| Message ID | 20200927194920.918550822@linutronix.de (mailing list archive) |
|---|---|
| State | Not Applicable |
| Delegated to: | Johannes Berg |
| Headers | show |
| Series | net: in_interrupt() cleanup and fixes | expand |
On 9/27/20 12:48 PM, Thomas Gleixner wrote: > From: Sebastian Andrzej Siewior <bigeasy@linutronix.de> > > The in_interrupt() usage in this driver tries to figure out which context > may sleep and which context may not sleep. in_interrupt() is not really > suitable as it misses both preemption disabled and interrupt disabled > invocations from task context. > > Conditionals like that in driver code are frowned upon in general because > invocations of functions from invalid contexts might not be detected > as the conditional papers over it. > > ionic_lif_addr() can be called from: > > 1) ->ndo_set_rx_mode() which is under netif_addr_lock_bh()) so it must not > sleep. > > 2) Init and setup functions which are in fully preemptible task context. > > _ionic_lif_rx_mode() has only one call path with BH disabled. > > ionic_link_status_check_request() has two call paths: > > 1) NAPI which obviously cannot sleep > > 2) Setup which is again fully preemptible task context > > Add 'can_sleep' arguments to the affected functions and let the callers > provide the context instead of letting the functions deduce it. > > Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> > Signed-off-by: Thomas Gleixner <tglx@linutronix.de> > Cc: Shannon Nelson <snelson@pensando.io> > Cc: Pensando Drivers <drivers@pensando.io> > Cc: "David S. Miller" <davem@davemloft.net> > Cc: Jakub Kicinski <kuba@kernel.org> > Cc: netdev@vger.kernel.org Acked-by: Shannon Nelson <snelson@pensando.io> > --- > > While reviewing the callpaths, a couple of things were observed which could > be improved: > > - ionic_lif_deferred_work() can iterate over the list. There is no need > to schedule the work item after each iteration I think the original writer's intent was to avoid monopolizing the work thread for very long on any one cycle, with the thought that we'd be making more use of this than we currently are. I'll address this. > > - ionic_link_status_check_request() could have ionic_deferred_work within > ionic_lif(). This would avoid memory allocation from NAPI. More > important, once IONIC_LIF_F_LINK_CHECK_REQUESTED is set and that alloc > fails, the link check never happens. Thanks, I'll fix up that error condition. > > - ionic_lif_handle_fw_down() sets IONIC_LIF_F_FW_RESET. Invokes then > ionic_lif_deinit() which only invokes cancel_work_sync() if > IONIC_LIF_F_FW_RESET is not set. I think the logic is wrong here as > the work must always be cancled. Also the list with ionic_deferred > work items needs a clean up. I'll look at that, thanks. sln
On 9/28/20 10:24 AM, Shannon Nelson wrote: > On 9/27/20 12:48 PM, Thomas Gleixner wrote: >> From: Sebastian Andrzej Siewior <bigeasy@linutronix.de> >> >> The in_interrupt() usage in this driver tries to figure out which >> context >> may sleep and which context may not sleep. in_interrupt() is not really >> suitable as it misses both preemption disabled and interrupt disabled >> invocations from task context. >> >> Conditionals like that in driver code are frowned upon in general >> because >> invocations of functions from invalid contexts might not be detected >> as the conditional papers over it. >> >> ionic_lif_addr() can be called from: >> >> 1) ->ndo_set_rx_mode() which is under netif_addr_lock_bh()) so it >> must not >> sleep. >> >> 2) Init and setup functions which are in fully preemptible task >> context. >> >> _ionic_lif_rx_mode() has only one call path with BH disabled. Now that I've had my coffee, let's look at this again - there are multiple paths that get us to _ionic_lif_rx_mode(): .ndo_set_rx_mode ionic_set_rx_mode, _ionic_lif_rx_mode { ionic_open, ionic_lif_handle_fw_up, ionic_start_queues_reconfig } ionic_txrx_init ionic_set_rx_mode _ionic_lif_rx_mode We may not want to change this one. sln >> >> ionic_link_status_check_request() has two call paths: >> >> 1) NAPI which obviously cannot sleep >> >> 2) Setup which is again fully preemptible task context >> >> Add 'can_sleep' arguments to the affected functions and let the callers >> provide the context instead of letting the functions deduce it. >> >> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> >> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> >> Cc: Shannon Nelson <snelson@pensando.io> >> Cc: Pensando Drivers <drivers@pensando.io> >> Cc: "David S. Miller" <davem@davemloft.net> >> Cc: Jakub Kicinski <kuba@kernel.org> >> Cc: netdev@vger.kernel.org > > Acked-by: Shannon Nelson <snelson@pensando.io> > >> --- >> >> While reviewing the callpaths, a couple of things were observed which >> could >> be improved: >> >> - ionic_lif_deferred_work() can iterate over the list. There is no need >> to schedule the work item after each iteration > > I think the original writer's intent was to avoid monopolizing the > work thread for very long on any one cycle, with the thought that we'd > be making more use of this than we currently are. I'll address this. > >> >> - ionic_link_status_check_request() could have ionic_deferred_work >> within >> ionic_lif(). This would avoid memory allocation from NAPI. More >> important, once IONIC_LIF_F_LINK_CHECK_REQUESTED is set and that >> alloc >> fails, the link check never happens. > > Thanks, I'll fix up that error condition. > >> >> - ionic_lif_handle_fw_down() sets IONIC_LIF_F_FW_RESET. Invokes then >> ionic_lif_deinit() which only invokes cancel_work_sync() if >> IONIC_LIF_F_FW_RESET is not set. I think the logic is wrong here as >> the work must always be cancled. Also the list with ionic_deferred >> work items needs a clean up. > > I'll look at that, thanks. > > sln > >
On Mon, Sep 28 2020 at 12:51, Shannon Nelson wrote: > On 9/28/20 10:24 AM, Shannon Nelson wrote: >>> ionic_lif_addr() can be called from: >>> >>> 1) ->ndo_set_rx_mode() which is under netif_addr_lock_bh()) so it >>> must not >>> sleep. >>> >>> 2) Init and setup functions which are in fully preemptible task >>> context. >>> >>> _ionic_lif_rx_mode() has only one call path with BH disabled. > > Now that I've had my coffee, let's look at this again - there are > multiple paths that get us to _ionic_lif_rx_mode(): > > .ndo_set_rx_mode > ionic_set_rx_mode, > _ionic_lif_rx_mode > > { ionic_open, ionic_lif_handle_fw_up, ionic_start_queues_reconfig } > ionic_txrx_init > ionic_set_rx_mode > _ionic_lif_rx_mode Hrm. Let me stare at it again...
--- a/drivers/net/ethernet/pensando/ionic/ionic_dev.c +++ b/drivers/net/ethernet/pensando/ionic/ionic_dev.c @@ -22,7 +22,7 @@ static void ionic_watchdog_cb(struct tim hb = ionic_heartbeat_check(ionic); if (hb >= 0 && ionic->master_lif) - ionic_link_status_check_request(ionic->master_lif); + ionic_link_status_check_request(ionic->master_lif, false); } void ionic_init_devinfo(struct ionic *ionic) --- a/drivers/net/ethernet/pensando/ionic/ionic_lif.c +++ b/drivers/net/ethernet/pensando/ionic/ionic_lif.c @@ -135,7 +135,7 @@ static void ionic_link_status_check(stru clear_bit(IONIC_LIF_F_LINK_CHECK_REQUESTED, lif->state); } -void ionic_link_status_check_request(struct ionic_lif *lif) +void ionic_link_status_check_request(struct ionic_lif *lif, bool can_sleep) { struct ionic_deferred_work *work; @@ -143,7 +143,7 @@ void ionic_link_status_check_request(str if (test_and_set_bit(IONIC_LIF_F_LINK_CHECK_REQUESTED, lif->state)) return; - if (in_interrupt()) { + if (!can_sleep) { work = kzalloc(sizeof(*work), GFP_ATOMIC); if (!work) return; @@ -751,7 +751,7 @@ static bool ionic_notifyq_service(struct switch (le16_to_cpu(comp->event.ecode)) { case IONIC_EVENT_LINK_CHANGE: - ionic_link_status_check_request(lif); + ionic_link_status_check_request(lif, false); break; case IONIC_EVENT_RESET: work = kzalloc(sizeof(*work), GFP_ATOMIC); @@ -928,7 +928,8 @@ static int ionic_lif_addr_del(struct ion return 0; } -static int ionic_lif_addr(struct ionic_lif *lif, const u8 *addr, bool add) +static int ionic_lif_addr(struct ionic_lif *lif, const u8 *addr, bool add, + bool can_sleep) { struct ionic *ionic = lif->ionic; struct ionic_deferred_work *work; @@ -957,7 +958,7 @@ static int ionic_lif_addr(struct ionic_l lif->nucast--; } - if (in_interrupt()) { + if (!can_sleep) { work = kzalloc(sizeof(*work), GFP_ATOMIC); if (!work) { netdev_err(lif->netdev, "%s OOM\n", __func__); @@ -983,12 +984,12 @@ static int ionic_lif_addr(struct ionic_l static int ionic_addr_add(struct net_device *netdev, const u8 *addr) { - return ionic_lif_addr(netdev_priv(netdev), addr, true); + return ionic_lif_addr(netdev_priv(netdev), addr, true, false); } static int ionic_addr_del(struct net_device *netdev, const u8 *addr) { - return ionic_lif_addr(netdev_priv(netdev), addr, false); + return ionic_lif_addr(netdev_priv(netdev), addr, false, false); } static void ionic_lif_rx_mode(struct ionic_lif *lif, unsigned int rx_mode) @@ -1032,19 +1033,15 @@ static void _ionic_lif_rx_mode(struct io { struct ionic_deferred_work *work; - if (in_interrupt()) { - work = kzalloc(sizeof(*work), GFP_ATOMIC); - if (!work) { - netdev_err(lif->netdev, "%s OOM\n", __func__); - return; - } - work->type = IONIC_DW_TYPE_RX_MODE; - work->rx_mode = rx_mode; - netdev_dbg(lif->netdev, "deferred: rx_mode\n"); - ionic_lif_deferred_enqueue(&lif->deferred, work); - } else { - ionic_lif_rx_mode(lif, rx_mode); + work = kzalloc(sizeof(*work), GFP_ATOMIC); + if (!work) { + netdev_err(lif->netdev, "%s OOM\n", __func__); + return; } + work->type = IONIC_DW_TYPE_RX_MODE; + work->rx_mode = rx_mode; + netdev_dbg(lif->netdev, "deferred: rx_mode\n"); + ionic_lif_deferred_enqueue(&lif->deferred, work); } static void ionic_set_rx_mode(struct net_device *netdev) @@ -1312,7 +1309,7 @@ static int ionic_set_mac_address(struct eth_commit_mac_addr_change(netdev, addr); netdev_info(netdev, "updating mac addr %pM\n", mac); - return ionic_addr_add(netdev, mac); + return ionic_lif_addr(netdev_priv(netdev), mac, true, true); } static int ionic_change_mtu(struct net_device *netdev, int new_mtu) @@ -2252,7 +2249,7 @@ static void ionic_lif_handle_fw_up(struc } clear_bit(IONIC_LIF_F_FW_RESET, lif->state); - ionic_link_status_check_request(lif); + ionic_link_status_check_request(lif, true); netif_device_attach(lif->netdev); dev_info(ionic->dev, "FW Up: LIFs restarted\n"); @@ -2468,7 +2465,7 @@ static int ionic_station_set(struct ioni */ if (!ether_addr_equal(ctx.comp.lif_getattr.mac, netdev->dev_addr)) - ionic_lif_addr(lif, netdev->dev_addr, true); + ionic_lif_addr(lif, netdev->dev_addr, true, true); } else { /* Update the netdev mac with the device's mac */ memcpy(addr.sa_data, ctx.comp.lif_getattr.mac, netdev->addr_len); @@ -2485,7 +2482,7 @@ static int ionic_station_set(struct ioni netdev_dbg(lif->netdev, "adding station MAC addr %pM\n", netdev->dev_addr); - ionic_lif_addr(lif, netdev->dev_addr, true); + ionic_lif_addr(lif, netdev->dev_addr, true, true); return 0; } --- a/drivers/net/ethernet/pensando/ionic/ionic_lif.h +++ b/drivers/net/ethernet/pensando/ionic/ionic_lif.h @@ -237,7 +237,7 @@ static inline u32 ionic_coal_usec_to_hw( typedef void (*ionic_reset_cb)(struct ionic_lif *lif, void *arg); -void ionic_link_status_check_request(struct ionic_lif *lif); +void ionic_link_status_check_request(struct ionic_lif *lif, bool can_sleep); void ionic_get_stats64(struct net_device *netdev, struct rtnl_link_stats64 *ns); void ionic_lif_deferred_enqueue(struct ionic_deferred *def,