Am Sonntag, 26. März 2017, 15:28:44 CEST schrieb Simon Glass:
Hi Heiko,
On 26 March 2017 at 15:00, Heiko Stuebner heiko@sntech.de wrote:
Am Sonntag, 26. März 2017, 22:52:16 CEST schrieb Heiko Stuebner:
Am Sonntag, 26. März 2017, 22:41:35 CEST schrieb Heiko Stuebner:
Am Sonntag, 26. März 2017, 22:13:08 CEST schrieb Heiko Stuebner:
Am Sonntag, 26. März 2017, 14:00:51 CEST schrieb Simon Glass:
Hi Heiko,
On 26 March 2017 at 13:59, Simon Glass sjg@chromium.org wrote: > Hi Heiko, > > On 26 March 2017 at 13:06, Heiko Stuebner heiko@sntech.de wrote: >> Hi Simon, >> >> Am Samstag, 25. März 2017, 20:39:08 CEST schrieb Simon Glass: >>> On 23 March 2017 at 17:41, Heiko Stuebner heiko@sntech.de wrote: >>> > The Rock is a RK3188 based single board computer by Radxa. >>> > Currently it still relies on the proprietary DDR init and >>> > cannot use the generic SPL, but at least is able to boot >>> > a linux kernel and system up to a regular login prompt. >>> > >>> > Signed-off-by: Heiko Stuebner heiko@sntech.de >>> > Reviewed-by: Simon Glass sjg@chromium.org >>> > Tested-by: Kever Yang kever.yang@rock-chips.com >>> > --- >>> > arch/arm/dts/Makefile | 1 + >>> > arch/arm/dts/rk3188-radxarock.dts | 382 ++++++++++++++++++++++++++++++++++ >>> > arch/arm/mach-rockchip/rk3188/Kconfig | 11 + >>> > board/radxa/rock/Kconfig | 15 ++ >>> > board/radxa/rock/MAINTAINERS | 6 + >>> > board/radxa/rock/Makefile | 7 + >>> > board/radxa/rock/rock.c | 7 + >>> > configs/rock_defconfig | 58 ++++++ >>> > include/configs/rock.h | 30 +++ >>> > 9 files changed, 517 insertions(+) >>> > create mode 100644 arch/arm/dts/rk3188-radxarock.dts >>> > create mode 100644 board/radxa/rock/Kconfig >>> > create mode 100644 board/radxa/rock/MAINTAINERS >>> > create mode 100644 board/radxa/rock/Makefile >>> > create mode 100644 board/radxa/rock/rock.c >>> > create mode 100644 configs/rock_defconfig >>> > create mode 100644 include/configs/rock.h >>> >>> I am still having trouble applying this patch. I get build errors: >>> >>> arm: + rock >>> +arch/arm/Makefile:22: CONFIG_CPU_V7 -march=armv7-a >>> +make[2]: *** No rule to make target 'dts/dt.dtb', needed by >>> 'tpl/u-boot-tpl.dtb'. Stop. >>> +make[1]: *** [tpl/u-boot-tpl.bin] Error 2 >>> +make: *** [sub-make] Error 2 >>> 0 0 1 /1 rock >>> >>> Also there seems to be a duplicate config: >>> >>> arm: + rock >>> +In file included from include/configs/rock.h:11:0, >>> + from include/config.h:5, >>> + from include/common.h:21, >>> + from arch/arm/lib/asm-offsets.c:15: >>> + #define CONFIG_SYS_THUMB_BUILD >>> + ^ >>> + from lib/asm-offsets.c:15: >>> +In file included from include/linux/kconfig.h:4:0, >>> + from <command-line>:0: >>> +include/generated/autoconf.h:10:0: note: this is the location of the >>> previous definition >>> + #define CONFIG_SYS_THUMB_BUILD 1 >> >> looks like this got run over by another Kconfig migration on march-18. >> New patches (migration + rock board) coming up shortly. > > Thanks - what toolchain are you using to test this?
Also I am still getting this error:
buildman rock$ boards.cfg is up to date. Nothing to do. Building current source for 1 boards (1 thread, 8 jobs per thread) arm: + rock +make[2]: *** No rule to make target 'dts/dt.dtb', needed by 'tpl/u-boot-tpl.dtb'. Stop. +make[1]: *** [tpl/u-boot-tpl.bin] Error 2 +make: *** [sub-make] Error 2 0 0 1 /1 rock
that is really strange.
I'm building with the armhf cross-compiler from Debian testing, which is
arm-linux-gnueabihf-gcc (Debian 6.3.0-5) 6.3.0 20170124
My git history also is up-to-date it seems: 14ef0b180b rockchip: rk3188: Add Radxa Rock board d0348986cc rockchip: rk3188: follow THUMB_BUILD Kconfig migration 3bffe88d68 rockchip: video: Split out HDMI controller code a188a5a35c rockchip: i2c: Add compatibles for Rockchip Cortex-A9 socs 903fae5666 rockchip: rk3188: Setup the armclk in spl 7957cc4bd0 rockchip: clk: rk3188: Allow configuration of the armclk
with 3bffe88d68 being your current head and my build commands being
make ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf- clean make ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf- rock_defconfig make ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
that also works with a "make mrproper" before everything else.
Strangely the tpl shouldn't require a dtb at all, as it needs to use OF_PLATDATA. Am I missing some config option somewhere?
I also did play around a bit with buildman just now:
At first I forgot the "$" after rock, so build everything with rock in the name and the radxarock failed because the somewhat old 4.9 toolchain used seems to produce a TPL image that is over the size constraints. I.e. 1020 bytes on 4.9 vs. 792 bytes on 6.3.0 .
When I drop the SPL_MAX_SIZE from rk3188_common.h for the TPL build tools/buildman/buildman -P rock builds 14 boards and finishes sucessfully including the radxarock.
Now when I try buildman -P rock$ I seem to also get the error about the missing dts/dt.dtb , which does not occur in my regular builds and also not when building the bigger number of boards. I guess I need to figure out what is different in that case.
sorry for spamming, but
also when entering .bm-work/rock where buildman failed and doing just my
make ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-in that directory created by buildman, produces perfectly fine images. (buildman does create the .config, so I'm just finishing the build, buildman failed at for unknown reasons)
and a tools/buildman/buildman -PVv -T1 -j1 rock$
also finishes sucessfully, so it looks like that is more some sort of build concurrency issue, with the platdata not being created in time for some reason.
Yes I think that is right. I will see if I can fix that.
Re the toolchain, if I pull in this patch then I will likely cause a build breakable on common toolchains. What do you think is the best option? Can the TPL be shrunk a little with gcc 4.9?
I've added Tom for comments, executive summary: - rk3188-tpl is size limited to 1020 bytes - gcc 6.3 produces a rk3188-tpl of 792 bytes - gcc 4.9 makes it 1020 bytes - buildman seems to always use gcc-4.9 - rk3188 board does not compile with buildman
Isn't holding on to a pretty old compiler for everything somewhat strange? ;-)
I think the recent memset change in commit c67c8c604b6c ("board_init.c: Always use memset()") did hurt us the most sizewise, as we're now using the libgeneric memset which is quite a bit bigger than the naive fallback that existed in board_init.c before.
Reverting and using the naive memset again brings us down to 952 bytes again even on gcc-4.9 .
I guess I could just remove the TPL_LIBGENERIC option again and provide a similarly small/naive memset in the rk3188 tpl board?
################
Or maybe just check if the size limit in u-boot-spl.lds is actually correct. In our example gcc-4.9 produces a 1020bytes result and the SPL_SIZE_LIMIT is set to (0x400 - 0x4) = 1020, so of course doing
diff --git a/arch/arm/cpu/u-boot-spl.lds b/arch/arm/cpu/u-boot-spl.lds index 068163b73a..c9643efd96 100644 --- a/arch/arm/cpu/u-boot-spl.lds +++ b/arch/arm/cpu/u-boot-spl.lds @@ -72,7 +72,7 @@ SECTIONS }
#if defined(CONFIG_SPL_MAX_SIZE) -ASSERT(__image_copy_end - __image_copy_start < (CONFIG_SPL_MAX_SIZE), \ +ASSERT(__image_copy_end - __image_copy_start <= (CONFIG_SPL_MAX_SIZE), \ "SPL image too big"); #endif
does look somewhat correct. But of course that rides way to much on the edge when every additional byte will break the build.
Heiko