[<prev] [next>] [day] [month] [year] [list]
Message-Id: <20191207.160457.297470232945357742.davem@davemloft.net>
Date: Sat, 07 Dec 2019 16:04:57 -0800 (PST)
From: David Miller <davem@...emloft.net>
To: bpf@...r.kernel.org, netdev@...r.kernel.org
CC: ast@...nel.org, daniel@...earbox.net, tglx@...utronix.de
Subject: [RFC v1 PATCH 7/7] bpf: Don't use tasklet in stackmap code when RT.
In the RT kernel we can't use up_read_non_owner().
So in such a configuration, simply elide the annotated stackmap and
just report the raw IPs.
In the longer term, we can make atomic friendly versions of the page
cache traversal which will at least provide the info if the pages are
resident and don't need to be paged in.
Signed-off-by: David S. Miller <davem@...emloft.net>
---
kernel/bpf/stackmap.c | 7 +++++++
1 file changed, 7 insertions(+)
diff --git a/kernel/bpf/stackmap.c b/kernel/bpf/stackmap.c
index caca752ee5e6..d49b5f184082 100644
--- a/kernel/bpf/stackmap.c
+++ b/kernel/bpf/stackmap.c
@@ -288,10 +288,17 @@ static void stack_map_get_build_id_offset(struct bpf_stack_build_id *id_offs,
struct stack_map_irq_work *work = NULL;
if (irqs_disabled()) {
+#ifdef CONFIG_PREEMPT_RT_FULL
+ /* We cannot use the up_read_non_owner() interface in RT
+ * kernels, so just force reporting of IPs.
+ */
+ irq_work_busy = true;
+#else
work = this_cpu_ptr(&up_read_work);
if (work->irq_work.flags & IRQ_WORK_BUSY)
/* cannot queue more up_read, fallback */
irq_work_busy = true;
+#endif
}
/*
--
2.20.1
Powered by blists - more mailing lists