Re: [PATCH] irq_domain/microblaze: Convert microblaze to use irq_domains

Grant Likely wrote:

On Mon, Jan 30, 2012 at 12:42:26PM +0100, Michal Simek wrote:

quoted

Hi Grant,

Grant Likely wrote:

quoted

***totally untested.  Don't apply yet***

Michal, I leave it to you to test this patch.  I've turned on SPARSE_IRQ, but
you may decide that isn't valuable.  I've compile tested it, but that's
about it (in fact, I had to hack head.S to get it to build because I don't have
an up-to-date compiler for microblaze.  It fails on an unknown opcode 'lwr')

I have looked at it and I there is problem with virq allocation in irq_create_mapping
which calles irq_alloc_desc(0) and then __irq_alloc_descs(-1, 0, 1, node, THIS_MODULE)
where bitmap_find_next_zero_area returns 0 (start = 0) which is passed to alloc_descs
which just return it.
It doesn't matter if SPARSE_IRQ is enabled or not.
Error log is below.

Can you give me hint how to fix it and what expected values are?

irq0 is getting allocated when it shouldn't be.  The fix is to change the
calls in irqdomain.c from irq_alloc_desc(0) to irq_alloc_desc_from(1, 0).

I've got the fix in my tree and I'm pushing it out to the following branch
in a few minutes:

git://git.secretlab.ca/git/linux-2.6 irqdomain/next

ok. I have tested your new branch I have fixed and tested microblaze patch.
I will disable SPARSE_IRQ because nothing bring to us. From my quick look it will
just additional code.

The last two thing before I send updated patch is why did you choosed NR_IRQ to 64.
We doesn't support cascades or any complicated IRQ subsystem and intc can provide just 32 lines.

And can you see any strong reason to keep get_irq in hardirq.h. I would move it to irq.h.

Thanks,
Michal


-- 
Michal Simek, Ing. (M.Eng)
w: www.monstr.eu p: +42-0-721842854
Maintainer of Linux kernel 2.6 Microblaze Linux - http://www.monstr.eu/fdt/
Microblaze U-BOOT custodian