Skip to contentSkip to conversation

only irq-safe atomic ops

PATCH Jens Axboe 3 participants 5 messages Feb 25, 2002 1 message not in the archive Block layer

lore
Share & tools
lore

Jens Axboe

author
#1 lore raw
In reply to a message not in the archive
Quoted 11 lines · Stephen Lord
Quoted 7 lines · Andrew Morton

Unfortunately I seem to have found a bug in existing ext2, a bug in existing block_write_full_page, a probable bug in the aic7xxx driver, and an oops in the aic7xxx driver. So progress has slowed down a bit :(

Try this for the aic driver:

Steve,

DaveM's alternate fix has been in the 2.4 and 2.5 kernels for some time, so given that Andrew is testing 2.5.5 this can't be the problem.

Quoted 2 lines

We hit this once bio got introduced and we maxed out the request size for the driver. Justin has the code in his next aic version.

Just for the record, this was/is not a bio bug, it was an aic7xxx bug that could trigger with or without bio.

Signature · 1 line
Jens Axboe
#2 lore raw
In reply to a message not in the archive
Quoted 10 lines · Jens Axboe
Quoted 8 lines · Stephen Lord
Quoted 6 lines · Andrew Morton

Unfortunately I seem to have found a bug in existing ext2, a bug in existing block_write_full_page, a probable bug in the aic7xxx driver, and an oops in the aic7xxx driver. So progress has slowed down a bit :(

Quoted 8 lines
Quoted 2 lines

Try this for the aic driver:

Steve,

DaveM's alternate fix has been in the 2.4 and 2.5 kernels for some time, so given that Andrew is testing 2.5.5 this can't be the problem.

OK, I was not aware that went in, thanks.

Quoted 9 lines
Quoted 5 lines

We hit this once bio got introduced and we maxed out the request size for the driver. Justin has the code in his next aic version.

Just for the record, this was/is not a bio bug, it was an aic7xxx bug that could trigger with or without bio.

Yep, bio just made it easier to get larger requests.

Steve

#3 lore raw
In reply to a message not in the archive
Quoted 3 lines · Stephen Lord

Yep, bio just made it easier to get larger requests.

Which promptly go kersplat when you feed them into submit_bio():

     BUG_ON(bio_sectors(bio) > q->max_sectors);

Given that I'm hand-rolling a monster bio, I need to know when to wrap it up and send it off, to avoid creating a bio which is larger than the target device will accept. I'm currently
using the below patch. Am I right that this is missing API
functionality, or did I miss something?

Also, I could not find a way of querying the size of the vector at *bi_io_vec. This is also information which would be helpful when building large scatter/gather lists.

drivers/block/ll_rw_blk.c renamed from drivers/block/ll_rw_blk.c~mpio-10-biobits +22 -0
@@ -1350,6 +1350,28 @@ static void end_bio_bh_io_sync(struct bi
 }
 
 /**
+ * bio_max_bytes: return the maximum number of bytes which can be
+ * placed in a single bio for a particular device.
+ *
+ * @dev: the device's kdev_t
+ *
+ * Each device has a maximum permissible queue size, and bios may
+ * not cover more data than that.
+ *
+ * Returns -ve on error.
+ */
+int bio_max_bytes(kdev_t dev)
+{
+	request_queue_t *q;
+	int ret = -1;
+
+	q = blk_get_queue(dev);
+	if (q)
+		ret = (q->max_sectors << 9);
+	return ret;
+}
+
+/**
  * submit_bio: submit a bio to the block device layer for I/O
  * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
  * @bio: The &struct bio which describes the I/O
include/linux/bio.h renamed from include/linux/bio.h~mpio-10-biobits +1 -0
@@ -204,5 +204,6 @@ extern struct bio *bio_copy(struct bio *
 extern inline void bio_init(struct bio *);
 
 extern int bio_ioctl(kdev_t, unsigned int, unsigned long);
+extern int bio_max_bytes(kdev_t dev);
 
 #endif /* __LINUX_BIO_H */

-

#4 lore raw
Quoted 8 lines · Andrew Morton
Quoted 3 lines · Stephen Lord

Yep, bio just made it easier to get larger requests.

Which promptly go kersplat when you feed them into submit_bio():

Quoted 8 lines
     BUG_ON(bio_sectors(bio) > q->max_sectors);

Given that I'm hand-rolling a monster bio, I need to know when to wrap it up and send it off, to avoid creating a bio which is larger than the target device will accept. I'm currently
using the below patch. Am I right that this is missing API
functionality, or did I miss something?

I don't run into that one, but probably because I limit xfs to use BIO_MAX_SECTORS, take a look at ll_rw_kio to see how that splits things up. This of course does not take into account any further restriction in an underlying queue.

Steve

Signature · 2 lines
Steve Lord                                      voice: +1-651-683-3511
Principal Engineer, Filesystem Software         email: [email protected]
#5 lore raw
In reply to Andrew Morton
Quoted 19 lines · Steve Lord
Quoted 8 lines · Andrew Morton
Quoted 3 lines · Stephen Lord

Yep, bio just made it easier to get larger requests.

Which promptly go kersplat when you feed them into submit_bio():

Quoted 8 lines
     BUG_ON(bio_sectors(bio) > q->max_sectors);

Given that I'm hand-rolling a monster bio, I need to know when to wrap it up and send it off, to avoid creating a bio which is larger than the target device will accept. I'm currently
using the below patch. Am I right that this is missing API
functionality, or did I miss something?

Quoted 4 lines

I don't run into that one, but probably because I limit xfs to use BIO_MAX_SECTORS, take a look at ll_rw_kio to see how that splits things up. This of course does not take into account any further restriction in an underlying queue.

Ah, yes. I looked at that, and promptly ignored it :)

Too small, too hard-wired. If the underlying device can cope with megabyte requests, why restrict it to 64k? Let's push the envelope a bit, rather than creating more must-fixes for 2.7.x.

-