On 08/12, Andrii Nakryiko wrote: > > Avoid taking refcount on uprobe in prepare_uretprobe(), instead take > uretprobe-specific SRCU lock and keep it active as kernel transfers > control back to user space. ... > include/linux/uprobes.h | 49 ++++++- > kernel/events/uprobes.c | 294 ++++++++++++++++++++++++++++++++++------ > 2 files changed, 301 insertions(+), 42 deletions(-) Oh. To be honest I don't like this patch. I would like to know what other reviewers think, but to me it adds too many complications that I can't even fully understand... And how much does it help performance-wise? I'll try to take another look, and I'll try to think about other approaches, not that I have something better in mind... But lets forgets this patch for the moment. The next one adds even more complications, and I think it doesn't make sense. As I have already mentioned in the previous discussions, we can simply kill utask->active_uprobe. And utask->auprobe. So can't we start with the patch below? On top of your 08/13. It doesn't kill utask->auprobe yet, this needs a bit more trivial changes. What do you think? Oleg. -------------------------------------------------------------------------------
