Skip to contentSkip to conversation

Infrastructure to mark exported symbols as unused-for-removal-soon

PATCH In review Arjan van de Ven 6 participants 10 messages May 2, 2006 – May 11, 2006

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/17 Infrastructure to mark exported symbols as unused-for-removal-soon

Hi,
As discussed on lkml before; the patch with the infrastructure to deprecate unused symbols

This is patch one in a series of 17; to not overload lkml the other 16 will be mailed direct; people who want to see them all can see them at http://www.fenrus.org/unused

This patch temporarily adds EXPORT_UNUSED_SYMBOL and EXPORT_UNUSED_SYMBOL_GPL. These will be used as transition measure for symbols that aren't used in the kernel and are on the way out. When a module uses such a symbol, a warning is printk'd at modprobe time.

The main reason for removing unused exports is size: eacho export takes roughly between 100 and 150 bytes of kernel space in the binary. This patch gives users the option to immediately get this size gain via a config option.

It would be nice to at least get this infrastructure into 2.6.17 even if the rest of this series won't get there.

---
 Documentation/feature-removal-schedule.txt |   10 +++
 include/asm-generic/vmlinux.lds.h          |   28 ++++++++++
 include/linux/module.h                     |   20 +++++++
 kernel/module.c                            |   79 +++++++++++++++++++++++++++--
 lib/Kconfig.debug                          |   14 +++++
 5 files changed, 148 insertions(+), 3 deletions(-)
Documentation/feature-removal-schedule.txt +10 -0
Index: linux-2.6.17-rc3-mm1-unused/Documentation/feature-removal-schedule.txt
===================================================================
@@ -22,6 +22,16 @@ Who:	Adrian Bunk <[email protected]>
 
 ---------------------------
 
+What:	Unused EXPORT_SYMBOL/EXPORT_SYMBOL_GPL exports
+	(temporary transition config option provided until then)
+	The transition config option will also be removed at the same time.
+When:	before 2.6.19
+Why:	Unused symbols are both increasing the size of the kernel binary
+	and are often a sign of "wrong API"
+Who:	Arjan van de Ven <[email protected]>
+
+---------------------------
+
 What:	RCU API moves to EXPORT_SYMBOL_GPL
 When:	April 2006
 Files:	include/linux/rcupdate.h, kernel/rcupdate.c
include/asm-generic/vmlinux.lds.h +28 -0
Index: linux-2.6.17-rc3-mm1-unused/include/asm-generic/vmlinux.lds.h
===================================================================
@@ -58,6 +58,20 @@
 		VMLINUX_SYMBOL(__stop___ksymtab_gpl) = .;		\
 	}								\
 									\
+	/* Kernel symbol table: Normal unused symbols */		\
+	__ksymtab_unused  : AT(ADDR(__ksymtab_unused) - LOAD_OFFSET) {	\
+		VMLINUX_SYMBOL(__start___ksymtab_unused) = .;		\
+		*(__ksymtab_unused)					\
+		VMLINUX_SYMBOL(__stop___ksymtab_unused) = .;		\
+	}								\
+									\
+	/* Kernel symbol table: GPL-only unused symbols */		\
+	__ksymtab_unused_gpl : AT(ADDR(__ksymtab_unused_gpl) - LOAD_OFFSET) { \
+		VMLINUX_SYMBOL(__start___ksymtab_unused_gpl) = .;	\
+		*(__ksymtab_unused_gpl)					\
+		VMLINUX_SYMBOL(__stop___ksymtab_unused_gpl) = .;	\
+	}								\
+									\
 	/* Kernel symbol table: GPL-future-only symbols */		\
 	__ksymtab_gpl_future : AT(ADDR(__ksymtab_gpl_future) - LOAD_OFFSET) { \
 		VMLINUX_SYMBOL(__start___ksymtab_gpl_future) = .;	\
@@ -79,6 +93,20 @@
 		VMLINUX_SYMBOL(__stop___kcrctab_gpl) = .;		\
 	}								\
 									\
+	/* Kernel symbol table: Normal unused symbols */		\
+	__kcrctab_unused  : AT(ADDR(__kcrctab_unused) - LOAD_OFFSET) {	\
+		VMLINUX_SYMBOL(__start___kcrctab_unused) = .;		\
+		*(__kcrctab_unused)					\
+		VMLINUX_SYMBOL(__stop___kcrctab_unused) = .;		\
+	}								\
+									\
+	/* Kernel symbol table: GPL-only unused symbols */		\
+	__kcrctab_unused_gpl : AT(ADDR(__kcrctab_unused_gpl) - LOAD_OFFSET) { \
+		VMLINUX_SYMBOL(__start___kcrctab_unused_gpl) = .;	\
+		*(__kcrctab_unused_gpl)					\
+		VMLINUX_SYMBOL(__stop___kcrctab_unused_gpl) = .;	\
+	}								\
+									\
 	/* Kernel symbol table: GPL-future-only symbols */		\
 	__kcrctab_gpl_future : AT(ADDR(__kcrctab_gpl_future) - LOAD_OFFSET) { \
 		VMLINUX_SYMBOL(__start___kcrctab_gpl_future) = .;	\
include/linux/module.h +20 -0
Index: linux-2.6.17-rc3-mm1-unused/include/linux/module.h
===================================================================
@@ -204,6 +204,15 @@ void *__symbol_get_gpl(const char *symbo
 #define EXPORT_SYMBOL_GPL_FUTURE(sym)				\
 	__EXPORT_SYMBOL(sym, "_gpl_future")
 
+
+#ifdef CONFIG_UNUSED_SYMBOLS
+#define EXPORT_UNUSED_SYMBOL(sym) __EXPORT_SYMBOL(sym, "_unused")
+#define EXPORT_UNUSED_SYMBOL_GPL(sym) __EXPORT_SYMBOL(sym, "_unused_gpl")
+#else
+#define EXPORT_UNUSED_SYMBOL(sym)
+#define EXPORT_UNUSED_SYMBOL_GPL(sym)
+#endif
+
 #endif
 
 struct module_ref
@@ -278,6 +287,15 @@ struct module
 	unsigned int num_gpl_syms;
 	const unsigned long *gpl_crcs;
 
+	/* unused exported symbols. */
+	const struct kernel_symbol *unused_syms;
+	unsigned int num_unused_syms;
+	const unsigned long *unused_crcs;
+	/* GPL-only, unused exported symbols. */
+	const struct kernel_symbol *unused_gpl_syms;
+	unsigned int num_unused_gpl_syms;
+	const unsigned long *unused_gpl_crcs;
+
 	/* symbols that will be GPL-only in the near future. */
 	const struct kernel_symbol *gpl_future_syms;
 	unsigned int num_gpl_future_syms;
@@ -470,6 +488,8 @@ void module_remove_driver(struct device_
 #define EXPORT_SYMBOL(sym)
 #define EXPORT_SYMBOL_GPL(sym)
 #define EXPORT_SYMBOL_GPL_FUTURE(sym)
+#define EXPORT_UNUSED_SYMBOL(sym)
+#define EXPORT_UNUSED_SYMBOL_GPL(sym)
 
 /* Given an address, look for it in the exception tables. */
 static inline const struct exception_table_entry *
kernel/module.c +76 -3

137 lines not on this page · raw message

lib/Kconfig.debug +14 -0
Index: linux-2.6.17-rc3-mm1-unused/lib/Kconfig.debug
===================================================================
@@ -32,6 +32,20 @@ config DEBUG_SHIRQ
 	 to be able to handle interrupts coming in at those points; some don't and
 	 need to be caught.
 
+config UNUSED_SYMBOLS
+	bool "Enable unused/obsolete exported symbols"
+	help
+	  Unused but exported symbols make the kernel needlessly bigger. For that
+	  reason most of these unused exports will soon be removed. This option is
+	  provided temporarily to provide a transition period in case some external
+	  kernel module needs one of these symbols anyway. If you encounter such
+	  a case in your module, consider if you are actually using the right API.
+	  (rationale: since nobody in the kernel is using this in a module, there is
+	  a pretty good chance it's actually the wrong interface to use)
+	  If you really need the symbol, please send a mail to the linux kernel
+	  mailing list mentioning the symbol and why you really need it, and what
+	  the merge plan to the mainline kernel for your module is.
+
 config DEBUG_KERNEL
 	bool "Kernel debugging"
 	help
#1 lore raw
Quoted 8 lines · On Tue, 02 May 2006 16:53:07 +0200 Arjan van de Ven wrote:

Hi,
As discussed on lkml before; the patch with the infrastructure to deprecate unused symbols

This is patch one in a series of 17; to not overload lkml the other 16 will be mailed direct; people who want to see them all can see them at http://www.fenrus.org/unused

Quoted 8 lines

This patch temporarily adds EXPORT_UNUSED_SYMBOL and EXPORT_UNUSED_SYMBOL_GPL. These will be used as transition measure for symbols that aren't used in the kernel and are on the way out. When a module uses such a symbol, a warning is printk'd at modprobe time.

The main reason for removing unused exports is size: eacho export takes roughly between 100 and 150 bytes of kernel space in the binary. This patch gives users the option to immediately get this size gain via a config option.

Do the exports take any space at runtime in RAM? or is this only on-disk or wherever the kernel image lives?

Quoted 6 lines

It would be nice to at least get this infrastructure into 2.6.17 even if the rest of this series won't get there.

Documentation/feature-removal-schedule.txt Who: Adrian Bunk <[email protected]> · line 31

Quoted 10 lines
Documentation/feature-removal-schedule.txt +7 -0
@@ -22,6 +22,16 @@ Who:	Adrian Bunk <[email protected]>
 
 ---------------------------
 
+What:	Unused EXPORT_SYMBOL/EXPORT_SYMBOL_GPL exports
+	(temporary transition config option provided until then)
+	The transition config option will also be removed at the same time.
+When:	before 2.6.19
+Why:	Unused symbols are both increasing the size of the kernel binary
+	and are often a sign of "wrong API"
+Who:	Arjan van de Ven <[email protected]>

scsi patch comments (only one that I have seen) say:
+EXPORT_UNUSED_SYMBOL(scsi_print_status); /* removal in 2.6.19 */

and When: above says "before 2.6.19". Those don't agree. Please fix. Thanks.

---
~Randy

#2 lore raw
Quoted 14 lines · Randy.Dunlap
Quoted 12 lines · On Tue, 02 May 2006 16:53:07 +0200 Arjan van de Ven wrote:

Hi,
As discussed on lkml before; the patch with the infrastructure to deprecate unused symbols

This is patch one in a series of 17; to not overload lkml the other 16 will be mailed direct; people who want to see them all can see them at http://www.fenrus.org/unused

This patch temporarily adds EXPORT_UNUSED_SYMBOL and EXPORT_UNUSED_SYMBOL_GPL.
These will be used as transition measure for symbols that aren't used in the kernel and are on the way out. When a module uses such a symbol, a warning is printk'd at modprobe time.

Quoted 5 lines
Quoted 3 lines

The main reason for removing unused exports is size: eacho export takes roughly between 100 and 150 bytes of kernel space in the binary. This patch gives users the option to immediately get this size gain via a config option.

Do the exports take any space at runtime in RAM?

yes; roughly 100 to 150 bytes or so

Quoted 5 lines

scsi patch comments (only one that I have seen) say:
+EXPORT_UNUSED_SYMBOL(scsi_print_status); /* removal in 2.6.19 */

and When: above says "before 2.6.19". Those don't agree. Please fix. Thanks.

there's no conflict actually; they'll be gone in 2.6.19, by removing them just before that ;)

#3 lore raw
In reply to Arjan van de Ven
Quoted 5 lines · Arjan van de Ven

As discussed on lkml before; the patch with the infrastructure to deprecate unused symbols

This is patch one in a series of 17; to not overload lkml the other 16 will be mailed direct; people who want to see them all can see them at http://www.fenrus.org/unused

A lot of these patches go through major APIs and seemingly-randomly prepare to unexport things based on whether they are presently used within modules.

So, for example, drivers/base/attribute_container.c gets a whole pile of exports scheduled for removal, regardless of whether the resulting module API makes *sense*. Ditto scsi core. And lib/*.

For example this:

EXPORT_SYMBOL(generic_getxattr);
-EXPORT_SYMBOL(generic_listxattr); +EXPORT_UNUSED_SYMBOL(generic_listxattr); /* removal in 2.6.19 */
EXPORT_SYMBOL(generic_setxattr); EXPORT_SYMBOL(generic_removexattr);

just seems random to me, and it's setting us up for later churn.

So hum. Don't you think it'd be better to look at each API as a whole, make decisions about what parts of it _should_ be offered to modules, rather then looking empirically at which parts presently _need_ to be exported?

#4 lore raw
Quoted 4 lines · Andrew Morton

So hum. Don't you think it'd be better to look at each API as a whole, make decisions about what parts of it _should_ be offered to modules, rather then looking empirically at which parts presently _need_ to be exported?

Well so far we as kernel developers have been rather bad at it, with the result that there are 900 unused ones roughly. Each export takes somewhere between 100 and 150 bytes. *WITHOUT ANY BENEFIT*. The reason to remove them all is to save that memory NOW. It's easy to add an export back later if it gets used. Yes that is churn, but it's minor churn. The price for not doing that is a bigger kernel for everyone, today, without any positive gain of that space..

(and this size excludes even those functions that aren't used at all, but are only there to be exported. Adrian has been working on removing the really unused functions in the kernel, via static marking and then gcc noticing the unusedness, but once they're exported that breaks down)

So I think personally it's worth biting the bullet. I expect 95% of those 900 to never ever come back. Those 5% will churn, sure. But, to a large degree, the fact that there's no user is an indication that the API may well not be right in the first place, or not in demand.

#5 lore raw
Quoted 6 lines · Arjan van de Ven
Quoted 4 lines · Andrew Morton

So hum. Don't you think it'd be better to look at each API as a whole, make decisions about what parts of it _should_ be offered to modules, rather then looking empirically at which parts presently _need_ to be exported?

Quoted 7 lines

So I think personally it's worth biting the bullet. I expect 95% of those 900 to
never ever come back. Those 5% will churn, sure. But, to a large degree, the fact
that there's no user is an indication that the API may well not be right in the
first place, or not in demand.

[ Your mailer... ]

Rusty once pointed out that Linux doesn't ever implement complete interfaces like students are usually told to do. Instead, only functions that are actually used are implemented. Among other things, this makes sure that implemented code is actually tested and works as advertised. Unused code, on the other hand, tends to bitrot.

Well, maybe Rusty was wrong and I shouldn't have listened. But his points made a lot of sense back then. Who knows, it might still make sense.

Jörn

Signature · 2 lines
This above all: to thine own self be true.
-- Shakespeare
#6 lore raw
In reply to Arjan van de Ven
Quoted 9 lines · Arjan van de Ven
Quoted 4 lines · Andrew Morton

So hum. Don't you think it'd be better to look at each API as a whole, make decisions about what parts of it _should_ be offered to modules, rather then looking empirically at which parts presently _need_ to be exported?

Well so far we as kernel developers have been rather bad at it, with the result that there are 900 unused ones roughly. Each export takes somewhere between 100 and 150 bytes.

Of course the more technically beneficial approach would be to stop exports taking such ludicrous amounts of memory.

#7 lore raw
In reply to Arjan van de Ven
Quoted 3 lines · Arjan

This is patch one in a series of 17; to not overload lkml the other 16 will be mailed direct; people who want to see them all can see them at http://www.fenrus.org/unused

Well ... here's one case where your patch series is broken.

Argh - I almost missed this one. My mailer is setup to tag all incoming lkml email that mentions the magic word 'cpuset'. But it is not setup to catch indirect patches, needless to say.

One of your proposed changes (the only one I reviewed) removed the only EXPORT_SYMBOL_GPL in kernel/cpuset.c. That EXPORT is needed because the routine in question is called from inlines which modules use.

I can't help but wonder how many more such cases your patches miss.

The details ...

Your patch includes this change:

+++++++++++++++++++++++ begin +++++++++++++++++++++++

kernel/cpuset.c +1 -1
Index: linux-2.6.17-rc3-mm1-unused/kernel/cpuset.c
===================================================================
@@ -2338,7 +2338,7 @@ int cpuset_mem_spread_node(void)
 	current->cpuset_mem_spread_rotor = node;
 	return node;
 }
-EXPORT_SYMBOL_GPL(cpuset_mem_spread_node);
+EXPORT_UNUSED_SYMBOL_GPL(cpuset_mem_spread_node); /* removal in 2.6.19 */
 
 /**
  * cpuset_excl_nodes_overlap - Do we overlap @p's mem_exclusive ancestors?

++++++++++++++++++++++++ end +++++++++++++++++++++++

Andrew added this EXPORT, with the following patch:

+++++++++++++++++++++++ begin +++++++++++++++++++++++
From: Andrew Morton <[email protected]>

It's called from inlines which modules use.

---

 kernel/cpuset.c |    1 +
 1 files changed, 1 insertion(+)

diff -puN kernel/cpuset.c~cpuset-memory-spread-basic-implementation-fix kernel/cpuset.c --- devel/kernel/cpuset.c~cpuset-memory-spread-basic-implementation-fix 2006-02-06 23:51:0 0.000000000 -0800
+++ devel-akpm/kernel/cpuset.c 2006-02-06 23:51:00.000000000 -0800
@@ -2220,6 +2220,7 @@ int cpuset_mem_spread_node(void)

        current->cpuset_mem_spread_rotor = node;
        return node;

}
+EXPORT_SYMBOL_GPL(cpuset_mem_spread_node);

/**
* cpuset_excl_nodes_overlap - Do we overlap @p's mem_exclusive ancestors?
++++++++++++++++++++++++ end +++++++++++++++++++++++

Signature · 3 lines
                  I won't rest till it's the best ...
                  Programmer, Linux Scalability
                  Paul Jackson <[email protected]> 1.925.600.0401
#8 lore raw
Quoted 7 lines · Paul Jackson
Quoted 3 lines · Arjan

This is patch one in a series of 17; to not overload lkml the other 16 will be mailed direct; people who want to see them all can see them at http://www.fenrus.org/unused

Well ... here's one case where your patch series is broken.

Quoted 7 lines

Argh - I almost missed this one. My mailer is setup to tag all incoming lkml email that mentions the magic word 'cpuset'. But it is not setup to catch indirect patches, needless to say.

One of your proposed changes (the only one I reviewed) removed the only EXPORT_SYMBOL_GPL in kernel/cpuset.c. That EXPORT is needed because the routine in question is called from inlines which modules use.

not in the configs I tested at least... but maybe I need to add a specific config to my set..

#9 lore raw
Quoted 2 lines · Arjan

not in the configs I tested at least... but maybe I need to add a specific config to my set..

Dang - I think you're right.

The original version of the patch that provided the new routine cpuset_mem_spread_node() on about Feb 2, 2006 required that EXPORT, as in that version cpuset_mem_spread_node() was called from the inline versions of page_cache_alloc() and page_cache_alloc_cold().

But the final version of the cpuset_mem_spread_node() patch, on about Feb 9, 2006, does not seem to require that EXPORT, because the callers page_cache_alloc() and page_cache_alloc_code() were taken -out-of-line- for the configurations that made use of cpuset_mem_spread_node().

However the EXPORT had already been added on about Feb 6 or 7, when everyone and his brother noticed that I had broken the build with my Feb 2 patch.

Signature · 3 lines
                  I won't rest till it's the best ...
                  Programmer, Linux Scalability
                  Paul Jackson <[email protected]> 1.925.600.0401