On Fri, Feb 12, 2016 at 03:54:18PM +1100, Alexey Kardashevskiy wrote:

On 02/11/2016 04:32 PM, Paul Mackerras wrote:

quoted

On Thu, Jan 21, 2016 at 06:39:37PM +1100, Alexey Kardashevskiy wrote:

quoted

+	rmap = (void *) vmalloc_to_phys(rmap);
+
+	lock_rmap(rmap);

A comment here explaining why we lock the rmap and what that achieves
would be useful for future generations.

/* This protects the guest page with the TCE list from going away while we
are reading TCE list */

?

By "going away" I mean H_ENTER/H_REMOVE executed on parallel CPUs, is this
roughly correct? as I did grep for "lock_rmap()" and did not find a single
comment next to it...

Actually, taking the rmap lock stops the guest real -> host real
mapping from changing.  For the comment, I suggest this:

	/*
	 * Synchronize with the MMU notifier callbacks in
	 * book3s_64_mmu_hv.c (kvm_unmap_hva_hv etc.).
	 * While we have the rmap lock, code running on other CPUs
	 * cannot finish unmapping the host real page that backs
	 * this guest real page, so we are OK to access the host
	 * real page.
	 */

Paul.