Apache performance etc.] |
| |
Benjamin Herrenschmidt wrote: > I beleive he though about using the kernel to mask/unmask the irq line, > which can cause trouble with shared IRQs. Of course, the solution is to > mask them on-chip.
Ok, this I understand. Barring interactions with RTLinux, which doesn't disable them on chip.
> Note that masking/unmasking wiht the kernel's disable_irq/enable_irq from > within the IRQ handler will not work on some platforms (notably PPC, > except if things change since the last time I hacked in this piece of > code). The problem is that if you mask the irq from within the handler, > it will be unmasked by the kernel when going out of the handler.
I think this is true on x86 also, at least for 8259 irqs.
I'm suggesting a change to the generic interrupt dispatch code rather than to the drivers, though the drivers might have to inform the dispatch code that it's ok to switch to timed polling.
-- Jamie
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |