Messages in this thread Patch in this message | | | Date | Tue, 5 May 2020 14:31:44 +0200 | From | Joerg Roedel <> | Subject | [PATCH] tracing: Call vmalloc_sync_mappings() after alloc_percpu() |
| |
On Mon, May 04, 2020 at 03:10:06PM -0400, Steven Rostedt wrote: > I'm fine with adding it to the tracing code (with that ridiculous > comment! ;-) > > I'll even tag is as stable, but again, it's uncertain what commit that it > "fixes".
Okay, so here it is. It fixes the issue for me and doesn't cause any lockdep splats on my machine. I am not sure how far this needs to be backported via stable, so I didn't specify it. Same with a Fixes tag. Oh, and I added your comment :-) (which makes up most of the added lines, so feel free to change authorship to you).
Regards,
Joerg
From a265290761dc9d87d35892dc821fd0f5c99f8b34 Mon Sep 17 00:00:00 2001 From: Joerg Roedel <jroedel@suse.de> Date: Tue, 5 May 2020 14:17:57 +0200 Subject: [PATCH] tracing: Call vmalloc_sync_mappings() after alloc_percpu()
Tracing code allocates percpu memory which is touched when trace events happen. When the events happen in the page-fault handler and per-cpu memory needs to be faulted in, the page-fault on per-cpu memory might never complete.
Call vmalloc_sync_mappings() after allocating the runtime per-cpu buffers to make sure they are mapped when the page-fault handler tries to access them.
This is not a nice solution, but the best fix until the vmalloc_sync_mappings/unmappings() interface can be ripped out of the kernel.
Cc: stable@vger.kernel.org Signed-off-by: Joerg Roedel <jroedel@suse.de> --- kernel/trace/trace.c | 13 +++++++++++++ 1 file changed, 13 insertions(+)
diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c index 8d2b98812625..0fc56063fcca 100644 --- a/kernel/trace/trace.c +++ b/kernel/trace/trace.c @@ -8525,6 +8525,19 @@ static int allocate_trace_buffers(struct trace_array *tr, int size) */ allocate_snapshot = false; #endif + + /* + * Because of some magic with the way alloc_percpu() works on + * x86_64, we need to synchronize the pgd of all the tables, + * otherwise the trace events that happen in x86_64 page fault + * handlers can't cope with accessing the chance that a + * alloc_percpu()'d memory might be touched in the page fault trace + * event. Oh, and we need to audit all alloc_percpu() and vmalloc() + * calls in tracing, because something might get triggered within a + * page fault trace event! + */ + vmalloc_sync_mappings(); + return 0; } -- 2.12.3
| |