[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Wed, 8 Aug 2018 22:05:54 +0200
From: Stefan Schmidt <stefan@...enfreihafen.org>
To: Alexander Aring <aring@...atatu.com>, netdev@...r.kernel.org
Cc: linux-wpan@...r.kernel.org, kernel@...atatu.com
Subject: Re: [PATCH net-next] ieee802154: hwsim: fix rcu handling
Hello David.
I want to apologize how the submission of the ieee802154_hwsim driver
worked out. The sparse warning finally fixed by this commit came due to
an outdated sparse used on our side while kbuild was up to date. I fixed
this now and sparse has no warnings for the ieee802154_hwsim driver
anymore with this patch applied.
The other two fixes from Wei Yongjun are problems I should have spotted
during code review, but did not. No excuse, I just overlooked them.
I would, once again, ask you to apply these three directly to net-next.
I you would prefer for me to pick them up and send you a pull request
this time just let me know and I will do so tomorrow.
On 08/08/2018 01:32 AM, Alexander Aring wrote:
> This patch adds missing rcu_assign_pointer()/rcu_dereference() to used rcu
> pointers. There was already a previous commit c5d99d2b35da ("ieee802154:
> hwsim: fix rcu address annotation"), but there was more which was
> pointed out on my side by using newest sparse version.
>
> Cc: Stefan Schmidt <stefan@...enfreihafen.org>
> Fixes: f25da51fdc38 ("ieee802154: hwsim: add replacement for fakelb")
> Signed-off-by: Alexander Aring <aring@...atatu.com>
> ---
> After I installed finally the newest sparse version I found more what kbuild
> was pointed out.
>
> I hope I did it right, not sure if I really need rcu functionality in these
> case because protection by mutex but I make sparse silent.
> At some places the resource isn't a shared resource at this moment.
>
> Anyway in my case I want to use this driver to create easily multi-hop
> scenarios, if somebody report things (or I hit it) which smells like some
> memory in this area - I will try and help to fix it.
>
> Newest sparse version from git is happy now and I hope kbuild bot as well.
>
> I tested this patch with RPOVE_RCU enabled and tried to update these settings
> while heavy loading in rcu protected xmit hotpath is occured, without any
> issues so far.
>
> Sorry again.
>
> drivers/net/ieee802154/mac802154_hwsim.c | 24 +++++++++++++++++++-----
> 1 file changed, 19 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/net/ieee802154/mac802154_hwsim.c b/drivers/net/ieee802154/mac802154_hwsim.c
> index f4e92054f7df..53f39435321e 100644
> --- a/drivers/net/ieee802154/mac802154_hwsim.c
> +++ b/drivers/net/ieee802154/mac802154_hwsim.c
> @@ -22,6 +22,7 @@
> #include <linux/module.h>
> #include <linux/timer.h>
> #include <linux/platform_device.h>
> +#include <linux/rtnetlink.h>
> #include <linux/netdevice.h>
> #include <linux/device.h>
> #include <linux/spinlock.h>
> @@ -110,7 +111,7 @@ static int hwsim_hw_channel(struct ieee802154_hw *hw, u8 page, u8 channel)
> pib->page = page;
> pib->channel = channel;
>
> - pib_old = phy->pib;
> + pib_old = rtnl_dereference(phy->pib);
> rcu_assign_pointer(phy->pib, pib);
> kfree_rcu(pib_old, rcu);
> return 0;
> @@ -406,7 +407,7 @@ static struct hwsim_edge *hwsim_alloc_edge(struct hwsim_phy *endpoint, u8 lqi)
> }
>
> einfo->lqi = 0xff;
> - e->info = einfo;
> + rcu_assign_pointer(e->info, einfo);
> e->endpoint = endpoint;
>
> return e;
> @@ -414,7 +415,13 @@ static struct hwsim_edge *hwsim_alloc_edge(struct hwsim_phy *endpoint, u8 lqi)
>
> static void hwsim_free_edge(struct hwsim_edge *e)
> {
> - kfree_rcu(e->info, rcu);
> + struct hwsim_edge_info *einfo;
> +
> + rcu_read_lock();
> + einfo = rcu_dereference(e->info);
> + rcu_read_unlock();
> +
> + kfree_rcu(einfo, rcu);
> kfree_rcu(e, rcu);
> }
>
> @@ -796,7 +803,7 @@ static int hwsim_add_one(struct genl_info *info, struct device *dev,
> goto err_pib;
> }
>
> - phy->pib = pib;
> + rcu_assign_pointer(phy->pib, pib);
> phy->idx = idx;
> INIT_LIST_HEAD(&phy->edges);
>
> @@ -829,10 +836,17 @@ static int hwsim_add_one(struct genl_info *info, struct device *dev,
>
> static void hwsim_del(struct hwsim_phy *phy)
> {
> + struct hwsim_pib *pib;
> +
> hwsim_edge_unsubscribe_me(phy);
>
> list_del(&phy->list);
> - kfree_rcu(phy->pib, rcu);
> +
> + rcu_read_lock();
> + pib = rcu_dereference(phy->pib);
> + rcu_read_unlock();
> +
> + kfree_rcu(pib, rcu);
>
> ieee802154_unregister_hw(phy->hw);
> ieee802154_free_hw(phy->hw);
>
Signed-off-by: Stefan Schmidt <stefan@...enfreihafen.org>
regards
Stefan Schmidt
Powered by blists - more mailing lists