On 08/13/2014 10:51 AM, Olof Johansson wrote:

On Wed, Aug 13, 2014 at 1:53 AM, Thierry Reding
[off-list ref] wrote:

quoted

On Tue, Aug 12, 2014 at 06:56:07PM -0700, Dylan Reid wrote:

quoted

The Acer Chromebook 13, codenamed "Big", contains an NVIDIA tegra124
processor and is similar to the Venice2 reference platform.

The keyboard, USB 2, audio, HDMI, sdcard and emmc have been tested
and work on the 1366x768 models.  I haven't tried on the HD systems
yet.

Presumably the HD systems will have a different compatible? Is it only
the panel that's different or are there other changes?

quoted

There is also an issue on reboot because the TPM isn't reset.  It will
cause the stock firmware to enter recovery mode.  This can be worked
around by an EC-reset, press refresh and power at the same time.

It's not clear to me what exactly this implies. Does it mean the system
won't boot unless you manually press "refresh" and "power"? Are those
keyboard buttons or keys on some debug board?

quoted

diff --git a/arch/arm/boot/dts/tegra124-big.dts b/arch/arm/boot/dts/tegra124-big.dts
new file mode 100644
index 0000000..aa0cd9f
--- /dev/null
+++ b/arch/arm/boot/dts/tegra124-big.dts

@@ -0,0 +1,1137 @@
+/dts-v1/;
+
+#include <dt-bindings/input/input.h>
+#include "tegra124.dtsi"
+
+/ {
+     model = "Google Big";

Perhaps this should at least mention that it's the Acer Chromebook 13.
Otherwise this could just as well be some random development board.

quoted

+     compatible = "google,nyan-big", "nvidia,tegra124";

What's "nyan"? Are there other variants of "big" other than "nyan"?

quoted

+     pinmux: pinmux at 0,70000868 {
+             pinctrl-names = "default";
+             pinctrl-0 = <&pinmux_default>;
+
+             pinmux_default: common {
+                     dap_mclk1_pw4 {
+                             nvidia,pins = "dap_mclk1_pw4";
+                             nvidia,function = "extperiph1";
+                             nvidia,enable-input = <TEGRA_PIN_DISABLE>;
+                             nvidia,pull = <TEGRA_PIN_PULL_NONE>;
+                             nvidia,tristate = <TEGRA_PIN_DISABLE>;
+                     };

Stephen proposed a patch not so long ago that removes the static pinmux
settings from DTS files[0]. I'd expect he'll have objections to this DTS
providing the static settings.

The assumption with that patch is that the bootloader will have to set
up the entire pinmux anyway (to prevent conflicting settings), therefore
there'd be no need to do it in the kernel. Is this something that your
bootloader already does (or could be modified to do)?

It's extra work to switch it over since the assumption is brand new.
It's something we can consider for future platforms but Stephen
dropped that patch like a bomb from a blue sky _after_ the final
versions of firmware had been cut -- or more importantly, tested. So,
we should have this in the kernel at this time.

We've always been developing the linux kernel on embedded platforms
with the mindset that firmware will usually get things wrong at some
point, so the safe thing is to not make assumptions about how it has
configured the hardware and do most of it from scratch in the kernel.
It's how we have handled all tegra pinmux up until now, and doing a
180-degree turn on this now, when there has been no _new_ concerns
that caused this, seems like a complication that is just throwing
everybody in a loop.

The same concerns about conflicting pinmux settings surely exists on
any platform that does muxing. Tegra isn't unique in this way, so I
wonder what is causing the need to remove this functionality from the
kernel?

It depends on the HW design, so it's quite possible HW-specific, or only 
affects a subset of HW with pinmux.

It's quite possible for pinmux HW to allow each pin/pad/... to support 1 
of N signals, yet only allow a particular signal to be routed to 1 
particular pin/pad. For such HW, there can never be conflicts, and hence 
there's no issue.

It's also possible that pinmux HW that allows a particular signal to be 
routed to various different pins/pads is designed in such a way that 
conflicts aren't possible, or aren't an issue. The registers could be 
designed to invert their sense, so that each signal gets a register 
saying where it's routed. Alternatively, the HW could have specific 
interlocks implemented so that if conflicts are programmed into the 
registers, the HW resolves them based on some priority, so that even if 
you don't get what you asked for, at least you get something 
well-defined. Admittedly the HW cost associated with either of these 
options is low.

I think the only way to move forward on not considering firmware buggy 
is to enforce it not to be, and to do that we have to draw a line in the 
sand at some point. I strongly want to make that point Tegra124 and 
later for upstream support.

Finally note while the patch in question was new to upstream and may 
seem like a bomb there, it's *absolutely* not something new in other 
contexts. However, as you say we can discuss this elsewhere.