On 13/07/16 23:26, Markus Mayer wrote:
> On 13 July 2016 at 10:19, Luis de Bethencourt <luisbg@xxxxxxxxxxxxxxx> wrote:
>> On 11/07/16 23:46, Markus Mayer wrote:
>>
>> Hi Markus,
>>
>> Amazing. I see this happening as well, but I know it shouldn't.
>>
>> The reason the #ifndef guards in headers are there is precisely to allow
>> circular dependencies.
>>
>> The problem in your output reads as:
>> strstr() is in string.h
>> #include string.h -> that includes kernel.h -> that includes string.h
>>
>> The third should do nothing based on _LINUX_STRING_H_ being defined already
>> and all code inside the #ifndef in string.h not being executed.
>> Yet it shouldn't block the first include above since that macro isn't defined,
>> which is what the error suggests since it doesn't have strstr()
>> If _LINUX_STRING_H is defined, strstr() should be available.
>>
>> Investigating this issue, it only happens when CONFIG_DYNAMIC_DEBUG is not
>> set and line 170 of dynamic_debug.h runs, but just above we have an
>> include of string.h.
>>
>> Very strange that #include <linux/string.h> isn't doing its job.
>>
>> The first thing I tried is to understand where dynamic_debug.h is used and
>> removed the unneeded ones:
>> --------------------------------------------------------------------------
>> diff --git a/include/linux/kernel.h b/include/linux/kernel.h
>> --- a/include/linux/kernel.h
>> +++ b/include/linux/kernel.h
>> @@ -11,7 +11,6 @@
>> #include <linux/log2.h>
>> #include <linux/typecheck.h>
>> #include <linux/printk.h>
>> -#include <linux/dynamic_debug.h>
>> #include <asm/byteorder.h>
>> #include <uapi/linux/kernel.h>
>>
>> diff --git a/include/linux/printk.h b/include/linux/printk.h
>> --- a/include/linux/printk.h
>> +++ b/include/linux/printk.h
>> @@ -307,10 +307,11 @@ asmlinkage __printf(1, 2) __cold void __pr_info(const char *fmt, ...);
>> no_printk(KERN_DEBUG pr_fmt(fmt), ##__VA_ARGS__)
>> #endif
>>
>> -#include <linux/dynamic_debug.h>
>>
>> /* If you are writing a driver, please use dev_dbg instead */
>> #if defined(CONFIG_DYNAMIC_DEBUG)
>> +#include <linux/dynamic_debug.h>
>> +
>> /* dynamic_pr_debug() uses pr_fmt() internally so we don't need it here */
>> #define pr_debug(fmt, ...) \
>> dynamic_pr_debug(fmt, ##__VA_ARGS__)
>> diff --git a/kernel/module.c b/kernel/module.c
>> index beaebea..e70a2fa 100644
>> --- a/kernel/module.c
>> +++ b/kernel/module.c
>> @@ -60,6 +60,7 @@
>> #include <linux/jump_label.h>
>> #include <linux/pfn.h>
>> #include <linux/bsearch.h>
>> +#include <linux/dynamic_debug.h>
>> #include <uapi/linux/module.h>
>> #include "module-internal.h"
>> --------------------------------------------------------------------------
>>
>> This diff [0] fixes the issue but it is a workaround for the original
>> issue about string.h not being properly included in dynamic_debug.h
>>
>> Puzzled by this and can't figure out what is happening wrong.
>>
>> The second thing I tried was adding
>> #warning "Linking to string header"
>> in include/linux/string.h, and I don't see any include path mentioning
>> kernel.h, where do you see the circular dependency? I might be missing
>> something.
>
> I did some more poking around and this is what I found.
>
> For starters, the problem happens with kernel/bounds.c. Without
> worrying about SIZE_MAX or making any other changes, I added a
> #warning line to kernel.h and string.h to see the include sequence.
>
> $ aarch64-linux-gcc -Wp,-MD,kernel/.bounds.s.d -nostdinc -isystem
> /opt/toolchain/stbgcc-4.8-1.5/bin/../lib/gcc/aarch64-linux-gnu/4.8.5/include
> -I./arch/arm64/include -Iarch/arm64/include/generated/uapi
> -Iarch/arm64/include/generated -Iinclude -I./arch/arm64/include/uapi
> -Iarch/arm64/include/generated/uapi -I./include/uapi
> -Iinclude/generated/uapi -include ./include/linux/kconfig.h
> -D__KERNEL__ -mlittle-endian -Wall -Wundef -Wstrict-prototypes
> -Wno-trigraphs -fno-strict-aliasing -fno-common
> -Werror-implicit-function-declaration -Wno-format-security -std=gnu89
> -mgeneral-regs-only -fno-asynchronous-unwind-tables
> -fno-delete-null-pointer-checks -O2 --param=allow-store-data-races=0
> -Wframe-larger-than=2048 -fno-stack-protector
> -Wno-unused-but-set-variable -fno-omit-frame-pointer
> -fno-optimize-sibling-calls -fno-var-tracking-assignments -g
> -Wdeclaration-after-statement -Wno-pointer-sign -fno-strict-overflow
> -fconserve-stack -Werror=implicit-int -Werror=strict-prototypes
> -DKBUILD_BASENAME='"bounds"' -DKBUILD_MODNAME='"bounds"'
> -fverbose-asm -E -o kernel/bounds.i kernel/bounds.c
> In file included from include/asm-generic/bug.h:13:0,
> from ./arch/arm64/include/asm/bug.h:62,
> from include/linux/bug.h:4,
> from include/linux/page-flags.h:9,
> from kernel/bounds.c:9:
> include/linux/kernel.h:4:2: warning: #warning In kernel.h [-Wcpp]
> #warning In kernel.h
> ^
> In file included from include/linux/dynamic_debug.h:111:0,
> from include/linux/printk.h:289,
> from include/linux/kernel.h:14,
> from include/asm-generic/bug.h:13,
> from ./arch/arm64/include/asm/bug.h:62,
> from include/linux/bug.h:4,
> from include/linux/page-flags.h:9,
> from kernel/bounds.c:9:
> include/linux/string.h:4:2: warning: #warning In string.h [-Wcpp]
> #warning In string.h
> ^
>
> So, kernel.h gets pulled in first. string.h gets pulled in by kernel.h
> (via dynamic_debug.h as you pointed out). The build still succeeds at
> this point (because I didn't change string.h yet).
>
> Now, if I reference SIZE_MAX in string.h, there's a problem: string.h
> gets pulled in at the top of kernel.h, before kernel.h defines
> SIZE_MAX. The compiler complains ("error: ‘SIZE_MAX’ undeclared").
>
> If I add "#include <linux/kernel.h>" to the top of string.h in an
> attempt to get SIZE_MAX sooner, nothing happens, because kernel.h has
> already been included (and therefore the include guard is defined,
> meaning that the new attempt to include it doesn't do anything). So, I
> am leaving it out at this point.
>
> To fix the SIZE_MAX issue, I quickly hacked up kernel.h, so SIZE_MAX
> gets defined before string.h gets pulled in.
>
> $ git diff
> diff --git a/include/linux/kernel.h b/include/linux/kernel.h
> index 94aa10f..6cd5269 100644
> --- a/include/linux/kernel.h
> +++ b/include/linux/kernel.h
> @@ -10,8 +10,6 @@
> #include <linux/bitops.h>
> #include <linux/log2.h>
> #include <linux/typecheck.h>
> -#include <linux/printk.h>
> -#include <linux/dynamic_debug.h>
> #include <asm/byteorder.h>
> #include <uapi/linux/kernel.h>
>
> @@ -29,6 +27,9 @@
> #define ULLONG_MAX (~0ULL)
> #define SIZE_MAX (~(size_t)0)
>
> +#include <linux/printk.h>
> +#include <linux/dynamic_debug.h>
> +
> #define U8_MAX ((u8)~0U)
> #define S8_MAX ((s8)(U8_MAX>>1))
> #define S8_MIN ((s8)(-S8_MAX - 1))
>
> This solves bounds.c. But it doesn't help with other files that don't
> automatically pull in kernel.h before string.h. They still complain
> about SIZE_MAX being undeclared. So, #include <linux/kernel.h> *does*
> need to go into string.h. At this point, we run into the next problem.
>
> $ aarch64-linux-gcc -Wp,-MD,scripts/mod/.devicetable-offsets.s.d
> -nostdinc -isystem
> /opt/toolchain/stbgcc-4.8-1.5/bin/../lib/gcc/aarch64-linux-gnu/4.8.5/include
> -I./arch/arm64/include -Iarch/arm64/include/generated/uapi
> -Iarch/arm64/include/generated -Iinclude -I./arch/arm64/include/uapi
> -Iarch/arm64/include/generated/uapi -I./include/uapi
> -Iinclude/generated/uapi -include ./include/linux/kconfig.h
> -D__KERNEL__ -mlittle-endian -Wall -Wundef -Wstrict-prototypes
> -Wno-trigraphs -fno-strict-aliasing -fno-common
> -Werror-implicit-function-declaration -Wno-format-security -std=gnu89
> -mgeneral-regs-only -fno-asynchronous-unwind-tables
> -fno-delete-null-pointer-checks -O2 --param=allow-store-data-races=0
> -Wframe-larger-than=2048 -fno-stack-protector
> -Wno-unused-but-set-variable -fno-omit-frame-pointer
> -fno-optimize-sibling-calls -fno-var-tracking-assignments -g
> -Wdeclaration-after-statement -Wno-pointer-sign -fno-strict-overflow
> -fconserve-stack -Werror=implicit-int -Werror=strict-prototypes
> -DKBUILD_BASENAME='"devicetable_offsets"'
> -DKBUILD_MODNAME='"devicetable_offsets"' -fverbose-asm -S -o
> scripts/mod/devicetable-offsets.s scripts/mod/devicetable-offsets.c
> In file included from include/uapi/linux/uuid.h:21:0,
> from include/linux/uuid.h:19,
> from include/linux/mod_devicetable.h:12,
> from scripts/mod/devicetable-offsets.c:2:
> include/linux/string.h:4:2: warning: #warning In string.h [-Wcpp]
> #warning In string.h
> ^
> In file included from include/linux/string.h:7:0,
> from include/uapi/linux/uuid.h:21,
> from include/linux/uuid.h:19,
> from include/linux/mod_devicetable.h:12,
> from scripts/mod/devicetable-offsets.c:2:
> include/linux/kernel.h:4:2: warning: #warning In kernel.h [-Wcpp]
> #warning In kernel.h
> ^
> In file included from include/linux/printk.h:289:0,
> from include/linux/kernel.h:31,
> from include/linux/string.h:7,
> from include/uapi/linux/uuid.h:21,
> from include/linux/uuid.h:19,
> from include/linux/mod_devicetable.h:12,
> from scripts/mod/devicetable-offsets.c:2:
> include/linux/dynamic_debug.h: In function ‘ddebug_dyndbg_module_param_cb’:
> include/linux/dynamic_debug.h:122:2: error: implicit declaration of
> function ‘strstr’ [-Werror=implicit-function-declaration]
> if (strstr(param, "dyndbg")) {
> ^
> include/linux/dynamic_debug.h:122:6: warning: incompatible implicit
> declaration of built-in function ‘strstr’ [enabled by default]
> if (strstr(param, "dyndbg")) {
> ^
> cc1: some warnings being treated as errors
>
> So, now we have the following sequence:
>
> string.h -> kernel.h -> dynamic_debug.h -> oops, we don't know strstr() yet.
>
> Yes, dynamic_debug.h does include string.h, but because of the include
> guard, that second attempt at including string.h is stubbed out and
> doesn't do anything.
>
> To get past that issue, I can move "#include <linux/kernel.h>" all the
> way down and only include it directly before my new functions. That
> way strstr() is declared when dynamic_debug.h wants it.
>
> diff --git a/include/linux/string.h b/include/linux/string.h
> index 6cc85dc..eb2d9a8 100644
> --- a/include/linux/string.h
> +++ b/include/linux/string.h
> @@ -171,6 +171,8 @@ static inline const char *kbasename(const char *path)
> return tail ? tail + 1 : path;
> }
>
> +#include <linux/kernel.h>
> +
> /**
> * strcpytoupper - Copy string and convert to uppercase.
> * @dst: The buffer to store the result.
>
> This is some scary stuff, but this way it seems to build. Not too sure
> how to fix this properly and without risking some major headaches when
> one least expects them.
>
> Regards,
> -Markus
>
>> Thanks,
>> Luis
Hi Markus,
Thank you very much for the thorough testing and solution searching.
Earlier today I sent a patch that removes dynamic_debug.h from
include/linux/kernel.h, since it isn't really needed. Sorry about this
since it changes what you were testing just a few hours later.
I am starting to think that getting access to SIZE_MAX isn't worth the
trouble, specially considering that moving an include out of the top
of the file makes the code less readable. The Linux kernel is growing
in complexity and trying to keep it readable for newcomers is very
important IMHO.
The maintainers have the last word on this, but for now I remove my
question about why use -1 instead of SIZE_MAX.
Apologies for that, the silver lining was that at least for me it was
interesting to explore this area of the code and its inclusions. I hope
it was for you as well.
Thank you very much,
Luis
PD: just a small comment, in case you end up sending a new version of
your patches. In the following documentation of your functions.
* @len: Maximum string length. May be SIZE_MAX (-1) to set no limit.
What do you think about "SIZE_MAX or -1" instead of "SIZE_MAX (-1)" to
avoid confusing any readers who might think SIZE_MAX is a function? As
I think you intend to mean your code accepts both -1 or SIZE_MAX, which
is ~0.
Just an idea, feel free to ignore it :)
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
