Hi, I had a few questions about the context switch code in seL4 for aarch64. I spent a couple of days trying to make sense of it, so I thought it would be better to ask as I might be missing something that is not obvious.
It looks like kernel_enter in traps.S [1] saves the user process's register immediately after an exception. And restore_user_context() in c_traps.c [2] is restoring them. Looks like the regs are stored in the TCB in the "tcbContext.registers" array.
The restore code makes sense to me. It puts the ptr value of "(NODE_STATE(ksCurThread)->tcbArch.tcbContext.registers" in to the sp and starts loading the regs from there.
However, the code in kernel_enter is a bit unclear to me. 1. What is the value of sp when executing the kernel_enter macro? Is that the SP_EL1, or somewhere else the sp was set to the address of the faulting thread's "tcbContext.registers". On reading the arm32's arm_swi_syscall [3] function, it looks like it is the latter, but I am unsure how that is being accomplished. It might be more to do with aarch64 than seL4. 2 Furthermore, when not specified, does sp mean SP_EL1 in EL1 and SP_EL0 in EL0, Or does lower case "sp" have a special meaning? I know that seL4 is configured to use SP_ELx at ELx based on value of SPEL in [4] 3 Lastly, what is lsp_i trying to do? Are we switching the value of sp from something else to the real kernel stack by storing the kernel's stack address in tpidr_el1?
Thanks a lot, as always! Sid
[1] https://github.com/seL4/seL4/blob/master/src/arch/arm/64/traps.S#L72 [2] https://github.com/seL4/seL4/blob/master/src/arch/arm/64/c_traps.c#L29 [3] https://github.com/seL4/seL4/blob/master/src/arch/arm/32/traps.S#L51 [4] https://github.com/seL4/seL4/blob/master/src/arch/arm/64/head.S#L78
I think I have understood how the sp(SP_EL1) was setup.
It was setup on the way out of the kernel by restore_user_context, and it just remained there until we entered the kernel again. Whatever may be the reason(timer interrupt, syscall) for entering the kernel again, it is certain that the same user process would still be running.
Thanks, Sid
On Fri, Jun 10, 2022 at 8:27 PM Sid Agrawal siagraw@cs.ubc.ca wrote:
Hi, I had a few questions about the context switch code in seL4 for aarch64. I spent a couple of days trying to make sense of it, so I thought it would be better to ask as I might be missing something that is not obvious.
It looks like kernel_enter in traps.S [1] saves the user process's register immediately after an exception. And restore_user_context() in c_traps.c [2] is restoring them. Looks like the regs are stored in the TCB in the "tcbContext.registers" array.
The restore code makes sense to me. It puts the ptr value of "(NODE_STATE(ksCurThread)->tcbArch.tcbContext.registers" in to the sp and starts loading the regs from there.
However, the code in kernel_enter is a bit unclear to me.
- What is the value of sp when executing the kernel_enter macro? Is that
the SP_EL1, or somewhere else the sp was set to the address of the faulting thread's "tcbContext.registers". On reading the arm32's arm_swi_syscall [3] function, it looks like it is the latter, but I am unsure how that is being accomplished. It might be more to do with aarch64 than seL4. 2 Furthermore, when not specified, does sp mean SP_EL1 in EL1 and SP_EL0 in EL0, Or does lower case "sp" have a special meaning? I know that seL4 is configured to use SP_ELx at ELx based on value of SPEL in [4] 3 Lastly, what is lsp_i trying to do? Are we switching the value of sp from something else to the real kernel stack by storing the kernel's stack address in tpidr_el1?
Thanks a lot, as always! Sid
[1] https://github.com/seL4/seL4/blob/master/src/arch/arm/64/traps.S#L72 [2] https://github.com/seL4/seL4/blob/master/src/arch/arm/64/c_traps.c#L29 [3] https://github.com/seL4/seL4/blob/master/src/arch/arm/32/traps.S#L51 [4] https://github.com/seL4/seL4/blob/master/src/arch/arm/64/head.S#L78
-
Sid Agrawal