On 8/9/21 11:26 PM, Kirill A. Shutemov wrote:
+void accept_memory(phys_addr_t start, phys_addr_t end)
+{
+ if (!boot_params.unaccepted_memory)
+ return;
+
+ spin_lock(&unaccepted_memory_lock);
+ __accept_memory(start, end);
+ spin_unlock(&unaccepted_memory_lock);
+}
Isn't this taken in the:
del_page_from_free_list()->
clear_page_offline()->
accept_memory()
call path?
That's underneath:
spin_lock_irqsave(&zone->lock, flags);
Which means that accept_memory() can happen from interrupt context. Is
it always covered by another spin_lock_irqsave() which means that it can
use a plain spin_lock()?
If so, it would be nice to call out that logic. It *looks* like a
spinlock that we would want to be spin_lock_irqsave().