Skip to contentSkip to conversation

Issue with Clock Parent Assignment on AM335x SoC (Watchdog Timer)

PATCH Mathieu Dubois-Briand 3 participants 5 messages Apr 28 – May 1 Common clk

lore
Share & tools
lore
#0 lore raw

Hi,

I believe I’ve identified an issue with clock management on the AM335x SoC and possibly other TI SoC. It can be seen while using the SoC watchdog timer, but I suspect other clocks or modules might be impacted.

While I partially understand the issue, I struggle to find a correct solution. I will describe my findings below.

First, here is a quick way to reproduce it.
- Hardware: Beagle Bone Black.
- Linux kernel version: v7.1-rc1 (or anything after 5.19).
- Configuration: omap2plus_defconfig + CONFIG_OMAP_WATCHDOG=y
- In the bootloader, I select the 32KHz clock derived from the 24MHz one as the watchdog clock source. This is controlled by bit 1 of CLKSEL_WDT1_CLK.
This is probably the point that differs from most use cases, and why it hasn't been observed before. On my final setup, I do not control the bootloader and so cannot change this value.
In u-boot, this is: mw.l 0x44e00538 1
- Once the kernel is booted, I open the watchdog device node. I'm using busybox "watchdog" command, but a simple "cat" will trigger the same error.
watchdog /dev/watchdog0
- This will result in the bus clock failing to be enabled and a failure to access the watchdog module:

  [   11.677752] l4-wkup-clkctrl:00d4:0: failed to enable
  [   11.686079] Unhandled fault: external abort on non-linefetch (0x1028) at 0xf9e35034
  I'm attaching the full trace at the very end of my email.

So, I tried to dig a bit into this issue and the reason why the l4-wkup-clkctrl:00d4:0 clock can't be enabled.

The parent clock of l4-wkup-clkctrl:00d4:0 is wdt1_fck, and the "struct clk_core" [1] representing this wdt1_fck looks incorrect:
- The "parents" list of all possible sources is set correctly. It contains pointers on both "clk_rc32k_ck" and "clk_24mhz_clkctrl" structures.
- The "parent" pointer is NULL, while it should point onto the "clk_24mhz_clkctrl" clk_core structure.

I believe this is because the value fails to be initialized correctly during boot. This should be set in clk_core_set_parent_nolock() [2], but fails as the call to clk_fetch_parent_index() [3] is failing.

The reason it fails is easy to explain:
- The clk_fetch_parent_index() function is used to find which entry in core->parents[] corresponds to the clock we try to assign as a parent.
- The name of the desired parent, as set in the "struct clk_core" is set to "clk-24mhz-clkctrl:0000:0".
- The expected name of the clock, as specified in core->parents[1] "struct clk_parent_map" is just "clk_24mhz_clkctrl".
- As a consequence, the name comparison [4] fails, and the clock choice is deemed invalid.

Now, where does these names come from? I believe it's partly from the device tree, partly from drivers/clk/ti/clkctrl.c and drivers/clk/ti/clk.c. I tried to play a bit with this code, but I'm a bit lost with the logic here.

Yet, I confirm I can get a working configuration with this horrible hack:

drivers/clk/clk.c +5 -0
@@ -2026,6 +2026,11 @@ static int clk_fetch_parent_index(struct clk_core *core,
                if (core->parents[i].name &&
                    !strcmp(parent->name, core->parents[i].name))
                        break;
+               if (core->parents[i].name &&
+                   !strcmp("clk_24mhz_clkctrl", core->parents[i].name) &&
+                   !strcmp("clk-24mhz-clkctrl:0000:0", parent->name)) {
+                       break;
+               }
        }
 

I can trace the issue back to commit ec7aa25fa483 [5] introduced in v5.19. Yet today, reverting this commit is not sufficient, as I believe the code did evolve significantly.

Does anyone have suggestions for fixing this issue? If any part of my analysis seems incorrect, please let me know.

[1]: https://elixir.bootlin.com/linux/v7.0.1/source/drivers/clk/clk.c#L66
[2]: https://elixir.bootlin.com/linux/v7.0.1/source/drivers/clk/clk.c#L2883
[3]: https://elixir.bootlin.com/linux/v7.0.1/source/drivers/clk/clk.c#L2004
[4]: https://elixir.bootlin.com/linux/v7.0.1/source/drivers/clk/clk.c#L2035
[5]: https://web.git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=ec7aa25fa483

Thanks,
Mathieu

The full log follows.

-------------- >8 -------------

# watchdog /dev/watchdog0
[   11.677752] l4-wkup-clkctrl:00d4:0: failed to enable
[   11.682999] 8<--- cut here ---
[   11.686079] Unhandled fault: external abort on non-linefetch (0x1028) at 0xf9e35034
[   11.693778] [f9e35034] *pgd=44e11452(bad)
[   11.697834] Internal error: : 1028 [#1] SMP ARM
[   11.702394] Modules linked in:
[   11.705474] CPU: 0 UID: 0 PID: 130 Comm: watchdog Not tainted 7.1.0-rc1-00001-g700d54ce1b97-dirty #2 VOLUNTARY 
[   11.715619] Hardware name: Generic AM33XX (Flattened Device Tree)
[   11.721742] PC is at omap_wdt_disable+0x14/0xb8
[   11.726327] LR is at omap_wdt_start+0x34/0xf8
[   11.730714] pc : [<c0944e08>]    lr : [<c0945068>]    psr: 600f0013
[   11.737010] sp : e0305d58  ip : f9e35048  fp : c1aaea80
[   11.742260] r10: c216b220  r9 : c1ec9ba0  r8 : c1f60c00
[   11.747509] r7 : b7cab00b  r6 : c1ec9bbc  r5 : f9e35000  r4 : c1ec9b40
[   11.754068] r3 : f9e35034  r2 : 00000000  r1 : 600f0013  r0 : f9e35000
[   11.760627] Flags: nZCv  IRQs on  FIQs on  Mode SVC_32  ISA ARM  Segment none
[   11.767800] Control: 10c5387d  Table: 82e68019  DAC: 00000051
[   11.773573] Register r0 information: 0-page vmalloc region starting at 0xf9c00000 allocated at iotable_init+0x0/0xc0
[   11.784171] Register r1 information: non-paged memory
[   11.789254] Register r2 information: NULL pointer
[   11.793986] Register r3 information: 0-page vmalloc region starting at 0xf9c00000 allocated at iotable_init+0x0/0xc0
[   11.804570] Register r4 information: slab kmalloc-256 start c1ec9b00 pointer offset 64 size 256
[   11.813342] Register r5 information: 0-page vmalloc region starting at 0xf9c00000 allocated at iotable_init+0x0/0xc0
[   11.823927] Register r6 information: slab kmalloc-256 start c1ec9b00 pointer offset 188 size 256
[   11.832781] Register r7 information: non-paged memory
[   11.837861] Register r8 information: slab kmalloc-1k start c1f60c00 pointer offset 0 size 1024
[   11.846539] Register r9 information: slab kmalloc-256 start c1ec9b00 pointer offset 160 size 256
[   11.855391] Register r10 information: slab dentry start c216b220 pointer offset 0 size 136
[   11.863721] Register r11 information: slab filp start c1aaea80 pointer offset 0 size 128
[   11.871876] Register r12 information: 0-page vmalloc region starting at 0xf9c00000 allocated at iotable_init+0x0/0xc0
[   11.882548] Process watchdog (pid: 130, stack limit = 0x(ptrval))
[   11.888674] Stack: (0xe0305d58 to 0xe0306000)
[   11.893055] 5d40:                                                       c1ec9b40 c1e44930
[   11.901275] 5d60: 00000002 c0943de8 0f900000 00000000 c18a5000 c1f60df0 c1ba7580 c1f60c00
[   11.909496] 5d80: c1e44930 c1aaea80 c1ec9b40 c1f60eac 00000000 c0943f4c c1f60df0 c1e44930
[   11.917715] 5da0: c0d90324 00000000 c1aaea80 c0370ba8 00000000 73a085ae e0305e78 c1aaea80
[   11.925935] 5dc0: 00000000 c1e44930 c0370ab8 c1aaeaa8 e0305e78 c0365070 c1aaea80 c1286208
[   11.934154] 5de0: c1aaeaa8 00000000 e0305f30 c0366ba4 e0305f30 c037986c e0305f30 c037a844
[   11.942374] 5e00: 00020001 c1286208 00000000 c0380710 e0305e68 b6e56000 b6fb7fff c1bc880c
[   11.950594] 5e20: 00000000 00000101 00000000 00000000 00000000 82e4a831 c2e79880 00100cca
[   11.958814] 5e40: 00000150 b6fa6000 b6fa6000 73a085ae c2e6adb8 c195bc00 e0305f30 ffffff9c
[   11.967034] 5e60: 00000001 c01002ec c1fe1500 00000142 005a6874 c0381170 c18f6f10 c216b220
[   11.975253] 5e80: 4a3c55ab 00000009 c195bc11 00000000 00000000 c21a2b28 c1e44930 00010001
[   11.983472] 5ea0: 00000000 00000000 00000000 00000098 00000002 00000000 00000000 00000000
[   11.991692] 5ec0: e0305ec4 00000000 00000000 00000000 00000000 00000000 00000000 00000000
[   11.999911] 5ee0: 00000000 00000000 00000000 00000000 00000000 c195bc00 c195bc0c 00000000
[   12.008131] 5f00: 00000002 ffffff9c 00000000 000041ed e0305f60 73a085ae ffffff9c 00000003
[   12.016351] 5f20: e0305f60 c195bc00 ffffff9c c0367010 00020001 00000000 00000002 00010000
[   12.024571] 5f40: 00000001 73a085ae bed20ab4 ffffff9c bed20f58 005a79bc 00000142 c03673e8
[   12.032790] 5f60: 00020001 00000000 00000000 00000000 00000000 00000000 00020001 00000000
[   12.041009] 5f80: 00000000 00000000 00000000 00000000 bed20b38 73a085ae bed20be4 00000000
[   12.049229] 5fa0: 00000000 c0100080 00000000 00000000 ffffff9c bed20f58 00020001 00000000
[   12.057449] 5fc0: 00000000 00000000 005a79bc 00000142 004dfc60 b6ffccf8 00000000 005a6874
[   12.065668] 5fe0: 00020001 bed20c28 00000001 b6ed2458 200f0010 ffffff9c 00000000 00000000
[   12.073883] Call trace: 
[   12.073895]  omap_wdt_disable from omap_wdt_start+0x34/0xf8
[   12.082052]  omap_wdt_start from watchdog_start+0x8c/0x184
[   12.087579]  watchdog_start from watchdog_open+0x6c/0xf0
[   12.092930]  watchdog_open from chrdev_open+0xf0/0x220
[   12.098113]  chrdev_open from do_dentry_open+0x120/0x4f8
[   12.103465]  do_dentry_open from vfs_open+0x24/0xe4
[   12.108378]  vfs_open from path_openat+0x828/0x1224
[   12.113293]  path_openat from do_file_open+0x64/0xd8
[   12.118290]  do_file_open from do_sys_openat2+0x80/0xe8
[   12.123550]  do_sys_openat2 from sys_openat+0x7c/0xd0
[   12.128637]  sys_openat from ret_fast_syscall+0x0/0x58
[   12.133810] Exception stack(0xe0305fa8 to 0xe0305ff0)
[   12.138891] 5fa0:                   00000000 00000000 ffffff9c bed20f58 00020001 00000000
[   12.147111] 5fc0: 00000000 00000000 005a79bc 00000142 004dfc60 b6ffccf8 00000000 005a6874
[   12.155329] 5fe0: 00020001 bed20c28 00000001 b6ed2458
[   12.160412] Code: e59f30a4 e280c048 e58c3000 e2803034 (e5902034) 
[   12.166540] ---[ end trace 0000000000000000 ]---
[   12.171183] note: watchdog[130] exited with irqs disabled
[   13.282448] omap_prm 44e00d00.prm: sync_state() pending due to 44d00000.cpu
Signature · 3 lines
Mathieu Dubois-Briand, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com
maintainer
#1 lore raw

+Tony (author of commit)

Quoting Mathieu Dubois-Briand (2026-04-28 05:03:16)

Quoted 174 lines

Hi,

I believe I’ve identified an issue with clock management on the AM335x SoC and possibly other TI SoC. It can be seen while using the SoC watchdog timer, but I suspect other clocks or modules might be impacted.

While I partially understand the issue, I struggle to find a correct solution. I will describe my findings below.

First, here is a quick way to reproduce it.
- Hardware: Beagle Bone Black.
- Linux kernel version: v7.1-rc1 (or anything after 5.19).
- Configuration: omap2plus_defconfig + CONFIG_OMAP_WATCHDOG=y
- In the bootloader, I select the 32KHz clock derived from the 24MHz one as the watchdog clock source. This is controlled by bit 1 of CLKSEL_WDT1_CLK.
This is probably the point that differs from most use cases, and why it hasn't been observed before. On my final setup, I do not control the bootloader and so cannot change this value.
In u-boot, this is: mw.l 0x44e00538 1
- Once the kernel is booted, I open the watchdog device node. I'm using busybox "watchdog" command, but a simple "cat" will trigger the same error.
watchdog /dev/watchdog0
- This will result in the bus clock failing to be enabled and a failure to access the watchdog module:

  [   11.677752] l4-wkup-clkctrl:00d4:0: failed to enable
  [   11.686079] Unhandled fault: external abort on non-linefetch (0x1028) at 0xf9e35034
  I'm attaching the full trace at the very end of my email.

So, I tried to dig a bit into this issue and the reason why the l4-wkup-clkctrl:00d4:0 clock can't be enabled.

The parent clock of l4-wkup-clkctrl:00d4:0 is wdt1_fck, and the "struct clk_core" [1] representing this wdt1_fck looks incorrect:
- The "parents" list of all possible sources is set correctly. It contains pointers on both "clk_rc32k_ck" and "clk_24mhz_clkctrl" structures.
- The "parent" pointer is NULL, while it should point onto the "clk_24mhz_clkctrl" clk_core structure.

I believe this is because the value fails to be initialized correctly during boot. This should be set in clk_core_set_parent_nolock() [2], but fails as the call to clk_fetch_parent_index() [3] is failing.

The reason it fails is easy to explain:
- The clk_fetch_parent_index() function is used to find which entry in core->parents[] corresponds to the clock we try to assign as a parent.
- The name of the desired parent, as set in the "struct clk_core" is set to "clk-24mhz-clkctrl:0000:0".
- The expected name of the clock, as specified in core->parents[1] "struct clk_parent_map" is just "clk_24mhz_clkctrl".
- As a consequence, the name comparison [4] fails, and the clock choice is deemed invalid.

Now, where does these names come from? I believe it's partly from the device tree, partly from drivers/clk/ti/clkctrl.c and drivers/clk/ti/clk.c. I tried to play a bit with this code, but I'm a bit lost with the logic here.

Yet, I confirm I can get a working configuration with this horrible hack:

drivers/clk/clk.c +5 -0
@@ -2026,6 +2026,11 @@ static int clk_fetch_parent_index(struct clk_core *core,
                if (core->parents[i].name &&
                    !strcmp(parent->name, core->parents[i].name))
                        break;
+               if (core->parents[i].name &&
+                   !strcmp("clk_24mhz_clkctrl", core->parents[i].name) &&
+                   !strcmp("clk-24mhz-clkctrl:0000:0", parent->name)) {
+                       break;
+               }
        }
 

I can trace the issue back to commit ec7aa25fa483 [5] introduced in v5.19. Yet today, reverting this commit is not sufficient, as I believe the code did evolve significantly.

Does anyone have suggestions for fixing this issue? If any part of my analysis seems incorrect, please let me know.

[1]: https://elixir.bootlin.com/linux/v7.0.1/source/drivers/clk/clk.c#L66
[2]: https://elixir.bootlin.com/linux/v7.0.1/source/drivers/clk/clk.c#L2883
[3]: https://elixir.bootlin.com/linux/v7.0.1/source/drivers/clk/clk.c#L2004
[4]: https://elixir.bootlin.com/linux/v7.0.1/source/drivers/clk/clk.c#L2035
[5]: https://web.git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=ec7aa25fa483

Thanks,
Mathieu

The full log follows.

-------------- >8 -------------

# watchdog /dev/watchdog0
[   11.677752] l4-wkup-clkctrl:00d4:0: failed to enable
[   11.682999] 8<--- cut here ---
[   11.686079] Unhandled fault: external abort on non-linefetch (0x1028) at 0xf9e35034
[   11.693778] [f9e35034] *pgd=44e11452(bad)
[   11.697834] Internal error: : 1028 [#1] SMP ARM
[   11.702394] Modules linked in:
[   11.705474] CPU: 0 UID: 0 PID: 130 Comm: watchdog Not tainted 7.1.0-rc1-00001-g700d54ce1b97-dirty #2 VOLUNTARY 
[   11.715619] Hardware name: Generic AM33XX (Flattened Device Tree)
[   11.721742] PC is at omap_wdt_disable+0x14/0xb8
[   11.726327] LR is at omap_wdt_start+0x34/0xf8
[   11.730714] pc : [<c0944e08>]    lr : [<c0945068>]    psr: 600f0013
[   11.737010] sp : e0305d58  ip : f9e35048  fp : c1aaea80
[   11.742260] r10: c216b220  r9 : c1ec9ba0  r8 : c1f60c00
[   11.747509] r7 : b7cab00b  r6 : c1ec9bbc  r5 : f9e35000  r4 : c1ec9b40
[   11.754068] r3 : f9e35034  r2 : 00000000  r1 : 600f0013  r0 : f9e35000
[   11.760627] Flags: nZCv  IRQs on  FIQs on  Mode SVC_32  ISA ARM  Segment none
[   11.767800] Control: 10c5387d  Table: 82e68019  DAC: 00000051
[   11.773573] Register r0 information: 0-page vmalloc region starting at 0xf9c00000 allocated at iotable_init+0x0/0xc0
[   11.784171] Register r1 information: non-paged memory
[   11.789254] Register r2 information: NULL pointer
[   11.793986] Register r3 information: 0-page vmalloc region starting at 0xf9c00000 allocated at iotable_init+0x0/0xc0
[   11.804570] Register r4 information: slab kmalloc-256 start c1ec9b00 pointer offset 64 size 256
[   11.813342] Register r5 information: 0-page vmalloc region starting at 0xf9c00000 allocated at iotable_init+0x0/0xc0
[   11.823927] Register r6 information: slab kmalloc-256 start c1ec9b00 pointer offset 188 size 256
[   11.832781] Register r7 information: non-paged memory
[   11.837861] Register r8 information: slab kmalloc-1k start c1f60c00 pointer offset 0 size 1024
[   11.846539] Register r9 information: slab kmalloc-256 start c1ec9b00 pointer offset 160 size 256
[   11.855391] Register r10 information: slab dentry start c216b220 pointer offset 0 size 136
[   11.863721] Register r11 information: slab filp start c1aaea80 pointer offset 0 size 128
[   11.871876] Register r12 information: 0-page vmalloc region starting at 0xf9c00000 allocated at iotable_init+0x0/0xc0
[   11.882548] Process watchdog (pid: 130, stack limit = 0x(ptrval))
[   11.888674] Stack: (0xe0305d58 to 0xe0306000)
[   11.893055] 5d40:                                                       c1ec9b40 c1e44930
[   11.901275] 5d60: 00000002 c0943de8 0f900000 00000000 c18a5000 c1f60df0 c1ba7580 c1f60c00
[   11.909496] 5d80: c1e44930 c1aaea80 c1ec9b40 c1f60eac 00000000 c0943f4c c1f60df0 c1e44930
[   11.917715] 5da0: c0d90324 00000000 c1aaea80 c0370ba8 00000000 73a085ae e0305e78 c1aaea80
[   11.925935] 5dc0: 00000000 c1e44930 c0370ab8 c1aaeaa8 e0305e78 c0365070 c1aaea80 c1286208
[   11.934154] 5de0: c1aaeaa8 00000000 e0305f30 c0366ba4 e0305f30 c037986c e0305f30 c037a844
[   11.942374] 5e00: 00020001 c1286208 00000000 c0380710 e0305e68 b6e56000 b6fb7fff c1bc880c
[   11.950594] 5e20: 00000000 00000101 00000000 00000000 00000000 82e4a831 c2e79880 00100cca
[   11.958814] 5e40: 00000150 b6fa6000 b6fa6000 73a085ae c2e6adb8 c195bc00 e0305f30 ffffff9c
[   11.967034] 5e60: 00000001 c01002ec c1fe1500 00000142 005a6874 c0381170 c18f6f10 c216b220
[   11.975253] 5e80: 4a3c55ab 00000009 c195bc11 00000000 00000000 c21a2b28 c1e44930 00010001
[   11.983472] 5ea0: 00000000 00000000 00000000 00000098 00000002 00000000 00000000 00000000
[   11.991692] 5ec0: e0305ec4 00000000 00000000 00000000 00000000 00000000 00000000 00000000
[   11.999911] 5ee0: 00000000 00000000 00000000 00000000 00000000 c195bc00 c195bc0c 00000000
[   12.008131] 5f00: 00000002 ffffff9c 00000000 000041ed e0305f60 73a085ae ffffff9c 00000003
[   12.016351] 5f20: e0305f60 c195bc00 ffffff9c c0367010 00020001 00000000 00000002 00010000
[   12.024571] 5f40: 00000001 73a085ae bed20ab4 ffffff9c bed20f58 005a79bc 00000142 c03673e8
[   12.032790] 5f60: 00020001 00000000 00000000 00000000 00000000 00000000 00020001 00000000
[   12.041009] 5f80: 00000000 00000000 00000000 00000000 bed20b38 73a085ae bed20be4 00000000
[   12.049229] 5fa0: 00000000 c0100080 00000000 00000000 ffffff9c bed20f58 00020001 00000000
[   12.057449] 5fc0: 00000000 00000000 005a79bc 00000142 004dfc60 b6ffccf8 00000000 005a6874
[   12.065668] 5fe0: 00020001 bed20c28 00000001 b6ed2458 200f0010 ffffff9c 00000000 00000000
[   12.073883] Call trace: 
[   12.073895]  omap_wdt_disable from omap_wdt_start+0x34/0xf8
[   12.082052]  omap_wdt_start from watchdog_start+0x8c/0x184
[   12.087579]  watchdog_start from watchdog_open+0x6c/0xf0
[   12.092930]  watchdog_open from chrdev_open+0xf0/0x220
[   12.098113]  chrdev_open from do_dentry_open+0x120/0x4f8
[   12.103465]  do_dentry_open from vfs_open+0x24/0xe4
[   12.108378]  vfs_open from path_openat+0x828/0x1224
[   12.113293]  path_openat from do_file_open+0x64/0xd8
[   12.118290]  do_file_open from do_sys_openat2+0x80/0xe8
[   12.123550]  do_sys_openat2 from sys_openat+0x7c/0xd0
[   12.128637]  sys_openat from ret_fast_syscall+0x0/0x58
[   12.133810] Exception stack(0xe0305fa8 to 0xe0305ff0)
[   12.138891] 5fa0:                   00000000 00000000 ffffff9c bed20f58 00020001 00000000
[   12.147111] 5fc0: 00000000 00000000 005a79bc 00000142 004dfc60 b6ffccf8 00000000 005a6874
[   12.155329] 5fe0: 00020001 bed20c28 00000001 b6ed2458
[   12.160412] Code: e59f30a4 e280c048 e58c3000 e2803034 (e5902034) 
[   12.166540] ---[ end trace 0000000000000000 ]---
[   12.171183] note: watchdog[130] exited with irqs disabled
[   13.282448] omap_prm 44e00d00.prm: sync_state() pending due to 44d00000.cpu
Signature · 3 lines
Mathieu Dubois-Briand, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com
#2 lore raw
Quoted 35 lines · Stephen Boyd

+Tony (author of commit)

Quoting Mathieu Dubois-Briand (2026-04-28 05:03:16)

Quoted 32 lines

Hi,

I believe I’ve identified an issue with clock management on the AM335x SoC and possibly other TI SoC. It can be seen while using the SoC watchdog timer, but I suspect other clocks or modules might be impacted.

While I partially understand the issue, I struggle to find a correct solution. I will describe my findings below.

First, here is a quick way to reproduce it.
- Hardware: Beagle Bone Black.
- Linux kernel version: v7.1-rc1 (or anything after 5.19).
- Configuration: omap2plus_defconfig + CONFIG_OMAP_WATCHDOG=y
- In the bootloader, I select the 32KHz clock derived from the 24MHz one as the watchdog clock source. This is controlled by bit 1 of CLKSEL_WDT1_CLK.
This is probably the point that differs from most use cases, and why it hasn't been observed before. On my final setup, I do not control the bootloader and so cannot change this value.
In u-boot, this is: mw.l 0x44e00538 1
- Once the kernel is booted, I open the watchdog device node. I'm using busybox "watchdog" command, but a simple "cat" will trigger the same error.
watchdog /dev/watchdog0
- This will result in the bus clock failing to be enabled and a failure to access the watchdog module:

  [   11.677752] l4-wkup-clkctrl:00d4:0: failed to enable
  [   11.686079] Unhandled fault: external abort on non-linefetch (0x1028) at 0xf9e35034
  I'm attaching the full trace at the very end of my email.

So, I tried to dig a bit into this issue and the reason why the l4-wkup-clkctrl:00d4:0 clock can't be enabled.

Just a quick guess.. You can use assigned-clocks in the dts file to configure how your board is wired up.

Please check the system timers configuration too, sounds like timers may not be configured properly either. Some timer examples can be seen with:

$ git grep -A10 "Preferred " arch/arm/boot/dts/

Regards,

Tony

#3 lore raw
Quoted 34 lines · Tony Lindgren
Quoted 33 lines · Stephen Boyd

+Tony (author of commit)

Quoting Mathieu Dubois-Briand (2026-04-28 05:03:16)

Quoted 30 lines

Hi,

I believe I’ve identified an issue with clock management on the AM335x SoC and possibly other TI SoC. It can be seen while using the SoC watchdog timer, but I suspect other clocks or modules might be impacted.

While I partially understand the issue, I struggle to find a correct solution. I will describe my findings below.

First, here is a quick way to reproduce it.
- Hardware: Beagle Bone Black.
- Linux kernel version: v7.1-rc1 (or anything after 5.19).
- Configuration: omap2plus_defconfig + CONFIG_OMAP_WATCHDOG=y
- In the bootloader, I select the 32KHz clock derived from the 24MHz one as the watchdog clock source. This is controlled by bit 1 of CLKSEL_WDT1_CLK.
This is probably the point that differs from most use cases, and why it hasn't been observed before. On my final setup, I do not control the bootloader and so cannot change this value.
In u-boot, this is: mw.l 0x44e00538 1
- Once the kernel is booted, I open the watchdog device node. I'm using busybox "watchdog" command, but a simple "cat" will trigger the same error.
watchdog /dev/watchdog0
- This will result in the bus clock failing to be enabled and a failure to access the watchdog module:

  [   11.677752] l4-wkup-clkctrl:00d4:0: failed to enable
  [   11.686079] Unhandled fault: external abort on non-linefetch (0x1028) at 0xf9e35034
  I'm attaching the full trace at the very end of my email.
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

So, I tried to dig a bit into this issue and the reason why the l4-wkup-clkctrl:00d4:0 clock can't be enabled.

Hi Tony,

Thanks for your reply.

Quoted 2 lines

Just a quick guess.. You can use assigned-clocks in the dts file to configure how your board is wired up.

Just to be clear: I believe this is entirely independent from how the board is wired up. On both my custom board and the Beagle Bone Black, I can successfully use any of the two muxing possibilities, if I manually write the correct data into the registers. My only issue is the kernel does not handle the clocks correctly when one these muxing possibilities is enabled.

Yet I tried to add an assigned-clocks property to the clock-wdt1-fck@538 node. Is that what you had in mind? This had no visible effect.

Quoted 4 lines

Please check the system timers configuration too, sounds like timers may not be configured properly either. Some timer examples can be seen with:

$ git grep -A10 "Preferred " arch/arm/boot/dts/

So I had a look at existing preferred timer configurations, but they are all applied to the "timer.*_target" [1][2] nodes, that do correspond to SoC timers (DMTIMER.*) and I fail to see how this is linked with the peripheral clocks involved here (in CM_PER module) and my current clock issues.

Nevertheless, I can see these being defined in arch/arm/boot/dts/ti/omap/am33xx.dtsi [3][4], so they are indeed applied in my situation.

So, I believe timers configuration are correct. Should I add something else?

Thanks again,
Mathieu

[1]: https://elixir.bootlin.com/linux/v7.0.1/source/arch/arm/boot/dts/ti/omap/am33xx-l4.dtsi#L357
[2]: https://elixir.bootlin.com/linux/v7.0.1/source/arch/arm/boot/dts/ti/omap/am33xx-l4.dtsi#L1280
[3]: https://elixir.bootlin.com/linux/v7.0.1/source/arch/arm/boot/dts/ti/omap/am33xx.dtsi#L702
[4]: https://elixir.bootlin.com/linux/v7.0.1/source/arch/arm/boot/dts/ti/omap/am33xx.dtsi#L715

Signature · 3 lines
Mathieu Dubois-Briand, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com
#4 lore raw
Quoted 10 lines · Mathieu Dubois-Briand
Quoted 2 lines · Tony Lindgren

Just a quick guess.. You can use assigned-clocks in the dts file to configure how your board is wired up.

Just to be clear: I believe this is entirely independent from how the board is wired up. On both my custom board and the Beagle Bone Black, I can successfully use any of the two muxing possibilities, if I manually write the correct data into the registers. My only issue is the kernel does not handle the clocks correctly when one these muxing possibilities is enabled.

OK

Quoted 2 lines

Yet I tried to add an assigned-clocks property to the clock-wdt1-fck@538 node. Is that what you had in mind? This had no visible effect.

Yes I was thinking adding assigned-clocks and assigned-clock-parents similar to what the timers do.

There's also the am33xx_dt_clk_init() reparenting of wdt1_fck that's worth checking for why the chosen parent is not getting set.

Quoted 11 lines
Quoted 4 lines

Please check the system timers configuration too, sounds like timers may not be configured properly either. Some timer examples can be seen with:

$ git grep -A10 "Preferred " arch/arm/boot/dts/

So I had a look at existing preferred timer configurations, but they are all applied to the "timer.*_target" [1][2] nodes, that do correspond to SoC timers (DMTIMER.*) and I fail to see how this is linked with the peripheral clocks involved here (in CM_PER module) and my current clock issues.

Quoted 6 lines

Nevertheless, I can see these being defined in arch/arm/boot/dts/ti/omap/am33xx.dtsi [3][4], so they are indeed applied in my situation.

So, I believe timers configuration are correct. Should I add something else?

OK no, so the issue is limited to wdt1 like you described earlier.

Quoted 4 lines

[1]: https://elixir.bootlin.com/linux/v7.0.1/source/arch/arm/boot/dts/ti/omap/am33xx-l4.dtsi#L357
[2]: https://elixir.bootlin.com/linux/v7.0.1/source/arch/arm/boot/dts/ti/omap/am33xx-l4.dtsi#L1280
[3]: https://elixir.bootlin.com/linux/v7.0.1/source/arch/arm/boot/dts/ti/omap/am33xx.dtsi#L702
[4]: https://elixir.bootlin.com/linux/v7.0.1/source/arch/arm/boot/dts/ti/omap/am33xx.dtsi#L715