On 07/16/2018 10:40 PM, Simon Glass wrote:
Hi Stephen,
On 12 July 2018 at 15:26, Stephen Warren swarren@wwwdotorg.org wrote:
On 07/12/2018 12:17 PM, Stephen Warren wrote:
On 07/12/2018 09:52 AM, Stephen Warren wrote:
On 07/11/2018 06:12 PM, Simon Glass wrote:
Hi Stephen,
On 11 July 2018 at 16:01, Stephen Warren swarren@wwwdotorg.org wrote:
On 07/10/2018 02:24 PM, Simon Glass wrote: > > > Hi Stephen, > > On 10 July 2018 at 13:53, Stephen Warren swarren@wwwdotorg.org wrote: >> >> >> On 07/10/2018 12:47 PM, Tom Rini wrote: >>> >>> >>> >>> On Mon, Jul 09, 2018 at 01:53:43PM -0600, Simon Glass wrote: >>> >>>> Hi Tom. >>>> >>>> Here are some test-coverage and DM core enhancements. Also it adds a >>>> way to access the binman definition from U-Boot. >>>> >>>> >>>> The following changes since commit >>>> 8c5d4fd0ec222701598a27b26ab7265d4cee45a3: >>>> >>>> Prepare v2018.07 (2018-07-09 10:24:14 -0400) >>>> >>>> are available in the Git repository at: >>>> >>>> git://git.denx.de/u-boot-dm.git >>>> >>>> for you to fetch changes up to >>>> 16b8d6b76992690c65c58dc8b0591496cc5e46ef: >>>> >>>> binman: Support updating the device tree with calc'd info >>>> (2018-07-09 09:11:00 -0600) >> >> >> >> >> This pull has caused intermittent/random build errors on my Jenkins >> system. >> The log shows: >> >>> LD spl/u-boot-spl >>> OBJCOPY spl/u-boot-spl-nodtb.bin >>> COPY spl/u-boot-spl.bin >>> BINMAN u-boot-tegra.bin >>> BINMAN u-boot-nodtb-tegra.bin >>> BINMAN u-boot-dtb-tegra.bin >>> binman: pylibfdt error -9: FDT_ERR_BADMAGIC >>> >>> >>> /var/lib/jenkins/workspace/u-boot-denx_uboot-master-build/src/u-boot/Makefile:1244: >>> recipe for target 'u-boot-tegra.bin' failed >>> make[1]: *** [u-boot-tegra.bin] Error 1 >>> make[1]: *** Waiting for unfinished jobs.... >>> make[1]: Leaving directory >>> >>> '/var/lib/jenkins/workspace/u-boot-denx_uboot-master-build/build/u-boot/beaver' >>> Makefile:148: recipe for target 'sub-make' failed >>> make: *** [sub-make] Error 2 >>> make: Leaving directory >>> '/var/lib/jenkins/workspace/u-boot-denx_uboot-master-build/src/u-boot' >> >> >> >> >> This doesn't happen every time; my Jenkins system builds 25 Tegra/sandbox >> boards, yet a varying set of boards fail each time I trigger the build: >> Just >> beaver the first time, then just colibri_t20 and ventana, then just >> medcom-wide. Note that the system performs incremental builds, if that >> matters. > > > > That might be the fdt_resize() problem which David Gibson has just > sorted out upstream. If you can run binman -D (to get a stack trace) > that might help. I should be able to do a patch if that is the > problem.
Is this the backtrace you're looking for?
> [swarren@swarren-lx1 u-boot]$ > CROSS_COMPILE=/home/swarren/shared/gcc-linaro-7.2.1-2017.11-i686_arm-linux-gnueabihf/bin/arm-linux-gnueabihf- > sh -c "make O=build-beaver -s beaver_defconfig && make O=build-beaver -s > -j8" > arch/arm/dts/tegra30-apalis.dtb: Warning (avoid_unnecessary_addr_size): > /i2c@7000d000/tps65911@2d/regulators: unnecessary #address-cells/#size-cells > without "ranges" or child "reg" property > arch/arm/dts/tegra30-beaver.dtb: Warning (avoid_unnecessary_addr_size): > /i2c@7000d000/tps65911@2d/regulators: unnecessary #address-cells/#size-cells > without "ranges" or child "reg" property > binman: pylibfdt error -9: FDT_ERR_BADMAGIC > > Traceback (most recent call last): > File "../tools/binman/binman", line 120, in RunBinman > ret_code = control.Binman(options, args) > File > "/home/swarren/shared/git_wa/tegra-uboot-flasher/u-boot/tools/binman/control.py", > line 128, in Binman > dtb = fdt.FdtScan(fname) > File > "/home/swarren/shared/git_wa/tegra-uboot-flasher/u-boot/tools/binman/../dtoc/fdt.py", > line 459, in FdtScan > dtb = Fdt(fname) > File > "/home/swarren/shared/git_wa/tegra-uboot-flasher/u-boot/tools/binman/../dtoc/fdt.py", > line 315, in __init__ > self._fdt_obj = libfdt.Fdt(fd.read()) > File "scripts/dtc/pylibfdt/libfdt.py", line 207, in __init__ > check_err(fdt_check_header(self._fdt)); > File "scripts/dtc/pylibfdt/libfdt.py", line 160, in check_err > raise FdtException(val) > FdtException: pylibfdt error -9: FDT_ERR_BADMAGIC > /home/swarren/shared/git_wa/tegra-uboot-flasher/u-boot/Makefile:1244: > recipe for target 'u-boot-tegra.bin' failed > make[1]: *** [u-boot-tegra.bin] Error 1 > make[1]: *** Waiting for unfinished jobs....
Thanks for that. But actually that is not something I have seen or can explain. Here it is reading the DT at the start and somehow failing in the check. That DT is created by the U-Boot build system. Is it possible that your system has its own libfdt installed?
As it happened, the pylibfdt changes have been applied upstream today, so I'll prepare a patch to sync U-Boot up with that. Hopefully that will resolve any issues, but I am not sure.
I have the standard Ubuntu 16.04 packages installed, but no non-standard local builds:
device-tree-compiler 1.4.0+dfsg-2 libfdt1:amd64 1.4.0+dfsg-2
I suspect this is a missing dependency in the makefiles, or perhaps some process isn't waiting for its child to exit. The code that's finding the bad FDT magic appears to be reading from a file that hasn't yet been written yet:
[pid 26251] execve("../tools/binman/binman", ["../tools/binman/binman", "-d", "u-boot.dtb", "-O", ".", "-I", ".", "-I", "../board/nvidia/beaver", "spl/u-boot-spl"], [/* 158 vars */] <unfinished ...>
[pid 26255] execve("../tools/binman/binman", ["../tools/binman/binman", "-d", "u-boot.dtb", "-O", ".", "-I", ".", "-I", "../board/nvidia/beaver", "spl/u-boot-spl"], [/* 158 vars */] <unfinished ...>
[pid 26255] open("./u-boot-out.dtb", O_WRONLY|O_CREAT|O_TRUNC, 0666 <unfinished ...> [pid 26251] <... close resumed> ) = 0 [pid 26251] close(3) = 0 [pid 26251] open("./u-boot-out.dtb", O_RDONLY) = 3 [pid 26251] fstat(3, {st_mode=S_IFREG|0600, st_size=0, ...}) = 0 [pid 26251] fstat(3, {st_mode=S_IFREG|0600, st_size=0, ...}) = 0 [pid 26251] lseek(3, 0, SEEK_CUR) = 0 [pid 26251] lseek(3, 0, SEEK_CUR) = 0 [pid 26251] fstat(3, {st_mode=S_IFREG|0600, st_size=0, ...}) = 0 [pid 26251] read(3, "", 4096) = 0 [pid 26251] close(3) = 0 [pid 26251] fstat(1, {st_mode=S_IFIFO|0600, st_size=0, ...}) = 0 [pid 26251] write(1, "binman: pylibfdt error -9: FDT_E"..., 44binman: pylibfdt error -9: FDT_ERR_BADMAGIC
[pid 26255] write(3, "\320\r\376\355\0\0We\0\0\0008\0\0Q\340\0\0\0(\0\0\0\21\0\0\0\20\0\0\0\0"..., 20480 <unfinished ...>
Specifically, this Makefile rule:
u-boot-dtb-tegra.bin u-boot-tegra.bin u-boot-nodtb-tegra.bin: \ spl/u-boot-spl u-boot.bin FORCE $(call if_changed,binman)
runs that same binman invocation for all 3 target binaries separately and in parallel, and tools/binman/control.py:Binman() uses the same hard-coded filename for an output (or temporary?) file in each invocation. I'm not sure if we can simply replace:
fname = tools.GetOutputFilename('u-boot-out.dtb')with a call to tempfile.mkstemp() or similar?
Thanks very much for digging into this. I actually think that rule is wrong and needs to be fixed. I will take a look.
Irrespective of whether the Makefile rule is wrong, and whether we change it at all, I still think we should apply the patch I sent for binman ("binman: ensure temp filenames don't collide") to solve this, since that's a legitimate fix irrespective of the Makefile rule.