On 25/09/2018 14:15, Balbir Singh wrote:

On Tue, Sep 25, 2018 at 11:14:56AM +0200, David Hildenbrand wrote:

quoted

Let's perform all checking + offlining + removing under
device_hotplug_lock, so nobody can mess with these devices via
sysfs concurrently.

Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <redacted>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Rashmica Gupta <redacted>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Michael Neuling <redacted>
Reviewed-by: Pavel Tatashin <redacted>
Reviewed-by: Rashmica Gupta <redacted>
Signed-off-by: David Hildenbrand <redacted>
---
 arch/powerpc/platforms/powernv/memtrace.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/arch/powerpc/platforms/powernv/memtrace.c b/arch/powerpc/platforms/powernv/memtrace.c
index fdd48f1a39f7..d84d09c56af9 100644
--- a/arch/powerpc/platforms/powernv/memtrace.c
+++ b/arch/powerpc/platforms/powernv/memtrace.c

@@ -70,6 +70,7 @@ static int change_memblock_state(struct memory_block *mem, void *arg)
 	return 0;
 }
 
+/* called with device_hotplug_lock held */
 static bool memtrace_offline_pages(u32 nid, u64 start_pfn, u64 nr_pages)
 {
 	u64 end_pfn = start_pfn + nr_pages - 1;

@@ -111,6 +112,7 @@ static u64 memtrace_alloc_node(u32 nid, u64 size)
 	end_pfn = round_down(end_pfn - nr_pages, nr_pages);
 
 	for (base_pfn = end_pfn; base_pfn > start_pfn; base_pfn -= nr_pages) {
+		lock_device_hotplug();

Why not grab the lock before the for loop? That way we can avoid bad cases like a
large node being scanned for a small number of pages (nr_pages). Ideally we need
a cond_resched() in the loop, but I guess offline_pages() has one.

Yes, it does.

I can move it out of the loop, thanks!

Acked-by: Balbir Singh <bsingharora@gmail.com>

-- 

Thanks,

David / dhildenb