Re: [PATCH 4/5] x86/mm: Provide helpers for unaccepted memory

[Dave Hansen]

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().