When we try to run RTL8187 driver on AD BlackFin platform, we got
messages from kernel about unaligned memory access at
compare_ether_addr() calls.
Replacing of compare_ether_addr() by memcmp() fixes this problem.
Signed-off-by: Yauhen Kharuzhy <[email protected]>
---
net/mac80211/ibss.c | 2 +-
net/wireless/scan.c | 2 +-
2 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/net/mac80211/ibss.c b/net/mac80211/ibss.c
index 0b30277..bcbb1ae 100644
--- a/net/mac80211/ibss.c
+++ b/net/mac80211/ibss.c
@@ -395,7 +395,7 @@ struct sta_info *ieee80211_ibss_add_sta(struct ieee80211_sub_if_data *sdata,
return NULL;
}
- if (compare_ether_addr(bssid, sdata->u.ibss.bssid))
+ if (memcmp(bssid, sdata->u.ibss.bssid, ETH_ALEN))
return NULL;
#ifdef CONFIG_MAC80211_VERBOSE_DEBUG
diff --git a/net/wireless/scan.c b/net/wireless/scan.c
index e95b638..6ad9b59 100644
--- a/net/wireless/scan.c
+++ b/net/wireless/scan.c
@@ -134,7 +134,7 @@ static bool is_bss(struct cfg80211_bss *a,
{
const u8 *ssidie;
- if (bssid && compare_ether_addr(a->bssid, bssid))
+ if (bssid && memcmp(a->bssid, bssid, ETH_ALEN))
return false;
if (!ssid)
--
1.6.3.1
On Sun, 2009-06-28 at 15:18 +0300, Ivan Kuten wrote:
> Hello,
>
> In net/wireless/scan.c : cfg80211_wext_siwscan there seems also unaligned allocations
> for creq->ssids and creq->channels. Should it be something like that?
Seems alright, but there is more than one instance of this, maybe you
can make a function to allocate a scan request properly and have it be
called from all the places it's needed.
johannes
> Modified: trunk/uClinux-dist-2008R1-RC8/compat-wireless-2009-06-11/net/wireless/scan.c
> ==============================================================================
> --- trunk/uClinux-dist-2008R1-RC8/compat-wireless-2009-06-11/net/wireless/scan.c (original)
> +++ trunk/uClinux-dist-2008R1-RC8/compat-wireless-2009-06-11/net/wireless/scan.c Fri Jun 26 14:00:52 2009
> @@ -619,7 +619,7 @@
> if (wiphy->bands[band])
> n_channels += wiphy->bands[band]->n_channels;
>
> - creq = kzalloc(sizeof(*creq) + sizeof(struct cfg80211_ssid) +
> + creq = kzalloc(roundup(sizeof(*creq), 4) + roundup(sizeof(struct cfg80211_ssid), 4) +
> n_channels * sizeof(void *),
> GFP_ATOMIC);
> if (!creq) {
> @@ -629,8 +629,8 @@
>
> creq->wiphy = wiphy;
> creq->ifidx = dev->ifindex;
> - creq->ssids = (void *)(creq + 1);
> - creq->channels = (void *)(creq->ssids + 1);
> + creq->ssids = (void *)creq + roundup(sizeof(*creq), 4);
> + creq->channels = (void *)creq->ssids + roundup(sizeof(*creq->ssids), 4);
> creq->n_channels = n_channels;
> creq->n_ssids = 1;
>
> Regards,
> Ivan
>
> > On Tue, 2009-06-16 at 13:54 +0300, Yauhen Kharuzhy wrote:
> >> When we try to run RTL8187 driver on AD BlackFin platform, we got
> >> messages from kernel about unaligned memory access at
> >> compare_ether_addr() calls.
> >>
> >> Replacing of compare_ether_addr() by memcmp() fixes this problem.
> >
> > This shouldn't be necessary. Which operand is unaligned?
> >
> >> --- a/net/mac80211/ibss.c
> >> +++ b/net/mac80211/ibss.c
> >> @@ -395,7 +395,7 @@ struct sta_info *ieee80211_ibss_add_sta(struct ieee80211_sub_if_data *sdata,
> >> return NULL;
> >> }
> >>
> >> - if (compare_ether_addr(bssid, sdata->u.ibss.bssid))
> >> + if (memcmp(bssid, sdata->u.ibss.bssid, ETH_ALEN))
> >> return NULL;
> >
> > So in this case it seems that it is possible that u.ibss.bssid is not
> > aligned, consider fixing by doing
> >
> > --- ieee80211_i.h
> > +++ ieee80211_i.h
> > - u8 bssid[ETH_ALEN];
> > + u8 bssid[ETH_ALEN] __align(2);
> >
> > or so instead.
> >
> >> --- a/net/wireless/scan.c
> >> +++ b/net/wireless/scan.c
> >> @@ -134,7 +134,7 @@ static bool is_bss(struct cfg80211_bss *a,
> >> {
> >> const u8 *ssidie;
> >>
> >> - if (bssid && compare_ether_addr(a->bssid, bssid))
> >> + if (bssid && memcmp(a->bssid, bssid, ETH_ALEN))
> >
> > Since a->bssid is after a pointer I can't see how it would be unaligned,
> > and bssid should be unaligned only if the call trace shows it's coming
> > from the above u.ibss.bssid.
> >
> > johannes
>
>
On Sun, 2009-06-28 at 15:18 +0300, Ivan Kuten wrote:
> Hello,
>
> In net/wireless/scan.c : cfg80211_wext_siwscan there seems also unaligned allocations
> for creq->ssids and creq->channels. Should it be something like that?
>
> Modified: trunk/uClinux-dist-2008R1-RC8/compat-wireless-2009-06-11/net/wireless/scan.c
> ==============================================================================
> --- trunk/uClinux-dist-2008R1-RC8/compat-wireless-2009-06-11/net/wireless/scan.c (original)
> +++ trunk/uClinux-dist-2008R1-RC8/compat-wireless-2009-06-11/net/wireless/scan.c Fri Jun 26 14:00:52 2009
> @@ -619,7 +619,7 @@
> if (wiphy->bands[band])
> n_channels += wiphy->bands[band]->n_channels;
>
> - creq = kzalloc(sizeof(*creq) + sizeof(struct cfg80211_ssid) +
> + creq = kzalloc(roundup(sizeof(*creq), 4) + roundup(sizeof(struct cfg80211_ssid), 4) +
> n_channels * sizeof(void *),
> GFP_ATOMIC);
Also, that should be sizeof(void *) instead of "4".
johannes
On Tue, 2009-06-16 at 13:54 +0300, Yauhen Kharuzhy wrote:
> When we try to run RTL8187 driver on AD BlackFin platform, we got
> messages from kernel about unaligned memory access at
> compare_ether_addr() calls.
>
> Replacing of compare_ether_addr() by memcmp() fixes this problem.
This shouldn't be necessary. Which operand is unaligned?
> --- a/net/mac80211/ibss.c
> +++ b/net/mac80211/ibss.c
> @@ -395,7 +395,7 @@ struct sta_info *ieee80211_ibss_add_sta(struct ieee80211_sub_if_data *sdata,
> return NULL;
> }
>
> - if (compare_ether_addr(bssid, sdata->u.ibss.bssid))
> + if (memcmp(bssid, sdata->u.ibss.bssid, ETH_ALEN))
> return NULL;
So in this case it seems that it is possible that u.ibss.bssid is not
aligned, consider fixing by doing
--- ieee80211_i.h
+++ ieee80211_i.h
- u8 bssid[ETH_ALEN];
+ u8 bssid[ETH_ALEN] __align(2);
or so instead.
> --- a/net/wireless/scan.c
> +++ b/net/wireless/scan.c
> @@ -134,7 +134,7 @@ static bool is_bss(struct cfg80211_bss *a,
> {
> const u8 *ssidie;
>
> - if (bssid && compare_ether_addr(a->bssid, bssid))
> + if (bssid && memcmp(a->bssid, bssid, ETH_ALEN))
Since a->bssid is after a pointer I can't see how it would be unaligned,
and bssid should be unaligned only if the call trace shows it's coming
from the above u.ibss.bssid.
johannes
Hello,
In net/wireless/scan.c : cfg80211_wext_siwscan there seems also unaligned allocations
for creq->ssids and creq->channels. Should it be something like that?
Modified: trunk/uClinux-dist-2008R1-RC8/compat-wireless-2009-06-11/net/wireless/scan.c
==============================================================================
--- trunk/uClinux-dist-2008R1-RC8/compat-wireless-2009-06-11/net/wireless/scan.c (original)
+++ trunk/uClinux-dist-2008R1-RC8/compat-wireless-2009-06-11/net/wireless/scan.c Fri Jun 26 14:00:52 2009
@@ -619,7 +619,7 @@
if (wiphy->bands[band])
n_channels += wiphy->bands[band]->n_channels;
- creq = kzalloc(sizeof(*creq) + sizeof(struct cfg80211_ssid) +
+ creq = kzalloc(roundup(sizeof(*creq), 4) + roundup(sizeof(struct cfg80211_ssid), 4) +
n_channels * sizeof(void *),
GFP_ATOMIC);
if (!creq) {
@@ -629,8 +629,8 @@
creq->wiphy = wiphy;
creq->ifidx = dev->ifindex;
- creq->ssids = (void *)(creq + 1);
- creq->channels = (void *)(creq->ssids + 1);
+ creq->ssids = (void *)creq + roundup(sizeof(*creq), 4);
+ creq->channels = (void *)creq->ssids + roundup(sizeof(*creq->ssids), 4);
creq->n_channels = n_channels;
creq->n_ssids = 1;
Regards,
Ivan
> On Tue, 2009-06-16 at 13:54 +0300, Yauhen Kharuzhy wrote:
>> When we try to run RTL8187 driver on AD BlackFin platform, we got
>> messages from kernel about unaligned memory access at
>> compare_ether_addr() calls.
>>
>> Replacing of compare_ether_addr() by memcmp() fixes this problem.
>
> This shouldn't be necessary. Which operand is unaligned?
>
>> --- a/net/mac80211/ibss.c
>> +++ b/net/mac80211/ibss.c
>> @@ -395,7 +395,7 @@ struct sta_info *ieee80211_ibss_add_sta(struct ieee80211_sub_if_data *sdata,
>> return NULL;
>> }
>>
>> - if (compare_ether_addr(bssid, sdata->u.ibss.bssid))
>> + if (memcmp(bssid, sdata->u.ibss.bssid, ETH_ALEN))
>> return NULL;
>
> So in this case it seems that it is possible that u.ibss.bssid is not
> aligned, consider fixing by doing
>
> --- ieee80211_i.h
> +++ ieee80211_i.h
> - u8 bssid[ETH_ALEN];
> + u8 bssid[ETH_ALEN] __align(2);
>
> or so instead.
>
>> --- a/net/wireless/scan.c
>> +++ b/net/wireless/scan.c
>> @@ -134,7 +134,7 @@ static bool is_bss(struct cfg80211_bss *a,
>> {
>> const u8 *ssidie;
>>
>> - if (bssid && compare_ether_addr(a->bssid, bssid))
>> + if (bssid && memcmp(a->bssid, bssid, ETH_ALEN))
>
> Since a->bssid is after a pointer I can't see how it would be unaligned,
> and bssid should be unaligned only if the call trace shows it's coming
> from the above u.ibss.bssid.
>
> johannes
> On Sun, 2009-06-28 at 15:18 +0300, Ivan Kuten wrote:
>> Hello,
>>
>> In net/wireless/scan.c : cfg80211_wext_siwscan there seems also unaligned allocations
>> for creq->ssids and creq->channels. Should it be something like that?
>
> Seems alright, but there is more than one instance of this, maybe you
> can make a function to allocate a scan request properly and have it be
> called from all the places it's needed.
>
> johannes
>
Hello Johannes,
Can you point to that multiple scan allocations? I see only one kzalloc
with followed possible alignment violation - it's in cfg80211_wext_siwscan in scan.c
Regards,
Ivan
>> Modified: trunk/uClinux-dist-2008R1-RC8/compat-wireless-2009-06-11/net/wireless/scan.c
>> ==============================================================================
>> --- trunk/uClinux-dist-2008R1-RC8/compat-wireless-2009-06-11/net/wireless/scan.c (original)
>> +++ trunk/uClinux-dist-2008R1-RC8/compat-wireless-2009-06-11/net/wireless/scan.c Fri Jun 26 14:00:52 2009
>> @@ -619,7 +619,7 @@
>> if (wiphy->bands[band])
>> n_channels += wiphy->bands[band]->n_channels;
>>
>> - creq = kzalloc(sizeof(*creq) + sizeof(struct cfg80211_ssid) +
>> + creq = kzalloc(roundup(sizeof(*creq), 4) + roundup(sizeof(struct cfg80211_ssid), 4) +
>> n_channels * sizeof(void *),
>> GFP_ATOMIC);
>> if (!creq) {
>> @@ -629,8 +629,8 @@
>>
>> creq->wiphy = wiphy;
>> creq->ifidx = dev->ifindex;
>> - creq->ssids = (void *)(creq + 1);
>> - creq->channels = (void *)(creq->ssids + 1);
>> + creq->ssids = (void *)creq + roundup(sizeof(*creq), 4);
>> + creq->channels = (void *)creq->ssids + roundup(sizeof(*creq->ssids), 4);
>> creq->n_channels = n_channels;
>> creq->n_ssids = 1;
>>
>> Regards,
>> Ivan
>>
>>> On Tue, 2009-06-16 at 13:54 +0300, Yauhen Kharuzhy wrote:
>>>> When we try to run RTL8187 driver on AD BlackFin platform, we got
>>>> messages from kernel about unaligned memory access at
>>>> compare_ether_addr() calls.
>>>>
>>>> Replacing of compare_ether_addr() by memcmp() fixes this problem.
>>> This shouldn't be necessary. Which operand is unaligned?
>>>
>>>> --- a/net/mac80211/ibss.c
>>>> +++ b/net/mac80211/ibss.c
>>>> @@ -395,7 +395,7 @@ struct sta_info *ieee80211_ibss_add_sta(struct ieee80211_sub_if_data *sdata,
>>>> return NULL;
>>>> }
>>>>
>>>> - if (compare_ether_addr(bssid, sdata->u.ibss.bssid))
>>>> + if (memcmp(bssid, sdata->u.ibss.bssid, ETH_ALEN))
>>>> return NULL;
>>> So in this case it seems that it is possible that u.ibss.bssid is not
>>> aligned, consider fixing by doing
>>>
>>> --- ieee80211_i.h
>>> +++ ieee80211_i.h
>>> - u8 bssid[ETH_ALEN];
>>> + u8 bssid[ETH_ALEN] __align(2);
>>>
>>> or so instead.
>>>
>>>> --- a/net/wireless/scan.c
>>>> +++ b/net/wireless/scan.c
>>>> @@ -134,7 +134,7 @@ static bool is_bss(struct cfg80211_bss *a,
>>>> {
>>>> const u8 *ssidie;
>>>>
>>>> - if (bssid && compare_ether_addr(a->bssid, bssid))
>>>> + if (bssid && memcmp(a->bssid, bssid, ETH_ALEN))
>>> Since a->bssid is after a pointer I can't see how it would be unaligned,
>>> and bssid should be unaligned only if the call trace shows it's coming
>>> from the above u.ibss.bssid.
>>>
>>> johannes
>>
On Thu, 2009-07-23 at 20:56 +0300, Ivan Kuten wrote:
> > On Sun, 2009-06-28 at 15:18 +0300, Ivan Kuten wrote:
> >> Hello,
> >>
> >> In net/wireless/scan.c : cfg80211_wext_siwscan there seems also unaligned allocations
> >> for creq->ssids and creq->channels. Should it be something like that?
> >
> > Seems alright, but there is more than one instance of this, maybe you
> > can make a function to allocate a scan request properly and have it be
> > called from all the places it's needed.
> >
> > johannes
> >
>
> Hello Johannes,
> Can you point to that multiple scan allocations? I see only one kzalloc
> with followed possible alignment violation - it's in cfg80211_wext_siwscan in scan.c
I'm sure there's one in nl80211.c too
johannes
On Thu, Jul 23, 2009 at 07:59:31PM +0200, Johannes Berg wrote:
> On Thu, 2009-07-23 at 20:56 +0300, Ivan Kuten wrote:
> > > On Sun, 2009-06-28 at 15:18 +0300, Ivan Kuten wrote:
> > >> Hello,
> > >>
> > >> In net/wireless/scan.c : cfg80211_wext_siwscan there seems also unaligned allocations
> > >> for creq->ssids and creq->channels. Should it be something like that?
> > >
> > > Seems alright, but there is more than one instance of this, maybe you
> > > can make a function to allocate a scan request properly and have it be
> > > called from all the places it's needed.
> > >
> > > johannes
> > >
> >
> > Hello Johannes,
> > Can you point to that multiple scan allocations? I see only one kzalloc
> > with followed possible alignment violation - it's in cfg80211_wext_siwscan in scan.c
>
> I'm sure there's one in nl80211.c too
Ivan/Yauhen,
Will you be posted an updated patch? Time is short before the next merge window...
John
--
John W. Linville Someday the world will need a hero, and you
[email protected] might be all we have. Be ready.