[PATCH] libbpf: fix warnings "'pad_type' 'pad_bits' 'new_off' may be used uninitialized"

Xiangyu Chen posted 1 patch 2 years, 6 months ago
tools/lib/bpf/btf_dump.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
[PATCH] libbpf: fix warnings "'pad_type' 'pad_bits' 'new_off' may be used uninitialized"
Posted by Xiangyu Chen 2 years, 6 months ago
From: Xiangyu Chen <xiangyu.chen@windriver.com>

When turn on the yocto DEBUG_BUILD flag, the build options for gcc would enable maybe-uninitialized,
and following warnings would be reported as below:

| btf_dump.c: In function 'btf_dump_emit_bit_padding':
| btf_dump.c:916:4: error: 'pad_type' may be used uninitialized in this function [-Werror=maybe-uninitialized]
|   916 |    btf_dump_printf(d, "\n%s%s: %d;", pfx(lvl), pad_type,
|       |    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
|   917 |      in_bitfield ? new_off - cur_off : 0);
|       |      ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
| btf_dump.c:929:6: error: 'pad_bits' may be used uninitialized in this function [-Werror=maybe-uninitialized]
|   929 |   if (bits == pad_bits) {
|       |      ^
| btf_dump.c:913:28: error: 'new_off' may be used uninitialized in this function [-Werror=maybe-uninitialized]
|   913 |       (new_off == next_off && roundup(cur_off, next_align * 8) != new_off) ||
|       |       ~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
|   HOSTLD  scripts/mod/modpost

Signed-off-by: Xiangyu Chen <xiangyu.chen@windriver.com>
---
 tools/lib/bpf/btf_dump.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/tools/lib/bpf/btf_dump.c b/tools/lib/bpf/btf_dump.c
index 4d9f30bf7f01..79923c3b8777 100644
--- a/tools/lib/bpf/btf_dump.c
+++ b/tools/lib/bpf/btf_dump.c
@@ -867,8 +867,8 @@ static void btf_dump_emit_bit_padding(const struct btf_dump *d,
 	} pads[] = {
 		{"long", d->ptr_sz * 8}, {"int", 32}, {"short", 16}, {"char", 8}
 	};
-	int new_off, pad_bits, bits, i;
-	const char *pad_type;
+	int new_off = 0, pad_bits = 0, bits, i;
+	const char *pad_type = NULL;
 
 	if (cur_off >= next_off)
 		return; /* no gap */
-- 
2.34.1
Re: [PATCH] libbpf: fix warnings "'pad_type' 'pad_bits' 'new_off' may be used uninitialized"
Posted by Jiri Olsa 2 years, 6 months ago
On Thu, Jul 27, 2023 at 04:25:36PM +0800, Xiangyu Chen wrote:
> From: Xiangyu Chen <xiangyu.chen@windriver.com>
> 
> When turn on the yocto DEBUG_BUILD flag, the build options for gcc would enable maybe-uninitialized,
> and following warnings would be reported as below:

curious, what's the gcc version? I can't reproduce that,
and we already have all warnings enabled:

  CFLAGS += -Werror -Wall

they seem like false warnings also, because ARRAY_SIZE(pads)
will be always > 0

jirka

> 
> | btf_dump.c: In function 'btf_dump_emit_bit_padding':
> | btf_dump.c:916:4: error: 'pad_type' may be used uninitialized in this function [-Werror=maybe-uninitialized]
> |   916 |    btf_dump_printf(d, "\n%s%s: %d;", pfx(lvl), pad_type,
> |       |    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> |   917 |      in_bitfield ? new_off - cur_off : 0);
> |       |      ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> | btf_dump.c:929:6: error: 'pad_bits' may be used uninitialized in this function [-Werror=maybe-uninitialized]
> |   929 |   if (bits == pad_bits) {
> |       |      ^
> | btf_dump.c:913:28: error: 'new_off' may be used uninitialized in this function [-Werror=maybe-uninitialized]
> |   913 |       (new_off == next_off && roundup(cur_off, next_align * 8) != new_off) ||
> |       |       ~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> |   HOSTLD  scripts/mod/modpost
> 
> Signed-off-by: Xiangyu Chen <xiangyu.chen@windriver.com>
> ---
>  tools/lib/bpf/btf_dump.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/tools/lib/bpf/btf_dump.c b/tools/lib/bpf/btf_dump.c
> index 4d9f30bf7f01..79923c3b8777 100644
> --- a/tools/lib/bpf/btf_dump.c
> +++ b/tools/lib/bpf/btf_dump.c
> @@ -867,8 +867,8 @@ static void btf_dump_emit_bit_padding(const struct btf_dump *d,
>  	} pads[] = {
>  		{"long", d->ptr_sz * 8}, {"int", 32}, {"short", 16}, {"char", 8}
>  	};
> -	int new_off, pad_bits, bits, i;
> -	const char *pad_type;
> +	int new_off = 0, pad_bits = 0, bits, i;
> +	const char *pad_type = NULL;
>  
>  	if (cur_off >= next_off)
>  		return; /* no gap */
> -- 
> 2.34.1
> 
>
Re: [PATCH] libbpf: fix warnings "'pad_type' 'pad_bits' 'new_off' may be used uninitialized"
Posted by Xiangyu Chen 2 years, 6 months ago
On 7/27/23 19:01, Jiri Olsa wrote:
> CAUTION: This email comes from a non Wind River email account!
> Do not click links or open attachments unless you recognize the sender and know the content is safe.
>
> On Thu, Jul 27, 2023 at 04:25:36PM +0800, Xiangyu Chen wrote:
>> From: Xiangyu Chen <xiangyu.chen@windriver.com>
>>
>> When turn on the yocto DEBUG_BUILD flag, the build options for gcc would enable maybe-uninitialized,
>> and following warnings would be reported as below:
> curious, what's the gcc version? I can't reproduce that,

Indeed, it's also strange to me, the compiler using gcc-11.3.0 which is 
generated by yocto 4.2.2(mickledore branch).

The warning happens on turn on DEBUG_BUILD flag in yocto configuration, 
but when disable the DEBUG_BUILD flag, this

warning disappeared.

I have compared w/wo the DEBUG_BUILD's kernel build parameters this 
morning, when enabled the flag, the kernel build

parameters as normal, the warning won't happen. When I enabled the 
DEBUG_BUILD, the cflags with -Og -g -feliminate-unused-debug-types

seems the debug cflags caused the false warning behavior.


> and we already have all warnings enabled:
>
>    CFLAGS += -Werror -Wall
>
> they seem like false warnings also, because ARRAY_SIZE(pads)
> will be always > 0
>
> jirka
>
>> | btf_dump.c: In function 'btf_dump_emit_bit_padding':
>> | btf_dump.c:916:4: error: 'pad_type' may be used uninitialized in this function [-Werror=maybe-uninitialized]
>> |   916 |    btf_dump_printf(d, "\n%s%s: %d;", pfx(lvl), pad_type,
>> |       |    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>> |   917 |      in_bitfield ? new_off - cur_off : 0);
>> |       |      ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>> | btf_dump.c:929:6: error: 'pad_bits' may be used uninitialized in this function [-Werror=maybe-uninitialized]
>> |   929 |   if (bits == pad_bits) {
>> |       |      ^
>> | btf_dump.c:913:28: error: 'new_off' may be used uninitialized in this function [-Werror=maybe-uninitialized]
>> |   913 |       (new_off == next_off && roundup(cur_off, next_align * 8) != new_off) ||
>> |       |       ~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>> |   HOSTLD  scripts/mod/modpost
>>
>> Signed-off-by: Xiangyu Chen <xiangyu.chen@windriver.com>
>> ---
>>   tools/lib/bpf/btf_dump.c | 4 ++--
>>   1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/tools/lib/bpf/btf_dump.c b/tools/lib/bpf/btf_dump.c
>> index 4d9f30bf7f01..79923c3b8777 100644
>> --- a/tools/lib/bpf/btf_dump.c
>> +++ b/tools/lib/bpf/btf_dump.c
>> @@ -867,8 +867,8 @@ static void btf_dump_emit_bit_padding(const struct btf_dump *d,
>>        } pads[] = {
>>                {"long", d->ptr_sz * 8}, {"int", 32}, {"short", 16}, {"char", 8}
>>        };
>> -     int new_off, pad_bits, bits, i;
>> -     const char *pad_type;
>> +     int new_off = 0, pad_bits = 0, bits, i;
>> +     const char *pad_type = NULL;
>>
>>        if (cur_off >= next_off)
>>                return; /* no gap */
>> --
>> 2.34.1
>>
>>
Re: [PATCH] libbpf: fix warnings "'pad_type' 'pad_bits' 'new_off' may be used uninitialized"
Posted by Yonghong Song 2 years, 6 months ago

On 7/27/23 4:01 AM, Jiri Olsa wrote:
> On Thu, Jul 27, 2023 at 04:25:36PM +0800, Xiangyu Chen wrote:
>> From: Xiangyu Chen <xiangyu.chen@windriver.com>
>>
>> When turn on the yocto DEBUG_BUILD flag, the build options for gcc would enable maybe-uninitialized,
>> and following warnings would be reported as below:
> 
> curious, what's the gcc version? I can't reproduce that,
> and we already have all warnings enabled:
> 
>    CFLAGS += -Werror -Wall
> 
> they seem like false warnings also, because ARRAY_SIZE(pads)
> will be always > 0

Agree. This definitely a false positive.
In kernel top Makefile, we have

# Enabled with W=2, disabled by default as noisy
ifdef CONFIG_CC_IS_GCC
KBUILD_CFLAGS += -Wno-maybe-uninitialized
endif

That means gcc -maybe-uninitialized is very noisy.

> 
> jirka
> 
>>
>> | btf_dump.c: In function 'btf_dump_emit_bit_padding':
>> | btf_dump.c:916:4: error: 'pad_type' may be used uninitialized in this function [-Werror=maybe-uninitialized]
>> |   916 |    btf_dump_printf(d, "\n%s%s: %d;", pfx(lvl), pad_type,
>> |       |    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>> |   917 |      in_bitfield ? new_off - cur_off : 0);
>> |       |      ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>> | btf_dump.c:929:6: error: 'pad_bits' may be used uninitialized in this function [-Werror=maybe-uninitialized]
>> |   929 |   if (bits == pad_bits) {
>> |       |      ^
>> | btf_dump.c:913:28: error: 'new_off' may be used uninitialized in this function [-Werror=maybe-uninitialized]
>> |   913 |       (new_off == next_off && roundup(cur_off, next_align * 8) != new_off) ||
>> |       |       ~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>> |   HOSTLD  scripts/mod/modpost
>>
>> Signed-off-by: Xiangyu Chen <xiangyu.chen@windriver.com>
>> ---
>>   tools/lib/bpf/btf_dump.c | 4 ++--
>>   1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/tools/lib/bpf/btf_dump.c b/tools/lib/bpf/btf_dump.c
>> index 4d9f30bf7f01..79923c3b8777 100644
>> --- a/tools/lib/bpf/btf_dump.c
>> +++ b/tools/lib/bpf/btf_dump.c
>> @@ -867,8 +867,8 @@ static void btf_dump_emit_bit_padding(const struct btf_dump *d,
>>   	} pads[] = {
>>   		{"long", d->ptr_sz * 8}, {"int", 32}, {"short", 16}, {"char", 8}
>>   	};
>> -	int new_off, pad_bits, bits, i;
>> -	const char *pad_type;
>> +	int new_off = 0, pad_bits = 0, bits, i;
>> +	const char *pad_type = NULL;
>>   
>>   	if (cur_off >= next_off)
>>   		return; /* no gap */
>> -- 
>> 2.34.1
>>
>>
>
Re: [PATCH] libbpf: fix warnings "'pad_type' 'pad_bits' 'new_off' may be used uninitialized"
Posted by Jiri Olsa 2 years, 6 months ago
On Thu, Jul 27, 2023 at 11:23:21AM -0700, Yonghong Song wrote:
> 
> 
> On 7/27/23 4:01 AM, Jiri Olsa wrote:
> > On Thu, Jul 27, 2023 at 04:25:36PM +0800, Xiangyu Chen wrote:
> > > From: Xiangyu Chen <xiangyu.chen@windriver.com>
> > > 
> > > When turn on the yocto DEBUG_BUILD flag, the build options for gcc would enable maybe-uninitialized,
> > > and following warnings would be reported as below:
> > 
> > curious, what's the gcc version? I can't reproduce that,
> > and we already have all warnings enabled:
> > 
> >    CFLAGS += -Werror -Wall
> > 
> > they seem like false warnings also, because ARRAY_SIZE(pads)
> > will be always > 0
> 
> Agree. This definitely a false positive.
> In kernel top Makefile, we have
> 
> # Enabled with W=2, disabled by default as noisy
> ifdef CONFIG_CC_IS_GCC
> KBUILD_CFLAGS += -Wno-maybe-uninitialized
> endif
> 
> That means gcc -maybe-uninitialized is very noisy.

nice, I think we should do the same then ;-)
bt not sure how to do the gcc check though

jirka

> 
> > 
> > jirka
> > 
> > > 
> > > | btf_dump.c: In function 'btf_dump_emit_bit_padding':
> > > | btf_dump.c:916:4: error: 'pad_type' may be used uninitialized in this function [-Werror=maybe-uninitialized]
> > > |   916 |    btf_dump_printf(d, "\n%s%s: %d;", pfx(lvl), pad_type,
> > > |       |    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > > |   917 |      in_bitfield ? new_off - cur_off : 0);
> > > |       |      ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > > | btf_dump.c:929:6: error: 'pad_bits' may be used uninitialized in this function [-Werror=maybe-uninitialized]
> > > |   929 |   if (bits == pad_bits) {
> > > |       |      ^
> > > | btf_dump.c:913:28: error: 'new_off' may be used uninitialized in this function [-Werror=maybe-uninitialized]
> > > |   913 |       (new_off == next_off && roundup(cur_off, next_align * 8) != new_off) ||
> > > |       |       ~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > > |   HOSTLD  scripts/mod/modpost
> > > 
> > > Signed-off-by: Xiangyu Chen <xiangyu.chen@windriver.com>
> > > ---
> > >   tools/lib/bpf/btf_dump.c | 4 ++--
> > >   1 file changed, 2 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/tools/lib/bpf/btf_dump.c b/tools/lib/bpf/btf_dump.c
> > > index 4d9f30bf7f01..79923c3b8777 100644
> > > --- a/tools/lib/bpf/btf_dump.c
> > > +++ b/tools/lib/bpf/btf_dump.c
> > > @@ -867,8 +867,8 @@ static void btf_dump_emit_bit_padding(const struct btf_dump *d,
> > >   	} pads[] = {
> > >   		{"long", d->ptr_sz * 8}, {"int", 32}, {"short", 16}, {"char", 8}
> > >   	};
> > > -	int new_off, pad_bits, bits, i;
> > > -	const char *pad_type;
> > > +	int new_off = 0, pad_bits = 0, bits, i;
> > > +	const char *pad_type = NULL;
> > >   	if (cur_off >= next_off)
> > >   		return; /* no gap */
> > > -- 
> > > 2.34.1
> > > 
> > > 
> >