2017-12-23 15:29:14

by Pravin Shedge

[permalink] [raw]
Subject: [PATCH 2/2] lib: Remove module auto-unloading which encourages inconsistent behaviour

The module auto-unload might seem like a nice optimization, but
it encourages inconsistent behaviour. And behaviour that is
different from all other normal modules.

rbtree_test.c and percpu_test.c returns -EAGAIN from module_init()
on successful completion. Normal module return 0 on success but
returning -EAGAIN perform auto module unloading which seem like
a nice optimization but it brings inconsistent behaviour as well.

I face the similar problem in my previous patch, "Paul Gortmaker"
gives nice review comment on module auto-unloading brings
inconsistent behaviour.

Imagine something as simple as this:

for i in $LIST ; do
modprobe $i
lsmod | grep -q $i
if [ $? != 0 ]; then echo Module $i is not present! ; fi
done

OK, not ideal code, ignoring the modprobe return -- but what it reports
is true --
rbtree_test.c and percpu_test.c test module (if it passed) will NOT be
present.

All other modules from linux/lib/* follow the same semantics that
module_init() returns 0 on successful completion & let external entity
like rmmod do the module unloading task.

Signed-off-by: Pravin Shedge <[email protected]>
---
lib/percpu_test.c | 3 ++-
lib/rbtree_test.c | 2 +-
2 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/lib/percpu_test.c b/lib/percpu_test.c
index 0b5d14d..57a637b 100644
--- a/lib/percpu_test.c
+++ b/lib/percpu_test.c
@@ -123,7 +123,8 @@ static int __init percpu_test_init(void)
preempt_enable();

pr_info("percpu test done\n");
- return -EAGAIN; /* Fail will directly unload the module */
+
+ return 0;
}

static void __exit percpu_test_exit(void)
diff --git a/lib/rbtree_test.c b/lib/rbtree_test.c
index 7d36c1e..5c52112 100644
--- a/lib/rbtree_test.c
+++ b/lib/rbtree_test.c
@@ -394,7 +394,7 @@ static int __init rbtree_test_init(void)

kfree(nodes);

- return -EAGAIN; /* Fail will directly unload the module */
+ return 0;
}

static void __exit rbtree_test_exit(void)
--
2.7.4


2017-12-29 16:47:07

by Andy Shevchenko

[permalink] [raw]
Subject: Re: [PATCH 2/2] lib: Remove module auto-unloading which encourages inconsistent behaviour

On Sat, Dec 23, 2017 at 5:28 PM, Pravin Shedge
<[email protected]> wrote:
> The module auto-unload might seem like a nice optimization, but
> it encourages inconsistent behaviour. And behaviour that is
> different from all other normal modules.
>
> rbtree_test.c and percpu_test.c returns -EAGAIN from module_init()
> on successful completion. Normal module return 0 on success but
> returning -EAGAIN perform auto module unloading which seem like
> a nice optimization but it brings inconsistent behaviour as well.
>
> I face the similar problem in my previous patch, "Paul Gortmaker"
> gives nice review comment on module auto-unloading brings
> inconsistent behaviour.
>
> Imagine something as simple as this:
>
> for i in $LIST ; do
> modprobe $i
> lsmod | grep -q $i
> if [ $? != 0 ]; then echo Module $i is not present! ; fi
> done
>
> OK, not ideal code, ignoring the modprobe return -- but what it reports
> is true --
> rbtree_test.c and percpu_test.c test module (if it passed) will NOT be
> present.
>
> All other modules from linux/lib/* follow the same semantics that
> module_init() returns 0 on successful completion & let external entity
> like rmmod do the module unloading task.
>

Suggested-by: Andrew ... ?

> Signed-off-by: Pravin Shedge <[email protected]>
> ---
> lib/percpu_test.c | 3 ++-
> lib/rbtree_test.c | 2 +-
> 2 files changed, 3 insertions(+), 2 deletions(-)
>
> diff --git a/lib/percpu_test.c b/lib/percpu_test.c
> index 0b5d14d..57a637b 100644
> --- a/lib/percpu_test.c
> +++ b/lib/percpu_test.c
> @@ -123,7 +123,8 @@ static int __init percpu_test_init(void)
> preempt_enable();
>
> pr_info("percpu test done\n");
> - return -EAGAIN; /* Fail will directly unload the module */
> +
> + return 0;
> }
>
> static void __exit percpu_test_exit(void)
> diff --git a/lib/rbtree_test.c b/lib/rbtree_test.c
> index 7d36c1e..5c52112 100644
> --- a/lib/rbtree_test.c
> +++ b/lib/rbtree_test.c
> @@ -394,7 +394,7 @@ static int __init rbtree_test_init(void)
>
> kfree(nodes);
>
> - return -EAGAIN; /* Fail will directly unload the module */
> + return 0;
> }
>
> static void __exit rbtree_test_exit(void)
> --
> 2.7.4
>



--
With Best Regards,
Andy Shevchenko

2018-01-01 17:10:39

by Pravin Shedge

[permalink] [raw]
Subject: Re: [PATCH 2/2] lib: Remove module auto-unloading which encourages inconsistent behaviour

On Fri, Dec 29, 2017 at 10:17 PM, Andy Shevchenko
<[email protected]> wrote:
> On Sat, Dec 23, 2017 at 5:28 PM, Pravin Shedge
> <[email protected]> wrote:
>> The module auto-unload might seem like a nice optimization, but
>> it encourages inconsistent behaviour. And behaviour that is
>> different from all other normal modules.
>>
>> rbtree_test.c and percpu_test.c returns -EAGAIN from module_init()
>> on successful completion. Normal module return 0 on success but
>> returning -EAGAIN perform auto module unloading which seem like
>> a nice optimization but it brings inconsistent behaviour as well.
>>
>> I face the similar problem in my previous patch, "Paul Gortmaker"
>> gives nice review comment on module auto-unloading brings
>> inconsistent behaviour.
>>
>> Imagine something as simple as this:
>>
>> for i in $LIST ; do
>> modprobe $i
>> lsmod | grep -q $i
>> if [ $? != 0 ]; then echo Module $i is not present! ; fi
>> done
>>
>> OK, not ideal code, ignoring the modprobe return -- but what it reports
>> is true --
>> rbtree_test.c and percpu_test.c test module (if it passed) will NOT be
>> present.
>>
>> All other modules from linux/lib/* follow the same semantics that
>> module_init() returns 0 on successful completion & let external entity
>> like rmmod do the module unloading task.
>>
>
> Suggested-by: Andrew ... ?
Hi Andy,

Please check the reference patch "[PATCH] lib: add module unload
support to sort tests" this patch was added for lib/test_sort.c file.
test_sort.c missing unloading feature so I added auto-unload feature
for the same. But after reviewers comment I am agree on the part that
module auto-unload feature will break the consistency from users point
of view.

The reviewers for this patch are Andrew, Paul Gortmaker.

Thanks & Regards,
PraviN



>
>> Signed-off-by: Pravin Shedge <[email protected]>
>> ---
>> lib/percpu_test.c | 3 ++-
>> lib/rbtree_test.c | 2 +-
>> 2 files changed, 3 insertions(+), 2 deletions(-)
>>
>> diff --git a/lib/percpu_test.c b/lib/percpu_test.c
>> index 0b5d14d..57a637b 100644
>> --- a/lib/percpu_test.c
>> +++ b/lib/percpu_test.c
>> @@ -123,7 +123,8 @@ static int __init percpu_test_init(void)
>> preempt_enable();
>>
>> pr_info("percpu test done\n");
>> - return -EAGAIN; /* Fail will directly unload the module */
>> +
>> + return 0;
>> }
>>
>> static void __exit percpu_test_exit(void)
>> diff --git a/lib/rbtree_test.c b/lib/rbtree_test.c
>> index 7d36c1e..5c52112 100644
>> --- a/lib/rbtree_test.c
>> +++ b/lib/rbtree_test.c
>> @@ -394,7 +394,7 @@ static int __init rbtree_test_init(void)
>>
>> kfree(nodes);
>>
>> - return -EAGAIN; /* Fail will directly unload the module */
>> + return 0;
>> }
>>
>> static void __exit rbtree_test_exit(void)
>> --
>> 2.7.4
>>
>
>
>
> --
> With Best Regards,
> Andy Shevchenko