Skip to contentSkip to conversation

possible fix for broken floppy driver since 2.5.13

RFC In review Mikael Pettersson 3 participants 5 messages May 20, 2002 – May 26, 2002

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

RFC 1/1 possible fix for broken floppy driver since 2.5.13

Since 2.5.13 I've been unable to use drivers/block/floppy.c. There were two symptoms: /dev/fd0 was read-only until after the first read, and writes wrote currupt data to the media.

The patch below against 2.5.16 fixes these problems for me:

- The read-only problem was caused by a getblk() call in floppy_revalidate() which had been commented out (2.5.13 did away with getblk() altogether.) This call is necessary, so the patch reintroduces a private getblk() in floppy.c.

- The data corruption on writes was caused by new code in fs/block_dev.c:do_open() which changed the block size after the call to bdev->bd_op->open().
Disabling this change to bdev->bd_block_size fixed the data corruption. I don't know what this code was trying to do, but my system works fine without it.

I did some tracing of the "struct request*"'s delivered to the floppy driver, and with vanilla 2.5.16 the parameters (all the '*sectors*' fields) looked very very different from what 2.5.12 passed to the driver.

Al: You did the block dev changes for 2.5.13 -- care to comment?

/Mikael

drivers/block/floppy.c +22 -15
diff -ruN linux-2.5.16/drivers/block/floppy.c linux-2.5.16.fix-floppy/drivers/block/floppy.c
@@ -3852,6 +3852,25 @@
 	return 0;
 }
 
+static struct buffer_head *floppy_getblk0(kdev_t dev)
+{
+	struct block_device *bdev;
+	struct buffer_head *bh;
+	int size;
+
+	bdev = bdget(kdev_t_to_nr(dev));
+	if (!bdev) {
+		printk("No block device for %s\n", __bdevname(dev));
+		BUG();
+	}
+	size = bdev->bd_block_size;
+	if (!size)
+		size = 1024;
+	bh = __getblk(bdev, 0, size);
+	atomic_dec(&bdev->bd_count);
+	return bh;
+}
+
 /* revalidate the floppy disk, i.e. trigger format autodetection by reading
  * the bootblock (block 0). "Autodetection" is also needed to check whether
  * there is a disk in the drive at all... Thus we also do it for fixed
@@ -3859,7 +3878,6 @@
 static int floppy_revalidate(kdev_t dev)
 {
 #define NO_GEOM (!current_type[drive] && !TYPE(dev))
-	struct buffer_head * bh;
 	int drive=DRIVE(dev);
 	int cf;
 
@@ -3886,29 +3904,18 @@
 		if (cf)
 			UDRS->generation++;
 		if (NO_GEOM){
-#if 0
-	/*
-	 * What the devil is going on here?  We are not guaranteed to do
-	 * any IO and ENXIO case is nothing but ENOMEM in disguise - it
-	 * happens if and only if buffer cache is out of memory.  WTF?
-	 */
 			/* auto-sensing */
-			int size = floppy_blocksizes[minor(dev)];
-			if (!size)
-				size = 1024;
-			if (!(bh = getblk(dev,0,size))){
+			struct buffer_head *bh = floppy_getblk0(dev);
+			if (!bh) {
 				process_fd_request();
 				return -ENXIO;
 			}
-			if (bh && !buffer_uptodate(bh))
+			if (!buffer_uptodate(bh))
 				ll_rw_block(READ, 1, &bh);
 			process_fd_request();
 			wait_on_buffer(bh);
 			brelse(bh);
 			return 0;
-#endif
-			process_fd_request();
-			return 0;
 		}
 		if (cf)
 			poll_drive(0, FD_RAW_NEED_DISK);
fs/block_dev.c +1 -10
diff -ruN linux-2.5.16/fs/block_dev.c linux-2.5.16.fix-floppy/fs/block_dev.c
@@ -606,16 +606,7 @@
 			goto out2;
 	}
 	bdev->bd_inode->i_size = blkdev_size(dev);
-	if (!bdev->bd_openers) {
-		unsigned bsize = bdev_hardsect_size(bdev);
-		while (bsize < PAGE_CACHE_SIZE) {
-			if (bdev->bd_inode->i_size & bsize)
-				break;
-			bsize <<= 1;
-		}
-		bdev->bd_block_size = bsize;
-		bdev->bd_inode->i_blkbits = blksize_bits(bsize);
-	}
+	bdev->bd_inode->i_blkbits = blksize_bits(block_size(bdev));
 	bdev->bd_openers++;
 	unlock_kernel();
 	up(&bdev->bd_sem);
#1 lore raw
Quoted 10 lines · Mikael Pettersson

Since 2.5.13 I've been unable to use drivers/block/floppy.c. There were two symptoms: /dev/fd0 was read-only until after the first read, and writes wrote currupt data to the media.

The patch below against 2.5.16 fixes these problems for me:

- The read-only problem was caused by a getblk() call in floppy_revalidate() which had been commented out (2.5.13 did away with getblk() altogether.) This call is necessary, so the patch reintroduces a private getblk() in floppy.c.

Please don't use getblk(), but go directly through the bio interface. In 2.5 the buffer_heads are just a legacy interface for filesystems and are not supposed to be used by lowlevel drivers.

#2 lore raw
Quoted 7 lines · Christoph Hellwig
Quoted 6 lines · Mikael Pettersson

Since 2.5.13 I've been unable to use drivers/block/floppy.c. There were two symptoms: /dev/fd0 was read-only until after the first read, and writes wrote currupt data to the media.

The patch below against 2.5.16 fixes these problems for me:

Quoted 8 lines
Quoted 4 lines

- The read-only problem was caused by a getblk() call in floppy_revalidate() which had been commented out (2.5.13 did away with getblk() altogether.) This call is necessary, so the patch reintroduces a private getblk() in floppy.c.

Please don't use getblk(), but go directly through the bio interface. In 2.5 the buffer_heads are just a legacy interface for filesystems and are not supposed to be used by lowlevel drivers.

I haven't got a clue on how to program Linux' block I/O interfaces. Show me how to do a modern equivalent of getblk(dev,0,1024) + waiting for the operation to complete and I'll update the patch ASAP.

/Mikael

#3 lore raw
In reply to Mikael Pettersson
Quoted 3 lines · Mikael Pettersson

I haven't got a clue on how to program Linux' block I/O interfaces. Show me how to do a modern equivalent of getblk(dev,0,1024) + waiting for the operation to complete and I'll update the patch ASAP.

I think something like the following should do it. (WARNING: entirely untested)

static void io_done(struct bio *bio)

{
	complete((struct completion *)bio->bi_private);
}

static int doio(struct block_device *bdev, size_t size)

{
	struct completion comp;
	struct page * page;
	struct bio * bio;

	bio = bio_alloc(GFP_KERNEL, 1);
	if (!bio)
		return -ENOMEM;

	page = alloc_page(GFP_KERNEL);
	if (!page) {
		bio_put(bio);
		return -ENOMEM;
	}

	bio->bi_sector = 0;
	bio->bi_bdev = bdev;
	bio->bi_io_vec[0].bv_page = page;
	bio->bi_io_vec[0].bv_len = size;
	bio->bi_io_vec[0].bv_offset = 0;

	bio->bi_vcnt = 1;
	bio->bi_idx = 0;
	bio->bi_size = LOGPSIZE;

	bio->bi_end_io = io_done;
	bio->bi_private = &complete;

	submit_bio(READ, bio);
	run_task_queue(&tq_disk);

	wait_for_completion(&comp);

	bio_put(bio);
	__free_page(page);

	return 0;
}
#4 lore raw
In reply to Mikael Pettersson

Hi!

Quoted 6 lines
Quoted 3 lines

I haven't got a clue on how to program Linux' block I/O interfaces. Show me how to do a modern equivalent of getblk(dev,0,1024) + waiting for the operation to complete and I'll update the patch ASAP.

I think something like the following should do it. (WARNING: entirely untested)

This looks pretty complicated... Maybe bread() is worth being put back for 2.5?
Pavel

Quoted 50 lines

static void io_done(struct bio *bio)

{
	complete((struct completion *)bio->bi_private);
}

static int doio(struct block_device *bdev, size_t size)

{
	struct completion comp;
	struct page * page;
	struct bio * bio;

	bio = bio_alloc(GFP_KERNEL, 1);
	if (!bio)
		return -ENOMEM;

	page = alloc_page(GFP_KERNEL);
	if (!page) {
		bio_put(bio);
		return -ENOMEM;
	}

	bio->bi_sector = 0;
	bio->bi_bdev = bdev;
	bio->bi_io_vec[0].bv_page = page;
	bio->bi_io_vec[0].bv_len = size;
	bio->bi_io_vec[0].bv_offset = 0;

	bio->bi_vcnt = 1;
	bio->bi_idx = 0;
	bio->bi_size = LOGPSIZE;

	bio->bi_end_io = io_done;
	bio->bi_private = &complete;

	submit_bio(READ, bio);
	run_task_queue(&tq_disk);

	wait_for_completion(&comp);

	bio_put(bio);
	__free_page(page);

	return 0;
}

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Signature · 2 lines
(about SSSCA) "I don't say this lightly.  However, I really think that the U.S.
no longer is classifiable as a democracy, but rather as a plutocracy." --hpa