lkml.org 
[lkml]   [2006]   [Sep]   [7]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH] FRV: do_gettimeofday() should no longer use tickadj
From
Date
On Thu, 2006-09-07 at 10:55 +0100, David Howells wrote:
> Benjamin Herrenschmidt <benh@kernel.crashing.org> wrote:
>
> > Well, genirq gives you more flexibility than the current mecanism so ...
>
> No, it doesn't because the FRV arch contains its own mechanism and I can do
> what ever I like in it.

But genirq allows you to have your own flow handlers... Which does mean
do whatever you like as well

> genirq's flexibility comes at a price. Count the number of hooks in struct
> irq_chip and struct irq_desc together.

You do not have to implement them all. Some of them are specific to a
given flow. For example, on XICS or MPIC, I only call eoi(), that is a
single hook, using the fasteoi handler. That's actually less than with
the previous generic code.

> > If I understand correctly, you need to do scray stuff to figure out your
> > toplevel irq, which shound't be a problem with either mecanisms...
>
> Yeah. I can't actually find out what source caused top-level IRQs. I have to
> guess from looking at the IRQ priority and poking around in the hardware. I
> got bitten that way too: at one point, I was peeking at the interrupt flag in
> the serial regs, only to realise this was causing the driver to go wrong
> because it cleared the interrupt requested flag in UART.
>
> Obviously I'd rather not use IRQ priorisation to help multiplex irqs, but
> unless I want a large polling set...

But neither the old mecanism nor genirq changes nor does anything in the
way of discovering what the toplevel irq is ... They only intervene
after you have found it...

> > Now, if you have funky cascades, then you can always group them into a
> > virtual irq cascade line and have a special chained flow handler that
> > does all the "figuring out" off those... it's up to you.
>
> You make it sound so easy, but it's not obvious how to do this, apart from
> installing interrupt handlers for the auxiliary PIC interrupts on the CPU and
> having those call back into __do_IRQ(). Chaining isn't mentioned in
> genericirq.tmpl.

No, you do a chain handler. Look at how I do it in
arch/powerpc/platform/pseries/setup.c for example. It's actually
trivial. You install a special flow handler (which means that there is
very little overhead, almost none, from the toplevel irq to the chained
irq). You can _also_ if you want just install an IRQ handler for the
cascaded controller and call generic_handle_irq (rather than __do_IRQ)
from it, but that has more overhead. A chained handler completely
relaces the flow handler for the cascade, and thus, if you don't need
all of the nits and bits of the other flow handlers for your cascade,
you can speed things up by hooking at that level.

> My code in the FRV arch has fewer indirection calls than the genirq code
> simply because it doesn't require tables of operations. I can trace through
> it with gdb and see them.
>
> I built all the stuff that genirq does in indirections directly into the
> handlers. It certainly has fewer hooks.

genirq allows you to do just that by using custom handlers.

> I attempted to convert it over to use genirq, and I came up with some numbers:
>
> The difference in kernel sizes:
>
> text data bss dec hex filename
> 1993023 77912 166964 2237899 2225cb vmlinux [with genirq]
> 1986511 76016 167908 2230435 2208a3 vmlinux [without genirq]
>
> The genirq subdir all wraps up into this:
>
> 10908 3272 12 14192 3770 kernel/irq/built-in.o
> 1548 64 4 1616 650 arch/frv/kernel/irq.o
> ---------------------------------------------------------------------
> 12456 3336 16 15808 3dc0 total
>
> My FRV-specific IRQ handling wraps up into these:
>
> 480 488 0 968 3c8 arch/frv/kernel/irq-mb93091.o
> 4688 16 520 5224 1468 arch/frv/kernel/irq.o
> 1576 1152 16 2744 ab8 arch/frv/kernel/irq-routing.o
> ---------------------------------------------------------------------
> 6744 1656 536 8936 22e8 total
>
> There's a difference in BSS size in the main kernel that I can't account for,
> but basically genirq uses 6.3KB more code and 1.8KB more initialised data, but
> 0.9KB less BSS. Overall, about 7.2KB more memory. I can shrink the BSS usage
> in the FRV specific version by reducing the amount of space in the IRQ sources
> table.
>
> So, again, why _should_ I use the generic IRQ stuff? It's bigger and very
> probably slower than what I already have.

Well, I would have to look precisely at how you did the port to
genirq... But in my case, it's definitely not slower, especially when
cascaded controllers are involved.

Ben.


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2006-09-08 00:59    [W:0.140 / U:1.204 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site