On 05/09/2019 12.39, Janosch Frank wrote:
> Let's add a rudimentary SMP library, which will scan for cpus and has
> helper functions that manage the cpu state.
>
> Signed-off-by: Janosch Frank <frankja@xxxxxxxxxxxxx>
> ---
[...]
> +/*Expected to be called from boot cpu */
> +extern uint64_t *stackptr;
> +void smp_setup(void)
> +{
> + int i = 0;
> + unsigned short cpu0_addr = stap();
> + struct ReadCpuInfo *info = (void *)cpu_info_buffer;
> +
> + spin_lock(&lock);
> + sclp_mark_busy();
> + info->h.length = PAGE_SIZE;
> + sclp_service_call(SCLP_READ_CPU_INFO, cpu_info_buffer);
> +
> + if (smp_query_num_cpus() > 1)
> + printf("SMP: Initializing, found %d cpus\n", info->nr_configured);
> +
> + cpus = calloc(info->nr_configured, sizeof(cpus));
> + for (i = 0; i < info->nr_configured; i++) {
> + if (info->entries[i].address == cpu0_addr) {
> + cpu0 = &cpus[i];
> + cpu0->stack = stackptr;
> + cpu0->lowcore = (void *)0;
> + cpu0->active = true;
So here cpus[i].active gets set to true ...
> + }
> + cpus[i].addr = info->entries[i].address;
> + cpus[i].active = false;
... but here it is set back to false.
Maybe move the if-statement below this line?
> + }
> + spin_unlock(&lock);
> +}
[...]
> diff --git a/s390x/cstart64.S b/s390x/cstart64.S
> index 36f7cab..a45ea8f 100644
> --- a/s390x/cstart64.S
> +++ b/s390x/cstart64.S
> @@ -172,6 +172,13 @@ diag308_load_reset:
> lhi %r2, 1
> br %r14
>
> +.globl smp_cpu_setup_state
> +smp_cpu_setup_state:
> + xgr %r1, %r1
> + lmg %r0, %r15, 512(%r1)
Can you use GEN_LC_SW_INT_GRS instead of 512?
> + lctlg %c0, %c0, 776(%r1)
... and here GEN_LC_SW_INT_CR0 instead of 776?
Apart from that, the patch looks fine to me.
Thomas
