[JFFS2] Don't pack on-medium structures, because GCC emits crappy code

If we use __attribute__((packed)), GCC will _also_ assume that the
structures aren't sensibly aligned, and it'll emit code to cope with
that instead of straight word load/save. This can be _very_ suboptimal
on architectures like ARM.

Ideally, we want an attribute which just tells GCC not to do any
padding, without the alignment side-effects. In the absense of that,
we'll just drop the 'packed' attribute and hope that everything stays as
it was (which to be fair is fairly much what we expect). And add some
paranoia checks in the initialisation code, which should be optimised
away completely in the normal case.

Signed-off-by: David Woodhouse <dwmw2@infradead.org>
This commit is contained in:
David Woodhouse
2006-05-15 00:49:43 +01:00
parent 52239da1b0
commit 3e68fbb59b
2 changed files with 20 additions and 8 deletions

View File

@@ -320,6 +320,18 @@ static int __init init_jffs2_fs(void)
{
int ret;
/* Paranoia checks for on-medium structures. If we ask GCC
to pack them with __attribute__((packed)) then it _also_
assumes that they're not aligned -- so it emits crappy
code on some architectures. Ideally we want an attribute
which means just 'no padding', without the alignment
thing. But GCC doesn't have that -- we have to just
hope the structs are the right sizes, instead. */
BUG_ON(sizeof(struct jffs2_unknown_node) != 12);
BUG_ON(sizeof(struct jffs2_raw_dirent) != 40);
BUG_ON(sizeof(struct jffs2_raw_inode) != 68);
BUG_ON(sizeof(struct jffs2_raw_summary) != 32);
printk(KERN_INFO "JFFS2 version 2.2."
#ifdef CONFIG_JFFS2_FS_WRITEBUFFER
" (NAND)"
@@ -327,7 +339,7 @@ static int __init init_jffs2_fs(void)
#ifdef CONFIG_JFFS2_SUMMARY
" (SUMMARY) "
#endif
" (C) 2001-2003 Red Hat, Inc.\n");
" (C) 2001-2006 Red Hat, Inc.\n");
jffs2_inode_cachep = kmem_cache_create("jffs2_i",
sizeof(struct jffs2_inode_info),