2018-04-24 15:53:09

by Jan Kiszka

Subject: [PATCH] kmemleak: Report if we need to tune KMEMLEAK_EARLY_LOG_SIZE

...rather than just mysteriously disabling it.

Signed-off-by: Jan Kiszka <[email protected]>
---
mm/kmemleak.c | 1 +
1 file changed, 1 insertion(+)

diff --git a/mm/kmemleak.c b/mm/kmemleak.c
index 9a085d525bbc..156c0c69cc5c 100644
--- a/mm/kmemleak.c
+++ b/mm/kmemleak.c
@@ -863,6 +863,7 @@ static void __init log_early(int op_type, const void *ptr, size_t size,

if (crt_early_log >= ARRAY_SIZE(early_log)) {
crt_early_log++;
+ pr_warn("Too many early logs\n");
kmemleak_disable();
return;
}
--
2.13.6

2018-04-24 15:56:34

by Catalin Marinas

Subject: Re: [PATCH] kmemleak: Report if we need to tune KMEMLEAK_EARLY_LOG_SIZE

On Tue, Apr 24, 2018 at 05:51:15PM +0200, Jan Kiszka wrote:
> ...rather than just mysteriously disabling it.
>
> Signed-off-by: Jan Kiszka <[email protected]>
> ---
> mm/kmemleak.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/mm/kmemleak.c b/mm/kmemleak.c
> index 9a085d525bbc..156c0c69cc5c 100644
> --- a/mm/kmemleak.c
> +++ b/mm/kmemleak.c
> @@ -863,6 +863,7 @@ static void __init log_early(int op_type, const void *ptr, size_t size,
>
> if (crt_early_log >= ARRAY_SIZE(early_log)) {
> crt_early_log++;
> + pr_warn("Too many early logs\n");

That's already printed, though later where we have an idea of how big the early
log needs to be:

if (crt_early_log > ARRAY_SIZE(early_log))
pr_warn("Early log buffer exceeded (%d), please increase DEBUG_KMEMLEAK_EARLY_LOG_SIZE\n",
crt_early_log);

--
Catalin

2018-04-24 16:12:14

by Jan Kiszka

Subject: Re: [PATCH] kmemleak: Report if we need to tune KMEMLEAK_EARLY_LOG_SIZE

On 2018-04-24 17:55, Catalin Marinas wrote:
> On Tue, Apr 24, 2018 at 05:51:15PM +0200, Jan Kiszka wrote:
>> ...rather than just mysteriously disabling it.
>>
>> Signed-off-by: Jan Kiszka <[email protected]>
>> ---
>> mm/kmemleak.c | 1 +
>> 1 file changed, 1 insertion(+)
>>
>> diff --git a/mm/kmemleak.c b/mm/kmemleak.c
>> index 9a085d525bbc..156c0c69cc5c 100644
>> --- a/mm/kmemleak.c
>> +++ b/mm/kmemleak.c
>> @@ -863,6 +863,7 @@ static void __init log_early(int op_type, const void *ptr, size_t size,
>>
>> if (crt_early_log >= ARRAY_SIZE(early_log)) {
>> crt_early_log++;
>> + pr_warn("Too many early logs\n");
>
> That's already printed, though later where we have an idea of how big the early
> log needs to be:
>
> if (crt_early_log > ARRAY_SIZE(early_log))
> pr_warn("Early log buffer exceeded (%d), please increase DEBUG_KMEMLEAK_EARLY_LOG_SIZE\n",
> crt_early_log);
>

Well, that's good, but where you read "detector disabled", there is no
hint on that. I missed that because subsystems tend to not do any
further actions after telling they are off.

BTW, my system (virtual ARM64 target) required 26035 entries, which is a
few orders of magnitude above the default and pretty close the the
limit. What's causing this? Other debug options?

Jan

--
Siemens AG, Corporate Technology, CT RDA IOT SES-DE
Corporate Competence Center Embedded Linux