Rename the deprecated attribute to __deprecated to make it obvious
this is something special and to avoid namespace clashes.
Mark more functionality deprecated:
- sleep_on & friends
- old module interfaces
This gives quite a lot warnings, I'll start to fix the ones in
core code and stuff I maintain now..
--- 1.7/include/linux/compiler.h Sat Dec 28 12:45:03 2002
+++ edited/include/linux/compiler.h Sun Dec 29 18:52:54 2002
@@ -21,9 +21,9 @@
* and then gcc will emit a warning for each usage of the function.
*/
#if __GNUC__ >= 3
-#define deprecated __attribute__((deprecated))
+#define __deprecated __attribute__((deprecated))
#else
-#define deprecated
+#define __deprecated
#endif
/* This macro obfuscates arithmetic on a variable address so that gcc
--- 1.5/include/linux/ioport.h Sat Dec 28 17:14:46 2002
+++ edited/include/linux/ioport.h Sun Dec 29 18:53:36 2002
@@ -108,7 +108,7 @@
#define check_mem_region(start,n) __check_region(&iomem_resource, (start), (n))
#define release_mem_region(start,n) __release_region(&iomem_resource, (start), (n))
-extern int deprecated __check_region(struct resource *, unsigned long, unsigned long);
+extern int __deprecated __check_region(struct resource *, unsigned long, unsigned long);
extern void __release_region(struct resource *, unsigned long, unsigned long);
#define get_ioport_list(buf) get_resource_list(&ioport_resource, buf, PAGE_SIZE)
--- 1.30/include/linux/module.h Mon Dec 2 01:44:43 2002
+++ edited/include/linux/module.h Sun Dec 29 19:50:27 2002
@@ -296,9 +296,20 @@
#define symbol_request(x) try_then_request_module(symbol_get(x), "symbol:" #x)
/* BELOW HERE ALL THESE ARE OBSOLETE AND WILL VANISH */
-#define __MOD_INC_USE_COUNT(mod) \
- do { __unsafe(mod); (void)try_module_get(mod); } while(0)
-#define __MOD_DEC_USE_COUNT(mod) module_put(mod)
+static inline void __deprecated __MOD_INC_USE_COUNT(struct module *module)
+{
+ __unsafe(module);
+ /*
+ * Yes, we ignore the retval here, that's why it's deprecated.
+ */
+ try_module_get(module);
+}
+
+static inline void __deprecated __MOD_DEC_USE_COUNT(struct module *module)
+{
+ module_put(module);
+}
+
#define SET_MODULE_OWNER(dev) ((dev)->owner = THIS_MODULE)
struct obsolete_modparm {
@@ -319,14 +330,21 @@
/* People do this inside their init routines, when the module isn't
"live" yet. They should no longer be doing that, but
meanwhile... */
+static inline void __deprecated _MOD_INC_USE_COUNT(struct module *module)
+{
+ __unsafe(module);
+
#if defined(CONFIG_MODULE_UNLOAD) && defined(MODULE)
-#define MOD_INC_USE_COUNT \
- do { __unsafe(THIS_MODULE); local_inc(&THIS_MODULE->ref[get_cpu()].count); put_cpu(); } while (0)
+ local_inc(&module->ref[get_cpu()].count);
+ put_cpu();
#else
-#define MOD_INC_USE_COUNT \
- do { __unsafe(THIS_MODULE); (void)try_module_get(THIS_MODULE); } while (0)
+ try_module_get(module);
#endif
-#define MOD_DEC_USE_COUNT module_put(THIS_MODULE)
+}
+#define MOD_INC_USE_COUNT \
+ _MOD_INC_USE_COUNT(THIS_MODULE)
+#define MOD_DEC_USE_COUNT \
+ __MOD_DEC_USE_COUNT(THIS_MODULE)
#define try_inc_mod_count(mod) try_module_get(mod)
#define EXPORT_NO_SYMBOLS
extern int module_dummy_usage;
@@ -364,5 +382,4 @@
extern const void *inter_module_get(const char *);
extern const void *inter_module_get_request(const char *, const char *);
extern void inter_module_put(const char *);
-
#endif /* _LINUX_MODULE_H */
--- 1.12/include/linux/wait.h Fri Nov 15 10:36:32 2002
+++ edited/include/linux/wait.h Sun Dec 29 19:05:26 2002
@@ -11,6 +11,7 @@
#ifdef __KERNEL__
#include <linux/config.h>
+#include <linux/compiler.h>
#include <linux/list.h>
#include <linux/stddef.h>
#include <linux/spinlock.h>
@@ -244,11 +245,11 @@
* They are racy. DO NOT use them, use the wait_event* interfaces above.
* We plan to remove these interfaces during 2.7.
*/
-extern void FASTCALL(sleep_on(wait_queue_head_t *q));
-extern long FASTCALL(sleep_on_timeout(wait_queue_head_t *q,
+extern void __deprecated FASTCALL(sleep_on(wait_queue_head_t *q));
+extern long __deprecated FASTCALL(sleep_on_timeout(wait_queue_head_t *q,
signed long timeout));
-extern void FASTCALL(interruptible_sleep_on(wait_queue_head_t *q));
-extern long FASTCALL(interruptible_sleep_on_timeout(wait_queue_head_t *q,
+extern void __deprecated FASTCALL(interruptible_sleep_on(wait_queue_head_t *q));
+extern long __deprecated FASTCALL(interruptible_sleep_on_timeout(wait_queue_head_t *q,
signed long timeout));
/*
--- 1.7/kernel/resource.c Sat Dec 28 17:05:55 2002
+++ edited/kernel/resource.c Sun Dec 29 18:53:07 2002
@@ -237,7 +237,7 @@
return res;
}
-int deprecated __check_region(struct resource *parent, unsigned long start, unsigned long n)
+int __deprecated __check_region(struct resource *parent, unsigned long start, unsigned long n)
{
struct resource * res;
===== kernel/sched.c 1.145 vs edited =====
--- 1.145/kernel/sched.c Mon Dec 2 08:06:13 2002
+++ edited/kernel/sched.c Sun Dec 29 18:57:50 2002
@@ -1238,7 +1238,7 @@
__remove_wait_queue(q, &wait); \
spin_unlock_irqrestore(&q->lock, flags);
-void interruptible_sleep_on(wait_queue_head_t *q)
+void __deprecated interruptible_sleep_on(wait_queue_head_t *q)
{
SLEEP_ON_VAR
@@ -1249,7 +1249,7 @@
SLEEP_ON_TAIL
}
-long interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout)
+long __deprecated interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout)
{
SLEEP_ON_VAR
@@ -1262,7 +1262,7 @@
return timeout;
}
-void sleep_on(wait_queue_head_t *q)
+void __deprecated sleep_on(wait_queue_head_t *q)
{
SLEEP_ON_VAR
@@ -1273,7 +1273,7 @@
SLEEP_ON_TAIL
}
-long sleep_on_timeout(wait_queue_head_t *q, long timeout)
+long __deprecated sleep_on_timeout(wait_queue_head_t *q, long timeout)
{
SLEEP_ON_VAR
Christoph Hellwig wrote:
>
> Rename the deprecated attribute to __deprecated to make it obvious
> this is something special and to avoid namespace clashes.
>
> Mark more functionality deprecated:
>
> - sleep_on & friends
Please do not make sleep_on() generate a warning. Unless you intend
to do the same to lock_kernel().
ext3 uses sleep_on(). It is perfectly safe. Weaning ext3 off lock_kernel()
is a large, delicate and thus-far undesigned body of work. I've been
working on other stuff and it is quite unlikely that ext3 locking will
be redesigned in the 2.5 timeframe.
> - old module interfaces
>
> This gives quite a lot warnings, I'll start to fix the ones in
> core code and stuff I maintain now..
Sounds like pointless churn to me. How about we fix some real stuff and get
this kernel out the door before we all die?
On Sun, Dec 29, 2002 at 01:38:51PM -0800, Andrew Morton wrote:
> Christoph Hellwig wrote:
> >
> > Rename the deprecated attribute to __deprecated to make it obvious
> > this is something special and to avoid namespace clashes.
> >
> > Mark more functionality deprecated:
> >
> > - sleep_on & friends
>
> Please do not make sleep_on() generate a warning. Unless you intend
> to do the same to lock_kernel().
>
> ext3 uses sleep_on(). It is perfectly safe.
Even if it's safe in that particular case, most code in the kernel runs
without BKL. This patch just makes the deprication of sleep_on
explicit.
> Weaning ext3 off lock_kernel()
> is a large, delicate and thus-far undesigned body of work. I've been
> working on other stuff and it is quite unlikely that ext3 locking will
> be redesigned in the 2.5 timeframe.
Then ext3 has to live with using depricated interfaces during 2.6,
what's the point?
Christoph Hellwig wrote:
>
> On Sun, Dec 29, 2002 at 01:38:51PM -0800, Andrew Morton wrote:
> > Christoph Hellwig wrote:
> > >
> > > Rename the deprecated attribute to __deprecated to make it obvious
> > > this is something special and to avoid namespace clashes.
> > >
> > > Mark more functionality deprecated:
> > >
> > > - sleep_on & friends
> >
> > Please do not make sleep_on() generate a warning. Unless you intend
> > to do the same to lock_kernel().
> >
> > ext3 uses sleep_on(). It is perfectly safe.
>
> Even if it's safe in that particular case, most code in the kernel runs
> without BKL. This patch just makes the deprication of sleep_on
> explicit.
This would be more appropriate:
--- 25/kernel/sched.c~a Sun Dec 29 13:53:24 2002
+++ 25-akpm/kernel/sched.c Sun Dec 29 13:54:27 2002
@@ -1264,7 +1264,14 @@ long interruptible_sleep_on_timeout(wait
void sleep_on(wait_queue_head_t *q)
{
+ static int count;
+
SLEEP_ON_VAR
+
+ if (count < 10 && !kernel_locked()) {
+ count++;
+ WARN_ON(1);
+ }
current->state = TASK_UNINTERRUPTIBLE;
> > Weaning ext3 off lock_kernel()
> > is a large, delicate and thus-far undesigned body of work. I've been
> > working on other stuff and it is quite unlikely that ext3 locking will
> > be redesigned in the 2.5 timeframe.
>
> Then ext3 has to live with using depricated interfaces during 2.6,
> what's the point?
We shouldn't generate tons of bogus warnings for something which
everyone knows about anyway.
On Sun, Dec 29, 2002 at 01:55:48PM -0800, Andrew Morton wrote:
> > Even if it's safe in that particular case, most code in the kernel runs
> > without BKL. This patch just makes the deprication of sleep_on
> > explicit.
>
> This would be more appropriate:
I don't think so. As you said before sleep_on is perfectly fine for
the small part of code still covered by BKL, so we should not impose
any runtime overhead (i.e. warnings) but rather remind at compile time.
That's what's so nice with the gcc extension (and you won't see it
anyway as long as you stay at egcs 1.1 :))
Christoph Hellwig wrote:
>
> On Sun, Dec 29, 2002 at 01:55:48PM -0800, Andrew Morton wrote:
> > > Even if it's safe in that particular case, most code in the kernel runs
> > > without BKL. This patch just makes the deprication of sleep_on
> > > explicit.
> >
> > This would be more appropriate:
>
> I don't think so. As you said before sleep_on is perfectly fine for
> the small part of code still covered by BKL, so we should not impose
> any runtime overhead (i.e. warnings) but rather remind at compile time.
>
> That's what's so nice with the gcc extension (and you won't see it
> anyway as long as you stay at egcs 1.1 :))
Others will. It will result in developers being distracted into "fixing"
non-bugs. It will introduce risk and it will delay the release of the
2.6 kernel.
Please concentrate on things which *matter*. Focus on getting this piece
of software into a deliverable state and do not be distracted into futzing
about with stuff which clearly was not addressed at the appropriate time.
> > > > Even if it's safe in that particular case, most code in the kernel runs
> > > > without BKL. This patch just makes the deprication of sleep_on
> > > > explicit.
> > >
> > > This would be more appropriate:
> >
> > I don't think so. As you said before sleep_on is perfectly fine for
> > the small part of code still covered by BKL, so we should not impose
> > any runtime overhead (i.e. warnings) but rather remind at compile time.
Especially as the fact that EXT-3 makes use of it doesn't affect
anybody writing user land apps. There isn't much point in warning
about things that are only relevant to readers of this list anyway.
> > That's what's so nice with the gcc extension (and you won't see it
> > anyway as long as you stay at egcs 1.1 :))
How long are we staying at 2.95.3 as the recommended compiler? A lot
of people have been using 3.x with the 2.5 tree successfully for a
while. What will be the recommended compiler for 2.6?
> Others will. It will result in developers being distracted into "fixing"
> non-bugs. It will introduce risk and it will delay the release of the
> 2.6 kernel.
featurefreeze()
{
> Please concentrate on things which *matter*. Focus on getting this piece
> of software into a deliverable state and do not be distracted into futzing
> about with stuff which clearly was not addressed at the appropriate time.
}
John.
On Sun, 29 Dec 2002, Andrew Morton wrote:
> We shouldn't generate tons of bogus warnings for something which
> everyone knows about anyway.
You could make the argument that we don't need to generate ANY bogus
warnings. For anything.
My feeling on the issue, there are lots of things in the to-do queue,
filling logs with warnings is a good way to get even more "what does this
warning mean" messages to the list and newsgroups.
--
bill davidsen <[email protected]>
CTO, TMR Associates, Inc
Doing interesting things with little computers since 1979.