Re: [PATCH] livepatch: Avoid CPU hogging with cond_resched

Adding modules + BPF list and maintainers to this thread.

David Vernet [off-list ref] wrote on Wed [2021-Dec-29 13:56:47 -0800]:

On Thu, Dec 30, 2021 at 09:46:50AM IST, David Vernet wrote:

Adding modules + BPF list and maintainers to this thread.

Thanks for the Cc, but I'm dropping my patch for the next revision, so there
should be no conflicts with this one.

[...]

--
Kartikeya

On Wed 2021-12-29 20:16:50, David Vernet wrote:

Adding modules + BPF list and maintainers to this thread.

David Vernet [off-list ref] wrote on Wed [2021-Dec-29 13:56:47 -0800]:

quoted

When initializing a 'struct klp_object' in klp_init_object_loaded(), and
performing relocations in klp_resolve_symbols(), klp_find_object_symbol()
is invoked to look up the address of a symbol in an already-loaded module
(or vmlinux). This, in turn, calls kallsyms_on_each_symbol() or
module_kallsyms_on_each_symbol() to find the address of the symbol that is
being patched.

It turns out that symbol lookups often take up the most CPU time when
enabling and disabling a patch, and may hog the CPU and cause other tasks
on that CPU's runqueue to starve -- even in paths where interrupts are
enabled.  For example, under certain workloads, enabling a KLP patch with
many objects or functions may cause ksoftirqd to be starved, and thus for

    ^^^^^^^^^^^^^^^^^^^^^^^^^
This suggests that a single kallsyms_on_each_symbol() is not a big
problem. cond_resched() might be called non-necessarily often there.
I wonder if it would be enough to add cond_resched() into the two
loops calling klp_find_object_symbol().

That said, kallsyms_on_each_symbol() is a slow path and there might
be many symbols. So, it might be the right place.

I am just thinking loudly. I do not have strong opinion. I am fine
with any cond_resched() location that solves the problem. Feel free
to use:

Acked-by: Petr Mladek <pmladek@suse.com>

Best Regards,
Petr

quoted

interrupts to be backlogged and delayed. This may end up causing TCP
retransmits on the host where the KLP patch is being applied, and in
general, may cause any interrupts serviced by softirqd to be delayed while
the patch is being applied.

So as to ensure that kallsyms_on_each_symbol() does not end up hogging the
CPU, this patch adds a call to cond_resched() in kallsyms_on_each_symbol()
and module_kallsyms_on_each_symbol(), which are invoked when doing a symbol
lookup in vmlinux and a module respectively.  Without this patch, if a
live-patch is applied on a 36-core Intel host with heavy TCP traffic, a
~10x spike is observed in TCP retransmits while the patch is being applied.
Additionally, collecting sched events with perf indicates that ksoftirqd is
awakened ~1.3 seconds before it's eventually scheduled.  With the patch, no
increase in TCP retransmit events is observed, and ksoftirqd is scheduled
shortly after it's awakened.

Signed-off-by: David Vernet <void@manifault.com>
---
 kernel/kallsyms.c | 1 +
 kernel/module.c   | 2 ++
 2 files changed, 3 insertions(+)

diff --git a/kernel/kallsyms.c b/kernel/kallsyms.c
index 0ba87982d017..2a9afe484aec 100644
--- a/kernel/kallsyms.c
+++ b/kernel/kallsyms.c

@@ -223,6 +223,7 @@ int kallsyms_on_each_symbol(int (*fn)(void *, const char *, struct module *,
 		ret = fn(data, namebuf, NULL, kallsyms_sym_address(i));
 		if (ret != 0)
 			return ret;
+		cond_resched();
 	}
 	return 0;
 }

diff --git a/kernel/module.c b/kernel/module.c
index 40ec9a030eec..c96160f7f3f5 100644
--- a/kernel/module.c
+++ b/kernel/module.c

@@ -4462,6 +4462,8 @@ int module_kallsyms_on_each_symbol(int (*fn)(void *, const char *,
 				 mod, kallsyms_symbol_value(sym));
 			if (ret != 0)
 				goto out;
+
+			cond_resched();
 		}
 	}
 out:

-- 
2.30.2

Apologies all for the delayed response -- I was still on holiday last week.

Petr Mladek [off-list ref] wrote on Mon [2022-Jan-03 17:04:31 +0100]:

quoted

quoted

It turns out that symbol lookups often take up the most CPU time when
enabling and disabling a patch, and may hog the CPU and cause other tasks
on that CPU's runqueue to starve -- even in paths where interrupts are
enabled.  For example, under certain workloads, enabling a KLP patch with
many objects or functions may cause ksoftirqd to be starved, and thus for

    ^^^^^^^^^^^^^^^^^^^^^^^^^
This suggests that a single kallsyms_on_each_symbol() is not a big
problem. cond_resched() might be called non-necessarily often there.
I wonder if it would be enough to add cond_resched() into the two
loops calling klp_find_object_symbol().

In the initial version of the patch I was intending to send out, I actually
had the cond_resched() in klp_find_object_symbol(). Having it there did
appear to fix the ksoftirqd starvation issue, but I elected to put it in
klp_find_object_symbol() after Chris (cc'd) suggested it because
cond_resched() is so lightweight, and it didn't affect the runtime for
livepatching in my experiments.

That said, kallsyms_on_each_symbol() is a slow path and there might
be many symbols. So, it might be the right place.

Yes, my thinking was that because it didn't seem to affect throughput, and
because it would could potentially cause the same ssue to occur if it were
ever called elsewhere, that this was the correct place for it.