clock problems with BeagleBone Black on 12.2BETA2

[John-Mark Gurney]

Emmanuel Vadot wrote this message on Fri, Sep 25, 2020 at 09:20 +0200:
> On Thu, 24 Sep 2020 22:52:43 +0000 (UTC)
> Oskar Holmlund via freebsd-arm <freebsd-arm at freebsd.org> wrote:
> 
> > > Den torsdag 24 september 2020 21:36:38 CEST, Ed Maste <emaste at freebsd.org> skrev: 
> > >
> > > On Thu, 24 Sep 2020 at 12:31, Ian Lepore <ian at freebsd.org> wrote:
> > > >
> > > > It was mentioned on irc a few days ago when someone noticed the BBB in
> > > > the CI setup was failing...
> > > >
> > > > <emaste> hrm, looks like BBB has been broken since end of July
> > > > <emaste> panic: Duplicated clock registration: clk at 4_0
> > > > <emaste> ah, mmel r363700
> > > > <strejda> emaste: im sorry, but this is (unfortunately) expected
> > > > collateral damage. im aware of this problem but right solution is not
> > > > trivial and it needs more time.
> > > >
> > > > (strejda == mmel@)
> > > >
> > > > Then it was mentioned a while later that that BBB is using a very old
> > > > u-boot, so maybe that has something to do with it.
> > > 
> > > I've updated uboot on eMMC on the the CI system's BBB to:
> > > U-Boot SPL 2020.07 (Sep 24 2020 - 04:58:48 +0000)^M
> > > 
> > > however it's still failing the same way:
> > > 
> > > clk_fixed7: <Fixed factor clock> on ofw_clkbus0^M
> > > clk_fixed7: Cannot FDT parameters.^M
> > > device_attach: clk_fixed7 attach returned 6^M
> > > clk_fixed7: <Fixed factor clock> on ofw_clkbus0^M
> > > clk_fixed7: Cannot FDT parameters.^M
> > > device_attach: clk_fixed7 attach returned 6^M
> > > ti_clkctrl0: <TI clkctrl> mem 0x14-0x14f on ti_omap4_cm0^M
> > > ti_clkctrl1: <TI clkctrl> mem 0x4-0xd7 on ti_omap4_cm1^M
> > > ti_clkctrl2: <TI clkctrl> mem 0x4-0x7 on ti_omap4_cm2^M
> > >
> > > panic: Duplicated clock registration: clk at 4_0
> > > ^M
> > > ^M
> > > cpuid = 0^M
> > > time = 1^M
> > >
> > > phk@ reported success booting 13-CURRENT snapshots though, so I'm not
> > > really sure what's going on.
> > 
> > Well this is u-boot playing a trick on us.
> > 
> > To have a common ground lets assume we use the snapshot from
> > https://download.freebsd.org/ftp/snapshots/arm/armv7/ISO-IMAGES/13.0/
> > FreeBSD-13.0-CURRENT-arm-armv7-BEAGLEBONE-20200924-3c514403bef.img.xz
> > 
> > DD to sd-card and press the "sysboot alter"-button it will boot to login prompt.
> > 
> > Hit [Enter] to boot immediately, or any other key for command prompt.
> > Booting [/boot/kernel/kernel]...               
> > Using DTB provided by EFI at 0x87ee7000.
> > Kernel entry at 0x97000200...
> > Kernel args: (null)
> > ---<<BOOT>>---
> > < remove some 30 lines of output >
> > ofwbus0: <Open Firmware Device Tree>
> > simplebus0: <Flattened device tree simple bus> on ofwbus0
> > simplebus1: <Flattened device tree simple bus> mem 0x44c00000-0x44c007ff,0x44c00800-0x44c00fff,0x44c01000-0x44c013ff,0x44c01400-0x44c017ff on0
> > simplebus2: <Flattened device tree simple bus> on simplebus1
> > simplebus3: <Flattened device tree simple bus> on simplebus1
> > simplebus4: <Flattened device tree simple bus> on simplebus1
> > ti_sysc0: <TI SYSC Interconnect> mem 0-0x3 on simplebus4
> > ti_prcm0: <TI Power and Clock Management> mem 0-0x1fff on ti_sysc0
> > ofw_clkbus0: <OFW clocks bus> on ti_prcm0
> > 
> > Login as root and run:
> > root at generic:~ # rm -rf /boot/msdos/dtb/
> > root at generic:~ # reboot
> > 
> > 
> > Now at boot we got the same problem as Ed Maste described.
> > At this boot simplebus1 has lost its ranges and ti_prcm0 are also a little bit different.
> > ...
> > ofwbus0: <Open Firmware Device Tree>
> > simplebus0: <Flattened device tree simple bus> on ofwbus0
> > simplebus1: <Flattened device tree simple bus> on simplebus0
> > ti_prcm0: <TI Power and Clock Management> mem 0x200000-0x203fff on simplebus1
> > ofw_clkbus0: <OFW clocks bus> on ti_prcm0
> > clk_fixed0: <Fixed clock> on ofw_clkbus0
> > clk_fixed1: <Fixed clock> on ofw_clkbus0
> > clk_fixed2: <Fixed clock> on ofw_clkbus0
> > clk_fixed3: <Fixed clock> on ofw_clkbus0
> > clk_fixed4: <Fixed clock> on ofw_clkbus0
> > clk_fixed5: <Fixed clock> on ofw_clkbus0
> > ....
> > device_attach: clk_fixed7 attach returned 6
> > clk_fixed7: <Fixed factor clock> on ofw_clkbus0
> > clk_fixed7: Cannot FDT parameters.
> > device_attach: clk_fixed7 attach returned 6
> > ti_clkctrl0: <TI clkctrl> mem 0x14-0x14f on ti_omap4_cm0
> > ti_clkctrl1: <TI clkctrl> mem 0x4-0xd7 on ti_omap4_cm1
> > ti_clkctrl2: <TI clkctrl> mem 0x4-0x7 on ti_omap4_cm2
> > panic: Duplicated clock registration: clk at 4_0
> > 
> > cpuid = 0
> > time = 1
> > KDB: stack backtrace:
> > db_trace_self() at db_trace_self
> >          pc = 0xc04e25f8  lr = 0xc00704ac (db_trace_self_wrapper+0x30)
> >          sp = 0xc0d14810  fp = 0xc0d14928
> > db_trace_self_wrapper() at db_trace_self_wrapper+0x30
> > 
> > U-Boot have its own version of devicetree used in the SPL. Later in the boot process if "the real" u-boot cant find any devicetree files in its partition it uses its default. The default are from Linux 4.20:
> > https://github.com/u-boot/u-boot/blob/v2020.07/arch/arm/dts/am33xx.dtsi
> > 
> > The devicetree expected by the clock implementation are Linux 5.7 and later.
> > 
> > To fix the problem in CI build - just copy devicetrees from base into the FAT partition for uboot to find.
> > 
> > //Oskar
> 
>  Thanks for testing on your side Oskar,
> 
>  As a reminder, using u-boot embedded dtb will almost always result in
> problems.
>  At best the dtb embedded is the same one from Linux (and so the same
> one from our tree) but stripped down, nodes are missing etc ...
>  At worse it's either from an old version of Linux or a completly
> different one specialy made for u-boot.

Can we add a FreeBSD specific node, and if it's not present print a
warning?

I mean, it's less than ideal, but provides some idea.  We could even
possibly add a FreeBSD kernel version to the node to do a compare
against...

I understand the idea of unversioned dtb, but the way that Linux
changes things enough, it seems like it's not really a valid option
in the long run...

>  There is no real way for us to warn the user as dts aren't versionned
> so all we can do is blindly accept the dtb passed to loader(8) via efi
> (or LINUX_BOOT_ABI in case of bootm/booti).
> 
>  Note that we need u-boot to preload the dtb for us for a few reasons :
>  - In a EFI based env we cannot query u-boot for the board name, we
> only have the full DTB.
>  - On some boards, like BBB, the same u-boot is used for multiple
> boards (BBB, PocketBeagle etc ...) so booting in a generic way is only
> possible if u-boot load the dtb for us.
>  - On some boards u-boot modify the dtb based on the version of the
> boards or something and will enable/disable some nodes or add some new
> ones.

-- 
  John-Mark Gurney				Voice: +1 415 225 5579

     "All that I will do, has been done, All that I have, has not."