Skip to contentSkip to conversation

blkfront: sector size > 512

PATCH In review Stefano Stabellini 3 participants 8 messages May 21, 2009 – May 22, 2009

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 blkfront: sector size > 512

The first and last sector as well as the sector number of the request is expressed in 512 bytes units, independently from the real sector size.

---

extras/mini-os/blkfront.c +4 -4
diff -r 61404d971c92 extras/mini-os/blkfront.c
@@ -310,14 +310,14 @@
     req->nr_segments = n;
     req->handle = dev->handle;
     req->id = (uintptr_t) aiocbp;
-    req->sector_number = aiocbp->aio_offset / dev->info.sector_size;
+    req->sector_number = aiocbp->aio_offset / 512;
 
     for (j = 0; j < n; j++) {
         req->seg[j].first_sect = 0;
-        req->seg[j].last_sect = PAGE_SIZE / dev->info.sector_size - 1;
+        req->seg[j].last_sect = PAGE_SIZE / 512 - 1;
     }
-    req->seg[0].first_sect = ((uintptr_t)aiocbp->aio_buf & ~PAGE_MASK) / dev->info.sector_size;
-    req->seg[n-1].last_sect = (((uintptr_t)aiocbp->aio_buf + aiocbp->aio_nbytes - 1) & ~PAGE_MASK) / dev->info.sector_size;
+    req->seg[0].first_sect = ((uintptr_t)aiocbp->aio_buf & ~PAGE_MASK) / 512;
+    req->seg[n-1].last_sect = (((uintptr_t)aiocbp->aio_buf + aiocbp->aio_nbytes - 1) & ~PAGE_MASK) / 512;
     for (j = 0; j < n; j++) {
 	uintptr_t data = start + j * PAGE_SIZE;
         if (!write) {
#1 lore raw

extras/mini-os/blkfront.c line 313

Quoted 6 lines · Stefano Stabellini, le Thu 21 May 2009 12:40:52 +0100, a écrit :

The first and last sector as well as the sector number of the request is expressed in 512 bytes units, independently from the real sector size.

     req->id = (uintptr_t) aiocbp;
extras/mini-os/blkfront.c +1 -1
-    req->sector_number = aiocbp->aio_offset / dev->info.sector_size;
+    req->sector_number = aiocbp->aio_offset / 512;

Oh?!

That needs to be better documented in xen/include/public/io/blkif.h, then. I'm however wondering whether this shouldn't actually be fixed in the blkback/cdrom interface instead.

Samuel

#2 lore raw
Quoted 8 lines · Samuel Thibault

extras/mini-os/blkfront.c line 313

Quoted 6 lines · Stefano Stabellini, le Thu 21 May 2009 12:40:52 +0100, a écrit :

The first and last sector as well as the sector number of the request is expressed in 512 bytes units, independently from the real sector size.

     req->id = (uintptr_t) aiocbp;
extras/mini-os/blkfront.c +1 -1
-    req->sector_number = aiocbp->aio_offset / dev->info.sector_size;
+    req->sector_number = aiocbp->aio_offset / 512;
Quoted 6 lines

Oh?!

That needs to be better documented in xen/include/public/io/blkif.h, then. I'm however wondering whether this shouldn't actually be fixed in the blkback/cdrom interface instead.

Yeah, I am surprised as you are, but the linux blkfront does the same thing so I prefer to avoid changing the protocol now.

#3 lore raw
Quoted 11 lines · Stefano Stabellini, le Thu 21 May 2009 14:17:11 +0100, a écrit :
Quoted 10 lines · Samuel Thibault

extras/mini-os/blkfront.c line 313

Quoted 6 lines · Stefano Stabellini, le Thu 21 May 2009 12:40:52 +0100, a écrit :

The first and last sector as well as the sector number of the request is expressed in 512 bytes units, independently from the real sector size.

     req->id = (uintptr_t) aiocbp;
extras/mini-os/blkfront.c +1 -1
-    req->sector_number = aiocbp->aio_offset / dev->info.sector_size;
+    req->sector_number = aiocbp->aio_offset / 512;

Oh?!

Quoted 7 lines
Quoted 4 lines

That needs to be better documented in xen/include/public/io/blkif.h, then. I'm however wondering whether this shouldn't actually be fixed in the blkback/cdrom interface instead.

Yeah, I am surprised as you are, but the linux blkfront does the same thing so I prefer to avoid changing the protocol now.

Well, it's not about changing the protocol, it's about fixing drivers doing the wrong thing according to the documentation. What is the use of the xenstore sector_size value if not here?

Samuel

#4 lore raw
Quoted 6 lines · Samuel Thibault
Quoted 2 lines

Yeah, I am surprised as you are, but the linux blkfront does the same thing so I prefer to avoid changing the protocol now.

Well, it's not about changing the protocol, it's about fixing drivers doing the wrong thing according to the documentation. What is the use of the xenstore sector_size value if not here?

It's due to the protocol having evolved a bit. The xenstore sector_size was added a bit later, specifically after testing on CD-ROM drives, to require that frontends issue sector_size-aligned and -sized requests. Otherwise Linux blkdev subsystem was dropping the reqs on the floor. But we did not change the wire request format at that time.

Yeah, I'd take a patch to add a nice clear comment about this to blkif.h. ;-)

-- Keir

#5 lore raw
Quoted 4 lines · Keir Fraser

Yeah, I'd take a patch to add a nice clear comment about this to blkif.h. ;-)

Is this comment clear enough?

---

xen/include/public/io/blkif.h +7 -0
diff -r 61404d971c92 xen/include/public/io/blkif.h
@@ -84,6 +84,13 @@
  */
 #define BLKIF_MAX_SEGMENTS_PER_REQUEST 11
 
+/* 
+ * first_sect and last_sect in blkif_request_segment, as well as
+ * sector_number in blkif_request are all expressed in 512 bytes units,
+ * however they must be properly aligned to the real sector size of
+ * the physical disk (as reported in the "sector-size" node in backend
+ * xenbus info).
+ */
 struct blkif_request_segment {
     grant_ref_t gref;        /* reference to I/O buffer frame        */
     /* @first_sect: first sector in frame to transfer (inclusive).   */
#6 lore raw
Quoted 5 lines · Stefano Stabellini, le Fri 22 May 2009 15:31:28 +0100, a écrit :
Quoted 2 lines · Keir Fraser

Yeah, I'd take a patch to add a nice clear comment about this to blkif.h. ;-)

Is this comment clear enough?

Could you also document whether the 'sectors' node in the xenstore is expressed in sector size or in 512 sectors?

Samuel

#7 lore raw
Quoted 8 lines · Samuel Thibault
Quoted 4 lines · Stefano Stabellini, le Fri 22 May 2009 15:31:28 +0100, a écrit :
Quoted 2 lines · Keir Fraser

Yeah, I'd take a patch to add a nice clear comment about this to blkif.h. ;-)

Is this comment clear enough?

Could you also document whether the 'sectors' node in the xenstore is expressed in sector size or in 512 sectors?

OK.

---

xen/include/public/io/blkif.h +8 -0
diff -r 61404d971c92 xen/include/public/io/blkif.h
@@ -84,6 +84,14 @@
  */
 #define BLKIF_MAX_SEGMENTS_PER_REQUEST 11
 
+/* 
+ * first_sect and last_sect in blkif_request_segment, as well as
+ * sector_number in blkif_request are all expressed in 512 bytes units,
+ * however they must be properly aligned to the real sector size of
+ * the physical disk, that is reported in the "sector-size" node in
+ * backend xenbus info.
+ * Also the "sectors" node on xenbus is expressed in 512 bytes units.
+ */
 struct blkif_request_segment {
     grant_ref_t gref;        /* reference to I/O buffer frame        */
     /* @first_sect: first sector in frame to transfer (inclusive).   */