From: Linus Walleij <redacted>
Instead of allocating descriptors on-the-fly for the device tree
initialization case, use irq_domain_add_simple() which will take
care of this if you pass negative as the first_irq.
Alter the signature of __vic_init() to pass the first_irq as
signed so this works as expected.
Switching the VIC to use irq_domain_add_simple() also has the
upside of displaying the same WARNING when you boot with
pre-allocated descriptors on systems using SPARSE_IRQ but
yet not using device tree.
Cc: Rob Herring <redacted>
Cc: Grant Likely <redacted>
Signed-off-by: Linus Walleij <redacted>
---
arch/arm/common/vic.c | 18 ++++++------------
arch/arm/include/asm/hardware/vic.h | 2 +-
2 files changed, 7 insertions(+), 13 deletions(-)
From: Rob Herring <hidden> Date: 2012-10-16 13:32:23
On 10/16/2012 08:06 AM, Linus Walleij wrote:
From: Linus Walleij <redacted>
Instead of allocating descriptors on-the-fly for the device tree
initialization case, use irq_domain_add_simple() which will take
care of this if you pass negative as the first_irq.
Alter the signature of __vic_init() to pass the first_irq as
signed so this works as expected.
Switching the VIC to use irq_domain_add_simple() also has the
upside of displaying the same WARNING when you boot with
pre-allocated descriptors on systems using SPARSE_IRQ but
yet not using device tree.
Cc: Rob Herring <redacted>
Cc: Grant Likely <redacted>
Signed-off-by: Linus Walleij <redacted>
"ARM: VIC: use irq_domain_add_simple()" causes a new warning:
arch/arm/common/vic.c: In function 'vic_of_init':
arch/arm/common/vic.c:410:6: warning: unused variable 'irq_base' [-Wunused-variable]
Since the irq_base variable is not used any more, it should be
removed now
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
On Tuesday 16 October 2012, Linus Walleij wrote:
From: Linus Walleij <redacted>
Instead of allocating descriptors on-the-fly for the device tree
initialization case, use irq_domain_add_simple() which will take
care of this if you pass negative as the first_irq.
Alter the signature of __vic_init() to pass the first_irq as
signed so this works as expected.
Switching the VIC to use irq_domain_add_simple() also has the
upside of displaying the same WARNING when you boot with
pre-allocated descriptors on systems using SPARSE_IRQ but
yet not using device tree.
Cc: Rob Herring <redacted>
Cc: Grant Likely <redacted>
Signed-off-by: Linus Walleij <redacted>
From: Russell King - ARM Linux <hidden> Date: 2012-11-08 13:55:37
On Thu, Nov 08, 2012 at 01:40:54PM +0000, Arnd Bergmann wrote:
"ARM: VIC: use irq_domain_add_simple()" causes a new warning:
arch/arm/common/vic.c: In function 'vic_of_init':
arch/arm/common/vic.c:410:6: warning: unused variable 'irq_base' [-Wunused-variable]
Since the irq_base variable is not used any more, it should be
removed now
I've just committed exactly the same patch to my tree...
On Thu, Nov 8, 2012 at 2:55 PM, Russell King - ARM Linux
[off-list ref] wrote:
On Thu, Nov 08, 2012 at 01:40:54PM +0000, Arnd Bergmann wrote:
quoted
"ARM: VIC: use irq_domain_add_simple()" causes a new warning:
arch/arm/common/vic.c: In function 'vic_of_init':
arch/arm/common/vic.c:410:6: warning: unused variable 'irq_base' [-Wunused-variable]
Since the irq_base variable is not used any more, it should be
removed now
I've just committed exactly the same patch to my tree...
From: Grant Likely <hidden> Date: 2012-12-18 23:34:35
Russell,
It looks to me like the below patch breaks versatile because it makes
it try to register a linear irq_domain instead of the legacy one that
it needs. Here are the relevant commits:
07c9249f1: "ARM: 7554/1: VIC: use irq_domain_add_simple()"
946c59a08: "ARM: vic: fix build warning caused by previous commit"
I'm working on getting it properly sorted out (and more importantly
*simplified*), but in the mean time I think the above two commits need
to be reverted. Reverting them on my tree fixes booting for me.
g.
On Tue, Oct 16, 2012 at 2:32 PM, Rob Herring [off-list ref] wrote:
On 10/16/2012 08:06 AM, Linus Walleij wrote:
quoted
From: Linus Walleij <redacted>
Instead of allocating descriptors on-the-fly for the device tree
initialization case, use irq_domain_add_simple() which will take
care of this if you pass negative as the first_irq.
Alter the signature of __vic_init() to pass the first_irq as
signed so this works as expected.
Switching the VIC to use irq_domain_add_simple() also has the
upside of displaying the same WARNING when you boot with
pre-allocated descriptors on systems using SPARSE_IRQ but
yet not using device tree.
Cc: Rob Herring <redacted>
Cc: Grant Likely <redacted>
Signed-off-by: Linus Walleij <redacted>
On Wed, Dec 19, 2012 at 12:34 AM, Grant Likely
[off-list ref] wrote:
It looks to me like the below patch breaks versatile because it makes
it try to register a linear irq_domain instead of the legacy one that
it needs. Here are the relevant commits:
07c9249f1: "ARM: 7554/1: VIC: use irq_domain_add_simple()"
946c59a08: "ARM: vic: fix build warning caused by previous commit"
I'm working on getting it properly sorted out (and more importantly
*simplified*), but in the mean time I think the above two commits need
to be reverted. Reverting them on my tree fixes booting for me.
Isn't that a bit violent, can't we just fix the real bug?
Does the below patch work for you, it does two things:
1) Bump all Versatile IRQs to offset at 32, because it is using
IRQ 0 which is NO_IRQ and illegal anyway so it's anyway
a bug that should be fixed.
2) Make sure we call irq_create_mapping() if the start IRQ is
anyway 0, as in the device tree case, and make sure to
actually pass zero in that case.
The initial patch set used negative IRQ as an indicator that
the linear domain should be used and not descriptors
allocated, but as that require changing a lot of singatures
to signed int and since IRQ 0 is now illegal anyway, I
let zero signify linear. Sorry for leftover mistakes like this...
I'll mail them out as separate patches too, and submit to
Russell's tracker if they fix the problem.
On Thu, Dec 20, 2012 at 7:45 PM, Linus Walleij [off-list ref] wrote:
On Wed, Dec 19, 2012 at 12:34 AM, Grant Likely
[off-list ref] wrote:
quoted
It looks to me like the below patch breaks versatile because it makes
it try to register a linear irq_domain instead of the legacy one that
it needs. Here are the relevant commits:
07c9249f1: "ARM: 7554/1: VIC: use irq_domain_add_simple()"
946c59a08: "ARM: vic: fix build warning caused by previous commit"
I'm working on getting it properly sorted out (and more importantly
*simplified*), but in the mean time I think the above two commits need
to be reverted. Reverting them on my tree fixes booting for me.
Isn't that a bit violent, can't we just fix the real bug?
Does the below patch work for you, it does two things:
I've submitted the fixes to Russell's patch tracker now ... would
appreciate some testing but I'm pretty sure they will fix the issue.
Yours,
Linus Walleij
From: Grant Likely <hidden> Date: 2013-01-11 18:57:58
On Thu, 20 Dec 2012 19:45:24 +0100, Linus Walleij [off-list ref] wrote:
On Wed, Dec 19, 2012 at 12:34 AM, Grant Likely
[off-list ref] wrote:
quoted
It looks to me like the below patch breaks versatile because it makes
it try to register a linear irq_domain instead of the legacy one that
it needs. Here are the relevant commits:
07c9249f1: "ARM: 7554/1: VIC: use irq_domain_add_simple()"
946c59a08: "ARM: vic: fix build warning caused by previous commit"
(sorry for the really late reply on this. I took a break at Christmas
and jetlag has been beating up my concentration since I got back. I
actually started writing this reply before Christmas, but the
investigation sidetracked me enough that I never finished it.)
quoted
I'm working on getting it properly sorted out (and more importantly
*simplified*), but in the mean time I think the above two commits need
to be reverted. Reverting them on my tree fixes booting for me.
Isn't that a bit violent, can't we just fix the real bug?
I did try to fix the bug first, but it became more involved that I would
like for a stablization change (and I went down a lot of rabbit trails
in the process). I preferred reverting because this was all contained
to a single driver which can be backed out easily to have another cycle
for a correct fix. I did try your change though...
Does the below patch work for you, it does two things:
1) Bump all Versatile IRQs to offset at 32, because it is using
IRQ 0 which is NO_IRQ and illegal anyway so it's anyway
a bug that should be fixed.
2) Make sure we call irq_create_mapping() if the start IRQ is
anyway 0, as in the device tree case, and make sure to
actually pass zero in that case.
In actual fact, only changing the offset to 32 is required to get
Versatile to boot with DT. That platform doesn't yet use vic_of_init().
It currently depends on pre-allocated irq_descs. Calling
irq_create_mapping() doesn't resolve this. I now have patches to rework
the versatile OF support to use irq_of_init(), but they aren't ready
for posting yet. Merging the bump to offset 32 would be fine for now.
Similarly here: #define IRQ_SIC_END (IRQ_SIC_START + 31)
And then you can add my acked by for the irq number changes.
Acked-by: Grant Likely <redacted>
g.
On Fri, Jan 11, 2013 at 7:57 PM, Grant Likely [off-list ref] wrote:
quoted
Does the below patch work for you, it does two things:
1) Bump all Versatile IRQs to offset at 32, because it is using
IRQ 0 which is NO_IRQ and illegal anyway so it's anyway
a bug that should be fixed.
2) Make sure we call irq_create_mapping() if the start IRQ is
anyway 0, as in the device tree case, and make sure to
actually pass zero in that case.
In actual fact, only changing the offset to 32 is required to get
Versatile to boot with DT. That platform doesn't yet use vic_of_init().
It currently depends on pre-allocated irq_descs.
OK both are already upstream, because there was someone
else verifying they solved the problem.
And the second patch was needed on the Integrator which
does use SPARSE_IRQ + vic_init_of().
Probably the DT-based SPEArs need it too.
Yours,
Linus Walleij