ieee80211_i.h: shouldn't struct ps_data.tim be aligned unsigned long?

ieee80211_i.h: shouldn't struct ps_data.tim be aligned unsigned long?

[Joe Perches]

about
net/mac80211/tx.c:              have_bits = !bitmap_empty((unsigned long*)ps->tim,
and
net/mac80211/ieee80211_i.h
d012a6051 (Marco Porsch        2012-10-10 12:39:50 -0700 244) struct ps_data {
f0706e828 (Jiri Benc           2007-05-05 11:45:53 -0700 245)   /* yes, this looks ugly, but guarantees that we can later use
f0706e828 (Jiri Benc           2007-05-05 11:45:53 -0700 246)    * bitmap_empty :)
004c872e7 (Johannes Berg       2008-02-20 11:21:35 +0100 247)    * NB: don't touch this bitmap, use sta_info_{set,clear}_tim_bit */
f0706e828 (Jiri Benc           2007-05-05 11:45:53 -0700 248)   u8 tim[sizeof(unsigned long) * BITS_TO_LONGS(IEEE80211_MAX_AID + 1)];
d012a6051 (Marco Porsch        2012-10-10 12:39:50 -0700 249)   struct sk_buff_head bc_buf;
056cdd599 (Johannes Berg       2008-03-26 23:21:47 +0100 250)   atomic_t num_sta_ps; /* number of stations in PS mode */
5dfdaf58d (Johannes Berg       2007-12-19 02:03:33 +0100 251)   int dtim_count;
512119b36 (Christian Lamparter 2011-01-31 20:48:44 +0200 252)   bool dtim_bc_mc;
f0706e828 (Jiri Benc           2007-05-05 11:45:53 -0700 253) };

Perhaps the
	u8 tim[sizeof...]
member should be marked
	__aligned(sizeof(unsigned long))
so that the cast to ulong when bitmap_empty
is actually forced to be ulong aligned?

Perhaps the "guarantee" isn't valid without it.

---

 net/mac80211/ieee80211_i.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/net/mac80211/ieee80211_i.h b/net/mac80211/ieee80211_i.h
index ed5bf8b..e5a9c51 100644
--- a/net/mac80211/ieee80211_i.h
+++ b/net/mac80211/ieee80211_i.h
@@ -245,7 +245,8 @@ struct ps_data {
 	/* yes, this looks ugly, but guarantees that we can later use
 	 * bitmap_empty :)
 	 * NB: don't touch this bitmap, use sta_info_{set,clear}_tim_bit */
-	u8 tim[sizeof(unsigned long) * BITS_TO_LONGS(IEEE80211_MAX_AID + 1)];
+	u8 tim[sizeof(unsigned long) * BITS_TO_LONGS(IEEE80211_MAX_AID + 1)]
+			__aligned(sizeof(unsigned long));
 	struct sk_buff_head bc_buf;
 	atomic_t num_sta_ps; /* number of stations in PS mode */
 	int dtim_count;

Re: ieee80211_i.h: shouldn't struct ps_data.tim be aligned unsigned long?

[Johannes Berg]

On Wed, 2013-12-18 at 00:16 -0800, Joe Perches wrote:

> Perhaps the
> 	u8 tim[sizeof...]
> member should be marked
> 	__aligned(sizeof(unsigned long))
> so that the cast to ulong when bitmap_empty
> is actually forced to be ulong aligned?

yeah, I suppose that's true.

> diff --git a/net/mac80211/ieee80211_i.h b/net/mac80211/ieee80211_i.h
> index ed5bf8b..e5a9c51 100644
> --- a/net/mac80211/ieee80211_i.h
> +++ b/net/mac80211/ieee80211_i.h
> @@ -245,7 +245,8 @@ struct ps_data {
>  	/* yes, this looks ugly, but guarantees that we can later use
>  	 * bitmap_empty :)
>  	 * NB: don't touch this bitmap, use sta_info_{set,clear}_tim_bit */
> -	u8 tim[sizeof(unsigned long) * BITS_TO_LONGS(IEEE80211_MAX_AID + 1)];
> +	u8 tim[sizeof(unsigned long) * BITS_TO_LONGS(IEEE80211_MAX_AID + 1)]
> +			__aligned(sizeof(unsigned long));

Care to send a proper patch? Otherwise I'll just re-do it and pick it up
that way ...

johannes

Re: ieee80211_i.h: shouldn't struct ps_data.tim be aligned unsigned long?

[Joe Perches]

On Wed, 2013-12-18 at 10:19 +0100, Johannes Berg wrote:
> On Wed, 2013-12-18 at 00:16 -0800, Joe Perches wrote:
> 
> > Perhaps the
> > 	u8 tim[sizeof...]
> > member should be marked
> > 	__aligned(sizeof(unsigned long))
> > so that the cast to ulong when bitmap_empty
> > is actually forced to be ulong aligned?
> 
> yeah, I suppose that's true.
> 
> > diff --git a/net/mac80211/ieee80211_i.h b/net/mac80211/ieee80211_i.h
> > index ed5bf8b..e5a9c51 100644
> > --- a/net/mac80211/ieee80211_i.h
> > +++ b/net/mac80211/ieee80211_i.h
> > @@ -245,7 +245,8 @@ struct ps_data {
> >  	/* yes, this looks ugly, but guarantees that we can later use
> >  	 * bitmap_empty :)
> >  	 * NB: don't touch this bitmap, use sta_info_{set,clear}_tim_bit */
> > -	u8 tim[sizeof(unsigned long) * BITS_TO_LONGS(IEEE80211_MAX_AID + 1)];
> > +	u8 tim[sizeof(unsigned long) * BITS_TO_LONGS(IEEE80211_MAX_AID + 1)]
> > +			__aligned(sizeof(unsigned long));
> 
> Care to send a proper patch? Otherwise I'll just re-do it and pick it up
> that way ...

Hi Johannes.

You can redo it easier than applying a patch
from me.

I just noticed it when the reformatting
patch went by.

No need to give a credit or anything.
It's just a nit.

cheers, Joe

[PATCH] ieee80211_i.h: Align struct ps_data.tim to unsigned long

[Joe Perches]

Its address is used as an unsigned long *, so make sure
that the tim u8 array is properly aligned.

Signed-off-by: Joe Perches <joe@perches.com
---
> Care to send a proper patch? Otherwise I'll just re-do it and pick it up
> that way ...

I had a passing thought about this when out today.
It should really be set to __aligned(__alignof(unsigned long))
so here's a proper patch.

 net/mac80211/ieee80211_i.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/net/mac80211/ieee80211_i.h b/net/mac80211/ieee80211_i.h
index ed5bf8b..9a61eef 100644
--- a/net/mac80211/ieee80211_i.h
+++ b/net/mac80211/ieee80211_i.h
@@ -245,7 +245,8 @@ struct ps_data {
 	/* yes, this looks ugly, but guarantees that we can later use
 	 * bitmap_empty :)
 	 * NB: don't touch this bitmap, use sta_info_{set,clear}_tim_bit */
-	u8 tim[sizeof(unsigned long) * BITS_TO_LONGS(IEEE80211_MAX_AID + 1)];
+	u8 tim[sizeof(unsigned long) * BITS_TO_LONGS(IEEE80211_MAX_AID + 1)]
+			__aligned(__alignof__(unsigned long));
 	struct sk_buff_head bc_buf;
 	atomic_t num_sta_ps; /* number of stations in PS mode */
 	int dtim_count;

Re: [PATCH] ieee80211_i.h: Align struct ps_data.tim to unsigned long

[Johannes Berg]

On Wed, 2013-12-18 at 16:52 -0800, Joe Perches wrote:
> Its address is used as an unsigned long *, so make sure
> that the tim u8 array is properly aligned.
> 
> Signed-off-by: Joe Perches <joe@perches.com
> ---
> > Care to send a proper patch? Otherwise I'll just re-do it and pick it up
> > that way ...
> 
> I had a passing thought about this when out today.
> It should really be set to __aligned(__alignof(unsigned long))
> so here's a proper patch.

Hmm, yeah, I guess that's true, though sizeof() is probably >= on
anything where Linux runs (and otherwise arrays get problematic anyway?)

Anyway, I've dropped my patch in favour of yours.

johannes