[PATCH] taskstats: remove unneeded dead assignment

[PATCH] taskstats: remove unneeded dead assignment

[Lukas Bulwahn]

make clang-analyzer on x86_64 defconfig caught my attention with:

  kernel/taskstats.c:120:2: warning: Value stored to 'rc' is never read \
  [clang-analyzer-deadcode.DeadStores]
          rc = 0;
          ^

Commit d94a041519f3 ("taskstats: free skb, avoid returns in
send_cpu_listeners") made send_cpu_listeners() not return a value and
hence, the rc variable remained only to be used within the loop where
it is always assigned before read and it does not need any other
initialisation.

So, simply remove this unneeded dead initializing assignment.

As compilers will detect this unneeded assignment and optimize this anyway,
the resulting object code is identical before and after this change.

No functional change. No change to object code.

Signed-off-by: Lukas Bulwahn <lukas.bulwahn@gmail.com>
---
applies cleanly on current master and next-20201105

Balbir, please pick this minor non-urgent clean-up patch.

 kernel/taskstats.c | 1 -
 1 file changed, 1 deletion(-)

diff --git a/kernel/taskstats.c b/kernel/taskstats.c
index a2802b6ff4bb..bd18a7bf5276 100644
--- a/kernel/taskstats.c
+++ b/kernel/taskstats.c
@@ -117,7 +117,6 @@ static void send_cpu_listeners(struct sk_buff *skb,
 
 	genlmsg_end(skb, reply);
 
-	rc = 0;
 	down_read(&listeners->sem);
 	list_for_each_entry(s, &listeners->list, list) {
 		skb_next = NULL;
-- 
2.17.1

Re: [PATCH] taskstats: remove unneeded dead assignment

[Nathan Chancellor]

On Fri, Nov 06, 2020 at 07:22:10AM +0100, Lukas Bulwahn wrote:
> make clang-analyzer on x86_64 defconfig caught my attention with:
> 
>   kernel/taskstats.c:120:2: warning: Value stored to 'rc' is never read \
>   [clang-analyzer-deadcode.DeadStores]
>           rc = 0;
>           ^
> 
> Commit d94a041519f3 ("taskstats: free skb, avoid returns in
> send_cpu_listeners") made send_cpu_listeners() not return a value and
> hence, the rc variable remained only to be used within the loop where
> it is always assigned before read and it does not need any other
> initialisation.
> 
> So, simply remove this unneeded dead initializing assignment.
> 
> As compilers will detect this unneeded assignment and optimize this anyway,
> the resulting object code is identical before and after this change.
> 
> No functional change. No change to object code.
> 
> Signed-off-by: Lukas Bulwahn <lukas.bulwahn@gmail.com>

Question below.

Reviewed-by: Nathan Chancellor <natechancellor@gmail.com>

> ---
> applies cleanly on current master and next-20201105
> 
> Balbir, please pick this minor non-urgent clean-up patch.
> 
>  kernel/taskstats.c | 1 -
>  1 file changed, 1 deletion(-)
> 
> diff --git a/kernel/taskstats.c b/kernel/taskstats.c
> index a2802b6ff4bb..bd18a7bf5276 100644
> --- a/kernel/taskstats.c
> +++ b/kernel/taskstats.c
> @@ -117,7 +117,6 @@ static void send_cpu_listeners(struct sk_buff *skb,
>  
>  	genlmsg_end(skb, reply);
>  
> -	rc = 0;
>  	down_read(&listeners->sem);
>  	list_for_each_entry(s, &listeners->list, list) {

Would it be worth moving the scope of rc into the for loop, now that it
is only used there? Looks like it used to be used in the main function
scope before commit 053c095a82cf ("netlink: make nlmsg_end() and
genlmsg_end() void") but if this is removed, it is only used to check
the return of genlmsg_unicast within the list_for_each_entry loop. Not
sure that buys us anything but I know you have done it in patches
before so I thought it was worth considering.

>  		skb_next = NULL;
> -- 
> 2.17.1
> 

Cheers,
Nathan

Re: [PATCH] taskstats: remove unneeded dead assignment

[Lukas Bulwahn]

On Fri, 6 Nov 2020, Nathan Chancellor wrote:

> On Fri, Nov 06, 2020 at 07:22:10AM +0100, Lukas Bulwahn wrote:
> > make clang-analyzer on x86_64 defconfig caught my attention with:
> > 
> >   kernel/taskstats.c:120:2: warning: Value stored to 'rc' is never read \
> >   [clang-analyzer-deadcode.DeadStores]
> >           rc = 0;
> >           ^
> > 
> > Commit d94a041519f3 ("taskstats: free skb, avoid returns in
> > send_cpu_listeners") made send_cpu_listeners() not return a value and
> > hence, the rc variable remained only to be used within the loop where
> > it is always assigned before read and it does not need any other
> > initialisation.
> > 
> > So, simply remove this unneeded dead initializing assignment.
> > 
> > As compilers will detect this unneeded assignment and optimize this anyway,
> > the resulting object code is identical before and after this change.
> > 
> > No functional change. No change to object code.
> > 
> > Signed-off-by: Lukas Bulwahn <lukas.bulwahn@gmail.com>
> 
> Question below.
> 
> Reviewed-by: Nathan Chancellor <natechancellor@gmail.com>
> 
> > ---
> > applies cleanly on current master and next-20201105
> > 
> > Balbir, please pick this minor non-urgent clean-up patch.
> > 
> >  kernel/taskstats.c | 1 -
> >  1 file changed, 1 deletion(-)
> > 
> > diff --git a/kernel/taskstats.c b/kernel/taskstats.c
> > index a2802b6ff4bb..bd18a7bf5276 100644
> > --- a/kernel/taskstats.c
> > +++ b/kernel/taskstats.c
> > @@ -117,7 +117,6 @@ static void send_cpu_listeners(struct sk_buff *skb,
> >  
> >  	genlmsg_end(skb, reply);
> >  
> > -	rc = 0;
> >  	down_read(&listeners->sem);
> >  	list_for_each_entry(s, &listeners->list, list) {
> 
> Would it be worth moving the scope of rc into the for loop, now that it
> is only used there? Looks like it used to be used in the main function
> scope before commit 053c095a82cf ("netlink: make nlmsg_end() and
> genlmsg_end() void") but if this is removed, it is only used to check
> the return of genlmsg_unicast within the list_for_each_entry loop. Not
> sure that buys us anything but I know you have done it in patches
> before so I thought it was worth considering.
>

I thought about moving it into the local scope, but it is a purely 
cosmetic matter. Compilers are smart enough to generate the same code no 
matter where it is defined.
So, I always look around in the same file to determine if there is some 
kind of strong preference for very locally scoped variable definition or 
if they are generally just all defined at the function entry.

Depending on my gut feeling in which style the file has mainly been 
written, I then go with the one or other option. In this case, I went 
with just keeping the definition at the function entry.

There is really no strong rule, though, that I see serving as good 
indicator.

Thanks for your review.

Lukas

Re: [linux-safety] [PATCH] taskstats: remove unneeded dead assignment

[Sudip Mukherjee]

Hi Lukas,

On 06/11/2020 06:22, Lukas Bulwahn wrote:
> make clang-analyzer on x86_64 defconfig caught my attention with:
> 
>   kernel/taskstats.c:120:2: warning: Value stored to 'rc' is never read \
>   [clang-analyzer-deadcode.DeadStores]
>           rc = 0;
>           ^
> 
> Commit d94a041519f3 ("taskstats: free skb, avoid returns in
> send_cpu_listeners") made send_cpu_listeners() not return a value and
> hence, the rc variable remained only to be used within the loop where
> it is always assigned before read and it does not need any other
> initialisation.
> 
> So, simply remove this unneeded dead initializing assignment.

Might be better to remove 'rc' completely as it is only used for the if
condition now.

diff --git a/kernel/taskstats.c b/kernel/taskstats.c
index a2802b6ff4bb..63541f1ae04a 100644
--- a/kernel/taskstats.c
+++ b/kernel/taskstats.c
@@ -113,11 +113,10 @@ static void send_cpu_listeners(struct sk_buff *skb,
        struct listener *s, *tmp;
        struct sk_buff *skb_next, *skb_cur = skb;
        void *reply = genlmsg_data(genlhdr);
-       int rc, delcount = 0;
+       int delcount = 0;

        genlmsg_end(skb, reply);

-       rc = 0;
        down_read(&listeners->sem);
        list_for_each_entry(s, &listeners->list, list) {
                skb_next = NULL;
@@ -126,8 +125,8 @@ static void send_cpu_listeners(struct sk_buff *skb,
                        if (!skb_next)
                                break;
                }
-               rc = genlmsg_unicast(&init_net, skb_cur, s->pid);
-               if (rc == -ECONNREFUSED) {
+               if (genlmsg_unicast(&init_net, skb_cur, s->pid) ==
+                   -ECONNREFUSED) {
                        s->valid = 0;
                        delcount++;
                }




-- 
Regards
Sudip

Re: [linux-safety] [PATCH] taskstats: remove unneeded dead assignment

[Lukas Bulwahn]

On Fri, 6 Nov 2020, Sudip Mukherjee wrote:

> Hi Lukas,
> 
> On 06/11/2020 06:22, Lukas Bulwahn wrote:
> > make clang-analyzer on x86_64 defconfig caught my attention with:
> > 
> >   kernel/taskstats.c:120:2: warning: Value stored to 'rc' is never read \
> >   [clang-analyzer-deadcode.DeadStores]
> >           rc = 0;
> >           ^
> > 
> > Commit d94a041519f3 ("taskstats: free skb, avoid returns in
> > send_cpu_listeners") made send_cpu_listeners() not return a value and
> > hence, the rc variable remained only to be used within the loop where
> > it is always assigned before read and it does not need any other
> > initialisation.
> > 
> > So, simply remove this unneeded dead initializing assignment.
> 
> Might be better to remove 'rc' completely as it is only used for the if
> condition now.
> 
> diff --git a/kernel/taskstats.c b/kernel/taskstats.c
> index a2802b6ff4bb..63541f1ae04a 100644
> --- a/kernel/taskstats.c
> +++ b/kernel/taskstats.c
> @@ -113,11 +113,10 @@ static void send_cpu_listeners(struct sk_buff *skb,
>         struct listener *s, *tmp;
>         struct sk_buff *skb_next, *skb_cur = skb;
>         void *reply = genlmsg_data(genlhdr);
> -       int rc, delcount = 0;
> +       int delcount = 0;
> 
>         genlmsg_end(skb, reply);
> 
> -       rc = 0;
>         down_read(&listeners->sem);
>         list_for_each_entry(s, &listeners->list, list) {
>                 skb_next = NULL;
> @@ -126,8 +125,8 @@ static void send_cpu_listeners(struct sk_buff *skb,
>                         if (!skb_next)
>                                 break;
>                 }
> -               rc = genlmsg_unicast(&init_net, skb_cur, s->pid);
> -               if (rc == -ECONNREFUSED) {
> +               if (genlmsg_unicast(&init_net, skb_cur, s->pid) ==
> +                   -ECONNREFUSED) {

I thought about that as well; and I did not like that because of the ugly
line break in this condition.

I did not try but I bet (a beverage of your choice) that the object code
remains the same also for your suggested patch. Try to disprove my claim 
and possibly earn yourself a beverage when we meet...

Lukas

Re: [PATCH] taskstats: remove unneeded dead assignment

[Lukas Bulwahn]

On Fri, 6 Nov 2020, Lukas Bulwahn wrote:

> 
> 
> On Fri, 6 Nov 2020, Nathan Chancellor wrote:
> 
> > On Fri, Nov 06, 2020 at 07:22:10AM +0100, Lukas Bulwahn wrote:
> > > make clang-analyzer on x86_64 defconfig caught my attention with:
> > > 
> > >   kernel/taskstats.c:120:2: warning: Value stored to 'rc' is never read \
> > >   [clang-analyzer-deadcode.DeadStores]
> > >           rc = 0;
> > >           ^
> > > 
> > > Commit d94a041519f3 ("taskstats: free skb, avoid returns in
> > > send_cpu_listeners") made send_cpu_listeners() not return a value and
> > > hence, the rc variable remained only to be used within the loop where
> > > it is always assigned before read and it does not need any other
> > > initialisation.
> > > 
> > > So, simply remove this unneeded dead initializing assignment.
> > > 
> > > As compilers will detect this unneeded assignment and optimize this anyway,
> > > the resulting object code is identical before and after this change.
> > > 
> > > No functional change. No change to object code.
> > > 
> > > Signed-off-by: Lukas Bulwahn <lukas.bulwahn@gmail.com>
> > 
> > Question below.
> > 
> > Reviewed-by: Nathan Chancellor <natechancellor@gmail.com>
> > 
> > > ---
> > > applies cleanly on current master and next-20201105
> > > 
> > > Balbir, please pick this minor non-urgent clean-up patch.
> > > 
> > >  kernel/taskstats.c | 1 -
> > >  1 file changed, 1 deletion(-)
> > > 
> > > diff --git a/kernel/taskstats.c b/kernel/taskstats.c
> > > index a2802b6ff4bb..bd18a7bf5276 100644
> > > --- a/kernel/taskstats.c
> > > +++ b/kernel/taskstats.c
> > > @@ -117,7 +117,6 @@ static void send_cpu_listeners(struct sk_buff *skb,
> > >  
> > >  	genlmsg_end(skb, reply);
> > >  
> > > -	rc = 0;
> > >  	down_read(&listeners->sem);
> > >  	list_for_each_entry(s, &listeners->list, list) {
> > 
> > Would it be worth moving the scope of rc into the for loop, now that it
> > is only used there? Looks like it used to be used in the main function
> > scope before commit 053c095a82cf ("netlink: make nlmsg_end() and
> > genlmsg_end() void") but if this is removed, it is only used to check
> > the return of genlmsg_unicast within the list_for_each_entry loop. Not
> > sure that buys us anything but I know you have done it in patches
> > before so I thought it was worth considering.
> >
> 
> I thought about moving it into the local scope, but it is a purely 
> cosmetic matter. Compilers are smart enough to generate the same code no 
> matter where it is defined.
> So, I always look around in the same file to determine if there is some 
> kind of strong preference for very locally scoped variable definition or 
> if they are generally just all defined at the function entry.
> 
> Depending on my gut feeling in which style the file has mainly been 
> written, I then go with the one or other option. In this case, I went 
> with just keeping the definition at the function entry.
> 
> There is really no strong rule, though, that I see serving as good 
> indicator.
> 
> Thanks for your review.
>

More specifically, if I think rc should be only defined locally, I would 
probably need to apply the same argument to skb_next in this function and 
put that in local scope as well. That did not happen in the past, so I am 
not going to change that now neither. Hence, the change stays minimal 
invasive but and that is important: it makes clang-analyzer happy.

And a happy clang-analyzer will eventually point to real bugs :)

There are a few examples of dead store warnings that in the end really 
point to missing or wrong paths in some functions...

Lukas

Re: [linux-safety] [PATCH] taskstats: remove unneeded dead assignment

[Sudip Mukherjee]

Hi Lukas,

On 06/11/2020 10:31, Lukas Bulwahn wrote:
> 
> 
> On Fri, 6 Nov 2020, Sudip Mukherjee wrote:
> 
>> Hi Lukas,
>>

<snip>

> 
> I did not try but I bet (a beverage of your choice) that the object code
> remains the same also for your suggested patch. Try to disprove my claim 
> and possibly earn yourself a beverage when we meet...

Lets decide which beverage.. ;-)

Using gcc-7.2.0 for MIPS:

original:- ab81d3305d578c2568fbc73aad2f9e61  kernel/taskstats.o
After your change:- ab81d3305d578c2568fbc73aad2f9e61  kernel/taskstats.o
After my change:- 0acae2c8d72abd3e15bf805fccdca711  kernel/taskstats.o



-- 
Regards
Sudip

Re: [linux-safety] [PATCH] taskstats: remove unneeded dead assignment

[Lukas Bulwahn]

On Fri, 6 Nov 2020, Sudip Mukherjee wrote:

> Hi Lukas,
> 
> On 06/11/2020 10:31, Lukas Bulwahn wrote:
> > 
> > 
> > On Fri, 6 Nov 2020, Sudip Mukherjee wrote:
> > 
> >> Hi Lukas,
> >>
> 
> <snip>
> 
> > 
> > I did not try but I bet (a beverage of your choice) that the object code
> > remains the same also for your suggested patch. Try to disprove my claim 
> > and possibly earn yourself a beverage when we meet...
> 
> Lets decide which beverage.. ;-)
> 
> Using gcc-7.2.0 for MIPS:
> 
> original:- ab81d3305d578c2568fbc73aad2f9e61  kernel/taskstats.o
> After your change:- ab81d3305d578c2568fbc73aad2f9e61  kernel/taskstats.o
> After my change:- 0acae2c8d72abd3e15bf805fccdca711  kernel/taskstats.o
> 
>

Interesting, can you share the diff of the objdump before and after?

I bet it found now a different assignment from variables to registers or 
so; with the beverage at hand, we can then discuss if that is effectively 
still the same object code or not.

Thanks for confirming that my patch here really remains the same compared 
to before, even on MIPS :) I only checked x86-64...

Lukas 

> 
> -- 
> Regards
> Sudip
> 
> 
> 
> 
> 
> 

Re: [linux-safety] [PATCH] taskstats: remove unneeded dead assignment

[Dan Carpenter]

On Fri, Nov 06, 2020 at 12:04:53PM +0000, Sudip Mukherjee wrote:
> Hi Lukas,
> 
> On 06/11/2020 10:31, Lukas Bulwahn wrote:
> > 
> > 
> > On Fri, 6 Nov 2020, Sudip Mukherjee wrote:
> > 
> >> Hi Lukas,
> >>
> 
> <snip>
> 
> > 
> > I did not try but I bet (a beverage of your choice) that the object code
> > remains the same also for your suggested patch. Try to disprove my claim 
> > and possibly earn yourself a beverage when we meet...
> 
> Lets decide which beverage.. ;-)
> 
> Using gcc-7.2.0 for MIPS:
> 
> original:- ab81d3305d578c2568fbc73aad2f9e61  kernel/taskstats.o
> After your change:- ab81d3305d578c2568fbc73aad2f9e61  kernel/taskstats.o
> After my change:- 0acae2c8d72abd3e15bf805fccdca711  kernel/taskstats.o

I'm surprised the line numbers from the printks aren't affecting it...

I personally prefer Lukas's patch.  "rc" should be function scope...

regards,
dan carpenter