linux内存未四字节对齐,linux的nand中计算ECC的时候buf地址不4字节对齐的问题

最近发现的 __nand_calculate_ecc(buf,eccsize,code)中buf地址非4字节对齐导致的异常问题， 经查发现是由于fs/jffs2/gc.c中

jffs2_garbage_collect_live()函数中获取的start值非对齐引起，这个start值传给给jffs2_garbage_collect_dnode()，

在jffs2_garbage_collect_dnode()中，会根据start值算出对应的frag(frag=jffs2_lookup_node_frag(&f->fragtree,start))，然

后找这个frag的前一个frag, 如下代码所示：

while((frag = frag_prev(frag)) && frag->ofs >= min) {

/* If the previous frag doesn't even reach the beginning, there's

excessive fragmentation. Just merge. */

if (frag->ofs > min) {

D1(printk(KERN_DEBUG "Expanding down to cover partial frag (0x%x-0x%x)\n",

frag->ofs, frag->ofs+frag->size));

start = frag->ofs;

continue;

}

/* OK. This frag holds the first byte of the page. */

if (!frag->node || !frag->node->raw) {

D1(printk(KERN_DEBUG "First frag in page is hole (0x%x-0x%x). Not expanding down.\n",

frag->ofs, frag->ofs+frag->size));

start = frag->ofs;

break;

} else {

/* OK, it's a frag which extends to the beginning of the page. Does it live

in a block which is still considered clean? If so, don't obsolete it.

If not, cover it anyway. */

struct jffs2_raw_node_ref *raw = frag->node->raw;

struct jffs2_eraseblock *jeb;

jeb = &c->blocks[raw->flash_offset / c->sector_size];

if (jeb == c->gcblock) {

D1(printk(KERN_DEBUG "Expanding down to cover frag (0x%x-0x%x) in gcblock at %08x\n",

frag->ofs, frag->ofs+frag->size, ref_offset(raw)));

start = frag->ofs;

break;

}

if (!ISDIRTY(jeb->dirty_size + jeb->wasted_size)) {

D1(printk(KERN_DEBUG "Not expanding down to cover frag (0x%x-0x%x) in clean block %08x\n",

frag->ofs, frag->ofs+frag->size, jeb->offset));

//add for fix bug--start

if(start%4){

start = frag->ofs;

printk("expanding down to cover frag (0x%x-0x%x) in clean block %08x, anyway, other wise unalligned exception

will be raised if start address unalligned\n",frag->ofs, frag->ofs+frag->size, jeb->offset);

}

//add for fix bug--start

break;

}

D1(printk(KERN_DEBUG "Expanding down to cover frag (0x%x-0x%x) in dirty block %08x\n",         frag->ofs, frag->ofs+frag->size, jeb->offset));     start = frag->ofs;     break;    }   }  正常情况下，找到前一个frag的frag->ofs是大于min的，而且是4字节对齐的，但现在的情况是frag->ofs刚好等于min， 而且又满足  if (!ISDIRTY(jeb->dirty_size + jeb->wasted_size)) 条件，导致start没有更新，start仍然未对齐，进而导致异常。  这个问题是更换toolchain(从uclibc到glibc)引起的，用uclibc时，while中的if (frag->ofs > min) 条件满足，start被更新成对齐的地址。  更深层的原因暂时没能找到，暂时作如此修改。    另外，发现linux内核跑起来后，过了大概10几秒钟，会跑到jffs2_garbage_collect_pass()中的以下代码，不断调用jffs2_erase_pending_blocks(),   /* If there are any blocks which need erasing, erase them now */   if (!list_empty(&c->erase_complete_list) ||    !list_empty(&c->erase_pending_list)) {    spin_unlock(&c->erase_completion_lock);    D1(printk(KERN_DEBUG "jffs2_garbage_collect_pass() erasing pending blocks\n"));    if (jffs2_erase_pending_blocks(c, 1)) {     mutex_unlock(&c->alloc_sem);     return 0;    }    D1(printk(KERN_DEBUG "No progress from erasing blocks; doing GC anyway\n"));    spin_lock(&c->erase_completion_lock);   }     在jffs2_erase_pending_blocks()中，也只是调用到jffs2_mark_erased_block()。这个过程会持续大概半分钟，在这半分钟内，几乎不能运行其它APP，  不知道是否正常，以前的版本也没有注意，使用uclibc toolchain也一样。