fix(core): fix handover to bootloader #4221
Conversation
cepetr
added
core
cepetr
force-pushed
the
cepetr/fix-handover-to-bl
branch
from
34ee671
to
01c08f7
Compare
[no changelog]
cepetr
force-pushed
the
cepetr/fix-handover-to-bl
branch
from
01c08f7
to
a2880ec
Compare
|
|
||
| "LDR R1, = 0xE000EF34 \n" // FPU->FPCCR | ||
| "LDR R0, [R1] \n" | ||
| "BIC R0, R0, #1 \n" // Clear LSPACT to suppress lazy |
There was a problem hiding this comment.
from the description of LSPACT bit in programming manual:
Lazy state preservation active. Indicates whether lazy preservation of the floating-point state is active.
i would be expecting that this bit is managed by hardware and should be used by software as read only. Are you sure about this? For disabling of lazy stacking, i would expect LSPEN bit to be used, or by suppression do you mean that this should be only one time suppressed and this actually achieves it?
There was a problem hiding this comment.
Yes, it's kind of a one-time suppression.
If we hadn't done it here, the first FPU instruction used in the bootloader might cause lazy stacking and an unexpected overwrite of a portion of SRAM. There are some possible alternatives: 1) use any FPU instruction here instead of writing to FPU->FPPCR, which causes immediate stacking of FPU registers and clears the LSPACT bit, 2) possibly returning from handler mode with a different EXC_RETURN code...
I don't like it much either and I understand it looks a bit hacky.
This PR fixes the firmware deadlock issue that occurred after the interaction-less update. Th bug was introduced in PR #4188, and only affects T and TS3 models.
The fix adds a transition from handler mode to thread mode before jumping to the bootloader.