Two bugs at once:
First the mask is backwards, so the it used to keeps the offset and
clears the page address, which is not what we need when we update the
offset.
Second the offset calculation is wrong in case head isn't page aligned.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
cpu_physical_memory_rw(head, q->buffer + bufpos, tail - head, rw);
bufpos += (tail - head);
+ offset += (tail - head);
bytes -= (tail - head);
if (bytes > 0) {
set_field(&q->qh.token, cpage, QTD_TOKEN_CPAGE);
/* save offset into cpage */
- offset = tail - head;
- q->qh.bufptr[0] &= ~QTD_BUFPTR_MASK;
+ q->qh.bufptr[0] &= QTD_BUFPTR_MASK;
q->qh.bufptr[0] |= offset;
return 0;