2020-06-15 04:03:22

by Jason Yan

[permalink] [raw]
Subject: [PATCH] erofs: Eliminate usage of uninitialized_var() macro

This is an effort to eliminate the uninitialized_var() macro[1].

The use of this macro is the wrong solution because it forces off ANY
analysis by the compiler for a given variable. It even masks "unused
variable" warnings.

Quoted from Linus[2]:

"It's a horrible thing to use, in that it adds extra cruft to the
source code, and then shuts up a compiler warning (even the _reliable_
warnings from gcc)."

The gcc option "-Wmaybe-uninitialized" has been disabled and this change
will not produce any warnnings even with "make W=1".

[1] https://github.com/KSPP/linux/issues/81
[2] https://lore.kernel.org/lkml/CA+55aFz2500WfbKXAx8s67wrm9=yVJu65TpLgN_ybYNv0VEOKA@mail.gmail.com/

Cc: Kees Cook <[email protected]>
Cc: Chao Yu <[email protected]>
Signed-off-by: Jason Yan <[email protected]>
---
fs/erofs/data.c | 4 ++--
fs/erofs/zdata.c | 2 +-
2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/fs/erofs/data.c b/fs/erofs/data.c
index 64b56c7df023..d0542151e8c4 100644
--- a/fs/erofs/data.c
+++ b/fs/erofs/data.c
@@ -265,7 +265,7 @@ static inline struct bio *erofs_read_raw_page(struct bio *bio,
*/
static int erofs_raw_access_readpage(struct file *file, struct page *page)
{
- erofs_off_t uninitialized_var(last_block);
+ erofs_off_t last_block;
struct bio *bio;

trace_erofs_readpage(page, true);
@@ -282,7 +282,7 @@ static int erofs_raw_access_readpage(struct file *file, struct page *page)

static void erofs_raw_access_readahead(struct readahead_control *rac)
{
- erofs_off_t uninitialized_var(last_block);
+ erofs_off_t last_block;
struct bio *bio = NULL;
struct page *page;

diff --git a/fs/erofs/zdata.c b/fs/erofs/zdata.c
index be50a4d9d273..24a26aaf847f 100644
--- a/fs/erofs/zdata.c
+++ b/fs/erofs/zdata.c
@@ -1161,7 +1161,7 @@ static void z_erofs_submit_queue(struct super_block *sb,
struct z_erofs_decompressqueue *q[NR_JOBQUEUES];
void *bi_private;
/* since bio will be NULL, no need to initialize last_index */
- pgoff_t uninitialized_var(last_index);
+ pgoff_t last_index;
unsigned int nr_bios = 0;
struct bio *bio = NULL;

--
2.25.4


2020-06-15 07:30:34

by Gao Xiang

[permalink] [raw]
Subject: Re: [PATCH] erofs: Eliminate usage of uninitialized_var() macro

Hi Jason,

On Mon, Jun 15, 2020 at 12:01:41PM +0800, Jason Yan wrote:
> This is an effort to eliminate the uninitialized_var() macro[1].
>
> The use of this macro is the wrong solution because it forces off ANY
> analysis by the compiler for a given variable. It even masks "unused
> variable" warnings.
>
> Quoted from Linus[2]:
>
> "It's a horrible thing to use, in that it adds extra cruft to the
> source code, and then shuts up a compiler warning (even the _reliable_
> warnings from gcc)."
>
> The gcc option "-Wmaybe-uninitialized" has been disabled and this change
> will not produce any warnnings even with "make W=1".
>
> [1] https://github.com/KSPP/linux/issues/81
> [2] https://lore.kernel.org/lkml/CA+55aFz2500WfbKXAx8s67wrm9=yVJu65TpLgN_ybYNv0VEOKA@mail.gmail.com/
>
> Cc: Kees Cook <[email protected]>
> Cc: Chao Yu <[email protected]>
> Signed-off-by: Jason Yan <[email protected]>
> ---

I'm fine with the patch since "-Wmaybe-uninitialized" has been disabled and
I've also asked Kees for it in private previously.

I still remembered that Kees sent out a treewide patch. Sorry about that
I don't catch up it... But what is wrong with the original patchset?

Thanks,
Gao Xiang

2020-06-15 07:45:54

by Jason Yan

[permalink] [raw]
Subject: Re: [PATCH] erofs: Eliminate usage of uninitialized_var() macro



?? 2020/6/15 15:25, Gao Xiang д??:
> Hi Jason,
>
> On Mon, Jun 15, 2020 at 12:01:41PM +0800, Jason Yan wrote:
>> This is an effort to eliminate the uninitialized_var() macro[1].
>>
>> The use of this macro is the wrong solution because it forces off ANY
>> analysis by the compiler for a given variable. It even masks "unused
>> variable" warnings.
>>
>> Quoted from Linus[2]:
>>
>> "It's a horrible thing to use, in that it adds extra cruft to the
>> source code, and then shuts up a compiler warning (even the _reliable_
>> warnings from gcc)."
>>
>> The gcc option "-Wmaybe-uninitialized" has been disabled and this change
>> will not produce any warnnings even with "make W=1".
>>
>> [1] https://github.com/KSPP/linux/issues/81
>> [2] https://lore.kernel.org/lkml/CA+55aFz2500WfbKXAx8s67wrm9=yVJu65TpLgN_ybYNv0VEOKA@mail.gmail.com/
>>
>> Cc: Kees Cook <[email protected]>
>> Cc: Chao Yu <[email protected]>
>> Signed-off-by: Jason Yan <[email protected]>
>> ---
>
> I'm fine with the patch since "-Wmaybe-uninitialized" has been disabled and
> I've also asked Kees for it in private previously.
>
> I still remembered that Kees sent out a treewide patch. Sorry about that
> I don't catch up it... But what is wrong with the original patchset?
>

Yes, Kees has remind me of that and I will let him handle it. So you can
ignore this patch.

Thanks,
Jason

> Thanks,
> Gao Xiang
>
>
> .
>

2020-06-15 08:10:18

by Gao Xiang

[permalink] [raw]
Subject: Re: [PATCH] erofs: Eliminate usage of uninitialized_var() macro

On Mon, Jun 15, 2020 at 03:43:09PM +0800, Jason Yan wrote:
>
>
> ?? 2020/6/15 15:25, Gao Xiang д??:
> > Hi Jason,
> >
> > On Mon, Jun 15, 2020 at 12:01:41PM +0800, Jason Yan wrote:
> > > This is an effort to eliminate the uninitialized_var() macro[1].
> > >
> > > The use of this macro is the wrong solution because it forces off ANY
> > > analysis by the compiler for a given variable. It even masks "unused
> > > variable" warnings.
> > >
> > > Quoted from Linus[2]:
> > >
> > > "It's a horrible thing to use, in that it adds extra cruft to the
> > > source code, and then shuts up a compiler warning (even the _reliable_
> > > warnings from gcc)."
> > >
> > > The gcc option "-Wmaybe-uninitialized" has been disabled and this change
> > > will not produce any warnnings even with "make W=1".
> > >
> > > [1] https://github.com/KSPP/linux/issues/81
> > > [2] https://lore.kernel.org/lkml/CA+55aFz2500WfbKXAx8s67wrm9=yVJu65TpLgN_ybYNv0VEOKA@mail.gmail.com/
> > >
> > > Cc: Kees Cook <[email protected]>
> > > Cc: Chao Yu <[email protected]>
> > > Signed-off-by: Jason Yan <[email protected]>
> > > ---
> >
> > I'm fine with the patch since "-Wmaybe-uninitialized" has been disabled and
> > I've also asked Kees for it in private previously.
> >
> > I still remembered that Kees sent out a treewide patch. Sorry about that
> > I don't catch up it... But what is wrong with the original patchset?
> >
>
> Yes, Kees has remind me of that and I will let him handle it. So you can
> ignore this patch.

Okay, I was just wondering if this part should be send out via EROFS tree
for this cycle. However if there was an automatic generated patch by Kees,
I think perhaps Linus could pick them out directly. But anyway, both ways
are fine with me. ;) Ping me when needed.

Thanks,
Gao Xiang

>
> Thanks,
> Jason
>
> > Thanks,
> > Gao Xiang
> >
> >
> > .
> >
>

2020-06-15 08:32:07

by Chao Yu

[permalink] [raw]
Subject: Re: [PATCH] erofs: Eliminate usage of uninitialized_var() macro

On 2020/6/15 16:07, Gao Xiang wrote:
> On Mon, Jun 15, 2020 at 03:43:09PM +0800, Jason Yan wrote:
>>
>>
>> ???2020/6/15 15:25, Gao Xiang 写道:
>>> Hi Jason,
>>>
>>> On Mon, Jun 15, 2020 at 12:01:41PM +0800, Jason Yan wrote:
>>>> This is an effort to eliminate the uninitialized_var() macro[1].
>>>>
>>>> The use of this macro is the wrong solution because it forces off ANY
>>>> analysis by the compiler for a given variable. It even masks "unused
>>>> variable" warnings.
>>>>
>>>> Quoted from Linus[2]:
>>>>
>>>> "It's a horrible thing to use, in that it adds extra cruft to the
>>>> source code, and then shuts up a compiler warning (even the _reliable_
>>>> warnings from gcc)."
>>>>
>>>> The gcc option "-Wmaybe-uninitialized" has been disabled and this change
>>>> will not produce any warnnings even with "make W=1".
>>>>
>>>> [1] https://github.com/KSPP/linux/issues/81
>>>> [2] https://lore.kernel.org/lkml/CA+55aFz2500WfbKXAx8s67wrm9=yVJu65TpLgN_ybYNv0VEOKA@mail.gmail.com/
>>>>
>>>> Cc: Kees Cook <[email protected]>
>>>> Cc: Chao Yu <[email protected]>
>>>> Signed-off-by: Jason Yan <[email protected]>
>>>> ---
>>>
>>> I'm fine with the patch since "-Wmaybe-uninitialized" has been disabled and
>>> I've also asked Kees for it in private previously.
>>>
>>> I still remembered that Kees sent out a treewide patch. Sorry about that
>>> I don't catch up it... But what is wrong with the original patchset?
>>>
>>
>> Yes, Kees has remind me of that and I will let him handle it. So you can
>> ignore this patch.
>
> Okay, I was just wondering if this part should be send out via EROFS tree
> for this cycle. However if there was an automatic generated patch by Kees,
> I think perhaps Linus could pick them out directly. But anyway, both ways
> are fine with me. ;) Ping me when needed.

Either way is okay to me.

Reviewed-by: Chao Yu <[email protected]>

Thanks,

>
> Thanks,
> Gao Xiang
>
>>
>> Thanks,
>> Jason
>>
>>> Thanks,
>>> Gao Xiang
>>>
>>>
>>> .
>>>
>>
>
> .
>