Xen devel - May 2012 - [PATCH] libxc: implement gnttab.set_max_grants for Linux

Legacy (non-pvops) gntdev drivers may require this operation to be
performed when the number of grants intended to be used simultaneously
exceeds a certain driver specific default limit, and qemu''s qdisk
driver is an example of needing to do so.

Signed-off-by: Jan Beulich <jbeulich@suse.com>

--- a/tools/libxc/xc_linux_osdep.c
+++ b/tools/libxc/xc_linux_osdep.c
@@ -541,6 +541,27 @@ static int linux_gnttab_close(xc_gnttab 
     return close(fd);
 }
 
+static int linux_gnttab_set_max_grants(xc_gnttab *xch, xc_osdep_handle h,
+                                       uint32_t count)
+{
+    int fd = (int)h, rc;
+    struct ioctl_gntdev_set_max_grants max_grants = { .count = count };
+
+    rc = ioctl(fd, IOCTL_GNTDEV_SET_MAX_GRANTS, &max_grants);
+    if (rc) {
+        /*
+         * Newer (e.g. pv-ops) kernels don''t implement this IOCTL,
+         * so ignore the resulting specific failure.
+         */
+        if (errno == ENOTTY)
+            rc = 0;
+        else
+            PERROR("linux_gnttab_set_max_grants: ioctl SET_MAX_GRANTS
failed");
+    }
+
+    return rc;
+}
+
 static void *linux_gnttab_grant_map(xc_gnttab *xch, xc_osdep_handle h,
                                     uint32_t count, int flags, int prot,
                                     uint32_t *domids, uint32_t *refs,
@@ -680,6 +701,7 @@ static struct xc_osdep_ops linux_gnttab_
     .close = &linux_gnttab_close,
 
     .u.gnttab = {
+        .set_max_grants = linux_gnttab_set_max_grants,
         .grant_map = &linux_gnttab_grant_map,
         .munmap = &linux_gnttab_munmap,
     },





_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

On Fri, 2012-05-11 at 08:23 +0100, Jan Beulich wrote:
> +        /*
> +         * Newer (e.g. pv-ops) kernels don''t implement this
IOCTL,
> +         * so ignore the resulting specific failure.
> +         */ 
pvops doesn''t implement this ioctl because it is already more dynamic
and therefore setting a max unnecessary?

(I''m pretty sure yes, but just wanted to confirm we shouldn''t
be fixing
anything on the pvops side).

>>> On 11.05.12 at 10:02, Ian Campbell <Ian.Campbell@citrix.com>
wrote:
> On Fri, 2012-05-11 at 08:23 +0100, Jan Beulich wrote:
>> +        /*
>> +         * Newer (e.g. pv-ops) kernels don''t implement this
IOCTL,
>> +         * so ignore the resulting specific failure.
>> +         */ 
> 
> pvops doesn''t implement this ioctl because it is already more
dynamic
> and therefore setting a max unnecessary?
> 
> (I''m pretty sure yes, but just wanted to confirm we
shouldn''t be fixing
> anything on the pvops side).
Correct. The only adjustment I would consider worth doing is to
accept the unnecessary ioctl without producing an error or logging
any message (perhaps one could in fact return an error for insanely
high input values).

Jan