Skip to contentSkip to conversation

x86: declaration fixes

GIT PULL Jaswinder Singh 2 participants 4 messages Jul 21, 2008 – Jul 22, 2008 ABI/API

lore
Share & tools
lore
#0 lore raw

Hello Ingo,

Please pull the declaration fixes for x86-tip from :

git://git.infradead.org/users/jaswinder/linux-2.6-tip.git

This includes following patches after commit 087115517d43c2 (Merge branch 'xen-64bit') :-

1. [PATCH] x86: Introducing asm/syscalls.h
2. [PATCH] x86: process_32.c declare cpu_number before they get used
3. [PATCH] x86: signal_XX.c declare do_notify_resume before they get used
4. [PATCH] x86: time_XX.c declare functions before they get used
5. [PATCH] x86: setup.c declare saved_video_mode before they get used
6. [PATCH] x86: e820.c declare pci_mem_start before they get used
7. [PATCH] x86: pci-dma.c declare iommu_bio_merge before they get used
8. [PATCH] x86: pci-nommu.c declare nommu_dma_ops before they get used
9. [PATCH] x86: i387.c declare dump_fpu() before they get used
10. [PATCH] x86: ptrace.c declare functions before they get used
11. [PATCH] x86: proc.c declare cpuinfo_op before they get used
12. [PATCH] x86: common.c declare idle_regs before they get used
13. [PATCH] x86: mtrr/main.c declare range_state as static

Thank you,

Jaswinder Singh

#1 lore raw
Quoted 5 lines · Jaswinder Singh

Hello Ingo,

Please pull the declaration fixes for x86-tip from :

git://git.infradead.org/users/jaswinder/linux-2.6-tip.git

pulled this into tip/x86/prototypes, thanks Jaswinder.

Btw., when sending pull requests it's best to use git-request-pull. I.e. something like:

git-request-pull tip/master git://git.infradead.org/users/jaswinder/linux-2.6-tip.git

would have generated the proper output.

The changes still need some fixing though. For example, this change:

Quoted 1 line

2. [PATCH] x86: process_32.c declare cpu_number before they get used

why is cpu_number moved over from smp.h to current.h? I think it's more logical in smp.h. If it's missing somewhere then add an #include <asm/smp.h> to that file?

also, this change:

+++ b/include/linux/timer.h
@@ -6,6 +6,10 @@
#include <linux/stddef.h>
#include <linux/debugobjects.h>

+#ifdef CONFIG_X86
+#include <asm/timer.h>
+#endif /* CONFIG_X86 */

is not really acceptable in a generic header, nor is:

include/linux/syscalls.h +3 -0
@@ -65,6 +67,10 @@ struct getcpu_cache;
 #include <linux/quota.h>
 #include <linux/key.h>
 
+#ifdef CONFIG_X86
+#include <asm/syscalls.h>
+#endif /* CONFIG_X86 */
 

i'd suggest we just do asm-x86/syscall.h and asm-x86/timer.h changes and add #include <asm/syscall.h> and #include <asm/timer.h> to the affected x86 .c files. Once every architecture has it, linux/syscall.h can be extended with an unconditional asm/syscall.h include.

Or, like Matthew has suggested, a 'complete' collection of all syscalls should be assembled into linux/syscall.h. In that case it has extra documentation value if there's an #ifdef CONFIG_X86 for certain syscalls. (say the LDT or VM86 syscalls make only sense on x86 platforms)

another thing:

--- a/include/asm-x86/signal.h
+++ b/include/asm-x86/signal.h

+void do_notify_resume(struct pt_regs *, void *, __u32); +

please add "extern" to each such prototype declaration. (the same is true for a handful of other prototypes you've added - please re-review and fix them)

+#ifdef CONFIG_X86_32
extern int timer_ack;
-extern int no_timer_check;
extern int recalibrate_cpu_khz(void);
+#endif /* CONFIG_X86_32 */
+
+extern int no_timer_check;

hm, why was this needed?

Ingo

#2 lore raw

Hello Ingo,

Quoted 10 lines · Ingo Molnar
Quoted 3 lines · Jaswinder Singh

Hello Ingo,

2. [PATCH] x86: process_32.c declare cpu_number before they get used

why is cpu_number moved over from smp.h to current.h? I think it's more logical in smp.h. If it's missing somewhere then add an #include <asm/smp.h> to that file?

Because cpu_number is used by everyone independent of smp.

In smp.h It was declared under #ifdef CONFIG_X86_32_SMP which was wrong.

Quoted 23 lines

also, this change:

+++ b/include/linux/timer.h

include/linux/syscalls.h +3 -0
@@ -6,6 +6,10 @@
 #include <linux/stddef.h>
 #include <linux/debugobjects.h>
 
+#ifdef CONFIG_X86
+#include <asm/timer.h>
+#endif /* CONFIG_X86 */
 

is not really acceptable in a generic header, nor is:

include/linux/syscalls.h +3 -0
@@ -65,6 +67,10 @@ struct getcpu_cache;
 #include <linux/quota.h>
 #include <linux/key.h>
 
+#ifdef CONFIG_X86
+#include <asm/syscalls.h>
+#endif /* CONFIG_X86 */
 
Quoted 5 lines

i'd suggest we just do asm-x86/syscall.h and asm-x86/timer.h changes and add #include <asm/syscall.h> and #include <asm/timer.h> to the affected x86 .c files. Once every architecture has it, linux/syscall.h can be extended with an unconditional asm/syscall.h include.

Because checkpatch.pl raise objection for it.

$ ./scripts/checkpatch.pl ../test.patch 
WARNING: Use #include <linux/timer.h> instead of <asm/timer.h>
#9: FILE: arch/x86/kernel/time_32.c:40:
+#include <asm/timer.h>
Quoted 10 lines

+#ifdef CONFIG_X86_32
extern int timer_ack;
-extern int no_timer_check;
extern int recalibrate_cpu_khz(void);
+#endif /* CONFIG_X86_32 */
+
+extern int no_timer_check;

hm, why was this needed?

Because timer_ack is only required for X86_32
and int recalibrate_cpu_khz(void) is declared inside #ifdef CONFIG_X86_32 in arch/x86/kernel/tsc.c

Thank you,

Jaswinder Singh.

#3 lore raw
In reply to Ingo Molnar

Hello Ingo,

Quoted 11 lines · Ingo Molnar

* Jaswinder Singh <[email protected]> wrote:

another thing:

include/asm-x86/signal.h +2 -0
 
+void do_notify_resume(struct pt_regs *, void *, __u32);
+
Quoted 4 lines

please add "extern" to each such prototype declaration. (the same is true for a handful of other prototypes you've added - please re-review and fix them)

I was trying to follow other functions declaration in the file.

function declaration without extern :-

http://git.infradead.org/users/jaswinder/linux-2.6-tip.git?a=commitdiff;h=f740a1e1bbee2ce8a0b1b97361771db40afc37d2
http://git.infradead.org/users/jaswinder/linux-2.6-tip.git?a=commitdiff;h=b7b5113220da95eaa8b75ebd5bbd5278eed122e5

function declaration with extern :-

http://git.infradead.org/users/jaswinder/linux-2.6-tip.git?a=commitdiff;h=8cac4294a5e9b036c30d30e2e894e74201ee1c6b
http://git.infradead.org/users/jaswinder/linux-2.6-tip.git?a=commitdiff;h=e7d3d48dbf3e517e2e827a2adc6a2f861b16e6a6
http://git.infradead.org/users/jaswinder/linux-2.6-tip.git?a=commitdiff;h=feb49274338051ddf11d16f67b08ee6cafb32dbf

But you are emphasizing to use extern for all function declarations.

With current gcc, I think using extern with function declaration is :-
1. Wastage of time
2. Wastage of energy
3. Increase file size with 7 bytes for each extern.

If there is any secret behind using extern in kernel header files. Please let me know.

Thank you,

Jaswinder Singh.