Hi guys!
I’m able to reproduce the issue on BSP 5.2.0 and I’m currently investigating this and debugging the kernel. I’ve informed the BSP team and I’ll try to escalate this properly so it gets fixed somehow.
Hello!
Sorry for long answering. I haven’t check rebuilt kernel 5.4 and dtb-file with BSP5 yet. I used kernel 5.4 from BSP5 only (with standart dtb-file for Iris v1.1 board). Thank you for information about problem. I look forward to further news.
Hello!
Sorry for long answering. I haven’t checked rebuilt kernel 5.4 and dtb-file with BSP 5 yet. I used standart kernel and dtb-file for Iris v1.1 from BSP 5. Thank you for information about problem. I look forward to further news.
Hello! I had tried to run Linux kernel 5.4 with BSP5.2 (uart2 was disabled) - kernel hung. Next I used Linux kernel 5.4 with BSP5.0 and kernel went in reset loop. It printed message before reset:
data abort
pc : [] lr : [<1fffd06c>]
reloc pc : [<878001a0>] lr : []
sp : bdf6809c ip : 20000964 fp : 00000002
r10: bdf985d0 r9 : bdf70eb0 r8 : 00000000
r7 : bd088000 r6 : f3af4803 r5 : 4903b11b r4 : 4b03b508
r3 : 1fffd06c r2 : 00004fd4 r1 : 80800320 r0 : 1fff8040
Flags: nzcv IRQs on FIQs on Mode SVC_32
Code: e3a0d013 e169f00d e1a0e00f e1b0f00e (e24dd048)
Resetting CPU ...
resetting ...
Hi guys,
if I understand correctly, with BSP 5 it was not possible to run the M4 firmware side by side with Linux. Linux stopped while loading the kernel. Am I correct? I’m experiencing the same in my system.
Did you figure out how to fix this on BSP 5?
Thank you
Marco
I haven’t solved this yet. But this problem is actual for me.
Hi @mortolaniEST, @alex.tx , @gustavo.tx and ALL
Thread subject is not about RPMSG. Perhaps it should be dedicated thread?
I’m using RPMSG on i.MX7D (and even remoteproc with my patch) for quite long. Since my DT for custom board deviates a lot from eval-v3, I didn’t understand your problem with RPMSG. I got some free time to check with untouched BSP, kernel and DT. After a lot of hair picking, what difference could break RPMSG, … try adding fdt rm spi2 to your fdt_fixup or adding to your *.dts
&ecspi3 {
status = "disabled";
};
I’m trying with this
U-Boot commands
setenv defargs quite clk_ignore_unused
fdt addr ${fdt_addr_r} && fdt rm serial1 && fdt rm spi2
tftp rpmsg_str_echo_freertos.elf
bootaux ${fileaddr}
run ubiboot
After linux boots modprobe imx_rpmsg_tty. As a result /dev/ttyRPMSG0 appears, rpmsg_str_echo_freertos reacts as it should.
Edward
Hi
I was puzzled why M4 could interfere with ecspi3. iMX7 M4 applications from FreeRTOS examples enable SYS_CLK / 2 = 240MHz clock and make M4 clocked from this clock. In Linux this clock is called pll_sys_main_240m. eCSPI driver takes the same pll_sys_main_240m for source. Nothing wrong with it and M4 would work with eCSPI driver enabled, if eCSPI driver (from downstream kernel) wasn’t turning root lock on/off before and after each transfer… So while SPI is idle M4 is unclocked. This is how clk_summary loocks. As you see pll_sys_main_240m is disabled.
cat /sys/kernel/debug/clk/clk_summary | grep -B7 ecspi3
pll_sys_main_240m 0 0 0 240000000 0 0 50000
pll_sys_main_240m_clk 0 0 0 240000000 0 0 50000
ecspi4_src 0 0 0 240000000 0 0 50000
ecspi4_cg 0 0 0 240000000 0 0 50000
ecspi4_pre_div 0 0 0 240000000 0 0 50000
ecspi4_post_div 0 0 0 60000000 0 0 50000
ecspi4_root_clk 0 0 0 60000000 0 0 50000
ecspi3_src 0 0 0 240000000 0 0 50000
ecspi3_cg 0 0 0 240000000 0 0 50000
ecspi3_pre_div 0 0 0 240000000 0 0 50000
ecspi3_post_div 0 0 0 60000000 0 0 50000
ecspi3_root_clk 0 0 0 60000000 0 0 50000
Is is possible to enable this clock without any driver? I mean just making clock ungated and its “enable count” > 0 using some magic in device tree?
It’s interesting why some spi-imx.c drivers toggle clock on/off before after transfer (see clk_prepare_enable() and matching clk_disable() call’s ), and others don’t do it, just enable at probe and leave enabled. Looking at Toradex git, it seems that even old 4.x imx kernels were toggling ECSPI clock between transfers. Did some old kernel work with RPMSG, M4 and ecspi enabled?
Edward
Thanks for answer! But I discovered that in BSP5 image there is one more problem that described here:
https://lists.denx.de/pipermail/u-boot/2011-July/095787.html
This bug is fixed in BSP3 but not in BSP5. So nessesary insert this strings in file u-boot-initial-env:
fdt_high=0xffffffff
initrd_high=0xffffffff
After this kernel boots normally, but M4 freeze remains.
Hi @mst31,
Yes, fdt_high is often required, as well without any M4.
Did you try to boot with this?:
setenv fdt_fixup 'fdt addr ${fdt_addr_r} && fdt rm serial1 && fdt rm spi2'
As I said eCSPI3 shares the same parent clock like M4, when eCSPI3 driver exits probe() or ends SPI transfer, it disables all unused clocks including parent clocks, including as well “unused” M4 clock and hence the freeze. It is so in unmodified DT of 512MB NAND Colibri, perhaps it is bit different on eMMC flavour. Just check out clocks summary as I wrote above.
Edward
Yes, I tried this command (serial1 is disabled in my dtb-file):
setenv fdt_fixup ‘fdt addr ${fdt_addr_r} && fdt rm spi2’
saveenv
and rebooted. But spi2 started all the same. Then I did this (I have spi_imx linked as kernel module):
setenv defargs clk_ignore_unused module_blacklist=spi_imx
saveenv
and rebooted. SPI interfaces are get disabled. Now Cortex-M4 doesn’t freeze.
Oh, on some thread I saw fdt_fixup removed, perhaps from eMMC variants?
Non working suggestion here
Unless someone suggests better solution, try adding this to DT to make eCSPI3 functional
&clks {
// assigned-clocks = <&clks IMX7D_ARM_M4_ROOT_CLK>;
// assigned-clock-rates = <240000000>;
protected-clocks = <&clks IMX7D_ARM_M4_ROOT_CLK>;
};
I expected this ^^ to work without clk_ignore_unused. Indeed clk_ignore_unused is still required. You may use either protected-clocks or assigned-clockXXXX. Protected clock is better, since you may set M4 clock in M4 FW and Linux shouldn’t touch its assignment. Assigned clock seems working as well, but there should be a moment when M4 clock is disabled, reconfigured and enabled again. During that time M4 is frozen.
Regards,
Edward
1 Like
Unfortunately this doesn’t work. I had to comment clk_disable_unprepare and clk_disable calls in spi-imx.c file. Now Cortex-M4 and spi works fine together.
Please disregard
Actually I didn’t try with spi functional, only boot issue with spi left enabled in DT with some devices to probe, but spi not actually used, no devices on spi bus. But spi certainly was probed and should disable clock if you are right. Are you sure you applied modified DT? I’ll recheck it later. What I’m really sure, without above addon to DT and spi enabled M4 freezes.
Yes, I played sooner as you with mods to spi-imx.c. This is as well working, but in contrast leads to enabled count non zero in clk_summary for 240MHz. Would be nice to achieve the same without SPI driver mods, with just DT mods.
Regards
Edward
Hi @mst31 ,
Indeed it was my fault, looks like I tftp’ed wrong dtb several times, sorry
Looking in code, it seems that protected-clocks is used only with QCOM architecture, though Linux kernel documentation seems suggesting this property is a common one. argh…
So it looks like dedicated driver is required to enable something like M4 root clock… Or launching M4 only from Linux.
Edward
Had to edit file linux-toradex/drivers/clk/imx/clk-imx7d.c. I added CLK_IS_CRITICAL property to IMX7D_PLL_SYS_MAIN_240M_CLK clock. Now clock is not disabled even when SPI driver disables it. Patch for toradex_5.4-2.1.x-imx branch is attached: clk-imx7d.patch (879 Bytes)
Hi there,
Thanks to the contributors to this thread for the solution but, Toradex people, could you patch this, provide and official solution or at least notice the users that this problem exist in your documentation? I am sure that your users will be happy to save time and prefer this over digging the community for a solution… We lost one week search for bugs in our code that were not there.
Thanks in advance.
V
Hello,
We have updated this document with the latest information about how to get the demos working.
We have tested the “hello world” and the “string echo” demos with one of the BSP 5.x nightly builds, but the fixes should be already included in our latest monthly (5.5.0-devel-202112+build.16) as well.
Everyone can subscribe to updates in our articles by inserting the e-mail address in the “Subscribe for this article updates” box.
Best regards,
Rafael Beims
Thanks a lot, Rafael
@ilan_figueiredo
I need talk to you about this. I have same issue here
Where I found u