inochisa commented on code in PR #12178: URL: https://github.com/apache/nuttx/pull/12178#discussion_r1580754406
########## arch/risc-v/src/common/riscv_macros.S: ########## @@ -364,3 +364,35 @@ csrr \out, CSR_MHARTID #endif .endm + +/**************************************************************************** + * Name: riscv_set_inital_sp + * + * Description: + * Set inital sp for riscv core. This function should be only called + * when initing. + * + * sp (stack top) = sp base + idle stack size * hart id + * sp (stack base) = sp (stack top) + idle stack size * - XCPTCONTEXT_SIZE + * + * Note: The XCPTCONTEXT_SIZE byte after stack base is reserved for + * up_initial_state since we are already running and using + * the per CPU idle stack. + * Parameter: + * base - Pointer to where the stack is allocated (e.g. _ebss) + * size - Stack size for pre cpu to allocate + * size - Hart id register of this hart (Usually a0) + * + ****************************************************************************/ +.macro riscv_set_inital_sp base, size, hartid + la t0, \base + li t1, \size + mul t1, \hartid, t1 + add t0, t0, t1 + + /* ensure the last XCPTCONTEXT_SIZE is reserved */ + + li t1, STACK_ALIGN_DOWN(\size - XCPTCONTEXT_SIZE) Review Comment: > > This is prepared for boot with cores other than 0. > > So, where is the idle stack for CPU0? I just reviewed your changes roughly and all XXX_RV_IDLESTACK_TOP should have been removed. > I think it needs another check when boot hart is not 0. But at least for now, I will take you advice and add a check for cpu 0, thanks. > > In addition, I have seen [`xcp->regs`](https://github.com/apache/nuttx/blob/master/arch/risc-v/src/common/riscv_initialstate.c#L135) always point to the XCPTCONTEXT_SIZE offset of the topstack. I have no idea about this truth you have pointed out, although I know it works. Do I miss something? > > CPU0 does not need to reserve XCPTCONTEXT_SIZE since the first idle task(CPU0) is running task, which does not to need to restore the context. the area of XCPTCONTEXT is mainly to allow the task to restore the correct parameters of initialization registers if context switch. Thanks. > Or you could compare the stack usage with/without this PR，I think cpu0 will take up more than before. You are right, I have noticed this. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
