This patch changes type of "pending" flag to uint32_t type.
Pointer to this flag is passed to hci_clear_bit function as void * and
then casted to uint32_t * again. Previously unsigned long type made some
unexpected performance which occured after violating strict aliasing
rule. This patch fixes this problem and this solution is much easier than
solution of changing types of many object (from uint32_t to unsigned long,
including changing type for castng in hci_*_bit function family), which
is overcomplicated. Also, not all objects can be changed, like
hci_dev_info struct, since it is violating compatibility rules between
kernel and user spaces.
---
plugins/hciops.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/plugins/hciops.c b/plugins/hciops.c
index 4225cd8..69627eb 100644
--- a/plugins/hciops.c
+++ b/plugins/hciops.c
@@ -111,7 +111,7 @@ static struct dev_info {
uint16_t did_version;
gboolean up;
- unsigned long pending;
+ uint32_t pending;
GIOChannel *io;
guint watch_id;
--
1.6.3.3
Hi Rafal,
On Wed, Mar 23, 2011, Rafal Michalski wrote:
> This patch changes type of "pending" flag to uint32_t type.
> Pointer to this flag is passed to hci_clear_bit function as void * and
> then casted to uint32_t * again. Previously unsigned long type made some
> unexpected performance which occured after violating strict aliasing
> rule. This patch fixes this problem and this solution is much easier than
> solution of changing types of many object (from uint32_t to unsigned long,
> including changing type for castng in hci_*_bit function family), which
> is overcomplicated. Also, not all objects can be changed, like
> hci_dev_info struct, since it is violating compatibility rules between
> kernel and user spaces.
> ---
> plugins/hciops.c | 2 +-
> 1 files changed, 1 insertions(+), 1 deletions(-)
Thanks. The patch has been pushed upstream after I did some heavy
cleanup on the commit message for it to be more readable :)
Johan