Hi Stefano, On 03/05/2023 00:55, Stefano Stabellini wrote:
+ { + printk("CPU%d: No release addr\n", cpu); + return -ENODEV; + } + + release = ioremap_nocache(cpu_release_addr[cpu], 4); + if ( !release ) + { + dprintk(XENLOG_ERR, "CPU%d: Unable to map release address\n", cpu); + return -EFAULT; + } + + writel(__pa(init_secondary), release); + + iounmap(release);I think we need a wmb() ?
I am not sure why we would need a wmb() here. Instead, looking at the Linux version, I think we are missing a cache flush (so does on arm64) which would be necessary if the CPU waiting for the release doesn't have cache enabled.
Cheers, -- Julien Grall
