| Message ID | 1464378815-23282-1-git-send-email-zajec5@gmail.com (mailing list archive) |
|---|---|
| State | Changes Requested |
| Delegated to: | Kalle Valo |
| Headers | show |
Rafał Miłecki <zajec5@gmail.com> writes: > Removing P2P interface is handled by sending a proper request to the > firmware. On success firmware triggers an event and driver's handler > removes a matching interface. > > However on event timeout we remove interface directly from the cfg80211 > callback. Current code doesn't handle this case correctly as it always > assumes rtnl to be unlocked. > > Fix it by adding an extra rtnl_locked parameter to functions and calling > unregister_netdevice when needed. > > Signed-off-by: Rafał Miłecki <zajec5@gmail.com> Arend, are you ok to push this to 4.7?
Rafał Miłecki <zajec5@gmail.com> writes: > Removing P2P interface is handled by sending a proper request to the > firmware. On success firmware triggers an event and driver's handler > removes a matching interface. > > However on event timeout we remove interface directly from the cfg80211 > callback. Current code doesn't handle this case correctly as it always > assumes rtnl to be unlocked. > > Fix it by adding an extra rtnl_locked parameter to functions and calling > unregister_netdevice when needed. > > Signed-off-by: Rafał Miłecki <zajec5@gmail.com> Failed to apply, please rebase: Applying: brcmfmac: fix lockup when removing P2P interface after event timeout Using index info to reconstruct a base tree... Falling back to patching base and 3-way merge... Auto-merging drivers/net/wireless/broadcom/brcm80211/brcmfmac/p2p.c Auto-merging drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.h CONFLICT (content): Merge conflict in drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.h Auto-merging drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.c Auto-merging drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c Failed to merge in the changes. Patch failed at 0001 brcmfmac: fix lockup when removing P2P interface after event timeout
On 16 June 2016 at 17:10, Kalle Valo <kvalo@codeaurora.org> wrote: > Rafał Miłecki <zajec5@gmail.com> writes: > >> Removing P2P interface is handled by sending a proper request to the >> firmware. On success firmware triggers an event and driver's handler >> removes a matching interface. >> >> However on event timeout we remove interface directly from the cfg80211 >> callback. Current code doesn't handle this case correctly as it always >> assumes rtnl to be unlocked. >> >> Fix it by adding an extra rtnl_locked parameter to functions and calling >> unregister_netdevice when needed. >> >> Signed-off-by: Rafał Miłecki <zajec5@gmail.com> > > Failed to apply, please rebase: > > Applying: brcmfmac: fix lockup when removing P2P interface after event timeout > Using index info to reconstruct a base tree... > Falling back to patching base and 3-way merge... > Auto-merging drivers/net/wireless/broadcom/brcm80211/brcmfmac/p2p.c > Auto-merging drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.h > CONFLICT (content): Merge conflict in drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.h > Auto-merging drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.c > Auto-merging drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c > Failed to merge in the changes. > Patch failed at 0001 brcmfmac: fix lockup when removing P2P interface after event timeout What tree did you try it on? I just went into a dir where I have cloned: git://git.kernel.org/pub/scm/linux/kernel/git/kvalo/wireless-drivers.git My HEAD commit is: 034fdd4 Merge ath-current from ath.git And I can apply this patch cleanly doing: curl https://patchwork.kernel.org/patch/9138925/mbox/ | git am
Rafał Miłecki <zajec5@gmail.com> writes: > On 16 June 2016 at 17:10, Kalle Valo <kvalo@codeaurora.org> wrote: >> Rafał Miłecki <zajec5@gmail.com> writes: >> >>> Removing P2P interface is handled by sending a proper request to the >>> firmware. On success firmware triggers an event and driver's handler >>> removes a matching interface. >>> >>> However on event timeout we remove interface directly from the cfg80211 >>> callback. Current code doesn't handle this case correctly as it always >>> assumes rtnl to be unlocked. >>> >>> Fix it by adding an extra rtnl_locked parameter to functions and calling >>> unregister_netdevice when needed. >>> >>> Signed-off-by: Rafał Miłecki <zajec5@gmail.com> >> >> Failed to apply, please rebase: >> >> Applying: brcmfmac: fix lockup when removing P2P interface after event timeout >> Using index info to reconstruct a base tree... >> Falling back to patching base and 3-way merge... >> Auto-merging drivers/net/wireless/broadcom/brcm80211/brcmfmac/p2p.c >> Auto-merging drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.h >> CONFLICT (content): Merge conflict in drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.h >> Auto-merging drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.c >> Auto-merging drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c >> Failed to merge in the changes. >> Patch failed at 0001 brcmfmac: fix lockup when removing P2P interface after event timeout > > What tree did you try it on? > > I just went into a dir where I have cloned: > git://git.kernel.org/pub/scm/linux/kernel/git/kvalo/wireless-drivers.git > > My HEAD commit is: > 034fdd4 Merge ath-current from ath.git > > And I can apply this patch cleanly doing: > curl https://patchwork.kernel.org/patch/9138925/mbox/ | git am I was trying to apply this to wireless-drivers-next. I didn't get a confirmation from Arend and I didn't consider the fix important enough for 4.7. But of course I can reconsider if needed.
On 17 June 2016 at 07:13, Kalle Valo <kvalo@codeaurora.org> wrote: > Rafał Miłecki <zajec5@gmail.com> writes: > >> On 16 June 2016 at 17:10, Kalle Valo <kvalo@codeaurora.org> wrote: >>> Rafał Miłecki <zajec5@gmail.com> writes: >>> >>>> Removing P2P interface is handled by sending a proper request to the >>>> firmware. On success firmware triggers an event and driver's handler >>>> removes a matching interface. >>>> >>>> However on event timeout we remove interface directly from the cfg80211 >>>> callback. Current code doesn't handle this case correctly as it always >>>> assumes rtnl to be unlocked. >>>> >>>> Fix it by adding an extra rtnl_locked parameter to functions and calling >>>> unregister_netdevice when needed. >>>> >>>> Signed-off-by: Rafał Miłecki <zajec5@gmail.com> >>> >>> Failed to apply, please rebase: >>> >>> Applying: brcmfmac: fix lockup when removing P2P interface after event timeout >>> Using index info to reconstruct a base tree... >>> Falling back to patching base and 3-way merge... >>> Auto-merging drivers/net/wireless/broadcom/brcm80211/brcmfmac/p2p.c >>> Auto-merging drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.h >>> CONFLICT (content): Merge conflict in drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.h >>> Auto-merging drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.c >>> Auto-merging drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c >>> Failed to merge in the changes. >>> Patch failed at 0001 brcmfmac: fix lockup when removing P2P interface after event timeout >> >> What tree did you try it on? >> >> I just went into a dir where I have cloned: >> git://git.kernel.org/pub/scm/linux/kernel/git/kvalo/wireless-drivers.git >> >> My HEAD commit is: >> 034fdd4 Merge ath-current from ath.git >> >> And I can apply this patch cleanly doing: >> curl https://patchwork.kernel.org/patch/9138925/mbox/ | git am > > I was trying to apply this to wireless-drivers-next. I didn't get a > confirmation from Arend and I didn't consider the fix important enough > for 4.7. But of course I can reconsider if needed. I think I agree it won't hurt to get it into -next. Noone earlier reported this bug and it seems to be there for a long time. Also applying it to -next will allow avoiding merge conflicts and immediate development work on -next. I'll resend this patch rebased on -next soon.
Rafał Miłecki <zajec5@gmail.com> writes: >>>> Failed to apply, please rebase: >>>> >>>> Applying: brcmfmac: fix lockup when removing P2P interface after event timeout >>>> Using index info to reconstruct a base tree... >>>> Falling back to patching base and 3-way merge... >>>> Auto-merging drivers/net/wireless/broadcom/brcm80211/brcmfmac/p2p.c >>>> Auto-merging drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.h >>>> CONFLICT (content): Merge conflict in drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.h >>>> Auto-merging drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.c >>>> Auto-merging drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c >>>> Failed to merge in the changes. >>>> Patch failed at 0001 brcmfmac: fix lockup when removing P2P interface after event timeout >>> >>> What tree did you try it on? >>> >>> I just went into a dir where I have cloned: >>> git://git.kernel.org/pub/scm/linux/kernel/git/kvalo/wireless-drivers.git >>> >>> My HEAD commit is: >>> 034fdd4 Merge ath-current from ath.git >>> >>> And I can apply this patch cleanly doing: >>> curl https://patchwork.kernel.org/patch/9138925/mbox/ | git am >> >> I was trying to apply this to wireless-drivers-next. I didn't get a >> confirmation from Arend and I didn't consider the fix important enough >> for 4.7. But of course I can reconsider if needed. > > I think I agree it won't hurt to get it into -next. Noone earlier > reported this bug and it seems to be there for a long time. Also > applying it to -next will allow avoiding merge conflicts and immediate > development work on -next. Exactly, there is a cost when taking patches to wireless-drivers.git and that's why I try to keep the bar high. Regressions and user reported bugs take priority, other fixes are handled case by case. > I'll resend this patch rebased on -next soon. Thanks.
diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c index 62f475e..ef09382 100644 --- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c +++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c @@ -5364,7 +5364,7 @@ brcmf_notify_connect_status_ap(struct brcmf_cfg80211_info *cfg, brcmf_dbg(CONN, "AP mode link down\n"); complete(&cfg->vif_disabled); if (ifp->vif->mbss) - brcmf_remove_interface(ifp); + brcmf_remove_interface(ifp, false); return 0; } diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.c b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.c index b590499..7f8ba48c 100644 --- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.c +++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.c @@ -548,12 +548,16 @@ fail: return -EBADE; } -static void brcmf_net_detach(struct net_device *ndev) +static void brcmf_net_detach(struct net_device *ndev, bool rtnl_locked) { - if (ndev->reg_state == NETREG_REGISTERED) - unregister_netdev(ndev); - else + if (ndev->reg_state == NETREG_REGISTERED) { + if (rtnl_locked) + unregister_netdevice(ndev); + else + unregister_netdev(ndev); + } else { brcmf_cfg80211_free_netdev(ndev); + } } void brcmf_net_setcarrier(struct brcmf_if *ifp, bool on) @@ -651,7 +655,7 @@ struct brcmf_if *brcmf_add_if(struct brcmf_pub *drvr, s32 bsscfgidx, s32 ifidx, brcmf_err("ERROR: netdev:%s already exists\n", ifp->ndev->name); netif_stop_queue(ifp->ndev); - brcmf_net_detach(ifp->ndev); + brcmf_net_detach(ifp->ndev, false); drvr->iflist[bsscfgidx] = NULL; } else { brcmf_dbg(INFO, "netdev:%s ignore IF event\n", @@ -699,7 +703,8 @@ struct brcmf_if *brcmf_add_if(struct brcmf_pub *drvr, s32 bsscfgidx, s32 ifidx, return ifp; } -static void brcmf_del_if(struct brcmf_pub *drvr, s32 bsscfgidx) +static void brcmf_del_if(struct brcmf_pub *drvr, s32 bsscfgidx, + bool rtnl_locked) { struct brcmf_if *ifp; @@ -729,7 +734,7 @@ static void brcmf_del_if(struct brcmf_pub *drvr, s32 bsscfgidx) cancel_work_sync(&ifp->multicast_work); cancel_work_sync(&ifp->ndoffload_work); } - brcmf_net_detach(ifp->ndev); + brcmf_net_detach(ifp->ndev, rtnl_locked); } else { /* Only p2p device interfaces which get dynamically created * end up here. In this case the p2p module should be informed @@ -743,14 +748,14 @@ static void brcmf_del_if(struct brcmf_pub *drvr, s32 bsscfgidx) } } -void brcmf_remove_interface(struct brcmf_if *ifp) +void brcmf_remove_interface(struct brcmf_if *ifp, bool rtnl_locked) { if (!ifp || WARN_ON(ifp->drvr->iflist[ifp->bsscfgidx] != ifp)) return; brcmf_dbg(TRACE, "Enter, bsscfgidx=%d, ifidx=%d\n", ifp->bsscfgidx, ifp->ifidx); brcmf_fws_del_interface(ifp); - brcmf_del_if(ifp->drvr, ifp->bsscfgidx); + brcmf_del_if(ifp->drvr, ifp->bsscfgidx, rtnl_locked); } int brcmf_get_next_free_bsscfgidx(struct brcmf_pub *drvr) @@ -1081,9 +1086,9 @@ fail: brcmf_fws_deinit(drvr); } if (ifp) - brcmf_net_detach(ifp->ndev); + brcmf_net_detach(ifp->ndev, false); if (p2p_ifp) - brcmf_net_detach(p2p_ifp->ndev); + brcmf_net_detach(p2p_ifp->ndev, false); drvr->iflist[0] = NULL; drvr->iflist[1] = NULL; if (drvr->settings->ignore_probe_fail) @@ -1152,7 +1157,7 @@ void brcmf_detach(struct device *dev) /* make sure primary interface removed last */ for (i = BRCMF_MAX_IFS-1; i > -1; i--) - brcmf_remove_interface(drvr->iflist[i]); + brcmf_remove_interface(drvr->iflist[i], false); brcmf_cfg80211_detach(drvr->config); diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.h b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.h index 647d3cc..5c5a392 100644 --- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.h +++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.h @@ -216,7 +216,7 @@ struct brcmf_if *brcmf_get_ifp(struct brcmf_pub *drvr, int ifidx); int brcmf_net_attach(struct brcmf_if *ifp, bool rtnl_locked); struct brcmf_if *brcmf_add_if(struct brcmf_pub *drvr, s32 bsscfgidx, s32 ifidx, bool is_p2pdev, char *name, u8 *mac_addr); -void brcmf_remove_interface(struct brcmf_if *ifp); +void brcmf_remove_interface(struct brcmf_if *ifp, bool rtnl_locked); int brcmf_get_next_free_bsscfgidx(struct brcmf_pub *drvr); void brcmf_txflowblock_if(struct brcmf_if *ifp, enum brcmf_netif_stop_reason reason, bool state); diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/fweh.c b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/fweh.c index b390561..9da7a4c 100644 --- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/fweh.c +++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/fweh.c @@ -183,7 +183,7 @@ static void brcmf_fweh_handle_if_event(struct brcmf_pub *drvr, err = brcmf_fweh_call_event_handler(ifp, emsg->event_code, emsg, data); if (ifp && ifevent->action == BRCMF_E_IF_DEL) - brcmf_remove_interface(ifp); + brcmf_remove_interface(ifp, false); } /** diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/p2p.c b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/p2p.c index a70cda6..c4dc6be 100644 --- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/p2p.c +++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/p2p.c @@ -2290,7 +2290,7 @@ int brcmf_p2p_del_vif(struct wiphy *wiphy, struct wireless_dev *wdev) err = 0; } if (err) - brcmf_remove_interface(vif->ifp); + brcmf_remove_interface(vif->ifp, true); brcmf_cfg80211_arm_vif_event(cfg, NULL); if (vif->wdev.iftype != NL80211_IFTYPE_P2P_DEVICE) @@ -2396,7 +2396,7 @@ void brcmf_p2p_detach(struct brcmf_p2p_info *p2p) if (vif != NULL) { brcmf_p2p_cancel_remain_on_channel(vif->ifp); brcmf_p2p_deinit_discovery(p2p); - brcmf_remove_interface(vif->ifp); + brcmf_remove_interface(vif->ifp, false); } /* just set it all to zero */ memset(p2p, 0, sizeof(*p2p));
Removing P2P interface is handled by sending a proper request to the firmware. On success firmware triggers an event and driver's handler removes a matching interface. However on event timeout we remove interface directly from the cfg80211 callback. Current code doesn't handle this case correctly as it always assumes rtnl to be unlocked. Fix it by adding an extra rtnl_locked parameter to functions and calling unregister_netdevice when needed. Signed-off-by: Rafał Miłecki <zajec5@gmail.com> --- .../broadcom/brcm80211/brcmfmac/cfg80211.c | 2 +- .../wireless/broadcom/brcm80211/brcmfmac/core.c | 29 +++++++++++++--------- .../wireless/broadcom/brcm80211/brcmfmac/core.h | 2 +- .../wireless/broadcom/brcm80211/brcmfmac/fweh.c | 2 +- .../net/wireless/broadcom/brcm80211/brcmfmac/p2p.c | 4 +-- 5 files changed, 22 insertions(+), 17 deletions(-)