Skip to contentSkip to conversation

: Make the in-kernel AFS filesystem use AF_RXRPC.

AF_RXRPC Geert Uytterhoeven 3 participants 7 messages May 1, 2007 1 message not in the archive

Subsystems (2)
lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw

fs/afs/fsclient.c

Quoted 9 lines · Linux Kernel Mailing List

[AF_RXRPC]: Make the in-kernel AFS filesystem use AF_RXRPC.

Make the in-kernel AFS filesystem use AF_RXRPC instead of the old RxRPC code.

fs/afs/fsclient.c +2 -0
 
+		if (call->count < PAGE_SIZE) {
+			buffer = kmap_atomic(call->reply3, KM_USER0);
                                             ^^^^^^^^^^^^

This causes a compile failure on m68k:

| linux/fs/afs/fsclient.c: In function 'afs_deliver_fs_fetch_data':
| linux/fs/afs/fsclient.c:269: warning: dereferencing 'void *' pointer
| linux/fs/afs/fsclient.c:269: error: request for member 'virtual' in something not a structure or union

Probably you wanted to assign call->reply3 to a struct page pointer first, like you do for the call->unmarshall == 2 case.

fs/afs/fsclient.c +2 -1
@@ -266,7 +266,8 @@ static int afs_deliver_fs_fetch_data(str
 		call->unmarshall++;
 
 		if (call->count < PAGE_SIZE) {
-			buffer = kmap_atomic(call->reply3, KM_USER0);
+			page = call->reply3;
+			buffer = kmap_atomic(page, KM_USER0);
 			memset(buffer + PAGE_SIZE - call->count, 0,
 			       call->count);
 			kunmap_atomic(buffer, KM_USER0);

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds

maintainer
#2 lore raw
Quoted 1 line · Geert Uytterhoeven

This causes a compile failure on m68k:

Hmmm... I couldn't actually test m68k as I couldn't get it to compile. Can you give me a good config I can use as a base?

Quoted 3 lines
| linux/fs/afs/fsclient.c: In function 'afs_deliver_fs_fetch_data':
| linux/fs/afs/fsclient.c:269: warning: dereferencing 'void *' pointer
| linux/fs/afs/fsclient.c:269: error: request for member 'virtual' in something not a structure or union

Ummm... Maybe m68k kmap_atomic() should cast it... I think everywhere else it must be an inline function or something.

However, I'll subsume your patch into mine if you don't mind.

David

#3 lore raw
Quoted 6 lines · David Howells
Quoted 1 line · Geert Uytterhoeven

This causes a compile failure on m68k:

Hmmm... I couldn't actually test m68k as I couldn't get it to compile. Can you give me a good config I can use as a base?

Quoted 6 lines
Quoted 3 lines
| linux/fs/afs/fsclient.c: In function 'afs_deliver_fs_fetch_data':
| linux/fs/afs/fsclient.c:269: warning: dereferencing 'void *' pointer
| linux/fs/afs/fsclient.c:269: error: request for member 'virtual' in something not a structure or union

Ummm... Maybe m68k kmap_atomic() should cast it... I think everywhere else it must be an inline function or something.

On architectures that #define WANT_PAGE_VIRTUAL (frv/m68k/mips/extensa), page_address(page) expands to (page)->virtual, which fails if page is a void *.

Quoted 1 line

However, I'll subsume your patch into mine if you don't mind.

Please apply ;-)

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds

maintainer
#4 lore raw
Quoted 3 lines · Geert Uytterhoeven

On architectures that #define WANT_PAGE_VIRTUAL (frv/m68k/mips/extensa), page_address(page) expands to (page)->virtual, which fails if page is a void *.

On FRV, kmap_atomic() should always be an inline function (I wrote it:-):

	static inline void *kmap_atomic(struct page *page, enum km_type type)
	{
	...
	}

David

#5 lore raw
Quoted 5 lines · David Howells
Quoted 3 lines · Geert Uytterhoeven

On architectures that #define WANT_PAGE_VIRTUAL (frv/m68k/mips/extensa), page_address(page) expands to (page)->virtual, which fails if page is a void *.

Quoted 8 lines

On FRV, kmap_atomic() should always be an inline function (I wrote it:-):

	static inline void *kmap_atomic(struct page *page, enum km_type type)
	{
	...
	}

David

That's why I was actually surprised to find out frv has WANT_PAGE_VIRTUAL...

However, upon a closer look, have you tried without CONFIG_HIGHMEM? ;-)

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds

#6 lore raw
In reply to Geert Uytterhoeven
Quoted 8 lines · Geert Uytterhoeven

This causes a compile failure on m68k:

| linux/fs/afs/fsclient.c: In function 'afs_deliver_fs_fetch_data':
| linux/fs/afs/fsclient.c:269: warning: dereferencing 'void *' pointer
| linux/fs/afs/fsclient.c:269: error: request for member 'virtual' in something not a structure or union

Probably you wanted to assign call->reply3 to a struct page pointer first, like you do for the call->unmarshall == 2 case.

Hmm. Wouldn't it be nicer to make "kmap_atomic()" be an inline function instead? That way, if you pass it a "void *", it still just works, and you get a much nicer/more readable error message too when the types actually don't match?

Of course, there may be some reason why <linux/highmem.h> does those with #define's rather than inline functions. Maybe "page_address()" or "pagefault_disabled()" is not declared at that point, and asking it to be declared might cause header file inclusion recursion?

Linus

#7 lore raw
Quoted 7 lines · Linus Torvalds
Quoted 6 lines · Geert Uytterhoeven

This causes a compile failure on m68k:

| linux/fs/afs/fsclient.c: In function 'afs_deliver_fs_fetch_data':
| linux/fs/afs/fsclient.c:269: warning: dereferencing 'void *' pointer
| linux/fs/afs/fsclient.c:269: error: request for member 'virtual' in something not a structure or union
Quoted 6 lines
Quoted 1 line

Probably you wanted to assign call->reply3 to a struct page pointer first, like you do for the call->unmarshall == 2 case.

Hmm. Wouldn't it be nicer to make "kmap_atomic()" be an inline function instead? That way, if you pass it a "void *", it still just works, and you get a much nicer/more readable error message too when the types actually don't match?

Of course.

Quoted 4 lines

Of course, there may be some reason why <linux/highmem.h> does those with #define's rather than inline functions. Maybe "page_address()" or "pagefault_disabled()" is not declared at that point, and asking it to be declared might cause header file inclusion recursion?

That's what I was afraid of, too...

However, this change seems to build fine (for one of my test configs):

Convert kmap_atomic() in the non-highmem case from a macro to a static inline function, for better type-checking and the ability to pass void pointers instead of struct page pointers.

include/linux/highmem.h +8 -2
@@ -42,8 +42,14 @@ static inline void *kmap(struct page *pa
 
 #define kunmap(page) do { (void) (page); } while (0)
 
-#define kmap_atomic(page, idx) \
-	({ pagefault_disable(); page_address(page); })
+#include <asm/kmap_types.h>
+
+static inline void *kmap_atomic(struct page *page, enum km_type idx)
+{
+	pagefault_disable();
+	return page_address(page);
+}
+
 #define kunmap_atomic(addr, idx)	do { pagefault_enable(); } while (0)
 #define kmap_atomic_pfn(pfn, idx)	kmap_atomic(pfn_to_page(pfn), (idx))
 #define kmap_atomic_to_page(ptr)	virt_to_page(ptr)

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds