ide-scsi doesn't compile currently in 2.5.x. This patch fixes it in a
rather hackish way by adding some kmap_atomic()s. It would be
probably better to kmap earlier in process context in the request
function instead, but I leave this to the people who know more
about IDE/block layer than me (but apparently not compile ide-scsi..)
I'm also not totally convinced that it is deadlock free here to use
KM_BOUNCE_READ here, it may be safer to add a new bounce type.
You have been warned.
I have only tested it without highmem and it works for me.
Hopefully there will be eventually a better fix, but for now it
allows to burn (and mount if you use /dev/scd* for ide) CDs under 2.5 again.
-Andi
Index: drivers/scsi/ide-scsi.c
===================================================================
RCS file: /cvs/linux/drivers/scsi/ide-scsi.c,v
retrieving revision 1.45
diff -u -u -r1.45 ide-scsi.c
--- drivers/scsi/ide-scsi.c 2002/01/16 20:58:39 1.45
+++ drivers/scsi/ide-scsi.c 2002/01/27 01:40:26
@@ -52,7 +52,7 @@
#include "sd.h"
#include <scsi/sg.h>
-#define IDESCSI_DEBUG_LOG 0
+/* #define IDESCSI_DEBUG_LOG 0 */
typedef struct idescsi_pc_s {
u8 c[12]; /* Actual packet bytes */
@@ -132,13 +132,20 @@
int count;
while (bcount) {
+ void *addr;
if (pc->sg - (struct scatterlist *) pc->scsi_cmd->request_buffer > pc->scsi_cmd->use_sg) {
printk (KERN_ERR "ide-scsi: scatter gather table too small, discarding data\n");
idescsi_discard_data (drive, bcount);
return;
}
count = IDE_MIN (pc->sg->length - pc->b_count, bcount);
- atapi_input_bytes (drive, pc->sg->address + pc->b_count, count);
+ /* This kmap should be moved earlier into process context
+ and use kmap instead of kmap_atomic. Not now.
+ It also should use an own KMAP_ type to avoid deadlocks,
+ butter better do the first should. b_count is hopefully <= PAGE_SIZE */
+ addr = kmap_atomic(pc->sg->page, KM_BOUNCE_READ);
+ atapi_input_bytes (drive, addr + pc->sg->offset + pc->b_count, count);
+ kunmap_atomic(addr, KM_BOUNCE_READ);
bcount -= count; pc->b_count += count;
if (pc->b_count == pc->sg->length) {
pc->sg++;
@@ -152,13 +159,16 @@
int count;
while (bcount) {
+ void *addr;
if (pc->sg - (struct scatterlist *) pc->scsi_cmd->request_buffer > pc->scsi_cmd->use_sg) {
printk (KERN_ERR "ide-scsi: scatter gather table too small, padding with zeros\n");
idescsi_output_zeros (drive, bcount);
return;
}
count = IDE_MIN (pc->sg->length - pc->b_count, bcount);
- atapi_output_bytes (drive, pc->sg->address + pc->b_count, count);
+ addr = kmap_atomic(pc->sg->page, KM_BOUNCE_READ);
+ atapi_output_bytes (drive, addr + pc->sg->offset + pc->b_count, count);
+ kunmap_atomic(addr, KM_BOUNCE_READ);
bcount -= count; pc->b_count += count;
if (pc->b_count == pc->sg->length) {
pc->sg++;
@@ -753,22 +763,21 @@
struct page *page = sg->page;
int offset = sg->offset;
+#if 0
if (!page) {
BUG_ON(!sg->address);
page = virt_to_page(sg->address);
offset = (unsigned long) sg->address & ~PAGE_MASK;
}
+#endif
bh->bi_io_vec[0].bv_page = page;
bh->bi_io_vec[0].bv_len = sg->length;
bh->bi_io_vec[0].bv_offset = offset;
bh->bi_size = sg->length;
bh = bh->bi_next;
- /*
- * just until scsi_merge is fixed up...
- */
- BUG_ON(PageHighMem(page));
- sg->address = page_address(page) + offset;
+ sg->page = page;
+ sg->offset = offset;
sg++;
}
} else {
On Sun, Jan 27 2002, Andi Kleen wrote:
>
> ide-scsi doesn't compile currently in 2.5.x. This patch fixes it in a
> rather hackish way by adding some kmap_atomic()s. It would be
> probably better to kmap earlier in process context in the request
> function instead, but I leave this to the people who know more
> about IDE/block layer than me (but apparently not compile ide-scsi..)
axboe@burns:~/src/linux> make drivers/scsi/ide-scsi.o
gcc -D__KERNEL__ -I/home/axboe/src/linux/include -Wall
-Wstrict-prototypes -Wno-trigraphs -O2 -fomit-frame-pointer
-fno-strict-aliasing -fno-common -pipe -mpreferred-stack-boundary=2
-march=i686 -c -o drivers/scsi/ide-scsi.o drivers/scsi/ide-scsi.c
axboe@burns:~/src/linux>
Works for me, what didn't compile?
Are you sure that you aren't hitting the sg address change? AFAICS,
there should not be a need for highmem mapping here. Requests coming out
of the block layer for /dev/scd* should be bounced for you, and sg
doesn't pass on highmem pages iirc.
Puzzled...
--
Jens Axboe
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Andi Kleen wrote:
>
> ide-scsi doesn't compile currently in 2.5.x. This patch fixes it in a
> rather hackish way by adding some kmap_atomic()s. It would be
> probably better to kmap earlier in process context in the request
> function instead, but I leave this to the people who know more
> about IDE/block layer than me (but apparently not compile ide-scsi..)
> I'm also not totally convinced that it is deadlock free here to use
> KM_BOUNCE_READ here, it may be safer to add a new bounce type.
> You have been warned.
>
> I have only tested it without highmem and it works for me.
>
> Hopefully there will be eventually a better fix, but for now it
> allows to burn (and mount if you use /dev/scd* for ide) CDs under 2.5 again.
There was a brief period when 2.5 was released and changes went in that
it didn't compile. Then that was fixed, but it couldn't mount or burn,
but that was also fixed.
I burned a CD a few days ago and it worked fine. Maybe you have a
partial patch hanging around or something.
- --
Skip ID: 0x7EDDDB0A
-----BEGIN PGP SIGNATURE-----
iEYEARECAAYFAjxUIewACgkQBMKxVH7d2wrDwACgrG0Nidkkl1uw6pSu33+NKxFR
9lwAnREktZYiYo2sKTSL4kpWLSnRK4BY
=43xD
-----END PGP SIGNATURE-----