Hi Kever,
On 11/5/24 6:21 PM, Quentin Schulz wrote:
I am the unfortunate owner of an RK3399 Puma that is failing DRAM init
every now and then with the following cryptic error message:
"""
pctl_start: Failed to init pctl channel 0
"""
When it fails, the current logic is implemented in such a way that the
board enters an infinite while loop. This is not ideal for embedded
systems that are not easily accessible and also an issue when having
devices in a CI lab for example.
Therefore, let's simply panic if the DRAM fails to init. This was tested
only on RK3399 Puma, similar changes may be required for other SoCs to
properly propagate the errors.
While this is all but a work-around instead of fixing the DRAM init
sequence, I believe it is important for resilience of devices in the
field.
Marking this as RFC because:
1) panic()s for all Rockchip SoCs but only tested on RK3399
2) unsure about error return values (ENODEV and the likes), not sure
what would be better than that though :/
Note that one can still hang() if they want by setting PANIC_HANG
symbol.
What do you think about this series?
I believe we should be able to merge everything but the panic() and fail
DRAM init on pctl patches without changing the current behavior?
I could keep the panic() and fail DRAM init on pctl patches in my
downstream fork if there's no interest but I think it makes sense to
have them merged.
Cheers,
Quentin
