| Message ID | YaV0pllJ5p/EuUat@google.com (mailing list archive) |
|---|---|
| State | Accepted |
| Commit | 1b8bb8919ef81bfc8873d223b9361f1685f2106d |
| Delegated to: | Kalle Valo |
| Headers | show |
| Series | mwifiex: Fix possible ABBA deadlock | expand |
Hi Brian, Thanks for your reply and explanation! The patch looks good to me :) Best wishes, Jia-Ju Bai On 2021/11/30 8:47, Brian Norris wrote: > Quoting Jia-Ju Bai <baijiaju1990@gmail.com>: > > mwifiex_dequeue_tx_packet() > spin_lock_bh(&priv->wmm.ra_list_spinlock); --> Line 1432 (Lock A) > mwifiex_send_addba() > spin_lock_bh(&priv->sta_list_spinlock); --> Line 608 (Lock B) > > mwifiex_process_sta_tx_pause() > spin_lock_bh(&priv->sta_list_spinlock); --> Line 398 (Lock B) > mwifiex_update_ralist_tx_pause() > spin_lock_bh(&priv->wmm.ra_list_spinlock); --> Line 941 (Lock A) > > Similar report for mwifiex_process_uap_tx_pause(). > > While the locking expectations in this driver are a bit unclear, the > Fixed commit only intended to protect the sta_ptr, so we can drop the > lock as soon as we're done with it. > > IIUC, this deadlock cannot actually happen, because command event > processing (which calls mwifiex_process_sta_tx_pause()) is > sequentialized with TX packet processing (e.g., > mwifiex_dequeue_tx_packet()) via the main loop (mwifiex_main_process()). > But it's good not to leave this potential issue lurking. > > Fixes: ("f0f7c2275fb9 mwifiex: minor cleanups w/ sta_list_spinlock in cfg80211.c") > Cc: Douglas Anderson <dianders@chromium.org> > Reported-by: TOTE Robot <oslab@tsinghua.edu.cn> > Link: https://lore.kernel.org/linux-wireless/0e495b14-efbb-e0da-37bd-af6bd677ee2c@gmail.com/ > Signed-off-by: Brian Norris <briannorris@chromium.org> > --- > > On Tue, Nov 23, 2021 at 11:31:34AM +0800, Jia-Ju Bai wrote: >> I am not quite sure whether these possible deadlocks are real and how to fix >> them if they are real. >> Any feedback would be appreciated, thanks :) > I think these are at least theoretically real, and so we should take > something like the $subject patch probably. But I don't believe we can > actually hit this due to the main-loop structure of this driver. > > Anyway, see the surrounding patch. > > Thanks, > Brian > > > drivers/net/wireless/marvell/mwifiex/sta_event.c | 8 ++++++-- > 1 file changed, 6 insertions(+), 2 deletions(-) > > diff --git a/drivers/net/wireless/marvell/mwifiex/sta_event.c b/drivers/net/wireless/marvell/mwifiex/sta_event.c > index 80e5d44bad9d..7d42c5d2dbf6 100644 > --- a/drivers/net/wireless/marvell/mwifiex/sta_event.c > +++ b/drivers/net/wireless/marvell/mwifiex/sta_event.c > @@ -365,10 +365,12 @@ static void mwifiex_process_uap_tx_pause(struct mwifiex_private *priv, > sta_ptr = mwifiex_get_sta_entry(priv, tp->peermac); > if (sta_ptr && sta_ptr->tx_pause != tp->tx_pause) { > sta_ptr->tx_pause = tp->tx_pause; > + spin_unlock_bh(&priv->sta_list_spinlock); > mwifiex_update_ralist_tx_pause(priv, tp->peermac, > tp->tx_pause); > + } else { > + spin_unlock_bh(&priv->sta_list_spinlock); > } > - spin_unlock_bh(&priv->sta_list_spinlock); > } > } > > @@ -400,11 +402,13 @@ static void mwifiex_process_sta_tx_pause(struct mwifiex_private *priv, > sta_ptr = mwifiex_get_sta_entry(priv, tp->peermac); > if (sta_ptr && sta_ptr->tx_pause != tp->tx_pause) { > sta_ptr->tx_pause = tp->tx_pause; > + spin_unlock_bh(&priv->sta_list_spinlock); > mwifiex_update_ralist_tx_pause(priv, > tp->peermac, > tp->tx_pause); > + } else { > + spin_unlock_bh(&priv->sta_list_spinlock); > } > - spin_unlock_bh(&priv->sta_list_spinlock); > } > } > }
Hi, On Mon, Nov 29, 2021 at 4:47 PM Brian Norris <briannorris@chromium.org> wrote: > > Quoting Jia-Ju Bai <baijiaju1990@gmail.com>: > > mwifiex_dequeue_tx_packet() > spin_lock_bh(&priv->wmm.ra_list_spinlock); --> Line 1432 (Lock A) > mwifiex_send_addba() > spin_lock_bh(&priv->sta_list_spinlock); --> Line 608 (Lock B) > > mwifiex_process_sta_tx_pause() > spin_lock_bh(&priv->sta_list_spinlock); --> Line 398 (Lock B) > mwifiex_update_ralist_tx_pause() > spin_lock_bh(&priv->wmm.ra_list_spinlock); --> Line 941 (Lock A) > > Similar report for mwifiex_process_uap_tx_pause(). > > While the locking expectations in this driver are a bit unclear, the > Fixed commit only intended to protect the sta_ptr, so we can drop the > lock as soon as we're done with it. > > IIUC, this deadlock cannot actually happen, because command event > processing (which calls mwifiex_process_sta_tx_pause()) is > sequentialized with TX packet processing (e.g., > mwifiex_dequeue_tx_packet()) via the main loop (mwifiex_main_process()). > But it's good not to leave this potential issue lurking. > > Fixes: ("f0f7c2275fb9 mwifiex: minor cleanups w/ sta_list_spinlock in cfg80211.c") > Cc: Douglas Anderson <dianders@chromium.org> > Reported-by: TOTE Robot <oslab@tsinghua.edu.cn> > Link: https://lore.kernel.org/linux-wireless/0e495b14-efbb-e0da-37bd-af6bd677ee2c@gmail.com/ > Signed-off-by: Brian Norris <briannorris@chromium.org> > --- > > On Tue, Nov 23, 2021 at 11:31:34AM +0800, Jia-Ju Bai wrote: > > I am not quite sure whether these possible deadlocks are real and how to fix > > them if they are real. > > Any feedback would be appreciated, thanks :) > > I think these are at least theoretically real, and so we should take > something like the $subject patch probably. But I don't believe we can > actually hit this due to the main-loop structure of this driver. > > Anyway, see the surrounding patch. > > Thanks, > Brian > > > drivers/net/wireless/marvell/mwifiex/sta_event.c | 8 ++++++-- > 1 file changed, 6 insertions(+), 2 deletions(-) Though I'm by no means an expert on this code and I wrote the patch in question a long time ago, this seems reasonable to me. Thanks for fixing. Reviewed-by: Douglas Anderson <dianders@chromium.org>
Brian Norris <briannorris@chromium.org> wrote: > Quoting Jia-Ju Bai <baijiaju1990@gmail.com>: > > mwifiex_dequeue_tx_packet() > spin_lock_bh(&priv->wmm.ra_list_spinlock); --> Line 1432 (Lock A) > mwifiex_send_addba() > spin_lock_bh(&priv->sta_list_spinlock); --> Line 608 (Lock B) > > mwifiex_process_sta_tx_pause() > spin_lock_bh(&priv->sta_list_spinlock); --> Line 398 (Lock B) > mwifiex_update_ralist_tx_pause() > spin_lock_bh(&priv->wmm.ra_list_spinlock); --> Line 941 (Lock A) > > Similar report for mwifiex_process_uap_tx_pause(). > > While the locking expectations in this driver are a bit unclear, the > Fixed commit only intended to protect the sta_ptr, so we can drop the > lock as soon as we're done with it. > > IIUC, this deadlock cannot actually happen, because command event > processing (which calls mwifiex_process_sta_tx_pause()) is > sequentialized with TX packet processing (e.g., > mwifiex_dequeue_tx_packet()) via the main loop (mwifiex_main_process()). > But it's good not to leave this potential issue lurking. > > Fixes: ("f0f7c2275fb9 mwifiex: minor cleanups w/ sta_list_spinlock in cfg80211.c") > Cc: Douglas Anderson <dianders@chromium.org> > Reported-by: TOTE Robot <oslab@tsinghua.edu.cn> > Link: https://lore.kernel.org/linux-wireless/0e495b14-efbb-e0da-37bd-af6bd677ee2c@gmail.com/ > Signed-off-by: Brian Norris <briannorris@chromium.org> > Reviewed-by: Douglas Anderson <dianders@chromium.org> Fixes tag is in wrong format, should be: Fixes: f0f7c2275fb9 ("mwifiex: minor cleanups w/ sta_list_spinlock in cfg80211.c") I'll fix it during commit.
Brian Norris <briannorris@chromium.org> wrote: > Quoting Jia-Ju Bai <baijiaju1990@gmail.com>: > > mwifiex_dequeue_tx_packet() > spin_lock_bh(&priv->wmm.ra_list_spinlock); --> Line 1432 (Lock A) > mwifiex_send_addba() > spin_lock_bh(&priv->sta_list_spinlock); --> Line 608 (Lock B) > > mwifiex_process_sta_tx_pause() > spin_lock_bh(&priv->sta_list_spinlock); --> Line 398 (Lock B) > mwifiex_update_ralist_tx_pause() > spin_lock_bh(&priv->wmm.ra_list_spinlock); --> Line 941 (Lock A) > > Similar report for mwifiex_process_uap_tx_pause(). > > While the locking expectations in this driver are a bit unclear, the > Fixed commit only intended to protect the sta_ptr, so we can drop the > lock as soon as we're done with it. > > IIUC, this deadlock cannot actually happen, because command event > processing (which calls mwifiex_process_sta_tx_pause()) is > sequentialized with TX packet processing (e.g., > mwifiex_dequeue_tx_packet()) via the main loop (mwifiex_main_process()). > But it's good not to leave this potential issue lurking. > > Fixes: f0f7c2275fb9 ("mwifiex: minor cleanups w/ sta_list_spinlock in cfg80211.c") > Cc: Douglas Anderson <dianders@chromium.org> > Reported-by: TOTE Robot <oslab@tsinghua.edu.cn> > Link: https://lore.kernel.org/linux-wireless/0e495b14-efbb-e0da-37bd-af6bd677ee2c@gmail.com/ > Signed-off-by: Brian Norris <briannorris@chromium.org> > Reviewed-by: Douglas Anderson <dianders@chromium.org> Patch applied to wireless-drivers-next.git, thanks. 1b8bb8919ef8 mwifiex: Fix possible ABBA deadlock
diff --git a/drivers/net/wireless/marvell/mwifiex/sta_event.c b/drivers/net/wireless/marvell/mwifiex/sta_event.c index 80e5d44bad9d..7d42c5d2dbf6 100644 --- a/drivers/net/wireless/marvell/mwifiex/sta_event.c +++ b/drivers/net/wireless/marvell/mwifiex/sta_event.c @@ -365,10 +365,12 @@ static void mwifiex_process_uap_tx_pause(struct mwifiex_private *priv, sta_ptr = mwifiex_get_sta_entry(priv, tp->peermac); if (sta_ptr && sta_ptr->tx_pause != tp->tx_pause) { sta_ptr->tx_pause = tp->tx_pause; + spin_unlock_bh(&priv->sta_list_spinlock); mwifiex_update_ralist_tx_pause(priv, tp->peermac, tp->tx_pause); + } else { + spin_unlock_bh(&priv->sta_list_spinlock); } - spin_unlock_bh(&priv->sta_list_spinlock); } } @@ -400,11 +402,13 @@ static void mwifiex_process_sta_tx_pause(struct mwifiex_private *priv, sta_ptr = mwifiex_get_sta_entry(priv, tp->peermac); if (sta_ptr && sta_ptr->tx_pause != tp->tx_pause) { sta_ptr->tx_pause = tp->tx_pause; + spin_unlock_bh(&priv->sta_list_spinlock); mwifiex_update_ralist_tx_pause(priv, tp->peermac, tp->tx_pause); + } else { + spin_unlock_bh(&priv->sta_list_spinlock); } - spin_unlock_bh(&priv->sta_list_spinlock); } } }
Quoting Jia-Ju Bai <baijiaju1990@gmail.com>: mwifiex_dequeue_tx_packet() spin_lock_bh(&priv->wmm.ra_list_spinlock); --> Line 1432 (Lock A) mwifiex_send_addba() spin_lock_bh(&priv->sta_list_spinlock); --> Line 608 (Lock B) mwifiex_process_sta_tx_pause() spin_lock_bh(&priv->sta_list_spinlock); --> Line 398 (Lock B) mwifiex_update_ralist_tx_pause() spin_lock_bh(&priv->wmm.ra_list_spinlock); --> Line 941 (Lock A) Similar report for mwifiex_process_uap_tx_pause(). While the locking expectations in this driver are a bit unclear, the Fixed commit only intended to protect the sta_ptr, so we can drop the lock as soon as we're done with it. IIUC, this deadlock cannot actually happen, because command event processing (which calls mwifiex_process_sta_tx_pause()) is sequentialized with TX packet processing (e.g., mwifiex_dequeue_tx_packet()) via the main loop (mwifiex_main_process()). But it's good not to leave this potential issue lurking. Fixes: ("f0f7c2275fb9 mwifiex: minor cleanups w/ sta_list_spinlock in cfg80211.c") Cc: Douglas Anderson <dianders@chromium.org> Reported-by: TOTE Robot <oslab@tsinghua.edu.cn> Link: https://lore.kernel.org/linux-wireless/0e495b14-efbb-e0da-37bd-af6bd677ee2c@gmail.com/ Signed-off-by: Brian Norris <briannorris@chromium.org> --- On Tue, Nov 23, 2021 at 11:31:34AM +0800, Jia-Ju Bai wrote: > I am not quite sure whether these possible deadlocks are real and how to fix > them if they are real. > Any feedback would be appreciated, thanks :) I think these are at least theoretically real, and so we should take something like the $subject patch probably. But I don't believe we can actually hit this due to the main-loop structure of this driver. Anyway, see the surrounding patch. Thanks, Brian drivers/net/wireless/marvell/mwifiex/sta_event.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)