Skip to content

Conversation

IxaKylin
Copy link

@IxaKylin IxaKylin commented Jul 3, 2025

Add the power-domains for the RK3576 SFC nodes according to the
TRM part 1. This fixes potential SErrors when accessing the SFC
registers without other peripherals (e.g. eMMC) doing a prior
power-domain enable. For example this is easy to trigger on the
Rock 4D, which enables the SFC0 interface, but does not enable
the eMMC interface at the moment.

Cc: [email protected]
Fixes: 3629975 ("arm64: dts: rockchip: Add SFC nodes for rk3576")
Signed-off-by: Sebastian Reichel [email protected]

I finally managed to get some RK3576 boards integrated to our CI
pipeline and promptly got some SError on Rock 4D with the extra
test coverage :( As we hope to get some of those boards for KernelCI,
it would be good to get this fixed in all affected trees. It seemed
enough to just describe the power-domain in DT (i.e. that fixed the
SError for the arm64 defconfig when booting the Rock 4D). If we see
further problems (I haven't so far), we might need something like
[0] for the FSPI driver.

[0] https://lore.kernel.org/all/[email protected]/

[ 15.248915] Kernel panic - not syncing: Asynchronous SError Interrupt
[ 15.248917] CPU: 7 UID: 0 PID: 142 Comm: (udev-worker) Not tainted 6.15.0-rc6-g51237a9145a9 #1 PREEMPT
[ 15.248921] Hardware name: Radxa ROCK 4D (DT)
[ 15.248923] Call trace:
[ 15.248924] show_stack+0x2c/0x84 (C)
[ 15.248937] dump_stack_lvl+0x60/0x80
[ 15.248941] dump_stack+0x18/0x24
[ 15.248944] panic+0x168/0x360
[ 15.248948] add_taint+0x0/0xbc
[ 15.248952] arm64_serror_panic+0x64/0x70
[ 15.248956] do_serror+0x3c/0x70
[ 15.248958] el1h_64_error_handler+0x30/0x48
[ 15.248964] el1h_64_error+0x6c/0x70
[ 15.248967] rockchip_sfc_init.isra.0+0x20/0x8c [spi_rockchip_sfc] (P)
[ 15.248972] platform_probe+0x68/0xdc
[ 15.248978] really_probe+0xc0/0x39c
[ 15.248982] __driver_probe_device+0x7c/0x14c
[ 15.248985] driver_probe_device+0x3c/0x120
[ 15.248989] __driver_attach+0xc4/0x200
[ 15.248992] bus_for_each_dev+0x7c/0xdc
[ 15.248995] driver_attach+0x24/0x30
[ 15.248998] bus_add_driver+0x110/0x240
[ 15.249001] driver_register+0x68/0x130
[ 15.249005] __platform_driver_register+0x24/0x30
[ 15.249010] rockchip_sfc_driver_init+0x20/0x1000 [spi_rockchip_sfc]
[ 15.249014] do_one_initcall+0x60/0x1e0
[ 15.249017] do_init_module+0x54/0x1fc
[ 15.249021] load_module+0x18f8/0x1e50
[ 15.249024] init_module_from_file+0x88/0xcc
[ 15.249027] __arm64_sys_finit_module+0x260/0x358
[ 15.249031] invoke_syscall+0x48/0x104
[ 15.249035] el0_svc_common.constprop.0+0x40/0xe0
[ 15.249040] do_el0_svc+0x1c/0x28
[ 15.249044] el0_svc+0x30/0xcc
[ 15.249048] el0t_64_sync_handler+0x10c/0x138
[ 15.249052] el0t_64_sync+0x198/0x19c
[ 15.249057] SMP: stopping secondary CPUs
[ 15.249064] Kernel Offset: 0x38f049600000 from 0xffff800080000000
[ 15.249066] PHYS_OFFSET: 0xfff0e21340000000
[ 15.249068] CPU features: 0x0400,00041250,01000400,0200421b
[ 15.249071] Memory Limit: none
[ 15.273962] ---[ end Kernel panic - not syncing: Asynchronous SError Interrupt ]---

TRM part 1. This fixes potential SErrors when accessing the SFC
registers without other peripherals (e.g. eMMC) doing a prior
power-domain enable. For example this is easy to trigger on the
Rock 4D, which enables the SFC0 interface, but does not enable
the eMMC interface at the moment.

Cc: [email protected]
Fixes: 3629975 ("arm64: dts: rockchip: Add SFC nodes for rk3576")
Signed-off-by: Sebastian Reichel <[email protected]>
---
I finally managed to get some RK3576 boards integrated to our CI
pipeline and promptly got some SError on Rock 4D with the extra
test coverage :( As we hope to get some of those boards for KernelCI,
it would be good to get this fixed in all affected trees. It seemed
enough to just describe the power-domain in DT (i.e. that fixed the
SError for the arm64 defconfig when booting the Rock 4D). If we see
further problems (I haven't so far), we might need something like
[0] for the FSPI driver.

[0] https://lore.kernel.org/all/[email protected]/
@IxaKylin IxaKylin changed the title Add the power-domains for the RK3576 SFC nodes according to the Add the power-domains for the RK3576 SFC device tree node Jul 3, 2025
@IxaKylin IxaKylin closed this Jul 3, 2025
@IxaKylin IxaKylin reopened this Jul 3, 2025
@RadxaStephen
Copy link
Member

In vendor rk kernel tree, the sfc1 power-domains is different.
See IxaKylin@5da753e

@IxaKylin IxaKylin closed this Jul 14, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants