Re: [Qemu-devel] [PATCH] Exynos4: added RTC device

Re: [Qemu-devel] [PATCH] Exynos4: added RTC device

[Paolo Bonzini]

Il 29/06/2012 14:26, Andreas Färber ha scritto:
>> > 
>> > Oh, I see. Should we place this device to hw/Makefile.objs in v2?
> That would've been nice, but I'll do it as a follow-up now.

Yes, so we can also use Anthony's new CONFIG_ARCH_ARM (introducing
CONFIG_EXYNOS can be done later).

Paolo

Re: [Qemu-devel] [PATCH] Exynos4: added RTC device

[Igor Mitsyanko]

On 07/01/2012 07:26 PM, Paolo Bonzini wrote:
> Il 29/06/2012 14:26, Andreas Färber ha scritto:
>>>>
>>>> Oh, I see. Should we place this device to hw/Makefile.objs in v2?
>> That would've been nice, but I'll do it as a follow-up now.
>
> Yes, so we can also use Anthony's new CONFIG_ARCH_ARM (introducing
> CONFIG_EXYNOS can be done later).
>
> Paolo
>
>
>

So, what's the consensus here, for now new devices go to 
hw/arm/Makefile.objs while somebody moves all new and old not 
cpu-specific devices (not just exynos-related) from hw/arm/Makefile.objs 
to hw/Makefile.objs with one commit? Because having one RTC device 
compile through hw/Makefile.objs while all other exynos devices are 
compiled through hw/arm/Makefile.objs doesn't makes much sense.

And also its not clear whether you're planning to compile these devices 
through obj-y or through hw-obj-y, because we still have 
target_phys_addr_t in our memory API.

Re: [Qemu-devel] [PATCH] Exynos4: added RTC device

[Peter Maydell]

On 2 July 2012 10:08, Igor Mitsyanko <i.mitsyanko@samsung.com> wrote:
> So, what's the consensus here, for now new devices go to
> hw/arm/Makefile.objs while somebody moves all new and old not cpu-specific
> devices (not just exynos-related) from hw/arm/Makefile.objs to
> hw/Makefile.objs with one commit? Because having one RTC device compile
> through hw/Makefile.objs while all other exynos devices are compiled through
> hw/arm/Makefile.objs doesn't makes much sense.

I don't want things moved piecemeal, especially not one file from
a whole board model.

I'd also like to see a nice clear summary of the ground rules first
(ie how you decide which makefile / target / whatever a file should
be in). At the moment I'm not really sure what the rules are, which
means I can't properly review those bits of patches.

-- PMM

Re: [Qemu-devel] [PATCH] Exynos4: added RTC device

[Paolo Bonzini]

Il 02/07/2012 11:31, Peter Maydell ha scritto:
> On 2 July 2012 10:08, Igor Mitsyanko <i.mitsyanko@samsung.com> wrote:
>> So, what's the consensus here, for now new devices go to
>> hw/arm/Makefile.objs while somebody moves all new and old not cpu-specific
>> devices (not just exynos-related) from hw/arm/Makefile.objs to
>> hw/Makefile.objs with one commit? Because having one RTC device compile
>> through hw/Makefile.objs while all other exynos devices are compiled through
>> hw/arm/Makefile.objs doesn't makes much sense.
> 
> I don't want things moved piecemeal, especially not one file from
> a whole board model.
> 
> I'd also like to see a nice clear summary of the ground rules first
> (ie how you decide which makefile / target / whatever a file should
> be in). At the moment I'm not really sure what the rules are, which
> means I can't properly review those bits of patches.

For now nothing should change compared to the past, except that obj-y
should appear in hw/ARCH/Makefile.objs rather than Makefile.target.
This is because there is still no mechanism to guard the build of
ARM-only devices from hw/Makefile.objs.  When Anthony's patch lands, we
can start moving files to hw/Makefile.objs using CONFIG_ARCH_ARM; I
understood Andreas is going to do that.

For new boards, you may want to request a new CONFIG_* symbol (e.g.
CONFIG_EXYNOS) so that new boards can already compile their devices from
hw/Makefile.objs.  However, that's up to you and definitely shouldn't be
done for one file from a whole board model, as you said.

Paolo

Re: [Qemu-devel] [PATCH] Exynos4: added RTC device

[Paolo Bonzini]

Il 02/07/2012 11:37, Paolo Bonzini ha scritto:
> Il 02/07/2012 11:31, Peter Maydell ha scritto:
>> On 2 July 2012 10:08, Igor Mitsyanko <i.mitsyanko@samsung.com> wrote:
>>> So, what's the consensus here, for now new devices go to
>>> hw/arm/Makefile.objs while somebody moves all new and old not cpu-specific
>>> devices (not just exynos-related) from hw/arm/Makefile.objs to
>>> hw/Makefile.objs with one commit? Because having one RTC device compile
>>> through hw/Makefile.objs while all other exynos devices are compiled through
>>> hw/arm/Makefile.objs doesn't makes much sense.
>>
>> I don't want things moved piecemeal, especially not one file from
>> a whole board model.
>>
>> I'd also like to see a nice clear summary of the ground rules first
>> (ie how you decide which makefile / target / whatever a file should
>> be in). At the moment I'm not really sure what the rules are, which
>> means I can't properly review those bits of patches.
> 
> For now nothing should change compared to the past, except that obj-y
> should appear in hw/ARCH/Makefile.objs rather than Makefile.target.
> This is because there is still no mechanism to guard the build of
> ARM-only devices from hw/Makefile.objs.  When Anthony's patch lands, we
> can start moving files to hw/Makefile.objs using CONFIG_ARCH_ARM; I
> understood Andreas is going to do that.

Small addendum: I think the idea should be to put source in hw/ARCH if
it has a hard dependency on target-ARCH, otherwise we can leave it in
hw/ and decide on a case-by-case basis.

Personally, I believe it'd be best if board descriptions were moved to
hw/ARCH, even if all the required hardware is in hw/ and even if the
file can be moved from obj-y to hw-obj-y.  However, this can be left
open to later discussion, and is complicated by the fact that some
boards (e.g. musicpal) include devices and machine models in the same file.

Paolo

Re: [Qemu-devel] [PATCH] Exynos4: added RTC device

[Peter Maydell]

On 2 July 2012 10:42, Paolo Bonzini <pbonzini@redhat.com> wrote:
> Personally, I believe it'd be best if board descriptions were moved to
> hw/ARCH, even if all the required hardware is in hw/ and even if the
> file can be moved from obj-y to hw-obj-y.  However, this can be left
> open to later discussion, and is complicated by the fact that some
> boards (e.g. musicpal) include devices and machine models in the same file.

I don't personally see the distinction between "device model" and
"machine model" as particularly interesting, and I hope/think that
with increased QOMification the implementation differences beween
the two will tend go away.

-- PMM

Re: [Qemu-devel] [PATCH] Exynos4: added RTC device

[Paolo Bonzini]

Il 02/07/2012 11:44, Peter Maydell ha scritto:
> On 2 July 2012 10:42, Paolo Bonzini <pbonzini@redhat.com> wrote:
>> > Personally, I believe it'd be best if board descriptions were moved to
>> > hw/ARCH, even if all the required hardware is in hw/ and even if the
>> > file can be moved from obj-y to hw-obj-y.  However, this can be left
>> > open to later discussion, and is complicated by the fact that some
>> > boards (e.g. musicpal) include devices and machine models in the same file.
> I don't personally see the distinction between "device model" and
> "machine model" as particularly interesting, and I hope/think that
> with increased QOMification the implementation differences beween
> the two will tend go away.

I agree.  The distinction is just that machine models often have a
dependency on target-ARCH/ files, and that would be (for me) the line
between hw/ and hw/ARCH.

Paolo

Re: [Qemu-devel] [PATCH] Exynos4: added RTC device

[Andreas Färber]

Am 02.07.2012 11:37, schrieb Paolo Bonzini:
> Il 02/07/2012 11:31, Peter Maydell ha scritto:
>> On 2 July 2012 10:08, Igor Mitsyanko <i.mitsyanko@samsung.com> wrote:
>>> So, what's the consensus here, for now new devices go to
>>> hw/arm/Makefile.objs while somebody moves all new and old not cpu-specific
>>> devices (not just exynos-related) from hw/arm/Makefile.objs to
>>> hw/Makefile.objs with one commit? Because having one RTC device compile
>>> through hw/Makefile.objs while all other exynos devices are compiled through
>>> hw/arm/Makefile.objs doesn't makes much sense.
>>
>> I don't want things moved piecemeal, especially not one file from
>> a whole board model.
>>
>> I'd also like to see a nice clear summary of the ground rules first
>> (ie how you decide which makefile / target / whatever a file should
>> be in). At the moment I'm not really sure what the rules are, which
>> means I can't properly review those bits of patches.
> 
> For now nothing should change compared to the past, except that obj-y
> should appear in hw/ARCH/Makefile.objs rather than Makefile.target.
> This is because there is still no mechanism to guard the build of
> ARM-only devices from hw/Makefile.objs.  When Anthony's patch lands, we
> can start moving files to hw/Makefile.objs using CONFIG_ARCH_ARM; I
> understood Andreas is going to do that.

Not quite. What I started long before Anthony's patch is making devices
(first Xilinx then ARM) less dependent on my CPU changes. That does not
need CONFIG_[ARCH_]* in our current system. With armeb in mind I've gone
for SoC-level and more fine-granular for generic devices.

Andreas

-- 
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer; HRB 16746 AG Nürnberg

Re: [Qemu-devel] [PATCH] Exynos4: added RTC device

[Andreas Färber]

Am 02.07.2012 11:31, schrieb Peter Maydell:
> On 2 July 2012 10:08, Igor Mitsyanko <i.mitsyanko@samsung.com> wrote:
>> So, what's the consensus here, for now new devices go to
>> hw/arm/Makefile.objs while somebody moves all new and old not cpu-specific
>> devices (not just exynos-related) from hw/arm/Makefile.objs to
>> hw/Makefile.objs with one commit? Because having one RTC device compile
>> through hw/Makefile.objs while all other exynos devices are compiled through
>> hw/arm/Makefile.objs doesn't makes much sense.
> 
> I don't want things moved piecemeal, especially not one file from
> a whole board model.
> 
> I'd also like to see a nice clear summary of the ground rules first
> (ie how you decide which makefile / target / whatever a file should
> be in). At the moment I'm not really sure what the rules are, which
> means I can't properly review those bits of patches.

We've been through that like three times lately, not sure what's left to
discuss there?

There's two things to review in the migration we're discussing:
1) Are files compiled before the conversion still compiled afterwards?
2) Does the build break due to use of CPU or explicit swapping?

For new patches you should keep on eye on whether a new device really
must go into hw/arm/Makefile.objs or whether it can go in
hw/Makefile.objs instead. I.e., hw/Makefile.objs (libhwX/) should be the
rule for devices, not the exception. Machines today are tightly coupled
to the CPU and need to go into hw/arch/Makefile.objs always. Whether the
files get placed into hw/arch/ or hw/ is a matter of taste.

Andreas

-- 
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer; HRB 16746 AG Nürnberg

Re: [Qemu-devel] [PATCH] Exynos4: added RTC device

[Peter Maydell]

On 2 July 2012 13:04, Andreas Färber <afaerber@suse.de> wrote:
> Am 02.07.2012 11:31, schrieb Peter Maydell:
>> On 2 July 2012 10:08, Igor Mitsyanko <i.mitsyanko@samsung.com> wrote:
>>> So, what's the consensus here, for now new devices go to
>>> hw/arm/Makefile.objs while somebody moves all new and old not cpu-specific
>>> devices (not just exynos-related) from hw/arm/Makefile.objs to
>>> hw/Makefile.objs with one commit? Because having one RTC device compile
>>> through hw/Makefile.objs while all other exynos devices are compiled through
>>> hw/arm/Makefile.objs doesn't makes much sense.
>>
>> I don't want things moved piecemeal, especially not one file from
>> a whole board model.
>>
>> I'd also like to see a nice clear summary of the ground rules first
>> (ie how you decide which makefile / target / whatever a file should
>> be in). At the moment I'm not really sure what the rules are, which
>> means I can't properly review those bits of patches.
>
> We've been through that like three times lately, not sure what's left to
> discuss there?

It's not in docs/ or in some makefile as a comment. I don't mean
"I want another mailing list thread"...

-- PMM

Re: [Qemu-devel] [PATCH] Exynos4: added RTC device

[Andreas Färber]

Am 01.07.2012 17:26, schrieb Paolo Bonzini:
> Il 29/06/2012 14:26, Andreas Färber ha scritto:
>>>>
>>>> Oh, I see. Should we place this device to hw/Makefile.objs in v2?
>> That would've been nice, but I'll do it as a follow-up now.
> 
> Yes, so we can also use Anthony's new CONFIG_ARCH_ARM (introducing
> CONFIG_EXYNOS can be done later).

I have a patch in the works, but exynos turned out more difficult than
other devices (also OMAP and PXA2xx) because it uses a shared
<socname>.h header that uses CPU. For tegra I fixed that by adding an
#ifdef NEED_CPU_H for the appropriate parts.

Andreas

-- 
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer; HRB 16746 AG Nürnberg

Re: [Qemu-devel] [PATCH] Exynos4: added RTC device

[Igor Mitsyanko]

On 07/02/2012 03:44 PM, Andreas Färber wrote:
> Am 01.07.2012 17:26, schrieb Paolo Bonzini:
>> Il 29/06/2012 14:26, Andreas Färber ha scritto:
>>>>>
>>>>> Oh, I see. Should we place this device to hw/Makefile.objs in v2?
>>> That would've been nice, but I'll do it as a follow-up now.
>>
>> Yes, so we can also use Anthony's new CONFIG_ARCH_ARM (introducing
>> CONFIG_EXYNOS can be done later).
>
> I have a patch in the works, but exynos turned out more difficult than
> other devices (also OMAP and PXA2xx) because it uses a shared
> <socname>.h header that uses CPU. For tegra I fixed that by adding an
> #ifdef NEED_CPU_H for the appropriate parts.
>
> Andreas
>

I think we can drop ARMCPU *cpu[] member from Exynos4210State and just 
use local variable in exynos4210_init() instead, like everyone else does.
exynos4210_write_secondary() prototype can be moved from exynos4210.h 
header to exynos4_boards.c, or we can pass arm_boot_info 
exynos4_board_binfo to Exynos4210 SoC init function exynos4210_init() to 
initialize .write_secondary_boot field there.