ARM kernel config for Linux 3.11

classic Classic list List threaded Threaded
8 messages Options
Reply | Threaded
Open this post in threaded view
|

ARM kernel config for Linux 3.11

Ben Hutchings-3
As discussed previously, I think PCI should be enabled for the
armhf/armmp flavour as the Marvell Armada SoCs have PCIe interfaces.  It
might be worthwhile to restrict which PCI drivers are built, if this
slows the build down a lot.

Ian also proposed to add an armhf/armmp-lpae flavour.  I think it's time
to do that, but I would also like to see the 3 platform-specific armhf
flavours removed.  Which of the platforms are sufficiently well
supported by armhf/armmp that we can do that?

The new imx-drm driver (CONFIG_DRM_IMX) should probably be enabled in
armhf/armmp and armhf/mx5.

After some cleanup of the top-level configuration, the armel/iop32x and
armel/ixp4xx flavours can again be built.  However these may still need
specific attention in the future.

Ben.

--
Ben Hutchings
In a hierarchy, every employee tends to rise to his level of incompetence.

signature.asc (845 bytes) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: ARM kernel config for Linux 3.11

Ian Campbell-2
On Sun, 2013-09-15 at 16:42 +0100, Ben Hutchings wrote:
> As discussed previously, I think PCI should be enabled for the
> armhf/armmp flavour as the Marvell Armada SoCs have PCIe interfaces.  It
> might be worthwhile to restrict which PCI drivers are built, if this
> slows the build down a lot.

I've started preparing patches for this against trunk. As well as
turning on CONFIG_PCI I'm also looking at enabling some additional NIC
and SCSI cards and adding the standard nic-modules and scsi-modules to
the udebs.

My armhf system is pretty fast so I can't make a good judgement on
slowing the build down (however that should be mitigated by the removal
of other flavours?)

> Ian also proposed to add an armhf/armmp-lpae flavour.  I think it's time
> to do that, but I would also like to see the 3 platform-specific armhf
> flavours removed.  Which of the platforms are sufficiently well
> supported by armhf/armmp that we can do that?

I've got patches to remove mx5 and vexpress in my queue, I don't know
how/why I skipped omap, but I'll add it in.

I cannot answer the question of how well they are actually supported (so
I'd be reluctant to push them). I do have some i.MX53 (loco) boards at
the office but I'm away at the minute (they are technically "production"
but are mostly obsolete and I could nick one or two for testing).

> The new imx-drm driver (CONFIG_DRM_IMX) should probably be enabled in
> armhf/armmp and armhf/mx5.

That reminds me, kernel-wedge is currently complaining about missing
sdhci-esdhc-imx from the mx5 flavour. I think this is fallout from
disabling MMC_SDHCI_PLTFM in 3.11-1~exp1.

> After some cleanup of the top-level configuration, the armel/iop32x and
> armel/ixp4xx flavours can again be built.  However these may still need
> specific attention in the future.
>
> Ben.
>



--
To UNSUBSCRIBE, email to [hidden email]
with a subject of "unsubscribe". Trouble? Contact [hidden email]
Archive: http://lists.debian.org/1379260396.11304.35.camel@...

Reply | Threaded
Open this post in threaded view
|

Re: ARM kernel config for Linux 3.11

Ben Hutchings-3
On Sun, 2013-09-15 at 16:53 +0100, Ian Campbell wrote:

> On Sun, 2013-09-15 at 16:42 +0100, Ben Hutchings wrote:
> > As discussed previously, I think PCI should be enabled for the
> > armhf/armmp flavour as the Marvell Armada SoCs have PCIe interfaces.  It
> > might be worthwhile to restrict which PCI drivers are built, if this
> > slows the build down a lot.
>
> I've started preparing patches for this against trunk. As well as
> turning on CONFIG_PCI I'm also looking at enabling some additional NIC
> and SCSI cards and adding the standard nic-modules and scsi-modules to
> the udebs.
>
> My armhf system is pretty fast so I can't make a good judgement on
> slowing the build down (however that should be mitigated by the removal
> of other flavours?)
Right.

> > Ian also proposed to add an armhf/armmp-lpae flavour.  I think it's time
> > to do that, but I would also like to see the 3 platform-specific armhf
> > flavours removed.  Which of the platforms are sufficiently well
> > supported by armhf/armmp that we can do that?
>
> I've got patches to remove mx5 and vexpress in my queue, I don't know
> how/why I skipped omap, but I'll add it in.
>
> I cannot answer the question of how well they are actually supported (so
> I'd be reluctant to push them). I do have some i.MX53 (loco) boards at
> the office but I'm away at the minute (they are technically "production"
> but are mostly obsolete and I could nick one or two for testing).
If no-one answers this then I think the way to find out is to go ahead
and remove them.

> > The new imx-drm driver (CONFIG_DRM_IMX) should probably be enabled in
> > armhf/armmp and armhf/mx5.
>
> That reminds me, kernel-wedge is currently complaining about missing
> sdhci-esdhc-imx from the mx5 flavour. I think this is fallout from
> disabling MMC_SDHCI_PLTFM in 3.11-1~exp1.

Right, I've just fixed that.  I also re-enabled REGULATOR properly (for
some reason I had changed it to =m which is not valid).

I also noticed that ARM configs mostly have CONFIG_CPU_THERMAL=m but
this symbol is now a boolean and the result is that it's disabled.  I
fixed that too.

Ben.

> > After some cleanup of the top-level configuration, the armel/iop32x and
> > armel/ixp4xx flavours can again be built.  However these may still need
> > specific attention in the future.
> >
> > Ben.

--
Ben Hutchings
In a hierarchy, every employee tends to rise to his level of incompetence.

signature.asc (845 bytes) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: ARM kernel config for Linux 3.11

Luke Kenneth Casson Leighton
In reply to this post by Ben Hutchings-3
On Sun, Sep 15, 2013 at 4:42 PM, Ben Hutchings <[hidden email]> wrote:
> As discussed previously, I think PCI should be enabled for the
> armhf/armmp flavour as the Marvell Armada SoCs have PCIe interfaces.

 and the freescale iMX6, and one of the samsung SoCs (the exynos5),
and the TI AM3389 (although that's a bit of a stretch / specialist
SoC), and... i'm sure there was one other... maybe it was the one in
my imagination oh look little blue fairies *wanders off*...

 sorry :)  yes.  definitely the iMX6 and the Exynos5.  the iMX6 is
only single-channel PCIe but the Exynos5 is 4 reconfigureable lanes
(1x4, 2x2).

l.


--
To UNSUBSCRIBE, email to [hidden email]
with a subject of "unsubscribe". Trouble? Contact [hidden email]
Archive: http://lists.debian.org/CAPweEDyZbubGO-Ro3Zr+GkjvV45-NNJXvaqFasMjx2yfPF==ww@...

Reply | Threaded
Open this post in threaded view
|

Re: ARM kernel config for Linux 3.11

Arnaud Patard (Rtp)
In reply to this post by Ben Hutchings-3
Ben Hutchings <[hidden email]> writes:

> As discussed previously, I think PCI should be enabled for the
> armhf/armmp flavour as the Marvell Armada SoCs have PCIe interfaces.  It
> might be worthwhile to restrict which PCI drivers are built, if this
> slows the build down a lot.

yeah, the problem is more about choosing what to enable or not rather
than enabling PCI or not. For instance, I remember reading some news
about running a nvidia pci express card on tegra host...

>
> Ian also proposed to add an armhf/armmp-lpae flavour.  I think it's time
> to do that, but I would also like to see the 3 platform-specific armhf
> flavours removed.  Which of the platforms are sufficiently well
> supported by armhf/armmp that we can do that?

all platforms should be working quite well with armmp kernels theses
days. Unfortunately, most (vendor-related) upstreams are not always
testing with drivers compiled as modules, so I don't think they're
testing with other SoCs enabled. Let's hope it'll go well.

At least, if we add LPAE flavour, the other non multiplatform should be
dropped to avoid waiting days for new kernels.

>
> The new imx-drm driver (CONFIG_DRM_IMX) should probably be enabled in
> armhf/armmp and armhf/mx5.

I'm not sure how well it's working. I've even seen recently patches for
using it on loco boards so some support may still be missing (no idea
about imx6 support). There has been also some lockdep warnings for ages.
There are maybe other bugs but I don't remember them for now. I've
already went through a lot of ups and downs with this driver.

>
> After some cleanup of the top-level configuration, the armel/iop32x and
> armel/ixp4xx flavours can again be built.  However these may still need
> specific attention in the future.

There's some work done to move from current gpio support on iop to
gpiolib. I fear it may increase vmlinux size (of course, it may also be
a decrease in size). Nevertheless, it proved that there are still some users
of iop32x (mostly n2100 users).

Arnaud


--
To UNSUBSCRIBE, email to [hidden email]
with a subject of "unsubscribe". Trouble? Contact [hidden email]
Archive: http://lists.debian.org/87zjrddw8e.fsf@...

Reply | Threaded
Open this post in threaded view
|

Re: ARM kernel config for Linux 3.11

Arnaud Patard (Rtp)
In reply to this post by Luke Kenneth Casson Leighton
Luke Kenneth Casson Leighton <[hidden email]> writes:

> On Sun, Sep 15, 2013 at 4:42 PM, Ben Hutchings <[hidden email]> wrote:
>> As discussed previously, I think PCI should be enabled for the
>> armhf/armmp flavour as the Marvell Armada SoCs have PCIe interfaces.
>
>  and the freescale iMX6, and one of the samsung SoCs (the exynos5),
> and the TI AM3389 (although that's a bit of a stretch / specialist
> SoC), and... i'm sure there was one other... maybe it was the one in
> my imagination oh look little blue fairies *wanders off*...
>
>  sorry :)  yes.  definitely the iMX6 and the Exynos5.  the iMX6 is
> only single-channel PCIe but the Exynos5 is 4 reconfigureable lanes
> (1x4, 2x2).

exynos in armmp won't happen anytime soon unless you bug upstream about
multiplatform support. iirc, even in -next, it's still not possible to
enable exynos in multiplatform builds.

Arnaud


--
To UNSUBSCRIBE, email to [hidden email]
with a subject of "unsubscribe". Trouble? Contact [hidden email]
Archive: http://lists.debian.org/87vc21dw3g.fsf@...

Reply | Threaded
Open this post in threaded view
|

Re: ARM kernel config for Linux 3.11

Ian Campbell-2
In reply to this post by Ben Hutchings-3
On Sun, 2013-09-15 at 16:42 +0100, Ben Hutchings wrote:
> As discussed previously, I think PCI should be enabled for the
> armhf/armmp flavour as the Marvell Armada SoCs have PCIe interfaces.  It
> might be worthwhile to restrict which PCI drivers are built, if this
> slows the build down a lot.
>
> Ian also proposed to add an armhf/armmp-lpae flavour.  I think it's time
> to do that, but I would also like to see the 3 platform-specific armhf
> flavours removed.  Which of the platforms are sufficiently well
> supported by armhf/armmp that we can do that?

Sounds like we have a rough consensus that we should try all this.

I've got patches which I'll push in the next couple of days.

I've also got a patch in my back pocket to add transitional/dummy
packages for the removed packages to linux-latest but I can't push that
until 3.11 moves out of trunk onto the sid branch.

Ian.

signature.asc (853 bytes) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: ARM kernel config for Linux 3.11

Mark brown-22
In reply to this post by Arnaud Patard (Rtp)
On Mon, Sep 16, 2013 at 10:11:47AM +0200, Arnaud Patard wrote:

> exynos in armmp won't happen anytime soon unless you bug upstream about
> multiplatform support. iirc, even in -next, it's still not possible to
> enable exynos in multiplatform builds.

It does mostly work with mainline, audio still needs doing but other
than that it should be good now.  Older Samsung SoCs still have work
needed but the newer ones should be multiplatform safe.

signature.asc (853 bytes) Download Attachment