Hi Marc,
On 2019/3/29 21:58, Marc Zyngier wrote:
Hi Zenghui,
On 29/03/2019 13:23, Zenghui Yu wrote:
Enable pseudo NMI together with function_graph tracer, will lead
the system to a hang. This is easy to reproduce,
1) Set "irqchip.gicv3_pseudo_nmi=1" on the kernel command line
2) echo function_graph > /sys/kernel/debug/tracing/current_tracer
This patch (RFC) set gic_handle_irq() as notrace and it seems works
fine now. But I have no idea about what the issue is exactly, and
you can regard this patch as a report then :)
Can someone give a look at it and provide some explanations ?
Thanks!
Cc: Julien Thierry <julien.thie...@arm.com>
Cc: Steven Rostedt <rost...@goodmis.org>
Signed-off-by: Zenghui Yu <yuzeng...@huawei.com>
---
drivers/irqchip/irq-gic-v3.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/irqchip/irq-gic-v3.c b/drivers/irqchip/irq-gic-v3.c
index 15e55d3..8d0c25f 100644
--- a/drivers/irqchip/irq-gic-v3.c
+++ b/drivers/irqchip/irq-gic-v3.c
@@ -487,7 +487,7 @@ static inline void gic_handle_nmi(u32 irqnr, struct pt_regs
*regs)
gic_deactivate_unhandled(irqnr);
}
-static asmlinkage void __exception_irq_entry gic_handle_irq(struct pt_regs *regs)
+static asmlinkage notrace void __exception_irq_entry gic_handle_irq(struct
pt_regs *regs)
{
u32 irqnr;
That's interesting. Do you have any out of tree patch that actually
makes use of the pseudo-NMI feature? Without those patches, the
behaviour should stay unchanged.
I am at commit 1a9df9e29c2afecf6e3089442d429b377279ca3c. No more
patches, and this is the most confusing. Just out of curiosity, I
wanted to run Julien's "Use NMI for perf interrupt" patch (posted
on the mailing list), so I have to enable NMI first.
That said, with
1) Select Kernel Feature -> Support for NMI-like interrupts
2) Set "irqchip.gicv3_pseudo_nmi=1" on the kernel command line
3) No pseudo-NMIs have been generated at all
and this issue was hit.
thanks,
zenghui
On the other hand, if you can generate pseudo-NMIs, you could end-up
tracing gic_handle_irq whilst being inside the tracing code with
interrupts being notionally disabled (and that could be pretty bad).
So, patches or no patches?
Thanks,
M.
