The merging of a number of clocksource drivers into fttmr010 means we
require clock-names to be specified in the Aspeed timer node, else the
clocksource fails to probe and boot hangs.
Signed-off-by: Andrew Jeffery <[email protected]>
---
arch/arm/boot/dts/aspeed-g4.dtsi | 1 +
arch/arm/boot/dts/aspeed-g5.dtsi | 1 +
2 files changed, 2 insertions(+)
diff --git a/arch/arm/boot/dts/aspeed-g4.dtsi b/arch/arm/boot/dts/aspeed-g4.dtsi
index 8c6bc29eb7f6..3e74929d3289 100644
--- a/arch/arm/boot/dts/aspeed-g4.dtsi
+++ b/arch/arm/boot/dts/aspeed-g4.dtsi
@@ -893,6 +893,7 @@
//interrupts = <16 17 18 35 36 37 38 39>;
interrupts = <16>;
clocks = <&clk_apb>;
+ clock-names = "PCLK";
};
wdt1: wdt@1e785000 {
diff --git a/arch/arm/boot/dts/aspeed-g5.dtsi b/arch/arm/boot/dts/aspeed-g5.dtsi
index a0bea4a6ec77..1e6c701da853 100644
--- a/arch/arm/boot/dts/aspeed-g5.dtsi
+++ b/arch/arm/boot/dts/aspeed-g5.dtsi
@@ -1000,6 +1000,7 @@
//interrupts = <16 17 18 35 36 37 38 39>;
interrupts = <16>;
clocks = <&clk_apb>;
+ clock-names = "PCLK";
};
--
2.11.0
On Mon, Jun 5, 2017 at 5:18 PM, Andrew Jeffery <[email protected]> wrote:
> The merging of a number of clocksource drivers into fttmr010 means we
> require clock-names to be specified in the Aspeed timer node, else the
> clocksource fails to probe and boot hangs.
Arnd,
Linus' reworked timer driver will go into 4.13.
Can we get this patch merged into 4.12 as a fix so we don't end up
with a broken boot at any stage?
Cheers,
Joel
>
> Signed-off-by: Andrew Jeffery <[email protected]>
> ---
> arch/arm/boot/dts/aspeed-g4.dtsi | 1 +
> arch/arm/boot/dts/aspeed-g5.dtsi | 1 +
> 2 files changed, 2 insertions(+)
>
> diff --git a/arch/arm/boot/dts/aspeed-g4.dtsi b/arch/arm/boot/dts/aspeed-g4.dtsi
> index 8c6bc29eb7f6..3e74929d3289 100644
> --- a/arch/arm/boot/dts/aspeed-g4.dtsi
> +++ b/arch/arm/boot/dts/aspeed-g4.dtsi
> @@ -893,6 +893,7 @@
> //interrupts = <16 17 18 35 36 37 38 39>;
> interrupts = <16>;
> clocks = <&clk_apb>;
> + clock-names = "PCLK";
> };
>
> wdt1: wdt@1e785000 {
> diff --git a/arch/arm/boot/dts/aspeed-g5.dtsi b/arch/arm/boot/dts/aspeed-g5.dtsi
> index a0bea4a6ec77..1e6c701da853 100644
> --- a/arch/arm/boot/dts/aspeed-g5.dtsi
> +++ b/arch/arm/boot/dts/aspeed-g5.dtsi
> @@ -1000,6 +1000,7 @@
> //interrupts = <16 17 18 35 36 37 38 39>;
> interrupts = <16>;
> clocks = <&clk_apb>;
> + clock-names = "PCLK";
> };
>
>
> --
> 2.11.0
>
On Mon, Jun 05, 2017 at 06:29:53PM +0930, Joel Stanley wrote:
> On Mon, Jun 5, 2017 at 5:18 PM, Andrew Jeffery <[email protected]> wrote:
> > The merging of a number of clocksource drivers into fttmr010 means we
> > require clock-names to be specified in the Aspeed timer node, else the
> > clocksource fails to probe and boot hangs.
>
> Arnd,
>
> Linus' reworked timer driver will go into 4.13.
>
> Can we get this patch merged into 4.12 as a fix so we don't end up
> with a broken boot at any stage?
I can take care of adding this patch in my branch before Linus' changes.
-- Daniel
On Tue, Jun 6, 2017 at 6:41 AM, Daniel Lezcano
<[email protected]> wrote:
> On Mon, Jun 05, 2017 at 06:29:53PM +0930, Joel Stanley wrote:
>> On Mon, Jun 5, 2017 at 5:18 PM, Andrew Jeffery <[email protected]> wrote:
>> > The merging of a number of clocksource drivers into fttmr010 means we
>> > require clock-names to be specified in the Aspeed timer node, else the
>> > clocksource fails to probe and boot hangs.
>>
>> Arnd,
>>
>> Linus' reworked timer driver will go into 4.13.
>>
>> Can we get this patch merged into 4.12 as a fix so we don't end up
>> with a broken boot at any stage?
>
> I can take care of adding this patch in my branch before Linus' changes.
That should be okay. Thanks Daniel!
Cheers,
Joel
On Mon, Jun 5, 2017 at 10:59 AM, Joel Stanley <[email protected]> wrote:
> On Mon, Jun 5, 2017 at 5:18 PM, Andrew Jeffery <[email protected]> wrote:
>> The merging of a number of clocksource drivers into fttmr010 means we
>> require clock-names to be specified in the Aspeed timer node, else the
>> clocksource fails to probe and boot hangs.
>
> Arnd,
>
> Linus' reworked timer driver will go into 4.13.
>
> Can we get this patch merged into 4.12 as a fix so we don't end up
> with a broken boot at any stage?
Hmm, can't we make the driver backward-compatible and have it fall
back on the first clock if no clk named "PCLK" is found? Otherwise
you still have an incompatible change in the DT binding and it will
break if someone uses an older dtb with a newer kernel.
Arnd
On Tue, Jun 06, 2017 at 11:41:11AM +0200, Arnd Bergmann wrote:
> On Mon, Jun 5, 2017 at 10:59 AM, Joel Stanley <[email protected]> wrote:
> > On Mon, Jun 5, 2017 at 5:18 PM, Andrew Jeffery <[email protected]> wrote:
> >> The merging of a number of clocksource drivers into fttmr010 means we
> >> require clock-names to be specified in the Aspeed timer node, else the
> >> clocksource fails to probe and boot hangs.
> >
> > Arnd,
> >
> > Linus' reworked timer driver will go into 4.13.
> >
> > Can we get this patch merged into 4.12 as a fix so we don't end up
> > with a broken boot at any stage?
>
> Hmm, can't we make the driver backward-compatible and have it fall
> back on the first clock if no clk named "PCLK" is found? Otherwise
> you still have an incompatible change in the DT binding and it will
> break if someone uses an older dtb with a newer kernel.
I would like to avoid to hack the kernel code for backward DT compatible
things.
--
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs
Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog
On Wed, Jun 7, 2017 at 2:52 PM, Daniel Lezcano
<[email protected]> wrote:
> On Tue, Jun 06, 2017 at 11:41:11AM +0200, Arnd Bergmann wrote:
>> On Mon, Jun 5, 2017 at 10:59 AM, Joel Stanley <[email protected]> wrote:
>> > On Mon, Jun 5, 2017 at 5:18 PM, Andrew Jeffery <[email protected]> wrote:
>> >> The merging of a number of clocksource drivers into fttmr010 means we
>> >> require clock-names to be specified in the Aspeed timer node, else the
>> >> clocksource fails to probe and boot hangs.
>> >
>> > Arnd,
>> >
>> > Linus' reworked timer driver will go into 4.13.
>> >
>> > Can we get this patch merged into 4.12 as a fix so we don't end up
>> > with a broken boot at any stage?
>>
>> Hmm, can't we make the driver backward-compatible and have it fall
>> back on the first clock if no clk named "PCLK" is found? Otherwise
>> you still have an incompatible change in the DT binding and it will
>> break if someone uses an older dtb with a newer kernel.
>
> I would like to avoid to hack the kernel code for backward DT compatible
> things.
How about a fixup in the platform code to add the property then?
Arnd
On Wed, Jun 07, 2017 at 04:29:50PM +0200, Arnd Bergmann wrote:
> On Wed, Jun 7, 2017 at 2:52 PM, Daniel Lezcano
> <[email protected]> wrote:
> > On Tue, Jun 06, 2017 at 11:41:11AM +0200, Arnd Bergmann wrote:
> >> On Mon, Jun 5, 2017 at 10:59 AM, Joel Stanley <[email protected]> wrote:
> >> > On Mon, Jun 5, 2017 at 5:18 PM, Andrew Jeffery <[email protected]> wrote:
> >> >> The merging of a number of clocksource drivers into fttmr010 means we
> >> >> require clock-names to be specified in the Aspeed timer node, else the
> >> >> clocksource fails to probe and boot hangs.
> >> >
> >> > Arnd,
> >> >
> >> > Linus' reworked timer driver will go into 4.13.
> >> >
> >> > Can we get this patch merged into 4.12 as a fix so we don't end up
> >> > with a broken boot at any stage?
> >>
> >> Hmm, can't we make the driver backward-compatible and have it fall
> >> back on the first clock if no clk named "PCLK" is found? Otherwise
> >> you still have an incompatible change in the DT binding and it will
> >> break if someone uses an older dtb with a newer kernel.
> >
> > I would like to avoid to hack the kernel code for backward DT compatible
> > things.
>
> How about a fixup in the platform code to add the property then?
Dunno, do you have an example?
--
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs
Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog
On Thu, Jun 8, 2017 at 3:26 PM, Daniel Lezcano
<[email protected]> wrote:
> On Wed, Jun 07, 2017 at 04:29:50PM +0200, Arnd Bergmann wrote:
>> On Wed, Jun 7, 2017 at 2:52 PM, Daniel Lezcano
>> <[email protected]> wrote:
>> > On Tue, Jun 06, 2017 at 11:41:11AM +0200, Arnd Bergmann wrote:
>> >> On Mon, Jun 5, 2017 at 10:59 AM, Joel Stanley <[email protected]> wrote:
>> >> > On Mon, Jun 5, 2017 at 5:18 PM, Andrew Jeffery <[email protected]> wrote:
>> >> >> The merging of a number of clocksource drivers into fttmr010 means we
>> >> >> require clock-names to be specified in the Aspeed timer node, else the
>> >> >> clocksource fails to probe and boot hangs.
>> >> >
>> >> > Arnd,
>> >> >
>> >> > Linus' reworked timer driver will go into 4.13.
>> >> >
>> >> > Can we get this patch merged into 4.12 as a fix so we don't end up
>> >> > with a broken boot at any stage?
>> >>
>> >> Hmm, can't we make the driver backward-compatible and have it fall
>> >> back on the first clock if no clk named "PCLK" is found? Otherwise
>> >> you still have an incompatible change in the DT binding and it will
>> >> break if someone uses an older dtb with a newer kernel.
>> >
>> > I would like to avoid to hack the kernel code for backward DT compatible
>> > things.
>>
>> How about a fixup in the platform code to add the property then?
>
> Dunno, do you have an example?
armada_375_380_coherency_init() adds a "arm,io-coherent" property on
some machines, when the DT might be lacking this.
Arnd
Add the clock-names property in init_timer() to work-around Aspeed
devicetrees from times prior to merging the Moxart/Aspeed and Faraday
drivers.
Signed-off-by: Andrew Jeffery <[email protected]>
---
Well, here's an implementation I knocked up. It's a fair chunk of code for
marginal benefit. Joel is against it.
At least it's something to debate.
Tested under QEMU for both AST2400 and AST2500 SoCs.
Cheers,
Andrew
arch/arm/Makefile | 1 +
arch/arm/mach-aspeed/Makefile | 1 +
arch/arm/mach-aspeed/aspeed.c | 73 +++++++++++++++++++++++++++++++++++++++++++
3 files changed, 75 insertions(+)
create mode 100644 arch/arm/mach-aspeed/Makefile
create mode 100644 arch/arm/mach-aspeed/aspeed.c
diff --git a/arch/arm/Makefile b/arch/arm/Makefile
index ab30cc634d02..f3ed359e5b28 100644
--- a/arch/arm/Makefile
+++ b/arch/arm/Makefile
@@ -154,6 +154,7 @@ textofs-$(CONFIG_ARCH_AXXIA) := 0x00308000
machine-$(CONFIG_ARCH_ALPINE) += alpine
machine-$(CONFIG_ARCH_ARTPEC) += artpec
machine-$(CONFIG_ARCH_AT91) += at91
+machine-$(CONFIG_ARCH_ASPEED) += aspeed
machine-$(CONFIG_ARCH_AXXIA) += axxia
machine-$(CONFIG_ARCH_BCM) += bcm
machine-$(CONFIG_ARCH_BERLIN) += berlin
diff --git a/arch/arm/mach-aspeed/Makefile b/arch/arm/mach-aspeed/Makefile
new file mode 100644
index 000000000000..de8cd76fcf5d
--- /dev/null
+++ b/arch/arm/mach-aspeed/Makefile
@@ -0,0 +1 @@
+obj-y += aspeed.o
diff --git a/arch/arm/mach-aspeed/aspeed.c b/arch/arm/mach-aspeed/aspeed.c
new file mode 100644
index 000000000000..feaac8eb5d2d
--- /dev/null
+++ b/arch/arm/mach-aspeed/aspeed.c
@@ -0,0 +1,73 @@
+/*
+ * Copyright 2017 IBM Corporation
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+
+#include <linux/clk-provider.h>
+#include <linux/clocksource.h>
+#include <linux/init.h>
+#include <linux/of.h>
+#include <linux/slab.h>
+#include <asm/mach/arch.h>
+
+const char *aspeed_timer_compatibles[] = {
+ "aspeed,ast2400-timer",
+ "aspeed,ast2500-timer",
+ NULL,
+};
+
+/*
+ * For backwards compatibility with pre-4.13 devicetrees, populate the
+ * clock-names property in the clocksource node
+ */
+static void __init aspeed_timer_set_clock_names(void)
+{
+ const char **compatible = aspeed_timer_compatibles;
+ struct device_node *np;
+
+ while (*compatible) {
+ for_each_compatible_node(np, NULL, *compatible) {
+ struct property *clock_names;
+ int rc;
+
+ rc = of_property_count_strings(np, "clock-names");
+ if (rc != -EINVAL)
+ continue;
+
+ clock_names = kzalloc(sizeof(*clock_names), GFP_KERNEL);
+
+ clock_names->name = kstrdup("clock-names", GFP_KERNEL);
+ clock_names->length = sizeof("PCLK");
+ clock_names->value = kstrdup("PCLK", GFP_KERNEL);
+
+ of_add_property(np, clock_names);
+ }
+
+ compatible++;
+ }
+}
+
+static void __init aspeed_init_time(void)
+{
+ aspeed_timer_set_clock_names();
+
+#ifdef CONFIG_COMMON_CLK
+ of_clk_init(NULL);
+#endif
+ timer_probe();
+}
+
+static const char *const aspeed_dt_match[] __initconst = {
+ "aspeed,ast2400",
+ "aspeed,ast2500",
+ NULL,
+};
+
+DT_MACHINE_START(aspeed_dt, "Aspeed SoC")
+ .init_time = aspeed_init_time,
+ .dt_compat = aspeed_dt_match,
+MACHINE_END
--
2.11.0
On Fri, Jun 9, 2017 at 9:30 AM, Andrew Jeffery <[email protected]> wrote:
> Add the clock-names property in init_timer() to work-around Aspeed
> devicetrees from times prior to merging the Moxart/Aspeed and Faraday
> drivers.
>
> Signed-off-by: Andrew Jeffery <[email protected]>
> ---
> Well, here's an implementation I knocked up. It's a fair chunk of code for
> marginal benefit. Joel is against it.
>
> At least it's something to debate.
>
> Tested under QEMU for both AST2400 and AST2500 SoCs.
Right, I think having the one-line fixup in the driver makes way more sense
here, and that is what we usually do, but we could do this if Daniel has good
reasons to keep the driver free of backwards-compatibility support.
It also depends a bit on how common the old binding version already
is, and if anyone is shipping systems with that.
Arnd
On Fri, 2017-06-09 at 11:19 +0200, Arnd Bergmann wrote:
> > On Fri, Jun 9, 2017 at 9:30 AM, Andrew Jeffery <[email protected]> wrote:
> > Add the clock-names property in init_timer() to work-around Aspeed
> > devicetrees from times prior to merging the Moxart/Aspeed and Faraday
> > drivers.
> >
> > > > Signed-off-by: Andrew Jeffery <[email protected]>
> > ---
> > Well, here's an implementation I knocked up. It's a fair chunk of code for
> > marginal benefit. Joel is against it.
> >
> > At least it's something to debate.
> >
> > Tested under QEMU for both AST2400 and AST2500 SoCs.
>
> Right, I think having the one-line fixup in the driver makes way more sense
> here, and that is what we usually do, but we could do this if Daniel has good
> reasons to keep the driver free of backwards-compatibility support.
Yeah, choosing between the two with no further information I'd prefer
adding support to the driver.
>
> It also depends a bit on how common the old binding version already
> is, and if anyone is shipping systems with that.
So the fttmr010 bindings describe the clocks and clock-names properties
as optional (a little confusingly, "Optionally required properties").
I guess keeping in mind the bindings describe the hardware and not the
driver this might be reasonable, but the driver fails init if they're
not present. arch/arm/boot/dts/moxart.dtsi doesn't specify clock-names
either so I would have thought systems based on it would also fail.
However, Linus' fttmr010 series has Tested-by tags from Jonas, so maybe
I've missed something.
Regardless, if it's the case that Moxa systems now fail to init the
clocksource then the Aspeed-specific init_time() solution is even less
attractive. moxart.dtsi dates back to December 2013 ("448e7edefa92 ARM:
moxart: add MOXA ART SoC device tree files").
The old binding is less of a problem for Aspeed systems as we don't yet
have a clk driver upstream. Joel only recently added fixed-clock nodes
in 4.12 so Aspeed systems could boot without DTS modifications.
Andrew
On Tue, Jun 6, 2017 at 11:41 AM, Arnd Bergmann <[email protected]> wrote:
> On Mon, Jun 5, 2017 at 10:59 AM, Joel Stanley <[email protected]> wrote:
>> On Mon, Jun 5, 2017 at 5:18 PM, Andrew Jeffery <[email protected]> wrote:
>>> The merging of a number of clocksource drivers into fttmr010 means we
>>> require clock-names to be specified in the Aspeed timer node, else the
>>> clocksource fails to probe and boot hangs.
>>
>> Arnd,
>>
>> Linus' reworked timer driver will go into 4.13.
>>
>> Can we get this patch merged into 4.12 as a fix so we don't end up
>> with a broken boot at any stage?
>
> Hmm, can't we make the driver backward-compatible and have it fall
> back on the first clock if no clk named "PCLK" is found? Otherwise
> you still have an incompatible change in the DT binding and it will
> break if someone uses an older dtb with a newer kernel.
I was under the impression that the Aspeed support is still in its
infancy and everyone working on it using DT will recompile and
bundle the DTB as part of their kernel build at this point in time.
Are there people shipping products with DTBs on this, really?
Yours,
Linus Walleij
On Fri, Jun 9, 2017 at 2:52 PM, Andrew Jeffery <[email protected]> wrote:
> So the fttmr010 bindings describe the clocks and clock-names properties
> as optional (a little confusingly, "Optionally required properties").
We should remove that. The timer frequency is strictly required.
> I guess keeping in mind the bindings describe the hardware and not the
> driver this might be reasonable, but the driver fails init if they're
> not present. arch/arm/boot/dts/moxart.dtsi doesn't specify clock-names
> either so I would have thought systems based on it would also fail.
It was added in
commit f46b563f2f270e451b2e1cee78573508cc1de256
"ARM: dts: augment Moxa and Aspeed DTS for FTTMR010"
Moxart only uses DT for boot, and Jonas controls all deployments of the
mainline kernel.
> Regardless, if it's the case that Moxa systems now fail to init the
> clocksource then the Aspeed-specific init_time() solution is even less
> attractive. moxart.dtsi dates back to December 2013 ("448e7edefa92 ARM:
> moxart: add MOXA ART SoC device tree files").
Moxart is deploying the DTBs with the kernel, they go hand-in-hand.
Backward compatibility with old DTBs here would just be an academic
exercise.
> The old binding is less of a problem for Aspeed systems as we don't yet
> have a clk driver upstream. Joel only recently added fixed-clock nodes
> in 4.12 so Aspeed systems could boot without DTS modifications.
The mentioned commit also adds the clocks to Aspeed AFAICT, is there
any problem in the real world, like did I miss some Aspeed platform?
Patching around drivers to make old DTBs work is something we should
only do when there are wide deployments for common users, such as people
buying products with a DTB inside them. Until such devices are shipped,
I consider the DT bindings still in flux.
Yours,
Linus Walleij
On 20 June 2017 at 10:06, Linus Walleij <[email protected]> wrote:
> Moxart is deploying the DTBs with the kernel, they go hand-in-hand.
> Backward compatibility with old DTBs here would just be an academic
> exercise.
I've used CONFIG_ARM_APPENDED_DTB=y since the beginning mainly because
UC-7112-LX's proprietary bootloader lacks DTB capabilities (that I'm
aware of).
Compiling both (kernel and dtb) at the same time always made sense to
me, and it does, even when the hardware has a sane bootloader (i.e.
DTB enabled).
Jonas