Re: [PATCH v3 14/14] lib/test_printf.c: test dentry printing

[Andrew Morton <akpm@linux-foundation.org>]

On Fri, 04 Dec 2015 09:16:02 +0100 Rasmus Villemoes <linux@rasmusvillemoes.dk> wrote:

> On Fri, Dec 04 2015, Andrew Morton <akpm@linux-foundation.org> wrote:
> 
> > On Thu,  3 Dec 2015 21:51:13 +0100 Rasmus Villemoes <linux@rasmusvillemoes.dk> wrote:
> >
> >> +static struct dentry test_dentry[4] __initdata = {
> >> +	{ .d_parent = &test_dentry[0],
> >> +	  .d_name = { .len = 3, .name = test_dentry[0].d_iname },
> >> +	  .d_iname = "foo" },
> >
> > Confused.  qstr has no .len.
> >
> > lib/test_printf.c:332: error: unknown field 'len' specified in initializer
> 
> Huh? It goes without saying that I've compiled and run this, and never
> seen this problem. struct qstr does have a len member, though it's
> hidden inside an anonymous struct inside an anonymous union. But at
> least my gcc (4.9) has no problem with those initializers - and
> fs/dcache.c happily accesses name->len all the time (how else would one
> get at the member). Also note the lack of 0day complaints. What tool
> gave that error?
> 

Ah, OK, that's the gcc-4.4.4 bug with initialization of anonymous
unions.

I really should get a new compiler, but running old compilers has value
- it finds problems with old compilers!

I've never found a workaround for this problem so the fix thus far has
been to do the initialization in regular old C code.

However it appears that the QSTR_INIT() macro (which you should have
used anyway!) contains some magic sauce.  This fixes it:

--- a/lib/test_printf.c~lib-test_printfc-test-dentry-printing-fix
+++ a/lib/test_printf.c
@@ -329,16 +329,16 @@ uuid(void)
 
 static struct dentry test_dentry[4] __initdata = {
 	{ .d_parent = &test_dentry[0],
-	  .d_name = { .len = 3, .name = test_dentry[0].d_iname },
+	  .d_name = QSTR_INIT(test_dentry[0].d_iname, 3),
 	  .d_iname = "foo" },
 	{ .d_parent = &test_dentry[0],
-	  .d_name = { .len = 5, .name = test_dentry[1].d_iname },
+	  .d_name = QSTR_INIT(test_dentry[1].d_iname, 5),
 	  .d_iname = "bravo" },
 	{ .d_parent = &test_dentry[1],
-	  .d_name = { .len = 4, .name = test_dentry[2].d_iname },
+	  .d_name = QSTR_INIT(test_dentry[2].d_iname, 4),
 	  .d_iname = "alfa" },
 	{ .d_parent = &test_dentry[2],
-	  .d_name = { .len = 5, .name = test_dentry[3].d_iname },
+	  .d_name = QSTR_INIT(test_dentry[3].d_iname, 5),
 	  .d_iname = "romeo" },
 };
 
I assume the code still works ;)



And while I'm there...

From: Andrew Morton <akpm@linux-foundation.org>
Subject: include/linux/dcache.h: remove semicolons from HASH_LEN_DECLARE

A little cleanup - the invocation site provdes the semicolon.

Cc: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
---

 include/linux/dcache.h |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff -puN include/linux/dcache.h~include-linux-dcacheh-remove-semicolons-from-hash_len_declare include/linux/dcache.h
--- a/include/linux/dcache.h~include-linux-dcacheh-remove-semicolons-from-hash_len_declare
+++ a/include/linux/dcache.h
@@ -27,10 +27,10 @@ struct vfsmount;
 
 /* The hash is always the low bits of hash_len */
 #ifdef __LITTLE_ENDIAN
- #define HASH_LEN_DECLARE u32 hash; u32 len;
+ #define HASH_LEN_DECLARE u32 hash; u32 len
  #define bytemask_from_count(cnt)	(~(~0ul << (cnt)*8))
 #else
- #define HASH_LEN_DECLARE u32 len; u32 hash;
+ #define HASH_LEN_DECLARE u32 len; u32 hash
  #define bytemask_from_count(cnt)	(~(~0ul >> (cnt)*8))
 #endif
 
_