[PATCH] test_bpf: add a schedule point

[PATCH] test_bpf: add a schedule point

[Eric Dumazet]

From: Eric Dumazet <edumazet@google.com>

test_bpf() is taking 1.6 seconds nowadays, it is time
to add a schedule point in it.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 lib/test_bpf.c |    2 ++
 1 file changed, 2 insertions(+)

diff --git a/lib/test_bpf.c b/lib/test_bpf.c
index b4e22345963f339ffe05c974bc111ae7da9dc58f..e6f550608d7220f434b45828549d36820923a51d 100644
--- a/lib/test_bpf.c
+++ b/lib/test_bpf.c
@@ -24,6 +24,7 @@
 #include <linux/if_vlan.h>
 #include <linux/random.h>
 #include <linux/highmem.h>
+#include <linux/sched.h>
 
 /* General test specific settings */
 #define MAX_SUBTESTS	3
@@ -6582,6 +6583,7 @@ static __init int test_bpf(void)
 		struct bpf_prog *fp;
 		int err;
 
+		cond_resched();
 		if (exclude_test(i))
 			continue;
 

Re: [PATCH] test_bpf: add a schedule point

[Daniel Borkmann]

On 02/26/2018 07:52 PM, Eric Dumazet wrote:
> From: Eric Dumazet <edumazet@google.com>
> 
> test_bpf() is taking 1.6 seconds nowadays, it is time
> to add a schedule point in it.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>

Applied to bpf tree, thanks Eric!

Re: [PATCH] test_bpf: add a schedule point

[Eric Dumazet]

On Mon, 2018-02-26 at 21:11 +0100, Daniel Borkmann wrote:
> On 02/26/2018 07:52 PM, Eric Dumazet wrote:
> > From: Eric Dumazet <edumazet@google.com>
> > 
> > test_bpf() is taking 1.6 seconds nowadays, it is time
> > to add a schedule point in it.
> > 
> > Signed-off-by: Eric Dumazet <edumazet@google.com>
> 
> Applied to bpf tree, thanks Eric!

Thanks Daniel

Note that some BPF programs are quite expensive

[  173.447471] test_bpf: #264 BPF_MAXINSNS: Call heavy transformations jited:1 19248 18548 PASS
jited:1 12519 PASS
[  173.509228] test_bpf: #269 BPF_MAXINSNS: ld_abs+get_processor_id jited:1 20896 PASS

So we can still consume ~200 ms per test, without cond_resched()

Maybe reducing MAX_TESTRUNS from 10000 to 1000 would be the next step ?

Re: [PATCH] test_bpf: add a schedule point

[Daniel Borkmann]

On 02/27/2018 01:12 AM, Eric Dumazet wrote:
> On Mon, 2018-02-26 at 21:11 +0100, Daniel Borkmann wrote:
>> On 02/26/2018 07:52 PM, Eric Dumazet wrote:
>>> From: Eric Dumazet <edumazet@google.com>
>>>
>>> test_bpf() is taking 1.6 seconds nowadays, it is time
>>> to add a schedule point in it.
>>>
>>> Signed-off-by: Eric Dumazet <edumazet@google.com>
>>
>> Applied to bpf tree, thanks Eric!
> 
> Thanks Daniel
> 
> Note that some BPF programs are quite expensive
> 
> [  173.447471] test_bpf: #264 BPF_MAXINSNS: Call heavy transformations jited:1 19248 18548 PASS
> jited:1 12519 PASS
> [  173.509228] test_bpf: #269 BPF_MAXINSNS: ld_abs+get_processor_id jited:1 20896 PASS
> 
> So we can still consume ~200 ms per test, without cond_resched()
> 
> Maybe reducing MAX_TESTRUNS from 10000 to 1000 would be the next step ?

Yeah, that's totally fine with me, please feel free to send a patch. Another step on
todo is to reduce the test cases from test_bpf and move them into the test_verifier's
run-time testing where applicable. Would be nice if at some point we can get rid of
test_bpf and have everything consolidated within test_verifier.

[PATCH net] test_bpf: reduce MAX_TESTRUNS

[Eric Dumazet]

From: Eric Dumazet <edumazet@google.com>

For tests that are using the maximal number of BPF instruction, each
run takes 20 usec. Looping 10,000 times on them totals 200 ms, which
is bad when the loop is not preemptible.

test_bpf: #264 BPF_MAXINSNS: Call heavy transformations jited:1 19248
18548 PASS
test_bpf: #269 BPF_MAXINSNS: ld_abs+get_processor_id jited:1 20896 PASS

Lets divide by ten the number of iterations, so that max latency is
20ms. We could use need_resched() to break the loop earlier if we
believe 20 ms is too much.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 lib/test_bpf.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/lib/test_bpf.c b/lib/test_bpf.c
index b4e22345963f339ffe05c974bc111ae7da9dc58f..d8dc618b223bb5168b0c7e9d9f4316eef3fbfa34 100644
--- a/lib/test_bpf.c
+++ b/lib/test_bpf.c
@@ -27,7 +27,7 @@
 
 /* General test specific settings */
 #define MAX_SUBTESTS	3
-#define MAX_TESTRUNS	10000
+#define MAX_TESTRUNS	1000
 #define MAX_DATA	128
 #define MAX_INSNS	512
 #define MAX_K		0xffffFFFF

Re: [PATCH net] test_bpf: reduce MAX_TESTRUNS

[Daniel Borkmann]

On 02/28/2018 05:39 PM, Eric Dumazet wrote:
> From: Eric Dumazet <edumazet@google.com>
> 
> For tests that are using the maximal number of BPF instruction, each
> run takes 20 usec. Looping 10,000 times on them totals 200 ms, which
> is bad when the loop is not preemptible.
> 
> test_bpf: #264 BPF_MAXINSNS: Call heavy transformations jited:1 19248
> 18548 PASS
> test_bpf: #269 BPF_MAXINSNS: ld_abs+get_processor_id jited:1 20896 PASS
> 
> Lets divide by ten the number of iterations, so that max latency is
> 20ms. We could use need_resched() to break the loop earlier if we
> believe 20 ms is too much.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>

Applied to bpf tree, thanks Eric!

Re: [PATCH net] test_bpf: reduce MAX_TESTRUNS

[David Miller]

From: Eric Dumazet <eric.dumazet@gmail.com>
Date: Wed, 28 Feb 2018 08:39:20 -0800

> From: Eric Dumazet <edumazet@google.com>
> 
> For tests that are using the maximal number of BPF instruction, each
> run takes 20 usec. Looping 10,000 times on them totals 200 ms, which
> is bad when the loop is not preemptible.
> 
> test_bpf: #264 BPF_MAXINSNS: Call heavy transformations jited:1 19248
> 18548 PASS
> test_bpf: #269 BPF_MAXINSNS: ld_abs+get_processor_id jited:1 20896 PASS
> 
> Lets divide by ten the number of iterations, so that max latency is
> 20ms. We could use need_resched() to break the loop earlier if we
> believe 20 ms is too much.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>

Acked-by: David S. Miller <davem@davemloft.net>