net: in_interrupt() cleanup and fixes
Subsystems (23)
cover net: in_interrupt() cleanup and fixes
Folks,
in the discussion about preempt count consistency accross kernel configurations:
https://lore.kernel.org/r/20200914204209.256266093 at linutronix.de/
Linus clearly requested that code in drivers and libraries which changes behaviour based on execution context should either be split up so that e.g. task context invocations and BH invocations have different interfaces or if that's not possible the context information has to be provided by the caller which knows in which context it is executing.
This includes conditional locking, allocation mode (GFP_*) decisions and avoidance of code paths which might sleep.
In the long run, usage of 'preemptible, in_*irq etc.' should be banned from driver code completely.
Our team started to dig through drivers and this it the first batch of cleanups in drivers/net/. It's not yet complete, so expect further patches in the next days.
The series contains:
- A couple of bug fixes
- Removal of the bitrotting CAIF SPI driver which has never had a matching driver providing the necessary platform device support.
- Removal of WARN/BUG(in_interrupt()) en masse as most of them are incomplete because they won't detect other non-preemptible context. All of the functions which have these WARN/BUG invoke core code functions which can sleep. These have plenty of checks to catch _all_ invalid contexts. So it's pointless to have incomplete WARN/BUG in the drivers.
If a driver wants to have such a check for paranoia reasons, then e.g. lockdep_assert_preemtion_enabled() is the right mechanism to chose because lockdep guarantees to catch all invalid contexts independent of kernel configuration while e.g. preemptible() does not.
- Conversion of in_interrupt() checks to use either different functions or to hand the context information in from the caller.
- For some drivers handing the context into functions which decided between netif_rx() and netif_rx_ni() turned out to be impossible due to lack of driver knowledge and convoluted code pathes with multiple indirections. For those a core code function netif_rx_any_context() is provided which contains an in_interrupt() check as a stop gap. This allows to make progess on the driver side cleanup and the function should go away once the driver wizards have fixed it up proper.
- Simplifcation and cleanups in various places where code pointlessly contains in_interrupt() conditionals which are mostly leftovers from calling conventions in older kernels and have never been cleaned up.
Along with removing if from the horrible DBG_FOO() macro mess which probably should be removed completely as the kernel today provides way more sensible mechanisms to do function tracing and similar.
- A few other cleanups which were obvious when chasing the in_interrupt() usage.
The pile is also available from:
git://git.kernel.org/pub/scm/linux/kernel/git/tglx/devel.git softirq
The diffstat summary is:
86 files changed, 300 insertions(+), 2233 deletions(-)
which is biased by the CAIF SPI removal. Without that it is:
79 files changed, 300 insertions(+), 697 deletions(-)
Thanks,
tglx
--- Documentation/networking/caif/spi_porting.rst | 229 -- b/Documentation/networking/caif/index.rst | 1 b/drivers/net/caif/Kconfig | 19 b/drivers/net/caif/Makefile | 4 b/drivers/net/caif/caif_hsi.c | 19 b/drivers/net/ethernet/amd/sun3lance.c | 11 b/drivers/net/ethernet/atheros/atl1c/atl1c_main.c | 1 b/drivers/net/ethernet/atheros/atl1e/atl1e_main.c | 2 b/drivers/net/ethernet/atheros/atlx/atl2.c | 1 b/drivers/net/ethernet/chelsio/cxgb3/adapter.h | 1 b/drivers/net/ethernet/chelsio/cxgb3/cxgb3_main.c | 2 b/drivers/net/ethernet/chelsio/cxgb3/sge.c | 44 b/drivers/net/ethernet/chelsio/cxgb4/sge.c | 3 b/drivers/net/ethernet/cisco/enic/enic.h | 1 b/drivers/net/ethernet/cisco/enic/enic_api.c | 6 b/drivers/net/ethernet/cisco/enic/enic_main.c | 27 b/drivers/net/ethernet/freescale/fec_mpc52xx.c | 10 b/drivers/net/ethernet/intel/e100.c | 4 b/drivers/net/ethernet/intel/e1000/e1000_main.c | 1 b/drivers/net/ethernet/intel/fm10k/fm10k_pci.c | 2 b/drivers/net/ethernet/intel/i40e/i40e_main.c | 4 b/drivers/net/ethernet/intel/ice/ice_main.c | 1 b/drivers/net/ethernet/intel/igb/igb_main.c | 1 b/drivers/net/ethernet/intel/igc/igc_main.c | 1 b/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c | 1 b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c | 2 b/drivers/net/ethernet/natsemi/sonic.c | 24 b/drivers/net/ethernet/natsemi/sonic.h | 2 b/drivers/net/ethernet/neterion/vxge/vxge-config.c | 9 b/drivers/net/ethernet/neterion/vxge/vxge-config.h | 7 b/drivers/net/ethernet/pensando/ionic/ionic_dev.c | 2 b/drivers/net/ethernet/pensando/ionic/ionic_lif.c | 43 b/drivers/net/ethernet/pensando/ionic/ionic_lif.h | 2 b/drivers/net/ethernet/pensando/ionic/ionic_main.c | 4 b/drivers/net/ethernet/sfc/ef10.c | 18 b/drivers/net/ethernet/sfc/ef100_nic.c | 3 b/drivers/net/ethernet/sfc/efx_common.c | 6 b/drivers/net/ethernet/sfc/ethtool_common.c | 2 b/drivers/net/ethernet/sfc/net_driver.h | 3 b/drivers/net/ethernet/sfc/siena.c | 3 b/drivers/net/ethernet/sun/sunbmac.c | 18 b/drivers/net/phy/mdio_bus.c | 15 b/drivers/net/usb/kaweth.c | 261 -- b/drivers/net/usb/net1080.c | 1 b/drivers/net/wan/lmc/lmc_debug.c | 18 b/drivers/net/wan/lmc/lmc_debug.h | 1 b/drivers/net/wan/lmc/lmc_main.c | 105 - b/drivers/net/wan/lmc/lmc_media.c | 4 b/drivers/net/wan/lmc/lmc_proto.c | 16 b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/bcmsdh.c | 4 b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/bus.h | 5 b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.c | 20 b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/fweh.c | 8 b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/fweh.h | 7 b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/msgbuf.c | 2 b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/sdio.c | 12 b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/sdio.h | 2 b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/usb.c | 2 b/drivers/net/wireless/intel/ipw2x00/ipw2100.c | 3 b/drivers/net/wireless/intel/ipw2x00/ipw2200.h | 6 b/drivers/net/wireless/intel/ipw2x00/libipw.h | 3 b/drivers/net/wireless/intel/iwlegacy/common.h | 4 b/drivers/net/wireless/intel/iwlwifi/iwl-debug.c | 5 b/drivers/net/wireless/intel/iwlwifi/iwl-devtrace-msg.h | 6 b/drivers/net/wireless/intersil/hostap/hostap_hw.c | 12 b/drivers/net/wireless/marvell/libertas/defs.h | 3 b/drivers/net/wireless/marvell/libertas/rx.c | 11 b/drivers/net/wireless/marvell/libertas_tf/deb_defs.h | 3 b/drivers/net/wireless/marvell/mwifiex/uap_txrx.c | 6 b/drivers/net/wireless/marvell/mwifiex/util.c | 6 b/drivers/net/wireless/realtek/rtlwifi/base.c | 47 b/drivers/net/wireless/realtek/rtlwifi/base.h | 3 b/drivers/net/wireless/realtek/rtlwifi/btcoexist/halbtcoutsrc.c | 12 b/drivers/net/wireless/realtek/rtlwifi/core.c | 6 b/drivers/net/wireless/realtek/rtlwifi/debug.c | 20 b/drivers/net/wireless/realtek/rtlwifi/debug.h | 6 b/drivers/net/wireless/realtek/rtlwifi/pci.c | 4 b/drivers/net/wireless/realtek/rtlwifi/ps.c | 27 b/drivers/net/wireless/realtek/rtlwifi/ps.h | 10 b/drivers/net/wireless/realtek/rtlwifi/wifi.h | 3 b/drivers/net/wireless/zydas/zd1211rw/zd_usb.c | 1 b/include/linux/netdevice.h | 1 b/net/core/dev.c | 15 drivers/net/caif/caif_spi.c | 874 ---------- drivers/net/caif/caif_spi_slave.c | 254 -- include/net/caif/caif_spi.h | 155 - 86 files changed, 300 insertions(+), 2233 deletions(-)
PATCH 1/35 net: enic: Cure the enic api locking trainwreck
From: Thomas Gleixner <[email protected]>
enic_dev_wait() has a BUG_ON(in_interrupt()).
Chasing the callers of enic_dev_wait() revealed the gems of enic_reset() and enic_tx_hang_reset() which are both invoked through work queues in order to be able to call rtnl_lock(). So far so good.
After locking rtnl both functions acquire enic::enic_api_lock which serializes against the (ab)use from infiniband. This is where the trainwreck starts.
enic::enic_api_lock is a spin_lock() which implicitly disables preemption, but both functions invoke a ton of functions under that lock which can sleep. The BUG_ON(in_interrupt()) does not trigger in that case because it can't detect the preempt disabled condition.
This clearly has never been tested with any of the mandatory debug options for 7+ years, which would have caught that for sure.
Cure it by adding a enic_api_busy member to struct enic, which is modified and evaluated with enic::enic_api_lock held.
If enic_api_devcmd_proxy_by_index() observes enic::enic_api_busy as true, it drops enic::enic_api_lock and busy waits for enic::enic_api_busy to become false.
It would be smarter to wait for a completion of that busy period, but enic_api_devcmd_proxy_by_index() is called with other spin locks held which obviously can't sleep.
Remove the BUG_ON(in_interrupt()) check as well because it's incomplete and with proper debugging enabled the problem would have been caught from the debug checks in schedule_timeout().
--- drivers/net/ethernet/cisco/enic/enic.h | 1 + drivers/net/ethernet/cisco/enic/enic_api.c | 6 ++++++ drivers/net/ethernet/cisco/enic/enic_main.c | 27 +++++++++++++++++++++------ 3 files changed, 28 insertions(+), 6 deletions(-)
drivers/net/ethernet/cisco/enic/enic.h +1 -0
@@ -169,6 +169,7 @@ struct enic {u16 num_vfs;#endif spinlock_t enic_api_lock; +bool enic_api_busy;struct enic_port_profile *pp;/* work queue cache line section */
drivers/net/ethernet/cisco/enic/enic_api.c +6 -0
@@ -34,6 +34,12 @@ int enic_api_devcmd_proxy_by_index(strucstruct vnic_dev *vdev = enic->vdev;spin_lock (&enic->enic_api_lock); +while (enic->enic_api_busy) { +spin_unlock (&enic->enic_api_lock); +cpu_relax (); +spin_lock (&enic->enic_api_lock); + } +spin_lock_bh (&enic->devcmd_lock);vnic_dev_cmd_proxy_by_index_start (vdev, vf);
drivers/net/ethernet/cisco/enic/enic_main.c +21 -6
@@ -2106,8 +2106,6 @@ static int enic_dev_wait(struct vnic_devint done;int err; -BUG_ON (in_interrupt ()); - err =start (vdev, arg);if (err)return err;
@@ -2295,6 +2293,13 @@ static int enic_set_rss_nic_cfg(struct e rss_hash_bits, rss_base_cpu, rss_enable); } +static void enic_set_api_busy (struct enic *enic,bool busy) +{ +spin_lock (&enic->enic_api_lock); + enic->enic_api_busy = busy; +spin_unlock (&enic->enic_api_lock); +} +static void enic_reset (struct work_struct *work) { struct enic *enic = container_of(work, struct enic, reset);
@@ -2304,7 +2309,9 @@ static void enic_reset(struct work_strucrtnl_lock (); -spin_lock (&enic->enic_api_lock); +/* Stop any activity from infiniband */ +enic_set_api_state (enic, true); +enic_stop (enic->netdev);enic_dev_soft_reset (enic);enic_reset_addr_lists (enic);
@@ -2312,7 +2319,10 @@ static void enic_reset(struct work_strucenic_set_rss_nic_cfg (enic);enic_dev_set_ig_vlan_rewrite_mode (enic);enic_open (enic->netdev); -spin_unlock (&enic->enic_api_lock); + +/* Allow infiniband to fiddle with the device again */ +enic_set_api_state (enic, false); +call_netdevice_notifiers (NETDEV_REBOOT , enic->netdev);rtnl_unlock ();
@@ -2324,7 +2334,9 @@ static void enic_tx_hang_reset(struct wortnl_lock (); -spin_lock (&enic->enic_api_lock); +/* Stop any activity from infiniband */ +enic_set_api_state (enic, true); +enic_dev_hang_notify (enic);enic_stop (enic->netdev);enic_dev_hang_reset (enic);
@@ -2333,7 +2345,10 @@ static void enic_tx_hang_reset(struct woenic_set_rss_nic_cfg (enic);enic_dev_set_ig_vlan_rewrite_mode (enic);enic_open (enic->netdev); -spin_unlock (&enic->enic_api_lock); + +/* Allow infiniband to fiddle with the device again */ +enic_set_api_state (enic, false); +call_netdevice_notifiers (NETDEV_REBOOT , enic->netdev);rtnl_unlock ();
kernel test robot
build botRe: [patch 01/35] net: enic: Cure the enic api locking trainwreck
I love your patch! Yet something to improve:
Hi Thomas,
I love your patch! Yet something to improve:
[auto build test ERROR on wireless-drivers/master]
[also build test ERROR on jkirsher-next-queue/dev-queue linus/master v5.9-rc6 next-20200925] [cannot apply to wireless-drivers-next/master]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch]
url: https://github.com/0day-ci/linux/commits/Thomas-Gleixner/net-in_interrupt-cleanup-and-fixes/20200928-040350 base: https://git.kernel.org/pub/scm/linux/kernel/git/kvalo/wireless-drivers.git master config: ia64-allmodconfig (attached as .config) compiler: ia64-linux-gcc (GCC) 9.3.0
reproduce (this is a W=1 build):
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross
# https://github.com/0day-ci/linux/commit/0faff202c0039d8da1baed3af43b0c1b4c67e36e
git remote add linux-review https://github.com/0day-ci/linux
git fetch --no-tags linux-review Thomas-Gleixner/net-in_interrupt-cleanup-and-fixes/20200928-040350 git checkout 0faff202c0039d8da1baed3af43b0c1b4c67e36e
# save the attached .config to linux build tree
COMPILER_INSTALL_PATH=$HOME/0day COMPILER=gcc-9.3.0 make.cross ARCH=ia64
If you fix the issue, kindly add following tag as appropriate
All errors (new ones prefixed by >>):
drivers/net/ethernet/cisco/enic/enic_main.c: In function 'enic_reset':
>> drivers/net/ethernet/cisco/enic/enic_main.c:2313:2: error: implicit declaration of function 'enic_set_api_state'; did you mean 'enic_set_api_busy'? [-Werror=implicit-function-declaration]
2313 | enic_set_api_state(enic, true);
| ^~~~~~~~~~~~~~~~~~
| enic_set_api_busy At top level:
drivers/net/ethernet/cisco/enic/enic_main.c:2296:13: warning: 'enic_set_api_busy' defined but not used [-Wunused-function]
2296 | static void enic_set_api_busy(struct enic *enic, bool busy)
| ^~~~~~~~~~~~~~~~~
cc1: some warnings being treated as errors
vim +2313 drivers/net/ethernet/cisco/enic/enic_main.c
2302
2303 static void enic_reset(struct work_struct *work)
2304 {
2305 struct enic *enic = container_of(work, struct enic, reset);
2306
2307 if (!netif_running(enic->netdev))
2308 return;
2309
2310 rtnl_lock();
2311
2312 /* Stop any activity from infiniband */
> 2313 enic_set_api_state(enic, true);
2314
2315 enic_stop(enic->netdev);
2316 enic_dev_soft_reset(enic);
2317 enic_reset_addr_lists(enic);
2318 enic_init_vnic_resources(enic);
2319 enic_set_rss_nic_cfg(enic);
2320 enic_dev_set_ig_vlan_rewrite_mode(enic);
2321 enic_open(enic->netdev);
2322
2323 /* Allow infiniband to fiddle with the device again */
2324 enic_set_api_state(enic, false);
2325
2326 call_netdevice_notifiers(NETDEV_REBOOT, enic->netdev);
2327
2328 rtnl_unlock();
2329 }
2330
---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all(a)lists.01.org
PATCH 2/35 net: caif: Remove unused caif SPI driver
In reply to Thomas GleixnerFrom: Thomas Gleixner <[email protected]>
While chasing in_interrupt() (ab)use in drivers it turned out that the caif_spi driver has never been in use since the driver was merged 10 years ago. There never was any matching code which provides a platform device.
The driver has not seen any update (asided of treewide changes and cleanups) since 8 years and the maintainers vanished from the planet.
So analysing the potential contexts and the (in)correctness of in_interrupt() usage is just a pointless exercise.
Remove the cruft.
--- Documentation/networking/caif/index.rst | 1 Documentation/networking/caif/spi_porting.rst | 229 ------ drivers/net/caif/Kconfig | 19 drivers/net/caif/Makefile | 4 drivers/net/caif/caif_spi.c | 874 -------------------------- drivers/net/caif/caif_spi_slave.c | 254 ------- include/net/caif/caif_spi.h | 155 ---- 7 files changed, 1536 deletions(-)
Documentation/networking/caif/index.rst +0 -1
@@ -10,4 +10,3 @@ CAIF linux_caif caif - spi_porting
Documentation/networking/caif/spi_porting.rst deleted +0 -229
229 lines not on this page · raw message
drivers/net/caif/Kconfig +0 -19
@@ -20,25 +20,6 @@ config CAIF_TTY identified as N_CAIF. When this ldisc is opened from user space it will redirect the TTY's traffic into the CAIF stack. -config CAIF_SPI_SLAVE - tristate "CAIF SPI transport driver for slave interface" - depends on CAIF && HAS_DMA - default n - help - The CAIF Link layer SPI Protocol driver for Slave SPI interface. - This driver implements a platform driver to accommodate for a - platform specific SPI device. A sample CAIF SPI Platform device is - provided in <file:Documentation/networking/caif/spi_porting.rst>. - -config CAIF_SPI_SYNC - bool "Next command and length in start of frame" - depends on CAIF_SPI_SLAVE - default n - help - Putting the next command and length in the start of the frame can - help to synchronize to the next transfer in case of over or under-runs. - This option also needs to be enabled on the modem. - config CAIF_HSI tristate "CAIF HSI transport driver" depends on CAIF
drivers/net/caif/Makefile +0 -4
@@ -4,10 +4,6 @@ ccflags-$(CONFIG_CAIF_DEBUG) := -DDEBUG # Serial interface obj-$(CONFIG_CAIF_TTY) += caif_serial.o -# SPI slave physical interfaces module -cfspi_slave-objs := caif_spi.o caif_spi_slave.o -obj-$(CONFIG_CAIF_SPI_SLAVE) += cfspi_slave.o - # HSI interface obj-$(CONFIG_CAIF_HSI) += caif_hsi.o
drivers/net/caif/caif_spi.c deleted +0 -874
874 lines not on this page · raw message
drivers/net/caif/caif_spi_slave.c deleted +0 -254
254 lines not on this page · raw message
include/net/caif/caif_spi.h deleted +0 -155
155 lines not on this page · raw message
PATCH 3/35 net: Add netif_rx_any_context()
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
Quite some drivers make conditional decisions based on in_interrupt() to invoke either netif_rx() or netif_rx_ni().
Conditionals based on in_interrupt() or other variants of preempt count checks in drivers should not exist for various reasons and Linus clearly requested to either split the code pathes or pass an argument to the common functions which provides the context.
This is obviously the correct solution, but for some of the affected drivers this needs a major rewrite due to their convoluted structure.
As in_interrupt() usage in drivers needs to be phased out, provide netif_rx_any_context() as a stop gap for these drivers.
This confines the in_interrupt() conditional to core code which in turn allows to remove the access to this check for driver code and provides one central place to do further modifications once the driver maze is cleaned up.
--- include/linux/netdevice.h | 1 + net/core/dev.c | 15 +++++++++++++++ 2 files changed, 16 insertions(+)
include/linux/netdevice.h +1 -0
@@ -3764,6 +3764,7 @@ void generic_xdp_tx(struct sk_buff *skb,int do_xdp_generic (struct bpf_prog *xdp_prog,struct sk_buff *skb);int netif_rx (struct sk_buff *skb);int netif_rx_ni (struct sk_buff *skb); +int netif_rx_any_context (struct sk_buff *skb);int netif_receive_skb (struct sk_buff *skb);int netif_receive_skb_core (struct sk_buff *skb);void netif_receive_skb_list (struct list_head *head);
net/core/dev.c +15 -0
@@ -4840,6 +4840,21 @@ int netif_rx_ni(struct sk_buff *skb) }EXPORT_SYMBOL (netif_rx_ni); +int netif_rx_any_context (struct sk_buff *skb) +{ +/* +* If invoked from contexts which do not invoke bottom half +* processing either at return from interrupt or when softrqs are +* reenabled, use netif_rx_ni() which invokes bottomhalf processing +* directly. +*/ +if (in_interrupt ()) +return netif_rx (skb); +else +return netif_rx_ni (skb); +} +EXPORT_SYMBOL (netif_rx_any_context); + static __latent_entropy void net_tx_action(struct softirq_action *h) {struct softnet_data *sd =this_cpu_ptr (&softnet_data);
PATCH 4/35 net: caif: Use netif_rx_any_context()
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
The usage of in_interrupt() in non-core code is phased out. Ideally the information of the calling context should be passed by the callers or the functions be split as appropriate.
cfhsi_rx_desc() and cfhsi_rx_pld() use in_interrupt() to distinguish if they should use netif_rx() or netif_rx_ni() for receiving packets.
The attempt to consolidate the code by passing an arguemnt or by distangling it failed due lack of knowledge about this driver and because the call chains are hard to follow.
As a stop gap use netif_rx_any_context() which invokes the correct code path depending on context and confines the in_interrupt() usage to core code.
--- drivers/net/caif/caif_hsi.c | 19 ++----------------- 1 file changed, 2 insertions(+), 17 deletions(-)
drivers/net/caif/caif_hsi.c +2 -17
@@ -458,15 +458,7 @@ static int cfhsi_rx_desc(struct cfhsi_deskb_reset_mac_header (skb); skb->dev = cfhsi->ndev; -/* -* We are in a callback handler and -* unfortunately we don't know what context we're -* running in. -*/ -if (in_interrupt ()) -netif_rx (skb); -else -netif_rx_ni (skb); +netif_rx_any_context (skb);/* Update network statistics. */ cfhsi->ndev->stats.rx_packets++;
@@ -587,14 +579,7 @@ static int cfhsi_rx_pld(struct cfhsi_desskb_reset_mac_header (skb); skb->dev = cfhsi->ndev; -/* -* We're called in callback from HSI -* and don't know the context we're running in. -*/ -if (in_interrupt ()) -netif_rx (skb); -else -netif_rx_ni (skb); +netif_rx_any_context (skb);/* Update network statistics. */ cfhsi->ndev->stats.rx_packets++;
PATCH 5/35 net: atheros: Remove WARN_ON(in_interrupt())
In reply to Thomas GleixnerFrom: Thomas Gleixner <[email protected]>
in_interrupt() is ill defined and does not provide what the name suggests. The usage especially in driver code is deprecated and a tree wide effort to clean up and consolidate the (ab)usage of in_interrupt() and related checks is happening.
In this case the check covers only parts of the contexts in which these functions cannot be called. It fails to detect preemption or interrupt disabled invocations.
As the functions which are invoked from at*_reinit_locked() contain a broad variety of checks (always enabled or debug option dependent) which cover all invalid conditions already, there is no point in having inconsistent warnings in those drivers.
Just remove them.
--- drivers/net/ethernet/atheros/atl1c/atl1c_main.c | 1 - drivers/net/ethernet/atheros/atl1e/atl1e_main.c | 2 -- drivers/net/ethernet/atheros/atlx/atl2.c | 1 - 3 files changed, 4 deletions(-)
drivers/net/ethernet/atheros/atl1c/atl1c_main.c +0 -1
@@ -220,7 +220,6 @@ static void atl1c_phy_config(struct timevoid atl1c_reinit_locked (struct atl1c_adapter *adapter) { -WARN_ON (in_interrupt ());atl1c_down (adapter);atl1c_up (adapter);clear_bit (__AT_RESETTING, &adapter->flags);
drivers/net/ethernet/atheros/atl1e/atl1e_main.c +0 -2
@@ -127,8 +127,6 @@ static void atl1e_phy_config(struct timevoid atl1e_reinit_locked (struct atl1e_adapter *adapter) { - -WARN_ON (in_interrupt ());while (test_and_set_bit (__AT_RESETTING, &adapter->flags))msleep (1 );atl1e_down (adapter);
drivers/net/ethernet/atheros/atlx/atl2.c +0 -1
@@ -1085,7 +1085,6 @@ static int atl2_up(struct atl2_adapter *static void atl2_reinit_locked (struct atl2_adapter *adapter) { -WARN_ON (in_interrupt ());while (test_and_set_bit (__ATL2_RESETTING, &adapter->flags))msleep (1 );atl2_down (adapter);
PATCH 6/35 net: cxgb3: Cleanup in_interrupt() usage
In reply to Thomas GleixnerFrom: Thomas Gleixner <[email protected]>
t3_sge_stop() is called from task context and from error handlers in interrupt context. It relies on in_interrupt() to differentiate the contexts.
in_interrupt() is deprecated as it is ill defined and does not provide what it suggests.
Instead of replacing it with some other construct, simply split the function into t3_sge_stop_dma(), which can be called from any context, and t3_sge_stop() which can be only called from task context.
This has the advantage that any bogus invocation of t3_sge_stop() from wrong contexts can be caught by debug kernels instead of being papered over by the conditional.
--- drivers/net/ethernet/chelsio/cxgb3/adapter.h | 1 drivers/net/ethernet/chelsio/cxgb3/cxgb3_main.c | 2 - drivers/net/ethernet/chelsio/cxgb3/sge.c | 44 ++++++++++++++---------- 3 files changed, 29 insertions(+), 18 deletions(-)
drivers/net/ethernet/chelsio/cxgb3/adapter.h +1 -0
@@ -313,6 +313,7 @@ void t3_os_link_fault(struct adapter *advoid t3_os_link_fault_handler (struct adapter *adapter,int port_id);void t3_sge_start (struct adapter *adap); +void t3_sge_stop_dma (struct adapter *adap);void t3_sge_stop (struct adapter *adap);void t3_start_sge_timers (struct adapter *adap);void t3_stop_sge_timers (struct adapter *adap);
drivers/net/ethernet/chelsio/cxgb3/cxgb3_main.c +1 -1
@@ -2996,7 +2996,7 @@ void t3_fatal_err(struct adapter *adapteunsigned int fw_status[4 ];if (adapter->flags &FULL_INIT_DONE ) { -t3_sge_stop (adapter); +t3_sge_stop_dma (adapter);t3_write_reg (adapter,A_XGM_TX_CTRL ,0 );t3_write_reg (adapter,A_XGM_RX_CTRL ,0 );t3_write_reg (adapter,XGM_REG (A_XGM_TX_CTRL ,1 ),0 );
drivers/net/ethernet/chelsio/cxgb3/sge.c +27 -17
57 lines not on this page · raw message
PATCH 7/35 net: cxbg4: Remove pointless in_interrupt() check
In reply to Thomas GleixnerFrom: Thomas Gleixner <[email protected]>
t4_sge_stop() is only ever called from task context and the in_interrupt() check is presumably a leftover from copying t3_sge_stop().
Aside of in_interrupt() being deprecated because it's not providing what it claims to provide, this check would paper over illegitimate callers.
The functions invoked from t4_sge_stop() contain already warnings to catch invocations from invalid contexts.
Remove it.
--- drivers/net/ethernet/chelsio/cxgb4/sge.c | 3 --- 1 file changed, 3 deletions(-)
drivers/net/ethernet/chelsio/cxgb4/sge.c +0 -3
@@ -4872,9 +4872,6 @@ void t4_sge_stop(struct adapter *adap)int i;struct sge *s = &adap->sge; -if (in_interrupt ())/* actions below require waiting */ -return ; -if (s->rx_timer.function)del_timer_sync (&s->rx_timer);if (s->tx_timer.function)
PATCH 8/35 net: e100: Remove in_interrupt() usage and pointless GFP_ATOMIC allocation
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
e100_hw_init() invokes e100_self_test() only if in_interrupt() returns false as e100_self_test() uses msleep() which requires sleepable task context. The in_interrupt() check is incomplete because in_interrupt() cannot catch callers from contexts which have just preemption or interrupts disabled.
e100_hw_init() is invoked from:
- e100_loopback_test() which clearly is sleepable task context as the function uses msleep() itself.
- e100_up() which clearly is sleepable task context as well because it invokes e100_alloc_cbs() abd request_irq() which both require sleepable task context due to GFP_KERNEL allocations and mutex_lock() operations.
Remove the pointless in_interrupt() check.
As a side effect of this analysis it turned out that e100_rx_alloc_list() which is only invoked from e100_loopback_test() and e100_up() pointlessly uses a GFP_ATOMIC allocation. The next invoked function e100_alloc_cbs() is using GFP_KERNEL already.
Change the allocation mode in e100_rx_alloc_list() to GFP_KERNEL as well.
--- drivers/net/ethernet/intel/e100.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/net/ethernet/intel/e100.c +2 -2
@@ -1531,7 +1531,7 @@ static int e100_hw_init(struct nic *nic)e100_hw_reset (nic);netif_err (nic, hw, nic->netdev,"e100_hw_init\n" ); -if (!in_interrupt () && (err =e100_self_test (nic))) +if ((err =e100_self_test (nic)))return err;if ((err =e100_phy_init (nic)))
@@ -2155,7 +2155,7 @@ static int e100_rx_alloc_list(struct nic nic->rx_to_use = nic->rx_to_clean =NULL ; nic->ru_running =RU_UNINITIALIZED ; -if (!(nic->rxs =kcalloc (count,sizeof (struct rx ),GFP_ATOMIC ))) +if (!(nic->rxs =kcalloc (count,sizeof (struct rx ),GFP_KERNEL )))return -ENOMEM ;for (rx = nic->rxs, i =0 ; i < count; rx++, i++) {
PATCH 9/35 net: fec_mpc52xx: Replace in_interrupt() usage
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
The usage of in_interrupt() in drivers is phased out and Linus clearly requested that code which changes behaviour depending on context should either be seperated or the context be conveyed in an argument passed by the caller, which usually knows the context.
mpc52xx_fec_stop() uses in_interrupt() to check if it is safe to sleep. All callers run in well defined contexts.
Pass an argument from the callers indicating whether it is safe to sleep.
--- drivers/net/ethernet/freescale/fec_mpc52xx.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)
drivers/net/ethernet/freescale/fec_mpc52xx.c +5 -5
@@ -74,7 +74,7 @@ struct mpc52xx_fec_priv {static irqreturn_t mpc52xx_fec_interrupt (int ,void *);static irqreturn_t mpc52xx_fec_rx_interrupt (int ,void *);static irqreturn_t mpc52xx_fec_tx_interrupt (int ,void *); -static void mpc52xx_fec_stop (struct net_device *dev); +static void mpc52xx_fec_stop (struct net_device *dev,bool may_sleep);static void mpc52xx_fec_start (struct net_device *dev);static void mpc52xx_fec_reset (struct net_device *dev);
@@ -283,7 +283,7 @@ static int mpc52xx_fec_close(struct net_netif_stop_queue (dev); -mpc52xx_fec_stop (dev); +mpc52xx_fec_stop (dev, true);mpc52xx_fec_free_rx_buffers (dev, priv->rx_dmatsk);
@@ -693,7 +693,7 @@ static void mpc52xx_fec_start(struct net* * stop all activity on fec and empty dma buffers */ -static void mpc52xx_fec_stop (struct net_device *dev) +static void mpc52xx_fec_stop (struct net_device *dev,bool may_sleep) {struct mpc52xx_fec_priv *priv =netdev_priv (dev);struct mpc52xx_fec __iomem *fec = priv->fec;
@@ -706,7 +706,7 @@ static void mpc52xx_fec_stop(struct net_bcom_disable (priv->rx_dmatsk);/* Wait for tx queue to drain, but only if we're in process context */ -if (!in_interrupt ()) { +if (may_sleep) { timeout = jiffies +msecs_to_jiffies (2000 );while (time_before (jiffies, timeout) && !bcom_queue_empty (priv->tx_dmatsk))
@@ -738,7 +738,7 @@ static void mpc52xx_fec_reset(struct netstruct mpc52xx_fec_priv *priv =netdev_priv (dev);struct mpc52xx_fec __iomem *fec = priv->fec; -mpc52xx_fec_stop (dev); +mpc52xx_fec_stop (dev, false);out_be32 (&fec->rfifo_status,in_be32 (&fec->rfifo_status));out_be32 (&fec->reset_cntrl,FEC_RESET_CNTRL_RESET_FIFO );
PATCH 10/35 net: intel: Remove in_interrupt() warnings
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
in_interrupt() is ill defined and does not provide what the name suggests. The usage especially in driver code is deprecated and a tree wide effort to clean up and consolidate the (ab)usage of in_interrupt() and related checks is happening.
In this case the checks cover only parts of the contexts in which these functions cannot be called. They fail to detect preemption or interrupt disabled invocations.
As the functions which are invoked from the various places contain already a broad variety of checks (always enabled or debug option dependent) cover all invalid conditions already, there is no point in having inconsistent warnings in those drivers.
Just remove them.
--- drivers/net/ethernet/intel/e1000/e1000_main.c | 1 - drivers/net/ethernet/intel/fm10k/fm10k_pci.c | 2 -- drivers/net/ethernet/intel/i40e/i40e_main.c | 4 ---- drivers/net/ethernet/intel/ice/ice_main.c | 1 - drivers/net/ethernet/intel/igb/igb_main.c | 1 - drivers/net/ethernet/intel/igc/igc_main.c | 1 - drivers/net/ethernet/intel/ixgbe/ixgbe_main.c | 1 - drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c | 2 -- 8 files changed, 13 deletions(-)
drivers/net/ethernet/intel/e1000/e1000_main.c +0 -1
@@ -534,7 +534,6 @@ void e1000_down(struct e1000_adapter *advoid e1000_reinit_locked (struct e1000_adapter *adapter) { -WARN_ON (in_interrupt ());while (test_and_set_bit (__E1000_RESETTING, &adapter->flags))msleep (1 );
drivers/net/ethernet/intel/fm10k/fm10k_pci.c +0 -2
@@ -221,8 +221,6 @@ static bool fm10k_prepare_for_reset(stru {struct net_device *netdev = interface->netdev; -WARN_ON (in_interrupt ()); -/* put off any impending NetWatchDogTimeout */ netif_trans_update (netdev);
drivers/net/ethernet/intel/i40e/i40e_main.c +0 -4
@@ -6689,7 +6689,6 @@ static void i40e_vsi_reinit_locked(struc {struct i40e_pf *pf = vsi->back; -WARN_ON (in_interrupt ());while (test_and_set_bit (__I40E_CONFIG_BUSY, pf->state))usleep_range (1000 ,2000 );i40e_down (vsi);
@@ -8462,9 +8461,6 @@ void i40e_do_reset(struct i40e_pf *pf, u {u32 val; -WARN_ON (in_interrupt ()); - -/* do the biggest reset indicated */ if (reset_flags &BIT_ULL (__I40E_GLOBAL_RESET_REQUESTED)) {
drivers/net/ethernet/intel/ice/ice_main.c +0 -1
@@ -486,7 +486,6 @@ static void ice_do_reset(struct ice_pf *struct ice_hw *hw = &pf->hw;dev_dbg (dev,"reset_type 0x%x requested\n" , reset_type); -WARN_ON (in_interrupt ());ice_prepare_for_reset (pf);
drivers/net/ethernet/intel/igb/igb_main.c +0 -1
@@ -2220,7 +2220,6 @@ void igb_down(struct igb_adapter *adaptevoid igb_reinit_locked (struct igb_adapter *adapter) { -WARN_ON (in_interrupt ());while (test_and_set_bit (__IGB_RESETTING, &adapter->state))usleep_range (1000 ,2000 );igb_down (adapter);
drivers/net/ethernet/intel/igc/igc_main.c +0 -1
@@ -3831,7 +3831,6 @@ void igc_down(struct igc_adapter *adaptevoid igc_reinit_locked (struct igc_adapter *adapter) { -WARN_ON (in_interrupt ());while (test_and_set_bit (__IGC_RESETTING, &adapter->state))usleep_range (1000 ,2000 );igc_down (adapter);
drivers/net/ethernet/intel/ixgbe/ixgbe_main.c +0 -1
@@ -5677,7 +5677,6 @@ static void ixgbe_up_complete(struct ixgvoid ixgbe_reinit_locked (struct ixgbe_adapter *adapter) { -WARN_ON (in_interrupt ());/* put off any impending NetWatchDogTimeout */ netif_trans_update (adapter->netdev);
drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c +0 -2
@@ -2526,8 +2526,6 @@ void ixgbevf_down(struct ixgbevf_adaptervoid ixgbevf_reinit_locked (struct ixgbevf_adapter *adapter) { -WARN_ON (in_interrupt ()); -while (test_and_set_bit (__IXGBEVF_RESETTING, &adapter->state))msleep (1 );
Re: [patch 10/35] net: intel: Remove in_interrupt() warnings
Quoted 17 lines · Thomas Gleixner
From: Sebastian Andrzej Siewior <[email protected]>
in_interrupt()is ill defined and does not provide what the name suggests. The usage especially in driver code is deprecated and a tree wide effort to clean up and consolidate the (ab)usage ofin_interrupt()and related checks is happening.In this case the checks cover only parts of the contexts in which these functions cannot be called. They fail to detect preemption or interrupt disabled invocations.
As the functions which are invoked from the various places contain already a broad variety of checks (always enabled or debug option dependent) cover all invalid conditions already, there is no point in having inconsistent warnings in those drivers.
Quoted 8 lines
Just remove them.
Signed-off-by Sebastian Andrzej Siewior <[email protected]>Signed-off-by Thomas Gleixner <[email protected]>Cc Jeff Kirsher <[email protected]>Cc "David S. Miller" <[email protected]>Cc Jakub Kicinski <[email protected]>Cc netdev at vger.kernel.org
The patch looks good to me.
Quoted 104 lines
--- drivers/net/ethernet/intel/e1000/e1000_main.c | 1 - drivers/net/ethernet/intel/fm10k/fm10k_pci.c | 2 -- drivers/net/ethernet/intel/i40e/i40e_main.c | 4 ---- drivers/net/ethernet/intel/ice/ice_main.c | 1 - drivers/net/ethernet/intel/igb/igb_main.c | 1 - drivers/net/ethernet/intel/igc/igc_main.c | 1 - drivers/net/ethernet/intel/ixgbe/ixgbe_main.c | 1 - drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c | 2 -- 8 files changed, 13 deletions(-)drivers/net/ethernet/intel/e1000/e1000_main.c +0
-1@@ -534,7 +534,6 @@ void e1000_down(struct e1000_adapter *advoid e1000_reinit_locked (struct e1000_adapter *adapter) { -WARN_ON (in_interrupt ());while (test_and_set_bit (__E1000_RESETTING, &adapter->flags))msleep (1 );drivers/net/ethernet/intel/fm10k/fm10k_pci.c +0
-2@@ -221,8 +221,6 @@ static bool fm10k_prepare_for_reset(stru {struct net_device *netdev = interface->netdev; -WARN_ON (in_interrupt ()); -/* put off any impending NetWatchDogTimeout */ netif_trans_update (netdev);drivers/net/ethernet/intel/i40e/i40e_main.c +0
-4@@ -6689,7 +6689,6 @@ static void i40e_vsi_reinit_locked(struc {struct i40e_pf *pf = vsi->back; -WARN_ON (in_interrupt ());while (test_and_set_bit (__I40E_CONFIG_BUSY, pf->state))usleep_range (1000 ,2000 );i40e_down (vsi);@@ -8462,9 +8461,6 @@ void i40e_do_reset(struct i40e_pf *pf, u {u32 val; -WARN_ON (in_interrupt ()); - -/* do the biggest reset indicated */ if (reset_flags &BIT_ULL (__I40E_GLOBAL_RESET_REQUESTED)) {drivers/net/ethernet/intel/ice/ice_main.c +0
-1@@ -486,7 +486,6 @@ static void ice_do_reset(struct ice_pf *struct ice_hw *hw = &pf->hw;dev_dbg (dev,"reset_type 0x%x requested\n" , reset_type); -WARN_ON (in_interrupt ());ice_prepare_for_reset (pf);drivers/net/ethernet/intel/igb/igb_main.c +0
-1@@ -2220,7 +2220,6 @@ void igb_down(struct igb_adapter *adaptevoid igb_reinit_locked (struct igb_adapter *adapter) { -WARN_ON (in_interrupt ());while (test_and_set_bit (__IGB_RESETTING, &adapter->state))usleep_range (1000 ,2000 );igb_down (adapter);drivers/net/ethernet/intel/igc/igc_main.c +0
-1@@ -3831,7 +3831,6 @@ void igc_down(struct igc_adapter *adaptevoid igc_reinit_locked (struct igc_adapter *adapter) { -WARN_ON (in_interrupt ());while (test_and_set_bit (__IGC_RESETTING, &adapter->state))usleep_range (1000 ,2000 );igc_down (adapter);drivers/net/ethernet/intel/ixgbe/ixgbe_main.c +0
-1@@ -5677,7 +5677,6 @@ static void ixgbe_up_complete(struct ixgvoid ixgbe_reinit_locked (struct ixgbe_adapter *adapter) { -WARN_ON (in_interrupt ());/* put off any impending NetWatchDogTimeout */ netif_trans_update (adapter->netdev);drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c +0
-2@@ -2526,8 +2526,6 @@ void ixgbevf_down(struct ixgbevf_adaptervoid ixgbevf_reinit_locked (struct ixgbevf_adapter *adapter) { -WARN_ON (in_interrupt ()); -while (test_and_set_bit (__IXGBEVF_RESETTING, &adapter->state))msleep (1 );
PATCH 11/35 net: ionic: Replace in_interrupt() usage.
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
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.
---
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
- 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.
- 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.
--- drivers/net/ethernet/pensando/ionic/ionic_dev.c | 2 - drivers/net/ethernet/pensando/ionic/ionic_lif.c | 43 +++++++++++------------- drivers/net/ethernet/pensando/ionic/ionic_lif.h | 2 - 3 files changed, 22 insertions(+), 25 deletions(-)
drivers/net/ethernet/pensando/ionic/ionic_dev.c +1 -1
@@ -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)
drivers/net/ethernet/pensando/ionic/ionic_lif.c +20 -23
114 lines not on this page · raw message
drivers/net/ethernet/pensando/ionic/ionic_lif.h +1 -1
@@ -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,
Re: [patch 11/35] net: ionic: Replace in_interrupt() usage.
Quoted 29 lines · Thomas Gleixner
From: Sebastian Andrzej Siewior <[email protected]>
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.
Quoted 7 lines
Signed-off-by Sebastian Andrzej Siewior <[email protected]>Signed-off-by Thomas Gleixner <[email protected]>Cc Shannon Nelson <[email protected]>Cc Pensando Drivers <[email protected]>Cc "David S. Miller" <[email protected]>Cc Jakub Kicinski <[email protected]>Cc netdev at vger.kernel.org
Quoted 7 lines
---
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.
Quoted 5 lines
-
ionic_link_status_check_request()could haveionic_deferred_workwithinionic_lif(). This would avoid memory allocation from NAPI. More important, onceIONIC_LIF_F_LINK_CHECK_REQUESTEDis set and that alloc fails, the link check never happens.
Thanks, I'll fix up that error condition.
Quoted 6 lines
-
ionic_lif_handle_fw_down()setsIONIC_LIF_F_FW_RESET. Invokes thenionic_lif_deinit()which only invokescancel_work_sync()ifIONIC_LIF_F_FW_RESETis not set. I think the logic is wrong here as the work must always be cancled. Also the list withionic_deferredwork items needs a clean up.
I'll look at that, thanks.
sln
Re: [patch 11/35] net: ionic: Replace in_interrupt() usage.
Quoted 16 lines · Shannon Nelson
Quoted 15 lines · Thomas Gleixner
From: Sebastian Andrzej Siewior <[email protected]>
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:
Quoted 8 lines
Quoted 8 lines
? 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
Quoted 55 lines
Quoted 10 lines
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.Quoted 7 lines
Signed-off-by Sebastian Andrzej Siewior <[email protected]>Signed-off-by Thomas Gleixner <[email protected]>Cc Shannon Nelson <[email protected]>Cc Pensando Drivers <[email protected]>Cc "David S. Miller" <[email protected]>Cc Jakub Kicinski <[email protected]>Cc netdev at vger.kernel.orgAcked-by Shannon Nelson <[email protected]>Quoted 8 lines
---
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 iterationI 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.
Quoted 7 lines
-
ionic_link_status_check_request()could haveionic_deferred_workwithin
??ionic_lif(). This would avoid memory allocation from NAPI. More ?? important, onceIONIC_LIF_F_LINK_CHECK_REQUESTEDis set and that alloc
?? fails, the link check never happens.Thanks, I'll fix up that error condition.
Quoted 6 lines
-
ionic_lif_handle_fw_down()setsIONIC_LIF_F_FW_RESET. Invokes then ??ionic_lif_deinit()which only invokescancel_work_sync()if ??IONIC_LIF_F_FW_RESETis not set. I think the logic is wrong here as ?? the work must always be cancled. Also the list withionic_deferred?? work items needs a clean up.I'll look at that, thanks.
sln
Re: [patch 11/35] net: ionic: Replace in_interrupt() usage.
Quoted 15 lines · Shannon Nelson
Quoted 10 lines · Shannon Nelson
Quoted 10 lines
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():
Quoted 8 lines
.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...
PATCH 12/35 net: ionic: Remove WARN_ON(in_interrupt()).
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
in_interrupt() is ill defined and does not provide what the name suggests. The usage especially in driver code is deprecated and a tree wide effort to clean up and consolidate the (ab)usage of in_interrupt() and related checks is happening.
In this case the check covers only parts of the contexts in which these functions cannot be called. It fails to detect preemption or interrupt disabled invocations.
As the functions which are invoked from ionic_adminq_post() and ionic_dev_cmd_wait() contain a broad variety of checks (always enabled or debug option dependent) which cover all invalid conditions already, there is no point in having inconsistent warnings in those drivers.
Just remove them.
--- drivers/net/ethernet/pensando/ionic/ionic_main.c | 4 ---- 1 file changed, 4 deletions(-)
drivers/net/ethernet/pensando/ionic/ionic_main.c +0 -4
@@ -248,8 +248,6 @@ static int ionic_adminq_post(struct ionistruct ionic_queue *adminq;int err =0 ; -WARN_ON (in_interrupt ()); -if (!lif->adminqcq)return -EIO ;
@@ -346,8 +344,6 @@ int ionic_dev_cmd_wait(struct ionic *ionint done;int err; -WARN_ON (in_interrupt ()); -/* Wait for dev cmd to complete, retrying if we get EAGAIN, * but don't wait any longer than max_seconds. */
Re: [patch 12/35] net: ionic: Remove WARN_ON(in_interrupt()).
Quoted 18 lines · Thomas Gleixner
From: Sebastian Andrzej Siewior <[email protected]>
in_interrupt()is ill defined and does not provide what the name suggests. The usage especially in driver code is deprecated and a tree wide effort to clean up and consolidate the (ab)usage ofin_interrupt()and related checks is happening.In this case the check covers only parts of the contexts in which these functions cannot be called. It fails to detect preemption or interrupt disabled invocations.
As the functions which are invoked from
ionic_adminq_post()andionic_dev_cmd_wait()contain a broad variety of checks (always enabled or debug option dependent) which cover all invalid conditions already, there is no point in having inconsistent warnings in those drivers.Just remove them.
Quoted 7 lines
Signed-off-by Sebastian Andrzej Siewior <[email protected]>Signed-off-by Thomas Gleixner <[email protected]>Cc Shannon Nelson <[email protected]>Cc Pensando Drivers <[email protected]>Cc "David S. Miller" <[email protected]>Cc Jakub Kicinski <[email protected]>Cc netdev at vger.kernel.org
Thanks.
Quoted 25 lines
--- drivers/net/ethernet/pensando/ionic/ionic_main.c | 4 ---- 1 file changed, 4 deletions(-)drivers/net/ethernet/pensando/ionic/ionic_main.c +0
-4@@ -248,8 +248,6 @@ static int ionic_adminq_post(struct ionistruct ionic_queue *adminq;int err =0 ; -WARN_ON (in_interrupt ()); -if (!lif->adminqcq)return -EIO ;@@ -346,8 +344,6 @@ int ionic_dev_cmd_wait(struct ionic *ionint done;int err; -WARN_ON (in_interrupt ()); -/* Wait for dev cmd to complete, retrying if we get EAGAIN, * but don't wait any longer than max_seconds. */
PATCH 13/35 net: mdiobus: Remove WARN_ON_ONCE(in_interrupt())
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
in_interrupt() is ill defined and does not provide what the name suggests. The usage especially in driver code is deprecated and a tree wide effort to clean up and consolidate the (ab)usage of in_interrupt() and related checks is happening.
In this case the check covers only parts of the contexts in which these functions cannot be called. It fails to detect preemption or interrupt disabled invocations.
As the functions which contain these warnings invoke mutex_lock() which contains a broad variety of checks (always enabled or debug option dependent) and therefore covers all invalid conditions already, there is no point in having inconsistent warnings in those drivers. The conditional return is not really valuable in practice either.
Just remove them.
--- drivers/net/phy/mdio_bus.c | 15 --------------- 1 file changed, 15 deletions(-)
drivers/net/phy/mdio_bus.c +0 -15
@@ -825,9 +825,6 @@ int mdiobus_read_nested(struct mii_bus * {int retval; -if (WARN_ON_ONCE (in_interrupt ())) -return -EINVAL ; -mutex_lock_nested (&bus->mdio_lock,MDIO_MUTEX_NESTED ); retval =__mdiobus_read (bus, addr, regnum);mutex_unlock (&bus->mdio_lock);
@@ -850,9 +847,6 @@ int mdiobus_read(struct mii_bus *bus, in {int retval; -if (WARN_ON_ONCE (in_interrupt ())) -return -EINVAL ; -mutex_lock (&bus->mdio_lock); retval =__mdiobus_read (bus, addr, regnum);mutex_unlock (&bus->mdio_lock);
@@ -879,9 +873,6 @@ int mdiobus_write_nested(struct mii_bus {int err; -if (WARN_ON_ONCE (in_interrupt ())) -return -EINVAL ; -mutex_lock_nested (&bus->mdio_lock,MDIO_MUTEX_NESTED ); err =__mdiobus_write (bus, addr, regnum, val);mutex_unlock (&bus->mdio_lock);
@@ -905,9 +896,6 @@ int mdiobus_write(struct mii_bus *bus, i {int err; -if (WARN_ON_ONCE (in_interrupt ())) -return -EINVAL ; -mutex_lock (&bus->mdio_lock); err =__mdiobus_write (bus, addr, regnum, val);mutex_unlock (&bus->mdio_lock);
@@ -929,9 +917,6 @@ int mdiobus_modify(struct mii_bus *bus, {int err; -if (WARN_ON_ONCE (in_interrupt ())) -return -EINVAL ; -mutex_lock (&bus->mdio_lock); err =__mdiobus_modify_changed (bus, addr, regnum, mask, set);mutex_unlock (&bus->mdio_lock);
Re: [patch 13/35] net: mdiobus: Remove WARN_ON_ONCE(in_interrupt())
Quoted 17 lines · Thomas Gleixner
From: Sebastian Andrzej Siewior <[email protected]>
in_interrupt()is ill defined and does not provide what the name suggests. The usage especially in driver code is deprecated and a tree wide effort to clean up and consolidate the (ab)usage ofin_interrupt()and related checks is happening.In this case the check covers only parts of the contexts in which these functions cannot be called. It fails to detect preemption or interrupt disabled invocations.
As the functions which contain these warnings invoke
mutex_lock()which contains a broad variety of checks (always enabled or debug option dependent) and therefore covers all invalid conditions already, there is no point in having inconsistent warnings in those drivers. The conditional return is not really valuable in practice either.
Quoted 4 lines
Just remove them.
Signed-off-by Sebastian Andrzej Siewior <[email protected]>Signed-off-by Thomas Gleixner <[email protected]>
Andrew
PATCH 14/35 net: natsemi: Replace in_interrupt() usage.
In reply to Thomas GleixnerThe usage of in_interrupt() in drivers is phased out and Linus clearly requested that code which changes behaviour depending on context should either be seperated or the context be conveyed in an argument passed by the caller, which usually knows the context.
sonic_quiesce() uses 'in_interrupt() || irqs_disabled()' to chose either udelay() or usleep_range() in the wait loop.
In all callchains leading to it the context is well defined and known.
Add a 'may_sleep' argument and pass it through the various callchains leading to this function.
--- drivers/net/ethernet/natsemi/sonic.c | 24 ++++++++++++------------ drivers/net/ethernet/natsemi/sonic.h | 2 +- 2 files changed, 13 insertions(+), 13 deletions(-)
drivers/net/ethernet/natsemi/sonic.c +12 -12
@@ -143,7 +143,7 @@ static int sonic_open(struct net_device/* * Initialize the SONIC */ -sonic_init (dev); +sonic_init (dev, true);netif_start_queue (dev);
@@ -153,7 +153,7 @@ static int sonic_open(struct net_device }/* Wait for the SONIC to become idle. */ -static void sonic_quiesce (struct net_device *dev,u16 mask) +static void sonic_quiesce (struct net_device *dev,u16 mask,bool may_sleep) {struct sonic_local *__maybe_unused lp =netdev_priv (dev);int i;
@@ -163,7 +163,7 @@ static void sonic_quiesce(struct net_dev bits =SONIC_READ (SONIC_CMD ) & mask;if (!bits)return ; -if (irqs_disabled () ||in_interrupt ()) +if (!may_sleep)udelay (20 );else usleep_range (100 ,200 );
@@ -187,7 +187,7 @@ static int sonic_close(struct net_device* stop the SONIC, disable interrupts */ SONIC_WRITE (SONIC_CMD ,SONIC_CR_RXDIS ); -sonic_quiesce (dev,SONIC_CR_ALL ); +sonic_quiesce (dev,SONIC_CR_ALL , true);SONIC_WRITE (SONIC_IMR ,0 );SONIC_WRITE (SONIC_ISR ,0x7fff );
@@ -229,7 +229,7 @@ static void sonic_tx_timeout(struct net_* disable all interrupts before releasing DMA buffers */ SONIC_WRITE (SONIC_CMD ,SONIC_CR_RXDIS ); -sonic_quiesce (dev,SONIC_CR_ALL ); +sonic_quiesce (dev,SONIC_CR_ALL , false);SONIC_WRITE (SONIC_IMR ,0 );SONIC_WRITE (SONIC_ISR ,0x7fff );
@@ -246,7 +246,7 @@ static void sonic_tx_timeout(struct net_ } }/* Try to restart the adaptor. */ -sonic_init (dev); +sonic_init (dev, false); lp->stats.tx_errors++;netif_trans_update (dev);/* prevent tx timeout */ netif_wake_queue (dev);
@@ -692,9 +692,9 @@ static void sonic_multicast_list(struct/* LCAM and TXP commands can't be used simultaneously */ spin_lock_irqsave (&lp->lock, flags); -sonic_quiesce (dev,SONIC_CR_TXP ); +sonic_quiesce (dev,SONIC_CR_TXP , false);SONIC_WRITE (SONIC_CMD ,SONIC_CR_LCAM ); -sonic_quiesce (dev,SONIC_CR_LCAM ); +sonic_quiesce (dev,SONIC_CR_LCAM , false);spin_unlock_irqrestore (&lp->lock, flags); } }
@@ -708,7 +708,7 @@ static void sonic_multicast_list(struct/* * Initialize the SONIC ethernet controller. */ -static int sonic_init (struct net_device *dev) +static int sonic_init (struct net_device *dev,bool may_sleep) {struct sonic_local *lp =netdev_priv (dev);int i;
@@ -730,7 +730,7 @@ static int sonic_init(struct net_device*/ SONIC_WRITE (SONIC_CMD ,0 );SONIC_WRITE (SONIC_CMD ,SONIC_CR_RXDIS |SONIC_CR_STP ); -sonic_quiesce (dev,SONIC_CR_ALL ); +sonic_quiesce (dev,SONIC_CR_ALL , may_sleep);/* * initialize the receive resource area
@@ -759,7 +759,7 @@ static int sonic_init(struct net_devicenetif_dbg (lp, ifup, dev,"%s: issuing RRRA command\n" , __func__);SONIC_WRITE (SONIC_CMD ,SONIC_CR_RRRA ); -sonic_quiesce (dev,SONIC_CR_RRRA ); +sonic_quiesce (dev,SONIC_CR_RRRA , may_sleep);/* * Initialize the receive descriptors so that they
@@ -834,7 +834,7 @@ static int sonic_init(struct net_device* load the CAM */ SONIC_WRITE (SONIC_CMD ,SONIC_CR_LCAM ); -sonic_quiesce (dev,SONIC_CR_LCAM ); +sonic_quiesce (dev,SONIC_CR_LCAM , may_sleep);/* * enable receiver, disable loopback
drivers/net/ethernet/natsemi/sonic.h +1 -1
@@ -338,7 +338,7 @@ static void sonic_rx(struct net_device *static int sonic_close (struct net_device *dev);static struct net_device_stats *sonic_get_stats (struct net_device *dev);static void sonic_multicast_list (struct net_device *dev); -static int sonic_init (struct net_device *dev); +static int sonic_init (struct net_device *dev,bool may_sleep);static void sonic_tx_timeout (struct net_device *dev,unsigned int txqueue);static void sonic_msg_init (struct net_device *dev);static int sonic_alloc_descriptors (struct net_device *dev);
PATCH 15/35 net: sfc: Replace in_interrupt() usage.
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
efx_ef10_try_update_nic_stats_vf() uses in_interrupt() to figure out whether it is safe to sleep or not.
The following callers are involved:
- efx_start_all() and efx_stop_all() are fully preemptible because a mutex is acquired near by.
- efx_ethtool_get_stats() is ivoked from ethtool_ops->get_ethtool_stats() and is fully preemptible.
- efx_net_stats() which can be invoked under dev_base_lock from net-sysfs::netstat_show(). dev_base_lock is a rwlock_t which disables preemption implicitly.
in_interrupt() cannot detect context which has only preemption disabled so the check fails to detect that this calling context is not safe to sleep.
Obviously this is a bug and clearly this has never been tested with any of the relevant and mandatory debug options enabled, which would have caught it.
Changing the condition to preemptible() is not useful either because on CONFIG_PREEMPT_COUNT=n kernels preemptible() is useless.
Aside of that Linus clearly requested that functions which change their behaviour depending on execution context should either be split up or the callers provide context information via an argument.
Add a 'can_sleep' argument to efx_ef10_try_update_nic_stats_vf() and let the callers indicate the context from which this is called.
Another oddity of that code is that it uses GFP_ATOMIC _after_ establishing that the context is safe to sleep.
Convert it to GFP_KERNEL while at it.
Note, that the fixes tag is empty as it is unclear which of the commits to blame.
--- drivers/net/ethernet/sfc/ef10.c | 18 +++++++++--------- drivers/net/ethernet/sfc/ef100_nic.c | 3 ++- drivers/net/ethernet/sfc/efx_common.c | 6 +++--- drivers/net/ethernet/sfc/ethtool_common.c | 2 +- drivers/net/ethernet/sfc/net_driver.h | 3 ++- drivers/net/ethernet/sfc/siena.c | 3 ++- 6 files changed, 19 insertions(+), 16 deletions(-)
drivers/net/ethernet/sfc/ef10.c +9 -9
@@ -1797,7 +1797,8 @@ static size_t efx_ef10_update_stats_comm }static size_t efx_ef10_update_stats_pf (struct efx_nic *efx,u64 *full_stats, -struct rtnl_link_stats64 *core_stats) +struct rtnl_link_stats64 *core_stats, +bool can_sleep) {struct efx_ef10_nic_data *nic_data = efx->nic_data;DECLARE_BITMAP (mask,EF10_STAT_COUNT );
@@ -1836,7 +1837,7 @@ static size_t efx_ef10_update_stats_pf(sreturn efx_ef10_update_stats_common (efx, full_stats, core_stats); } -static int efx_ef10_try_update_nic_stats_vf (struct efx_nic *efx) +static int efx_ef10_try_update_nic_stats_vf (struct efx_nic *efx,bool can_sleep) __must_hold(&efx->stats_lock) {MCDI_DECLARE_BUF (inbuf,MC_CMD_MAC_STATS_IN_LEN );
@@ -1849,20 +1850,18 @@ static int efx_ef10_try_update_nic_stats__le64 *dma_stats;int rc; -spin_unlock_bh (&efx->stats_lock); - -if (in_interrupt ()) { +if (!can_sleep) {/* If in atomic context, cannot update stats. Just update the * software stats and return so the caller can continue. */ -spin_lock_bh (&efx->stats_lock);efx_update_sw_stats (efx, stats);return 0 ; } +spin_unlock_bh (&efx->stats_lock);efx_ef10_get_stat_mask (efx, mask); - rc =efx_nic_alloc_buffer (efx, &stats_buf, dma_len,GFP_ATOMIC ); + rc =efx_nic_alloc_buffer (efx, &stats_buf, dma_len,GFP_KERNEL );if (rc) {spin_lock_bh (&efx->stats_lock);return rc;
@@ -1910,9 +1909,10 @@ static int efx_ef10_try_update_nic_stats }static size_t efx_ef10_update_stats_vf (struct efx_nic *efx,u64 *full_stats, -struct rtnl_link_stats64 *core_stats) +struct rtnl_link_stats64 *core_stats, +bool can_sleep) { -if (efx_ef10_try_update_nic_stats_vf (efx)) +if (efx_ef10_try_update_nic_stats_vf (efx, can_sleep))return 0 ;return efx_ef10_update_stats_common (efx, full_stats, core_stats);
drivers/net/ethernet/sfc/ef100_nic.c +2 -1
@@ -599,7 +599,8 @@ static size_t ef100_update_stats_common(static size_t ef100_update_stats (struct efx_nic *efx,u64 *full_stats, -struct rtnl_link_stats64 *core_stats) +struct rtnl_link_stats64 *core_stats, +bool can_sleep) {__le64 *mc_stats =kmalloc (array_size (efx->num_mac_stats,sizeof (__le64)),GFP_ATOMIC );struct ef100_nic_data *nic_data = efx->nic_data;
drivers/net/ethernet/sfc/efx_common.c +3 -3
@@ -552,7 +552,7 @@ void efx_start_all(struct efx_nic *efx) efx->type->start_stats (efx); efx->type->pull_stats (efx);spin_lock_bh (&efx->stats_lock); - efx->type->update_stats (efx,NULL ,NULL ); + efx->type->update_stats (efx,NULL ,NULL , true);spin_unlock_bh (&efx->stats_lock); } }
@@ -576,7 +576,7 @@ void efx_stop_all(struct efx_nic *efx)*/ efx->type->pull_stats (efx);spin_lock_bh (&efx->stats_lock); - efx->type->update_stats (efx,NULL ,NULL ); + efx->type->update_stats (efx,NULL ,NULL , true);spin_unlock_bh (&efx->stats_lock); efx->type->stop_stats (efx); }
@@ -600,7 +600,7 @@ void efx_net_stats(struct net_device *nestruct efx_nic *efx =netdev_priv (net_dev);spin_lock_bh (&efx->stats_lock); - efx->type->update_stats (efx,NULL , stats); + efx->type->update_stats (efx,NULL , stats, false);spin_unlock_bh (&efx->stats_lock); }
drivers/net/ethernet/sfc/ethtool_common.c +1 -1
@@ -502,7 +502,7 @@ void efx_ethtool_get_stats(struct net_despin_lock_bh (&efx->stats_lock);/* Get NIC statistics */ - data += efx->type->update_stats (efx, data,NULL ); + data += efx->type->update_stats (efx, data,NULL , true);/* Get software statistics */ for (i =0 ; i <EFX_ETHTOOL_SW_STAT_COUNT ; i++) {
drivers/net/ethernet/sfc/net_driver.h +2 -1
@@ -1358,7 +1358,8 @@ struct efx_nic_type {void (*finish_flr)(struct efx_nic *efx);size_t (*describe_stats)(struct efx_nic *efx,u8 *names);size_t (*update_stats)(struct efx_nic *efx,u64 *full_stats, -struct rtnl_link_stats64 *core_stats); +struct rtnl_link_stats64 *core_stats, +bool can_sleep);void (*start_stats)(struct efx_nic *efx);void (*pull_stats)(struct efx_nic *efx);void (*stop_stats)(struct efx_nic *efx);
drivers/net/ethernet/sfc/siena.c +2 -1
@@ -587,7 +587,8 @@ static int siena_try_update_nic_stats(st }static size_t siena_update_nic_stats (struct efx_nic *efx,u64 *full_stats, -struct rtnl_link_stats64 *core_stats) +struct rtnl_link_stats64 *core_stats, +bool can_sleep) {struct siena_nic_data *nic_data = efx->nic_data;u64 *stats = nic_data->stats;
Re: [patch 15/35] net: sfc: Replace in_interrupt() usage.
Quoted 2 lines · Thomas Gleixner
Note, that the fixes tag is empty as it is unclear which of the commits to blame.
Seems like it should be
?since that adds the in_interrupt() check and the code concerned ?doesn't seemto have changed a great deal since.
Anyway, this fix looks correct, and you can have my
?but I thinkit might be cleaner to avoid having to have this unused ?can_sleep argument on all the NICs that don't need it, by instead ?adding an update_stats_atomic() member to struct efx_nic_type, which ?could be set to the same as update_stats() for everything except ?EF10 VFs which would just do the call to efx_update_sw_stats(). I'll send an rfc patch embodying the above shortly...
-ed
RFC PATCH 1/1 sfc: replace in_interrupt() usage
efx_ef10_try_update_nic_stats_vf() used in_interrupt() to figure out
whether it is safe to sleep (for MCDI) or not.
The only caller from which it was not is efx_net_stats(), which can be
invoked under dev_base_lock from net-sysfs::netstat_show().
So add a new update_stats_atomic() method to struct efx_nic_type, and
call it from efx_net_stats(), removing the need for efx_ef10_try_update_nic_stats_vf() to behave differently for this case (which it wasn't doing correctly anyway).
For all nic_types other than EF10 VF, this method is NULL and so we
call the regular update_stats() methods, which are happy with being called from atomic contexts.
---
Only compile-tested so far, because I'm waiting for my kernel to
finish rebuilding with CONFIG_DEBUG_ATOMIC_SLEEP which I'm hoping is the right thing to detect the bug in the existing code.
I also wasn't quite sure how to give credit to the thorough analysis
in the commit message of Sebastian's patch. I don't think we have a Whatever-by: tag to cover that, do we?
And this doesn't include your GFP_KERNEL change, which should
probably go in separately if you take this.
drivers/net/ethernet/sfc/ef10.c | 22 +++++++++++++--------- drivers/net/ethernet/sfc/efx_common.c | 2 +- drivers/net/ethernet/sfc/net_driver.h | 5 +++++ drivers/net/ethernet/sfc/nic_common.h | 7 +++++++ 4 files changed, 26 insertions(+), 10 deletions(-)
drivers/net/ethernet/sfc/ef10.c +13 -9
@@ -1871,15 +1871,6 @@ static int efx_ef10_try_update_nic_stats_vf(struct efx_nic *efx)spin_unlock_bh (&efx->stats_lock); -if (in_interrupt ()) { -/* If in atomic context, cannot update stats. Just update the -* software stats and return so the caller can continue. -*/ -spin_lock_bh (&efx->stats_lock); -efx_update_sw_stats (efx, stats); -return 0 ; - } -efx_ef10_get_stat_mask (efx, mask); rc =efx_nic_alloc_buffer (efx, &stats_buf, dma_len,GFP_ATOMIC );
@@ -1938,6 +1929,18 @@ static size_t efx_ef10_update_stats_vf(struct efx_nic *efx, u64 *full_stats,return efx_ef10_update_stats_common (efx, full_stats, core_stats); } +static size_t efx_ef10_update_stats_atomic_vf (struct efx_nic *efx,u64 *full_stats, +struct rtnl_link_stats64 *core_stats) +{ +struct efx_ef10_nic_data *nic_data = efx->nic_data; + +/* In atomic context, cannot update HW stats. Just update the +* software stats and return so the caller can continue. +*/ +efx_update_sw_stats (efx, nic_data->stats); +return efx_ef10_update_stats_common (efx, full_stats, core_stats); +} +static void efx_ef10_push_irq_moderation (struct efx_channel *channel) {struct efx_nic *efx = channel->efx;
@@ -3998,6 +4001,7 @@ const struct efx_nic_type efx_hunt_a0_vf_nic_type = { .finish_flr = efx_port_dummy_op_void, .describe_stats = efx_ef10_describe_stats, .update_stats = efx_ef10_update_stats_vf, + .update_stats_atomic = efx_ef10_update_stats_atomic_vf, .start_stats = efx_port_dummy_op_void, .pull_stats = efx_port_dummy_op_void, .stop_stats = efx_port_dummy_op_void,
drivers/net/ethernet/sfc/efx_common.c +1 -1
@@ -602,7 +602,7 @@ void efx_net_stats(struct net_device *net_dev, struct rtnl_link_stats64 *stats)struct efx_nic *efx =netdev_priv (net_dev);spin_lock_bh (&efx->stats_lock); - efx->type->update_stats (efx,NULL , stats); +efx_nic_update_stats_atomic (efx,NULL , stats);spin_unlock_bh (&efx->stats_lock); }
drivers/net/ethernet/sfc/net_driver.h +5 -0
@@ -1172,6 +1172,9 @@ struct efx_udp_tunnel {* @describe_stats: Describe statistics for ethtool * @update_stats: Update statistics not provided by event handling. * Either argument may be %NULL. +* @update_stats_atomic: Update statistics while in atomic context, if that +* is more limiting than @update_stats. Otherwise, leave %NULL and +* driver core will call @update_stats. * @start_stats: Start the regular fetching of statistics * @pull_stats: Pull stats from the NIC and wait until they arrive. * @stop_stats: Stop the regular fetching of statistics
@@ -1316,6 +1319,8 @@ struct efx_nic_type {size_t (*describe_stats)(struct efx_nic *efx,u8 *names);size_t (*update_stats)(struct efx_nic *efx,u64 *full_stats,struct rtnl_link_stats64 *core_stats); +size_t (*update_stats_atomic)(struct efx_nic *efx,u64 *full_stats, +struct rtnl_link_stats64 *core_stats);void (*start_stats)(struct efx_nic *efx);void (*pull_stats)(struct efx_nic *efx);void (*stop_stats)(struct efx_nic *efx);
drivers/net/ethernet/sfc/nic_common.h +7 -0
@@ -244,6 +244,13 @@ void efx_nic_update_stats(const struct efx_hw_stat_desc *desc, size_t count, const unsigned long *mask, u64 *stats, const void *dma_buf, bool accumulate);void efx_nic_fix_nodesc_drop_stat (struct efx_nic *efx,u64 *stat); +static inline size_t efx_nic_update_stats_atomic (struct efx_nic *efx,u64 *full_stats, +struct rtnl_link_stats64 *core_stats) +{ +if (efx->type->update_stats_atomic) +return efx->type->update_stats_atomic (efx, full_stats, core_stats); +return efx->type->update_stats (efx, full_stats, core_stats); +}#define EFX_MAX_FLUSH_TIME 5000
Re: [RFC PATCH net-next] sfc: replace in_interrupt() usage
Quoted 12 lines · Edward Cree
efx_ef10_try_update_nic_stats_vf()usedin_interrupt()to figure out
whether it is safe to sleep (for MCDI) or not.
The only caller from which it was not isefx_net_stats(), which can be
invoked underdev_base_lockfrom net-sysfs::netstat_show().
So add a newupdate_stats_atomic()method to structefx_nic_type, and
call it fromefx_net_stats(), removing the need forefx_ef10_try_update_nic_stats_vf()to behave differently for this case (which it wasn't doing correctly anyway).
For allnic_typesother than EF10 VF, this method is NULL and so we
call the regularupdate_stats()methods, which are happy with being called from atomic contexts.
Quoted 3 lines
Fixes f00bf2305cab ("sfc: don't update stats on VF when called in atomic context")Reported-by Sebastian Andrzej Siewior <[email protected]>Signed-off-by Edward Cree <[email protected]>
That's much nicer.
Quoted 7 lines
---
Only compile-tested so far, because I'm waiting for my kernel to
finish rebuilding withCONFIG_DEBUG_ATOMIC_SLEEPwhich I'm hoping is the right thing to detect the bug in the existing code.
I also wasn't quite sure how to give credit to the thorough analysis
in the commit message of Sebastian's patch. I don't think we have a Whatever-by: tag to cover that, do we?
Sebastian did the analysis and I did some word polishing, but the credit surely goes to him.
Thanks,
tglx
Re: [RFC PATCH net-next] sfc: replace in_interrupt() usage
Quoted 3 lines · Edward Cree
Quoted 2 lines
Only compile-tested so far, because I'm waiting for my kernel to
finish rebuilding withCONFIG_DEBUG_ATOMIC_SLEEP
I've now tested and confirmed that the might_sleep warning goes
away with this patch.
Thomas, do you want to pull it into v2 of your series, or should
I submit it separately to David?
-ed
Re: [RFC PATCH net-next] sfc: replace in_interrupt() usage
Quoted 9 lines · Edward Cree
Quoted 3 lines · Edward Cree
Quoted 2 lines
Only compile-tested so far, because I'm waiting for my kernel to
finish rebuilding withCONFIG_DEBUG_ATOMIC_SLEEPI've now tested and confirmed that the
might_sleepwarning goes
away with this patch.Thomas, do you want to pull it into v2 of your series, or should
I submit it separately to David?
I have it already, but if Dave applies it right away, that's fine.
Thanks,
tglx
Re: [RFC PATCH net-next] sfc: replace in_interrupt() usage
In reply to Edward CreeQuoted 12 lines · Edward Cree
efx_ef10_try_update_nic_stats_vf()usedin_interrupt()to figure out
whether it is safe to sleep (for MCDI) or not.
The only caller from which it was not isefx_net_stats(), which can be
invoked underdev_base_lockfrom net-sysfs::netstat_show().
So add a newupdate_stats_atomic()method to structefx_nic_type, and
call it fromefx_net_stats(), removing the need forefx_ef10_try_update_nic_stats_vf()to behave differently for this case (which it wasn't doing correctly anyway).
For allnic_typesother than EF10 VF, this method is NULL and so we
call the regularupdate_stats()methods, which are happy with being called from atomic contexts.
Quoted 3 lines
Fixes f00bf2305cab ("sfc: don't update stats on VF when called in atomic context")Reported-by Sebastian Andrzej Siewior <[email protected]>Signed-off-by Edward Cree <[email protected]>
Quoted 117 lines
---
Only compile-tested so far, because I'm waiting for my kernel to
finish rebuilding withCONFIG_DEBUG_ATOMIC_SLEEPwhich I'm hoping is the right thing to detect the bug in the existing code.
I also wasn't quite sure how to give credit to the thorough analysis
in the commit message of Sebastian's patch. I don't think we have a Whatever-by: tag to cover that, do we?
And this doesn't include yourGFP_KERNELchange, which should
probably go in separately if you take this.drivers/net/ethernet/sfc/ef10.c | 22 +++++++++++++--------- drivers/net/ethernet/sfc/efx_common.c | 2 +- drivers/net/ethernet/sfc/net_driver.h | 5 +++++ drivers/net/ethernet/sfc/nic_common.h | 7 +++++++ 4 files changed, 26 insertions(+), 10 deletions(-)drivers/net/ethernet/sfc/ef10.c +13
-9@@ -1871,15 +1871,6 @@ static int efx_ef10_try_update_nic_stats_vf(struct efx_nic *efx)spin_unlock_bh (&efx->stats_lock); -if (in_interrupt ()) { -/* If in atomic context, cannot update stats. Just update the -* software stats and return so the caller can continue. -*/ -spin_lock_bh (&efx->stats_lock); -efx_update_sw_stats (efx, stats); -return 0 ; - } -efx_ef10_get_stat_mask (efx, mask); rc =efx_nic_alloc_buffer (efx, &stats_buf, dma_len,GFP_ATOMIC );@@ -1938,6 +1929,18 @@ static size_t efx_ef10_update_stats_vf(struct efx_nic *efx, u64 *full_stats,return efx_ef10_update_stats_common (efx, full_stats, core_stats); } +static size_t efx_ef10_update_stats_atomic_vf (struct efx_nic *efx,u64 *full_stats, +struct rtnl_link_stats64 *core_stats) +{ +struct efx_ef10_nic_data *nic_data = efx->nic_data; + +/* In atomic context, cannot update HW stats. Just update the +* software stats and return so the caller can continue. +*/ +efx_update_sw_stats (efx, nic_data->stats); +return efx_ef10_update_stats_common (efx, full_stats, core_stats); +} +static void efx_ef10_push_irq_moderation (struct efx_channel *channel) {struct efx_nic *efx = channel->efx;@@ -3998,6 +4001,7 @@ const struct efx_nic_type efx_hunt_a0_vf_nic_type = { .finish_flr = efx_port_dummy_op_void, .describe_stats = efx_ef10_describe_stats, .update_stats = efx_ef10_update_stats_vf, + .update_stats_atomic = efx_ef10_update_stats_atomic_vf, .start_stats = efx_port_dummy_op_void, .pull_stats = efx_port_dummy_op_void, .stop_stats = efx_port_dummy_op_void,drivers/net/ethernet/sfc/efx_common.c +1
-1@@ -602,7 +602,7 @@ void efx_net_stats(struct net_device *net_dev, struct rtnl_link_stats64 *stats)struct efx_nic *efx =netdev_priv (net_dev);spin_lock_bh (&efx->stats_lock); - efx->type->update_stats (efx,NULL , stats); +efx_nic_update_stats_atomic (efx,NULL , stats);spin_unlock_bh (&efx->stats_lock); }drivers/net/ethernet/sfc/net_driver.h +5
-0@@ -1172,6 +1172,9 @@ struct efx_udp_tunnel {* @describe_stats: Describe statistics for ethtool * @update_stats: Update statistics not provided by event handling. * Either argument may be %NULL. +* @update_stats_atomic: Update statistics while in atomic context, if that +* is more limiting than @update_stats. Otherwise, leave %NULL and +* driver core will call @update_stats. * @start_stats: Start the regular fetching of statistics * @pull_stats: Pull stats from the NIC and wait until they arrive. * @stop_stats: Stop the regular fetching of statistics @@ -1316,6 +1319,8 @@ struct efx_nic_type {size_t (*describe_stats)(struct efx_nic *efx,u8 *names);size_t (*update_stats)(struct efx_nic *efx,u64 *full_stats,struct rtnl_link_stats64 *core_stats); +size_t (*update_stats_atomic)(struct efx_nic *efx,u64 *full_stats, +struct rtnl_link_stats64 *core_stats);void (*start_stats)(struct efx_nic *efx);void (*pull_stats)(struct efx_nic *efx);void (*stop_stats)(struct efx_nic *efx);drivers/net/ethernet/sfc/nic_common.h +7
-0@@ -244,6 +244,13 @@ void efx_nic_update_stats(const struct efx_hw_stat_desc *desc, size_t count, const unsigned long *mask, u64 *stats, const void *dma_buf, bool accumulate);void efx_nic_fix_nodesc_drop_stat (struct efx_nic *efx,u64 *stat); +static inline size_t efx_nic_update_stats_atomic (struct efx_nic *efx,u64 *full_stats, +struct rtnl_link_stats64 *core_stats) +{ +if (efx->type->update_stats_atomic) +return efx->type->update_stats_atomic (efx, full_stats, core_stats); +return efx->type->update_stats (efx, full_stats, core_stats); +}#define EFX_MAX_FLUSH_TIME 5000
Signature · 1 line
Martin Habets <[email protected]>
PATCH 16/35 net: sunbmac: Replace in_interrupt() usage
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
bigmac_init_rings() has an argument signaling if it is called from the interrupt handler. This is used to decide between GFP_KERNEL and GFP_ATOMIC for memory allocations.
But it also checks in_interrupt() to handle invocations which come from the timer callback bigmac_timer() via bigmac_hw_init(), which is invoked with 'in_irq = 0'. While the timer callback is clearly not in hard interrupt context it is still not sleepable context.
Rename the argument to `non_blocking' and set it to true if invoked from the timer callback or the interrupt handler which allows to remove the in_interrupt() check and makes the code consistent.
--- drivers/net/ethernet/sun/sunbmac.c | 18 +++++++++--------- 1 file changed, 9 insertions(+), 9 deletions(-)
drivers/net/ethernet/sun/sunbmac.c +9 -9
@@ -209,13 +209,13 @@ static void bigmac_clean_rings(struct bi } } -static void bigmac_init_rings (struct bigmac *bp,int from_irq) +static void bigmac_init_rings (struct bigmac *bp,bool non_blocking) {struct bmac_init_block *bb = bp->bmac_block;int i;gfp_t gfp_flags =GFP_KERNEL ; -if (from_irq ||in_interrupt ()) +if (non_blocking) gfp_flags =GFP_ATOMIC ; bp->rx_new = bp->rx_old = bp->tx_new = bp->tx_old =0 ;
@@ -489,7 +489,7 @@ static void bigmac_tcvr_init(struct bigm } } -static int bigmac_init_hw (struct bigmac *,int ); +static int bigmac_init_hw (struct bigmac *,bool );static int try_next_permutation (struct bigmac *bp,void __iomem *tregs) {
@@ -549,7 +549,7 @@ static void bigmac_timer(struct timer_liif (ret ==-1 ) {printk (KERN_ERR "%s: Link down, cable problem?\n" , bp->dev->name); - ret =bigmac_init_hw (bp,0 ); + ret =bigmac_init_hw (bp, true);if (ret) {printk (KERN_ERR "%s: Error, cannot re-init the " "BigMAC.\n" , bp->dev->name);
@@ -617,7 +617,7 @@ static void bigmac_begin_auto_negotiatioadd_timer (&bp->bigmac_timer); } -static int bigmac_init_hw (struct bigmac *bp,int from_irq) +static int bigmac_init_hw (struct bigmac *bp,bool non_blocking) {void __iomem *gregs = bp->gregs;void __iomem *cregs = bp->creg;
@@ -635,7 +635,7 @@ static int bigmac_init_hw(struct bigmacqec_init (bp);/* Alloc and reset the tx/rx descriptor chains. */ -bigmac_init_rings (bp, from_irq); +bigmac_init_rings (bp, non_blocking);/* Initialize the PHY. */ bigmac_tcvr_init (bp);
@@ -749,7 +749,7 @@ static void bigmac_is_medium_rare(struct }printk (" RESET\n" ); -bigmac_init_hw (bp,1 ); +bigmac_init_hw (bp, true); }/* BigMAC transmit complete service routines. */
@@ -921,7 +921,7 @@ static int bigmac_open(struct net_devicereturn ret; }timer_setup (&bp->bigmac_timer, bigmac_timer,0 ); - ret =bigmac_init_hw (bp,0 ); + ret =bigmac_init_hw (bp, false);if (ret)free_irq (dev->irq, bp);return ret;
@@ -945,7 +945,7 @@ static void bigmac_tx_timeout(struct net {struct bigmac *bp =netdev_priv (dev); -bigmac_init_hw (bp,0 ); +bigmac_init_hw (bp, true);netif_wake_queue (dev); }
PATCH 17/35 net: sun3lance: Remove redundant checks in interrupt handler
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
lance_interrupt() contains two pointless checks:
- A check whether the 'dev_id' argument is NULL. 'dev_id' is the pointer which was handed in to request_irq() and the interrupt handler will always be invoked with that pointer as 'dev_id' argument by the core code.
- A check for interrupt reentrancy. The core code already guarantees non-reentrancy of interrupt handlers.
Remove these check.
--- drivers/net/ethernet/amd/sun3lance.c | 11 ----------- 1 file changed, 11 deletions(-)
drivers/net/ethernet/amd/sun3lance.c +0 -11
@@ -657,16 +657,6 @@ static irqreturn_t lance_interrupt( intstruct net_device *dev = dev_id;struct lance_private *lp =netdev_priv (dev);int csr0; -static int in_interrupt; - -if (dev ==NULL ) { -DPRINTK (1 , ("lance_interrupt(): invalid dev_id\n" )); -return IRQ_NONE ; - } - -if (in_interrupt) -DPRINTK (2 , ("%s: Re-entering the interrupt handler.\n" , dev->name )); - in_interrupt =1 ; still_more:flush_cache_all ();
@@ -774,7 +764,6 @@ static irqreturn_t lance_interrupt( intDPRINTK (2 , ("%s: exiting interrupt, csr0=%#04x.\n" , dev->name,DREG )); - in_interrupt =0 ;return IRQ_HANDLED ; }
PATCH 18/35 net: vxge: Remove in_interrupt() conditionals
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
vxge_os_dma_malloc() and vxge_os_dma_malloc_async() are both called from callchains which use GFP_KERNEL allocations unconditionally or have other requirements to be called from fully preemptible task context..
vxge_os_dma_malloc():
1) __vxge_hw_blockpool_create() <- GFP_KERNEL
2) __vxge_hw_mempool_grow() <- vzalloc()
__vxge_hw_blockpool_malloc()
vxge_os_dma_malloc_async():
1 __vxge_hw_mempool_grow() <- vzalloc()
__vxge_hw_blockpool_malloc()
__vxge_hw_blockpool_blocks_add()
2) vxge_hw_vpath_open() <- vzalloc()
__vxge_hw_blockpool_block_allocate()
That means neither of these functions needs a conditional allocation mode.
Remove the in_interrupt() conditional and use GFP_KERNEL.
--- drivers/net/ethernet/neterion/vxge/vxge-config.c | 9 +-------- drivers/net/ethernet/neterion/vxge/vxge-config.h | 7 +------ 2 files changed, 2 insertions(+), 14 deletions(-)
drivers/net/ethernet/neterion/vxge/vxge-config.c +1 -8
@@ -2303,16 +2303,9 @@ static void vxge_hw_blockpool_block_add(static inline void vxge_os_dma_malloc_async (struct pci_dev *pdev,void *devh,unsigned long size) { -gfp_t flags;void *vaddr; -if (in_interrupt ()) - flags =GFP_ATOMIC |GFP_DMA ; -else - flags =GFP_KERNEL |GFP_DMA ; - - vaddr =kmalloc ((size), flags); - + vaddr =kmalloc (size,GFP_KERNEL |GFP_DMA );vxge_hw_blockpool_block_add (devh, vaddr, size, pdev, pdev); }
drivers/net/ethernet/neterion/vxge/vxge-config.h +1 -6
@@ -1899,18 +1899,13 @@ static inline void *vxge_os_dma_malloc(sstruct pci_dev **p_dmah, struct pci_dev **p_dma_acch) { - gfp_t flags; void *vaddr;unsigned long misaligned =0 ;int realloc_flag =0 ; *p_dma_acch = *p_dmah =NULL ; -if (in_interrupt ()) - flags =GFP_ATOMIC |GFP_DMA ; -else - flags =GFP_KERNEL |GFP_DMA ; realloc: - vaddr =kmalloc ((size), flags); + vaddr =kmalloc (size,GFP_KERNEL |GFP_DMA );if (vaddr ==NULL )return vaddr; misaligned = (unsigned long)VXGE_ALIGN((unsigned long)vaddr,
PATCH 19/35 net: zd1211rw: Remove ZD_ASSERT(in_interrupt())
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
in_interrupt() is ill defined and does not provide what the name suggests. The usage especially in driver code is deprecated and a tree wide effort to clean up and consolidate the (ab)usage of in_interrupt() and related checks is happening.
handle_regs_int() is always invoked as part of URB callback which is either invoked from hard or soft interrupt context.
Remove the magic assertion.
--- drivers/net/wireless/zydas/zd1211rw/zd_usb.c | 1 - 1 file changed, 1 deletion(-)
drivers/net/wireless/zydas/zd1211rw/zd_usb.c +0 -1
@@ -378,7 +378,6 @@ static inline void handle_regs_int(strucint len;u16 int_num; -ZD_ASSERT (in_interrupt ());spin_lock_irqsave (&intr->lock, flags); int_num =le16_to_cpu (*(__le16 *)(urb->transfer_buffer+2 ));
PATCH 20/35 net: usb: kaweth: Replace kaweth_control() with usb_control_msg()
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
kaweth_control() is almost the same as usb_control_msg() except for the memory allocation mode (GFP_ATOMIC vs GFP_NOIO) and the in_interrupt() check.
All the invocations of kaweth_control() are within the probe function in fully preemtible context so there is no reason to use atomic allocations, GFP_NOIO which is used by usb_control_msg() is perfectly fine.
Replace kaweth_control() invocations from probe with usb_control_msg().
--- drivers/net/usb/kaweth.c | 93 +++++++++++++++-------------------------------- 1 file changed, 30 insertions(+), 63 deletions(-)
drivers/net/usb/kaweth.c +30 -63
@@ -282,19 +282,13 @@ static int kaweth_control(struct kaweth_****************************************************************/ static int kaweth_read_configuration (struct kaweth_device *kaweth) { -int retval; - - retval =kaweth_control (kaweth, -usb_rcvctrlpipe (kaweth->dev,0 ), +return usb_control_msg (kaweth->dev,usb_rcvctrlpipe (kaweth->dev,0 ),KAWETH_COMMAND_GET_ETHERNET_DESC ,USB_TYPE_VENDOR |USB_DIR_IN |USB_RECIP_DEVICE , -0 , -0 , - (void *)&kaweth->configuration, +0 ,0 , + &kaweth->configuration,sizeof (kaweth->configuration),KAWETH_CONTROL_TIMEOUT ); - -return retval; }/****************************************************************
@@ -302,21 +296,14 @@ static int kaweth_read_configuration(str****************************************************************/ static int kaweth_set_urb_size (struct kaweth_device *kaweth,__u16 urb_size) { -int retval; -netdev_dbg (kaweth->net,"Setting URB size to %d\n" , (unsigned )urb_size); - retval =kaweth_control (kaweth, -usb_sndctrlpipe (kaweth->dev,0 ), -KAWETH_COMMAND_SET_URB_SIZE , -USB_TYPE_VENDOR |USB_DIR_OUT |USB_RECIP_DEVICE , - urb_size, -0 , - (void *)&kaweth->scratch, -0 , -KAWETH_CONTROL_TIMEOUT ); - -return retval; +return usb_control_msg (kaweth->dev,usb_sndctrlpipe (kaweth->dev,0 ), +KAWETH_COMMAND_SET_URB_SIZE , +USB_TYPE_VENDOR |USB_DIR_OUT |USB_RECIP_DEVICE , + urb_size,0 , + &kaweth->scratch,0 , +KAWETH_CONTROL_TIMEOUT ); }/****************************************************************
@@ -324,21 +311,14 @@ static int kaweth_set_urb_size(struct ka****************************************************************/ static int kaweth_set_sofs_wait (struct kaweth_device *kaweth,__u16 sofs_wait) { -int retval; -netdev_dbg (kaweth->net,"Set SOFS wait to %d\n" , (unsigned )sofs_wait); - retval =kaweth_control (kaweth, -usb_sndctrlpipe (kaweth->dev,0 ), -KAWETH_COMMAND_SET_SOFS_WAIT , -USB_TYPE_VENDOR |USB_DIR_OUT |USB_RECIP_DEVICE , - sofs_wait, -0 , - (void *)&kaweth->scratch, -0 , -KAWETH_CONTROL_TIMEOUT ); - -return retval; +return usb_control_msg (kaweth->dev,usb_sndctrlpipe (kaweth->dev,0 ), +KAWETH_COMMAND_SET_SOFS_WAIT , +USB_TYPE_VENDOR |USB_DIR_OUT |USB_RECIP_DEVICE , + sofs_wait,0 , + &kaweth->scratch,0 , +KAWETH_CONTROL_TIMEOUT ); }/****************************************************************
@@ -347,22 +327,15 @@ static int kaweth_set_sofs_wait(struct kstatic int kaweth_set_receive_filter (struct kaweth_device *kaweth,__u16 receive_filter) { -int retval; -netdev_dbg (kaweth->net,"Set receive filter to %d\n" , (unsigned )receive_filter); - retval =kaweth_control (kaweth, -usb_sndctrlpipe (kaweth->dev,0 ), -KAWETH_COMMAND_SET_PACKET_FILTER , -USB_TYPE_VENDOR |USB_DIR_OUT |USB_RECIP_DEVICE , - receive_filter, -0 , - (void *)&kaweth->scratch, -0 , -KAWETH_CONTROL_TIMEOUT ); - -return retval; +return usb_control_msg (kaweth->dev,usb_sndctrlpipe (kaweth->dev,0 ), +KAWETH_COMMAND_SET_PACKET_FILTER , +USB_TYPE_VENDOR |USB_DIR_OUT |USB_RECIP_DEVICE , + receive_filter,0 , + &kaweth->scratch,0 , +KAWETH_CONTROL_TIMEOUT ); }/****************************************************************
@@ -407,14 +380,11 @@ static int kaweth_download_firmware(stru kaweth->firmware_buf, kaweth);netdev_dbg (kaweth->net,"Firmware length: %d\n" , data_len); -return kaweth_control (kaweth, -usb_sndctrlpipe (kaweth->dev,0 ), +return usb_control_msg (kaweth->dev,usb_sndctrlpipe (kaweth->dev,0 ),KAWETH_COMMAND_SCAN ,USB_TYPE_VENDOR |USB_DIR_OUT |USB_RECIP_DEVICE , -0 , -0 , - (void *)kaweth->firmware_buf, - data_len, +0 ,0 , + kaweth->firmware_buf, data_len,KAWETH_CONTROL_TIMEOUT ); }
@@ -433,15 +403,12 @@ static int kaweth_trigger_firmware(struc kaweth->firmware_buf[6 ] =0x00 ; kaweth->firmware_buf[7 ] =0x00 ; -return kaweth_control (kaweth, -usb_sndctrlpipe (kaweth->dev,0 ), -KAWETH_COMMAND_SCAN , -USB_TYPE_VENDOR |USB_DIR_OUT |USB_RECIP_DEVICE , -0 , -0 , - (void *)kaweth->firmware_buf, -8 , -KAWETH_CONTROL_TIMEOUT ); +return usb_control_msg (kaweth->dev,usb_sndctrlpipe (kaweth->dev,0 ), +KAWETH_COMMAND_SCAN , +USB_TYPE_VENDOR |USB_DIR_OUT |USB_RECIP_DEVICE , +0 ,0 , + (void *)kaweth->firmware_buf,8 , +KAWETH_CONTROL_TIMEOUT ); }/****************************************************************
PATCH 21/35 net: usb: kaweth: Remove last user of kaweth_control()
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
kaweth_async_set_rx_mode() invokes kaweth_contol() and has two callers:
- kaweth_open() which is invoked from preemptible context
.
- kaweth_start_xmit() which holds a spinlock and has bottom halfs disabled.
If called from kaweth_start_xmit() kaweth_async_set_rx_mode() obviously cannot block, which means it can't call kaweth_control(). This is detected with an in_interrupt() check.
Replace the in_interrupt() check in kaweth_async_set_rx_mode() with an argument which is set true by the caller if the context is safe to sleep, otherwise false.
Now kaweth_control() is only called from preemptible context which means there is no need for GFP_ATOMIC allocations anymore. Replace it with usb_control_msg(). Cleanup the code a bit while at it.
Finally remove kaweth_control() since the last user is gone.
--- drivers/net/usb/kaweth.c | 168 ++++------------------------------------------- 1 file changed, 17 insertions(+), 151 deletions(-)
drivers/net/usb/kaweth.c +17 -151
@@ -103,10 +103,6 @@ static int kaweth_probe(const struct usb_device_id *id/* from id_table */ );static void kaweth_disconnect (struct usb_interface *intf); -static int kaweth_internal_control_msg (struct usb_device *usb_dev, -unsigned int pipe, -struct usb_ctrlrequest *cmd,void *data, -int len,int timeout);static int kaweth_suspend (struct usb_interface *intf,pm_message_t message);static int kaweth_resume (struct usb_interface *intf);
@@ -236,48 +232,6 @@ struct kaweth_device };/**************************************************************** -* kaweth_control -****************************************************************/ -static int kaweth_control (struct kaweth_device *kaweth, -unsigned int pipe, -__u8 request, -__u8 requesttype, -__u16 value, -__u16 index, -void *data, -__u16 size, -int timeout) -{ -struct usb_ctrlrequest *dr; -int retval; - -if (in_interrupt ()) { -netdev_dbg (kaweth->net,"in_interrupt()\n" ); -return -EBUSY ; - } - - dr =kmalloc (sizeof (struct usb_ctrlrequest ),GFP_ATOMIC ); -if (!dr) -return -ENOMEM ; - - dr->bRequestType = requesttype; - dr->bRequest = request; - dr->wValue =cpu_to_le16 (value); - dr->wIndex =cpu_to_le16 (index); - dr->wLength =cpu_to_le16 (size); - - retval =kaweth_internal_control_msg (kaweth->dev, - pipe, - dr, - data, - size, - timeout); - -kfree (dr); -return retval; -} - -/**************************************************************** * kaweth_read_configuration ****************************************************************/ static int kaweth_read_configuration (struct kaweth_device *kaweth)
@@ -531,7 +485,8 @@ static int kaweth_resubmit_rx_urb(structreturn result; } -static void kaweth_async_set_rx_mode (struct kaweth_device *kaweth); +static void kaweth_async_set_rx_mode (struct kaweth_device *kaweth, +bool may_sleep);/**************************************************************** * kaweth_usb_receive
@@ -661,7 +616,7 @@ static int kaweth_open(struct net_devicenetif_start_queue (net); -kaweth_async_set_rx_mode (kaweth); +kaweth_async_set_rx_mode (kaweth, true);return 0 ; err_out:
@@ -749,7 +704,7 @@ static netdev_tx_t kaweth_start_xmit(strspin_lock_irq (&kaweth->device_lock); -kaweth_async_set_rx_mode (kaweth); +kaweth_async_set_rx_mode (kaweth, false);netif_stop_queue (net);if (IS_BLOCKED (kaweth->status)) { goto skip;
@@ -826,36 +781,31 @@ static void kaweth_set_rx_mode(struct ne/**************************************************************** * kaweth_async_set_rx_mode ****************************************************************/ -static void kaweth_async_set_rx_mode (struct kaweth_device *kaweth) +static void kaweth_async_set_rx_mode (struct kaweth_device *kaweth, +bool may_sleep) { -int result; +int ret;__u16 packet_filter_bitmap = kaweth->packet_filter_bitmap; kaweth->packet_filter_bitmap =0 ;if (packet_filter_bitmap ==0 )return ; -if (in_interrupt ()) +if (!may_sleep)return ; - result =kaweth_control (kaweth, -usb_sndctrlpipe (kaweth->dev,0 ), -KAWETH_COMMAND_SET_PACKET_FILTER , -USB_TYPE_VENDOR |USB_DIR_OUT |USB_RECIP_DEVICE , - packet_filter_bitmap, -0 , - (void *)&kaweth->scratch, -0 , -KAWETH_CONTROL_TIMEOUT ); - -if (result <0 ) { + ret =usb_control_msg (kaweth->dev,usb_sndctrlpipe (kaweth->dev,0 ), +KAWETH_COMMAND_SET_PACKET_FILTER , +USB_TYPE_VENDOR |USB_DIR_OUT |USB_RECIP_DEVICE , + packet_filter_bitmap,0 , + &kaweth->scratch,0 , +KAWETH_CONTROL_TIMEOUT ); +if (ret <0 )dev_err (&kaweth->intf->dev,"Failed to set Rx mode: %d\n" , - result); - } -else { + ret); +else netdev_dbg (kaweth->net,"Set Rx mode to %d\n" , packet_filter_bitmap); - } }/****************************************************************
@@ -1163,88 +1113,4 @@ static void kaweth_disconnect(struct usb } -// FIXME this completion stuff is a modified clone of -// an OLD version of some stuff in usb.c ... -struct usb_api_data { -wait_queue_head_t wqh; -int done; -}; - -/*-------------------------------------------------------------------* -* completion handler for compatibility wrappers (sync control/bulk) * -*-------------------------------------------------------------------*/ -static void usb_api_blocking_completion (struct urb *urb) -{ -struct usb_api_data *awd = (struct usb_api_data *)urb->context; - - awd->done=1 ; -wake_up (&awd->wqh); -} - -/*-------------------------------------------------------------------* -* COMPATIBILITY STUFF * -*-------------------------------------------------------------------*/ - -// Starts urb and waits for completion or timeout -static int usb_start_wait_urb (struct urb *urb,int timeout,int * actual_length) -{ -struct usb_api_data awd; -int status; - -init_waitqueue_head (&awd.wqh); - awd.done =0 ; - - urb->context = &awd; - status =usb_submit_urb (urb,GFP_ATOMIC ); -if (status) { -// something went wrong -usb_free_urb (urb); -return status; - } - -if (!wait_event_timeout (awd.wqh, awd.done, timeout)) { -// timeout -dev_warn (&urb->dev->dev,"usb_control/bulk_msg: timeout\n" ); -usb_kill_urb (urb);// remove urb safely - status = -ETIMEDOUT ; - } -else { - status = urb->status; - } - -if (actual_length) { - *actual_length = urb->actual_length; - } - -usb_free_urb (urb); -return status; -} - -/*-------------------------------------------------------------------*/ -// returns status (negative) or length (positive) -static int kaweth_internal_control_msg (struct usb_device *usb_dev, -unsigned int pipe, -struct usb_ctrlrequest *cmd,void *data, -int len,int timeout) -{ -struct urb *urb; -int retv; -int length =0 ;/* shut up GCC */ - - urb =usb_alloc_urb (0 ,GFP_ATOMIC ); -if (!urb) -return -ENOMEM ; - -usb_fill_control_urb (urb, usb_dev, pipe, (unsigned char *)cmd, data, - len, usb_api_blocking_completion,NULL ); - - retv =usb_start_wait_urb (urb, timeout, &length); -if (retv <0 ) { -return retv; - } -else { -return length; - } -} -module_usb_driver (kaweth_driver);
Re: [patch 21/35] net: usb: kaweth: Remove last user of kaweth_control()
Quoted 22 lines · Thomas Gleixner
From: Sebastian Andrzej Siewior <[email protected]>
kaweth_async_set_rx_mode()invokeskaweth_contol()and has two callers:-
kaweth_open()which is invoked from preemptible context
.
-kaweth_start_xmit()which holds a spinlock and has bottom halfs disabled.If called from
kaweth_start_xmit()kaweth_async_set_rx_mode()obviously cannot block, which means it can't callkaweth_control(). This is detected with anin_interrupt()check.Replace the
in_interrupt()check inkaweth_async_set_rx_mode()with an argument which is set true by the caller if the context is safe to sleep, otherwise false.Now
kaweth_control()is only called from preemptible context which means there is no need forGFP_ATOMICallocations anymore. Replace it withusb_control_msg(). Cleanup the code a bit while at it.Finally remove
kaweth_control()since the last user is gone.
Quoted 7 lines
Signed-off-by Sebastian Andrzej Siewior <[email protected]>Signed-off-by Thomas Gleixner <[email protected]>Cc "David S. Miller" <[email protected]>Cc Jakub Kicinski <[email protected]>Cc Greg Kroah-Hartman <[email protected]>
Thanks for the cleanup, that driver really needed it!
PATCH 22/35 net: usb: net1080: Remove in_interrupt() comment
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
The comment above nc_vendor_write() suggests that the function could become async so that is usable in `in_interrupt()' context or that it already is safe to be called from such a context.
Eitherway: The function did not become async since v2.4.9.2 (2002) and it must be not be called from `in_interrupt()' context because it sleeps on mutltiple occations.
Remove the misleading comment.
--- drivers/net/usb/net1080.c | 1 - 1 file changed, 1 deletion(-)
drivers/net/usb/net1080.c +0 -1
@@ -113,7 +113,6 @@ nc_register_read(struct usbnet *dev, u8return nc_vendor_read (dev,REQUEST_REGISTER , regnum, retval_ptr); } -// no retval ... can become async, usable in_interrupt() static void nc_vendor_write (struct usbnet *dev,u8 req,u8 regnum,u16 value) {
PATCH 23/35 net: wan/lmc: Remove lmc_trace()
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
lmc_trace() was first introduced in commit e7a392d5158af ("Import 2.3.99pre6-5") and was not touched ever since.
The reason for looking at this was to get rid of the in_interrupt() usage, but while looking at it the following observations were made:
- At least lmc_get_stats() (->ndo_get_stats()) is invoked with disabled preemption which is not detected by the in_interrupt() check, which would cause schedule() to be called from invalid context.
- The code is hidden behind #ifdef LMC_TRACE which is not defined within the kernel and wasn't at the time it was introduced.
- Three jiffies don't match 50ms. msleep() would be a better match which would also avoid the schedule() invocation. But why have it to begin with?
- Nobody would do something like this today. Either netdev_dbg() or trace_printk() or a trace event would be used. If only the functions related to this driver are interesting then ftrace can be used with filtering.
As it is obviously broken for years, simply remove it.
--- drivers/net/wan/lmc/lmc_debug.c | 18 ------ drivers/net/wan/lmc/lmc_debug.h | 1 drivers/net/wan/lmc/lmc_main.c | 105 +++------------------------------------- drivers/net/wan/lmc/lmc_media.c | 4 - drivers/net/wan/lmc/lmc_proto.c | 16 ------ 5 files changed, 8 insertions(+), 136 deletions(-)
drivers/net/wan/lmc/lmc_debug.c +0 -18
@@ -62,22 +62,4 @@ void lmcEventLog(u32 EventNum, u32 arg2, }#endif /* DEBUG */ -void lmc_trace (struct net_device *dev,char *msg){ -#ifdef LMC_TRACE -unsigned long j = jiffies +3 ;/* Wait for 50 ms */ - -if (in_interrupt ()){ -printk ("%s: * %s\n" , dev->name, msg); -// while(time_before(jiffies, j+10)) -// ; - } -else { -printk ("%s: %s\n" , dev->name, msg); -while (time_before (jiffies, j)) -schedule (); - } -#endif -} - -/* --------------------------- end if_lmc_linux.c ------------------------ */
drivers/net/wan/lmc/lmc_debug.h +0 -1
@@ -48,6 +48,5 @@ extern u32 lmcEventLogBuf[LMC_EVENTLOGSIvoid lmcConsoleLog (char *type,unsigned char *ucData,int iLen);void lmcEventLog (u32 EventNum,u32 arg2,u32 arg3); -void lmc_trace (struct net_device *dev,char *msg);#endif
drivers/net/wan/lmc/lmc_main.c +8 -97
369 lines not on this page · raw message
drivers/net/wan/lmc/lmc_media.c +0 -4
@@ -1026,7 +1026,6 @@ lmc_t1_get_link_status (lmc_softc_t * co* led3 red = Loss of Signal (LOS) or out of frame (OOF) * conditions detected on T3 receive signal */ -lmc_trace (sc->lmc_device,"lmc_t1_get_link_status in" );lmc_led_on (sc,LMC_DS3_LED2 );lmc_mii_writereg (sc,0 ,17 ,T1FRAMER_ALARM1_STATUS );
@@ -1120,9 +1119,6 @@ lmc_t1_get_link_status (lmc_softc_t * colmc_mii_writereg (sc,0 ,17 ,T1FRAMER_ALARM2_STATUS ); sc->lmc_xinfo.t1_alarm2_status =lmc_mii_readreg (sc,0 ,18 ); - -lmc_trace (sc->lmc_device,"lmc_t1_get_link_status out" ); -return ret; }
drivers/net/wan/lmc/lmc_proto.c +0 -16
@@ -47,7 +47,6 @@// attach void lmc_proto_attach (lmc_softc_t *sc)/*FOLD00*/ { -lmc_trace (sc->lmc_device,"lmc_proto_attach in" );if (sc->if_type ==LMC_NET ) {struct net_device *dev = sc->lmc_device;/*
@@ -57,12 +56,10 @@ void lmc_proto_attach(lmc_softc_t *sc) / dev->hard_header_len =0 ; dev->addr_len =0 ; } -lmc_trace (sc->lmc_device,"lmc_proto_attach out" ); }int lmc_proto_ioctl (lmc_softc_t *sc,struct ifreq *ifr,int cmd) { -lmc_trace (sc->lmc_device,"lmc_proto_ioctl" );if (sc->if_type ==LMC_PPP )return hdlc_ioctl (sc->lmc_device, ifr, cmd);return -EOPNOTSUPP ;
@@ -72,32 +69,23 @@ int lmc_proto_open(lmc_softc_t *sc) {int ret =0 ; -lmc_trace (sc->lmc_device,"lmc_proto_open in" ); -if (sc->if_type ==LMC_PPP ) { ret =hdlc_open (sc->lmc_device);if (ret <0 )printk (KERN_WARNING "%s: HDLC open failed: %d\n" , sc->name, ret); } - -lmc_trace (sc->lmc_device,"lmc_proto_open out" );return ret; }void lmc_proto_close (lmc_softc_t *sc) { -lmc_trace (sc->lmc_device,"lmc_proto_close in" ); -if (sc->if_type ==LMC_PPP )hdlc_close (sc->lmc_device); - -lmc_trace (sc->lmc_device,"lmc_proto_close out" ); }__be16 lmc_proto_type (lmc_softc_t *sc,struct sk_buff *skb)/*FOLD00*/ { -lmc_trace (sc->lmc_device,"lmc_proto_type in" );switch (sc->if_type){case LMC_PPP :return hdlc_type_trans (skb, sc->lmc_device);
@@ -113,13 +101,10 @@ void lmc_proto_close(lmc_softc_t *sc)return htons (ETH_P_802_2 );break ; } -lmc_trace (sc->lmc_device,"lmc_proto_tye out" ); - }void lmc_proto_netif (lmc_softc_t *sc,struct sk_buff *skb)/*FOLD00*/ { -lmc_trace (sc->lmc_device,"lmc_proto_netif in" );switch (sc->if_type){case LMC_PPP :case LMC_NET :
@@ -129,5 +114,4 @@ void lmc_proto_netif(lmc_softc_t *sc, stcase LMC_RAW :break ; } -lmc_trace (sc->lmc_device,"lmc_proto_netif out" ); }
PATCH 24/35 net: brcmfmac: Replace in_interrupt()
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
brcmf_sdio_isr() is using in_interrupt() to distinguish if it is called from a interrupt service routine or from a worker thread.
Passing such information from the calling context is preferred and requested by Linus, so add an argument `in_isr' to brcmf_sdio_isr() and let the callers pass the information about the calling context.
--- drivers/net/wireless/broadcom/brcm80211/brcmfmac/bcmsdh.c | 4 ++-- drivers/net/wireless/broadcom/brcm80211/brcmfmac/sdio.c | 4 ++-- drivers/net/wireless/broadcom/brcm80211/brcmfmac/sdio.h | 2 +- 3 files changed, 5 insertions(+), 5 deletions(-)
drivers/net/wireless/broadcom/brcm80211/brcmfmac/bcmsdh.c +2 -2
@@ -73,7 +73,7 @@ static irqreturn_t brcmf_sdiod_oob_irqha sdiodev->irq_en = false; } -brcmf_sdio_isr (sdiodev->bus); +brcmf_sdio_isr (sdiodev->bus, true);return IRQ_HANDLED ; }
@@ -85,7 +85,7 @@ static void brcmf_sdiod_ib_irqhandler(stbrcmf_dbg (INTR ,"IB intr triggered\n" ); -brcmf_sdio_isr (sdiodev->bus); +brcmf_sdio_isr (sdiodev->bus, false); }/* dummy handler for SDIO function 2 interrupt */
drivers/net/wireless/broadcom/brcm80211/brcmfmac/sdio.c +2 -2
@@ -3625,7 +3625,7 @@ void brcmf_sdio_trigger_dpc(struct brcmf } } -void brcmf_sdio_isr (struct brcmf_sdio *bus) +void brcmf_sdio_isr (struct brcmf_sdio *bus,bool in_isr) {brcmf_dbg (TRACE ,"Enter\n" );
@@ -3636,7 +3636,7 @@ void brcmf_sdio_isr(struct brcmf_sdio *b/* Count the interrupt call */ bus->sdcnt.intrcount++; -if (in_interrupt ()) +if (in_isr)atomic_set (&bus->ipend,1 );else if (brcmf_sdio_intr_rstatus (bus)) {
drivers/net/wireless/broadcom/brcm80211/brcmfmac/sdio.h +1 -1
@@ -372,7 +372,7 @@ int brcmf_sdiod_remove(struct brcmf_sdiostruct brcmf_sdio *brcmf_sdio_probe (struct brcmf_sdio_dev *sdiodev);void brcmf_sdio_remove (struct brcmf_sdio *bus); -void brcmf_sdio_isr (struct brcmf_sdio *bus); +void brcmf_sdio_isr (struct brcmf_sdio *bus,bool in_isr);void brcmf_sdio_wd_timer (struct brcmf_sdio *bus,bool active);void brcmf_sdio_wowl_config (struct device *dev,bool enabled);
Arend Van Spriel
maintainerRe: [patch 24/35] net: brcmfmac: Replace in_interrupt()
+ Uffe
drivers/net/wireless/broadcom/brcm80211/brcmfmac/bcmsdh.c static void brcmf_sdiod_ib_irqhandler(st · line 89
Quoted 7 lines · Thomas Gleixner
drivers/net/wireless/broadcom/brcm80211/brcmfmac/bcmsdh.c +1
-1@@ -85,7 +85,7 @@ static void brcmf_sdiod_ib_irqhandler(stbrcmf_dbg (INTR ,"IB intr triggered\n" ); -brcmf_sdio_isr (sdiodev->bus); +brcmf_sdio_isr (sdiodev->bus, false); }
Hi Uffe,
I assume the above code is okay, but want to confirm. Is the SDIO interrupt guaranteed to be on a worker thread?
Thanks,
Arend
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 4176 bytes
Desc: S/MIME Cryptographic Signature
URL: <http://lists.osuosl.org/pipermail/intel-wired-lan/attachments/20200928/6c6f49f1/attachment-0001.p7s>
Re: [patch 24/35] net: brcmfmac: Replace in_interrupt()
Quoted 8 lines · Arend Van Spriel
+ Uffe
Quoted 4 lines · Thomas Gleixner
drivers/net/wireless/broadcom/brcm80211/brcmfmac/bcmsdh.c +0
-0@@ -85,7 +85,7 @@ static void brcmf_sdiod_ib_irqhandler(stbrcmf_dbg (INTR ,"IB intr triggered\n" );
Quoted 8 lines
drivers/net/wireless/broadcom/brcm80211/brcmfmac/bcmsdh.c static void brcmf_sdiod_ib_irqhandler(st · line 89
Quoted 3 lines
drivers/net/wireless/broadcom/brcm80211/brcmfmac/bcmsdh.c +1
-1-brcmf_sdio_isr (sdiodev->bus); +brcmf_sdio_isr (sdiodev->bus, false); }Hi Uffe,
I assume the above code is okay, but want to confirm. Is the SDIO interrupt guaranteed to be on a worker thread?
Correct.
As a matter of fact, the sdio irqs can be delivered through a couple of different paths. The legacy (scheduled for removal), is from a dedicated kthread. The more "modern" way is either from the context of a threaded IRQ handler or via a workqueue.
However, there are also so-called out of band SDIO irqs, typically routed via a separate GPIO line. This isn't managed by the MMC/SDIO subsystem, but the SDIO functional driver itself.
Kind regards
Uffe
Arend Van Spriel
maintainerRe: [patch 24/35] net: brcmfmac: Replace in_interrupt()
Quoted 21 lines · Ulf Hansson
Quoted 8 lines · Arend Van Spriel
+ Uffe
Quoted 4 lines · Thomas Gleixner
drivers/net/wireless/broadcom/brcm80211/brcmfmac/bcmsdh.c +0
-0@@ -85,7 +85,7 @@ static void brcmf_sdiod_ib_irqhandler(stbrcmf_dbg (INTR ,"IB intr triggered\n" );Quoted 8 lines
drivers/net/wireless/broadcom/brcm80211/brcmfmac/bcmsdh.c static void brcmf_sdiod_ib_irqhandler(st · line 89
Quoted 3 lines
drivers/net/wireless/broadcom/brcm80211/brcmfmac/bcmsdh.c +1
-1-brcmf_sdio_isr (sdiodev->bus); +brcmf_sdio_isr (sdiodev->bus, false); }Hi Uffe,
I assume the above code is okay, but want to confirm. Is the SDIO interrupt guaranteed to be on a worker thread?
Correct.
Quoted 8 lines
As a matter of fact, the sdio irqs can be delivered through a couple of different paths. The legacy (scheduled for removal), is from a dedicated kthread. The more "modern" way is either from the context of a threaded IRQ handler or via a workqueue.
However, there are also so-called out of band SDIO irqs, typically routed via a separate GPIO line. This isn't managed by the MMC/SDIO subsystem, but the SDIO functional driver itself.
In brcmfmac we indeed support the out-of-band interrupt. That scnenario is also taken care of in this patch. Thanks for the confirmation.
Regards,
Arend
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 4176 bytes
Desc: S/MIME Cryptographic Signature
URL: <http://lists.osuosl.org/pipermail/intel-wired-lan/attachments/20200928/c06b9ee1/attachment-0001.p7s>
Re: [patch 24/35] net: brcmfmac: Replace in_interrupt()
In reply to Thomas GleixnerQuoted 8 lines · Thomas Gleixner
From: Sebastian Andrzej Siewior <[email protected]>
brcmf_sdio_isr()is usingin_interrupt()to distinguish if it is called from a interrupt service routine or from a worker thread.Passing such information from the calling context is preferred and requested by Linus, so add an argument `in_isr' to
brcmf_sdio_isr()and let the callers pass the information about the calling context.
After getting confirmation here is my ...
Quoted 14 lines
Signed-off-by Sebastian Andrzej Siewior <[email protected]>Signed-off-by Thomas Gleixner <[email protected]>Cc Arend van Spriel <[email protected]>Cc Franky Lin <[email protected]>Cc Hante Meuleman <[email protected]>Cc Chi-Hsien Lin <[email protected]>Cc Wright Feng <[email protected]>Cc Kalle Valo <[email protected]>Cc "David S. Miller" <[email protected]>Cc Jakub Kicinski <[email protected]>Cc linux-wireless at vger.kernel.orgCc brcm80211-dev-list.pdl at broadcom.comCc brcm80211-dev-list at cypress.comCc netdev at vger.kernel.org
Quoted 5 lines
--- drivers/net/wireless/broadcom/brcm80211/brcmfmac/bcmsdh.c | 4 ++-- drivers/net/wireless/broadcom/brcm80211/brcmfmac/sdio.c | 4 ++-- drivers/net/wireless/broadcom/brcm80211/brcmfmac/sdio.h | 2 +- 3 files changed, 5 insertions(+), 5 deletions(-)
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 4176 bytes
Desc: S/MIME Cryptographic Signature
URL: <http://lists.osuosl.org/pipermail/intel-wired-lan/attachments/20200928/04397eb9/attachment-0001.p7s>
PATCH 25/35 net: brcmfmac: Use netif_rx_any_context().
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
The usage of in_interrupt() in non-core code is phased out. Ideally the information of the calling context should be passed by the callers or the functions be split as appropriate.
brcmfmac uses in_interupt() to select the netif_rx*() variant which matches the calling context. The attempt to consolidate the code by passing an arguemnt or by distangling it failed due lack of knowledge about this driver and because the call chains are convoluted and hard to follow.
As a stop gap use netif_rx_any_context() which invokes the correct code path depending on context which confines the usage of in_interrupt() to core code.
--- drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.c | 9 +-------- 1 file changed, 1 insertion(+), 8 deletions(-)
drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.c +1 -8
@@ -418,14 +418,7 @@ void brcmf_netif_rx(struct brcmf_if *ifp ifp->ndev->stats.rx_packets++;brcmf_dbg (DATA ,"rx proto=0x%X\n" ,ntohs (skb->protocol)); -if (in_interrupt ()) -netif_rx (skb); -else -/* If the receive is not processed inside an ISR, -* the softirqd must be woken explicitly to service -* the NET_RX_SOFTIRQ. This is handled by netif_rx_ni(). -*/ -netif_rx_ni (skb); +netif_rx_any_context (skb); }void brcmf_netif_mon_rx (struct brcmf_if *ifp,struct sk_buff *skb)
Arend Van Spriel
maintainerRe: [patch 25/35] net: brcmfmac: Use netif_rx_any_context().
Quoted 10 lines · Thomas Gleixner
From: Sebastian Andrzej Siewior <[email protected]>
The usage of
in_interrupt()in non-core code is phased out. Ideally the information of the calling context should be passed by the callers or the functions be split as appropriate.brcmfmac uses
in_interupt()to select the netif_rx*() variant which matches the calling context. The attempt to consolidate the code by passing an arguemnt or by distangling it failed due lack of knowledge about this driver and because the call chains are convoluted and hard to follow.
I think it is only for USB devices that the function can be called in "interrupt" context. PCIe devices call it from thread context for sure. The function brcmf_netif_rx() is (in)directly called by brcmf_rx_frame(), which is used by SDIO and USB. Anyway, it will be a bit more work, but doable. Let me see what I can come up with.
Regards,
Arend
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 4176 bytes
Desc: S/MIME Cryptographic Signature
URL: <http://lists.osuosl.org/pipermail/intel-wired-lan/attachments/20200928/2c81f2e0/attachment-0001.p7s>
PATCH 26/35 net: brcmfmac: Convey allocation mode as argument
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
The usage of in_interrupt() in drivers is phased out and Linus clearly requested that code which changes behaviour depending on context should either be seperated or the context be conveyed in an argument passed by the caller, which usually knows the context.
brcmf_fweh_process_event() uses in_interrupt() to select the allocation mode GFP_KERNEL/GFP_ATOMIC. Aside of the above reasons this check is incomplete as it cannot detect contexts which just have preemption or interrupts disabled.
All callchains leading to brcmf_fweh_process_event() can clearly identify the calling context. Convey a 'gfp' argument through the callchains and let the callers hand in the appropriate GFP mode.
This has also the advantage that any change of execution context or preemption/interrupt state in these callchains will be detected by the memory allocator for all GFP_KERNEL allocations.
--- drivers/net/wireless/broadcom/brcm80211/brcmfmac/bus.h | 5 +++-- drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.c | 11 ++++++----- drivers/net/wireless/broadcom/brcm80211/brcmfmac/fweh.c | 8 ++------ drivers/net/wireless/broadcom/brcm80211/brcmfmac/fweh.h | 7 ++++--- drivers/net/wireless/broadcom/brcm80211/brcmfmac/msgbuf.c | 2 +- drivers/net/wireless/broadcom/brcm80211/brcmfmac/sdio.c | 8 ++++---- drivers/net/wireless/broadcom/brcm80211/brcmfmac/usb.c | 2 +- 7 files changed, 21 insertions(+), 22 deletions(-)
drivers/net/wireless/broadcom/brcm80211/brcmfmac/bus.h +3 -2
@@ -249,9 +249,10 @@ int brcmf_bus_reset(struct brcmf_bus *bu*/ /* Receive frame for delivery to OS. Callee disposes of rxp. */ -void brcmf_rx_frame (struct device *dev,struct sk_buff *rxp,bool handle_event); +void brcmf_rx_frame (struct device *dev,struct sk_buff *rxp,bool handle_event, +gfp_t gfp);/* Receive async event packet from firmware. Callee disposes of rxp. */ -void brcmf_rx_event (struct device *dev,struct sk_buff *rxp); +void brcmf_rx_event (struct device *dev,struct sk_buff *rxp,gfp_t gfp);int brcmf_alloc (struct device *dev,struct brcmf_mp_device *settings);/* Indication from bus module regarding presence/insertion of dongle. */
drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.c +6 -5
@@ -489,7 +489,8 @@ static int brcmf_rx_hdrpull(struct brcmfreturn 0 ; } -void brcmf_rx_frame (struct device *dev,struct sk_buff *skb,bool handle_event) +void brcmf_rx_frame (struct device *dev,struct sk_buff *skb,bool handle_event, +gfp_t gfp) {struct brcmf_if *ifp;struct brcmf_bus *bus_if =dev_get_drvdata (dev);
@@ -506,13 +507,13 @@ void brcmf_rx_frame(struct device *dev,/* Process special event packets */ if (handle_event)brcmf_fweh_process_skb (ifp->drvr, skb, -BCMILCP_SUBTYPE_VENDOR_LONG ); - +BCMILCP_SUBTYPE_VENDOR_LONG , + gfp);brcmf_netif_rx (ifp, skb); } } -void brcmf_rx_event (struct device *dev,struct sk_buff *skb) +void brcmf_rx_event (struct device *dev,struct sk_buff *skb,gfp_t gfp) {struct brcmf_if *ifp;struct brcmf_bus *bus_if =dev_get_drvdata (dev);
@@ -523,7 +524,7 @@ void brcmf_rx_event(struct device *dev,if (brcmf_rx_hdrpull (drvr, skb, &ifp))return ; -brcmf_fweh_process_skb (ifp->drvr, skb,0 ); +brcmf_fweh_process_skb (ifp->drvr, skb,0 , gfp);brcmu_pkt_buf_free_skb (skb); }
drivers/net/wireless/broadcom/brcm80211/brcmfmac/fweh.c +2 -6
@@ -387,12 +387,11 @@ int brcmf_fweh_activate_events(struct br*/ void brcmf_fweh_process_event (struct brcmf_pub *drvr,struct brcmf_event *event_packet, -u32 packet_len) +u32 packet_len,gfp_t gfp) {enum brcmf_fweh_event_code code;struct brcmf_fweh_info *fweh = &drvr->fweh;struct brcmf_fweh_queue_item *event; -gfp_t alloc_flag =GFP_KERNEL ;void *data;u32 datalen;
@@ -411,10 +410,7 @@ void brcmf_fweh_process_event(struct brc datalen + sizeof(*event_packet) > packet_len)return ; -if (in_interrupt ()) - alloc_flag =GFP_ATOMIC ; - - event =kzalloc (sizeof (*event) + datalen, alloc_flag); + event =kzalloc (sizeof (*event) + datalen, gfp);if (!event)return ;
drivers/net/wireless/broadcom/brcm80211/brcmfmac/fweh.h +4 -3
@@ -319,11 +319,12 @@ void brcmf_fweh_unregister(struct brcmf_int brcmf_fweh_activate_events (struct brcmf_if *ifp);void brcmf_fweh_process_event (struct brcmf_pub *drvr,struct brcmf_event *event_packet, -u32 packet_len); +u32 packet_len,gfp_t gfp);void brcmf_fweh_p2pdev_setup (struct brcmf_if *ifp,bool ongoing);static inline void brcmf_fweh_process_skb (struct brcmf_pub *drvr, -struct sk_buff *skb,u16 stype) +struct sk_buff *skb,u16 stype, +gfp_t gfp) {struct brcmf_event *event_packet;u16 subtype, usr_stype;
@@ -354,7 +355,7 @@ static inline void brcmf_fweh_process_skif (usr_stype !=BCMILCP_BCM_SUBTYPE_EVENT )return ; - brcmf_fweh_process_event(drvr, event_packet, skb->len + ETH_HLEN); + brcmf_fweh_process_event(drvr, event_packet, skb->len + ETH_HLEN, gfp); }#endif /* FWEH_H_ */
drivers/net/wireless/broadcom/brcm80211/brcmfmac/msgbuf.c +1 -1
@@ -1128,7 +1128,7 @@ static void brcmf_msgbuf_process_event(s skb->protocol =eth_type_trans (skb, ifp->ndev); -brcmf_fweh_process_skb (ifp->drvr, skb,0 ); +brcmf_fweh_process_skb (ifp->drvr, skb,0 ,GFP_KERNEL ); exit:brcmu_pkt_buf_free_skb (skb);
drivers/net/wireless/broadcom/brcm80211/brcmfmac/sdio.c +4 -4
@@ -1701,10 +1701,10 @@ static u8 brcmf_sdio_rxglom(struct brcmf pfirst->prev);skb_unlink (pfirst, &bus->glom);if (brcmf_sdio_fromevntchan (&dptr[SDPCM_HWHDR_LEN ])) -brcmf_rx_event (bus->sdiodev->dev, pfirst); +brcmf_rx_event (bus->sdiodev->dev, pfirst,GFP_KERNEL );else brcmf_rx_frame (bus->sdiodev->dev, pfirst, - false); + false,GFP_KERNEL ); bus->sdcnt.rxglompkts++; }
@@ -2035,10 +2035,10 @@ static uint brcmf_sdio_readframes(structif (pkt->len ==0 )brcmu_pkt_buf_free_skb (pkt);else if (rd->channel ==SDPCM_EVENT_CHANNEL ) -brcmf_rx_event (bus->sdiodev->dev, pkt); +brcmf_rx_event (bus->sdiodev->dev, pkt,GFP_KERNEL );else brcmf_rx_frame (bus->sdiodev->dev, pkt, - false); + false,GFP_KERNEL );/* prepare the descriptor for the next read */ rd->len = rd->len_nxtfrm <<4 ;
drivers/net/wireless/broadcom/brcm80211/brcmfmac/usb.c +1 -1
@@ -532,7 +532,7 @@ static void brcmf_usb_rx_complete(structif (devinfo->bus_pub.state ==BRCMFMAC_USB_STATE_UP || devinfo->bus_pub.state ==BRCMFMAC_USB_STATE_SLEEP ) {skb_put (skb, urb->actual_length); -brcmf_rx_frame (devinfo->dev, skb, true); +brcmf_rx_frame (devinfo->dev, skb, true,GFP_ATOMIC );brcmf_usb_rx_refill (devinfo, req);usb_mark_last_busy (urb->dev); }else {
Arend Van Spriel
maintainerRe: [patch 26/35] net: brcmfmac: Convey allocation mode as argument
Quoted 12 lines · Thomas Gleixner
From: Sebastian Andrzej Siewior <[email protected]>
The usage of
in_interrupt()in drivers is phased out and Linus clearly requested that code which changes behaviour depending on context should either be seperated or the context be conveyed in an argument passed by the caller, which usually knows the context.
brcmf_fweh_process_event()usesin_interrupt()to select the allocation mode GFP_KERNEL/GFP_ATOMIC. Aside of the above reasons this check is incomplete as it cannot detect contexts which just have preemption or interrupts disabled.
Quoted 7 lines
All callchains leading to
brcmf_fweh_process_event()can clearly identify the calling context. Convey a 'gfp' argument through the callchains and let the callers hand in the appropriate GFP mode.This has also the advantage that any change of execution context or preemption/interrupt state in these callchains will be detected by the memory allocator for all
GFP_KERNELallocations.
Looking at the functions changed I see overlap with patch 25/35 so maybe change to gfp flag to a 'in_isr' bool. The flag can then be used for dealing with netif_rx as well.
Regards,
Arend
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 4176 bytes
Desc: S/MIME Cryptographic Signature
URL: <http://lists.osuosl.org/pipermail/intel-wired-lan/attachments/20200928/ff2058e2/attachment-0001.p7s>
PATCH 27/35 net: ipw2x00,iwlegacy,iwlwifi: Remove in_interrupt() from debug macros
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
The usage of in_interrupt() in non-core code is phased out.
The debugging macros in these drivers use in_interrupt() to print 'I' or 'U' depending on the return value of in_interrupt(). While 'U' is confusing at best and 'I' is not really describing the actual context (hard interupt, soft interrupt, bottom half disabled section) these debug macros originate from the pre ftrace kernel era and their value today is questionable. They probably should be removed completely.
The macros weere added initially for ipw2100 and then spreaded when the driver was forked.
Remove the in_interrupt() usage at least..
--- drivers/net/wireless/intel/ipw2x00/ipw2100.c | 3 +-- drivers/net/wireless/intel/ipw2x00/ipw2200.h | 6 ++---- drivers/net/wireless/intel/ipw2x00/libipw.h | 3 +-- drivers/net/wireless/intel/iwlegacy/common.h | 4 ++-- drivers/net/wireless/intel/iwlwifi/iwl-debug.c | 3 +-- 5 files changed, 7 insertions(+), 12 deletions(-)
drivers/net/wireless/intel/ipw2x00/ipw2100.c +1 -2
@@ -201,8 +201,7 @@ static u32 ipw2100_debug_level = IPW_DL_ #define IPW_DEBUG(level, message...) \ do { \ if (ipw2100_debug_level & (level)) { \ - printk(KERN_DEBUG "ipw2100: %c %s ", \ - in_interrupt() ? 'I' : 'U', __func__); \ + printk(KERN_DEBUG "ipw2100: %s ", __func__); \ printk(message); \ } \ } while (0)
drivers/net/wireless/intel/ipw2x00/ipw2200.h +2 -4
@@ -1382,14 +1382,12 @@ BIT_ARG16(x) #define IPW_DEBUG(level, fmt, args...) \ do { if (ipw_debug_level & (level)) \ - printk(KERN_DEBUG DRV_NAME": %c %s " fmt, \ - in_interrupt() ? 'I' : 'U', __func__ , ## args); } while (0) + printk(KERN_DEBUG DRV_NAME": %s " fmt, __func__ , ## args); } while (0)#ifdef CONFIG_IPW2200_DEBUG #define IPW_LL_DEBUG(level, fmt, args...) \ do { if (ipw_debug_level & (level)) \ - printk(KERN_DEBUG DRV_NAME": %c %s " fmt, \ - in_interrupt() ? 'I' : 'U', __func__ , ## args); } while (0) + printk(KERN_DEBUG DRV_NAME": %s " fmt, __func__ , ## args); } while (0)#else #define IPW_LL_DEBUG(level, fmt, args...) do {} while (0)#endif /* CONFIG_IPW2200_DEBUG */
drivers/net/wireless/intel/ipw2x00/libipw.h +1 -2
@@ -60,8 +60,7 @@extern u32 libipw_debug_level; #define LIBIPW_DEBUG(level, fmt, args...) \ do { if (libipw_debug_level & (level)) \ - printk(KERN_DEBUG "libipw: %c %s " fmt, \ - in_interrupt() ? 'I' : 'U', __func__ , ## args); } while (0) + printk(KERN_DEBUG "libipw: %s " fmt, __func__ , ## args); } while (0)#else #define LIBIPW_DEBUG(level, fmt, args...) do {} while (0)#endif /* CONFIG_LIBIPW_DEBUG */
drivers/net/wireless/intel/iwlegacy/common.h +2 -2
@@ -2925,8 +2925,8 @@ do { \ #define IL_DBG(level, fmt, args...) \ do { \ if (il_get_debug_level(il) & level) \ - dev_err(&il->hw->wiphy->dev, "%c %s " fmt, \ - in_interrupt() ? 'I' : 'U', __func__ , ##args); \ + dev_err(&il->hw->wiphy->dev, "%s " fmt, __func__, \ + ##args); \ } while (0)#define il_print_hex_dump (il, level, p, len) \
drivers/net/wireless/intel/iwlwifi/iwl-debug.c +1 -2
@@ -121,8 +121,7 @@ void __iwl_dbg(struct device *dev,#ifdef CONFIG_IWLWIFI_DEBUG if (iwl_have_debug_level (level) && (!limit ||net_ratelimit ())) -dev_printk (KERN_DEBUG , dev,"%c %s %pV" , -in_interrupt () ?'I' :'U' , function, &vaf); +dev_printk (KERN_DEBUG , dev,"%s %pV" , function, &vaf);#endif trace_iwlwifi_dbg (level,in_interrupt (), function, &vaf);va_end (args);
PATCH 28/35 net: iwlwifi: Remove in_interrupt() from tracing macro.
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
The usage of in_interrupt) in driver code is phased out.
The iwlwifi_dbg tracepoint records in_interrupt() seperately, but that's superfluous because the trace header already records all kind of state and context information like hardirq status, softirq status, preemption count etc.
Aside of that the recording of in_interrupt() as boolean does not allow to distinguish between the possible contexts (hard interrupt, soft interrupt, bottom half disabled) while the trace header gives precise information.
Remove the duplicate information from the tracepoint and fixup the caller.
--- drivers/net/wireless/intel/iwlwifi/iwl-debug.c | 2 +- drivers/net/wireless/intel/iwlwifi/iwl-devtrace-msg.h | 6 ++---- 2 files changed, 3 insertions(+), 5 deletions(-)
drivers/net/wireless/intel/iwlwifi/iwl-debug.c +1 -1
@@ -123,7 +123,7 @@ void __iwl_dbg(struct device *dev, (!limit || net_ratelimit()))dev_printk (KERN_DEBUG , dev,"%s %pV" , function, &vaf);#endif -trace_iwlwifi_dbg (level,in_interrupt (), function, &vaf); +trace_iwlwifi_dbg (level, function, &vaf);va_end (args); }IWL_EXPORT_SYMBOL (__iwl_dbg);
drivers/net/wireless/intel/iwlwifi/iwl-devtrace-msg.h +2 -4
@@ -54,18 +54,16 @@ DEFINE_EVENT(iwlwifi_msg_event, iwlwifi_ ); TRACE_EVENT(iwlwifi_dbg, - TP_PROTO(u32 level, bool in_interrupt, const char *function, + TP_PROTO(u32 level, const char *function, struct va_format *vaf), - TP_ARGS(level, in_interrupt, function, vaf), + TP_ARGS(level, function, vaf), TP_STRUCT__entry( __field(u32, level) - __field(u8, in_interrupt) __string(function, function) __dynamic_array(char, msg, MAX_MSG_LEN) ), TP_fast_assign( __entry->level = level; - __entry->in_interrupt = in_interrupt; __assign_str(function, function); WARN_ON_ONCE(vsnprintf(__get_dynamic_array(msg), MAX_MSG_LEN, vaf->fmt,
Re: [patch 28/35] net: iwlwifi: Remove in_interrupt() from tracing macro.
Quoted 15 lines · Thomas Gleixner
From: Sebastian Andrzej Siewior <[email protected]>
The usage of
in_interrupt) in driver code is phased out.The
iwlwifi_dbgtracepoint recordsin_interrupt()seperately, but that's superfluous because the trace header already records all kind of state and context information like hardirq status, softirq status, preemption count etc.Aside of that the recording of
in_interrupt()as boolean does not allow to distinguish between the possible contexts (hard interrupt, soft interrupt, bottom half disabled) while the trace header gives precise information.Remove the duplicate information from the tracepoint and fixup the caller.
Quoted 11 lines
Signed-off-by Sebastian Andrzej Siewior <[email protected]>Signed-off-by Thomas Gleixner <[email protected]>Cc Johannes Berg <[email protected]>Cc Emmanuel Grumbach <[email protected]>Cc Luca Coelho <[email protected]>Cc Intel Linux Wireless <[email protected]>Cc Kalle Valo <[email protected]>Cc "David S. Miller" <[email protected]>Cc Jakub Kicinski <[email protected]>Cc linux-wireless at vger.kernel.orgCc netdev at vger.kernel.org
--
Luca.
PATCH 29/35 net: hostap: Remove in_interrupt() usage
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
in_interrupt() is ill defined and does not provide what the name suggests. The usage especially in driver code is deprecated and a tree wide effort to clean up and consolidate the (ab)usage of in_interrupt() and related checks is happening.
hfa384x_cmd() and prism2_hw_reset() check in_interrupt() at function entry and if true emit a printk at debug loglevel and return. This is clearly debug code.
Both functions invoke functions which can sleep. These functions already have appropriate debug checks which cover all invalid contexts, while in_interrupt() fails to detect context which just has preemption or interrupts disabled.
Remove both checks as they are incomplete, debug only and already covered by the subsequently invoked functions properly. If called from invalid context the resulting back trace is definitely more helpful to analyze the problem than a printk at debug loglevel.
--- drivers/net/wireless/intersil/hostap/hostap_hw.c | 12 ------------ 1 file changed, 12 deletions(-)
drivers/net/wireless/intersil/hostap/hostap_hw.c +0 -12
@@ -320,12 +320,6 @@ static int hfa384x_cmd(struct net_device iface =netdev_priv (dev); local = iface->local; -if (in_interrupt ()) { -printk (KERN_DEBUG "%s: hfa384x_cmd called from interrupt " -"context\n" , dev->name); -return -1 ; - } -if (local->cmd_queue_len >=HOSTAP_CMD_QUEUE_MAX_LEN ) {printk (KERN_DEBUG "%s: hfa384x_cmd: cmd_queue full\n" , dev->name);
@@ -1560,12 +1554,6 @@ static void prism2_hw_reset(struct net_d iface =netdev_priv (dev); local = iface->local; -if (in_interrupt ()) { -printk (KERN_DEBUG "%s: driver bug - prism2_hw_reset() called " -"in interrupt context\n" , dev->name); -return ; - } -if (local->hw_downloading)return ;
PATCH 30/35 net: mwifiex: Use netif_rx_any_context().
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
The usage of in_interrupt() in non-core code is phased out. Ideally the information of the calling context should be passed by the callers or the functions be split as appropriate.
mwifiex uses in_interupt() to select the netif_rx*() variant which matches the calling context. The attempt to consolidate the code by passing an arguemnt or by distangling it failed due lack of knowledge about this driver and because the call chains are hard to follow.
As a stop gap use netif_rx_any_context() which invokes the correct code path depending on context and confines the in_interrupt() usage to core code.
--- drivers/net/wireless/marvell/mwifiex/uap_txrx.c | 6 +----- drivers/net/wireless/marvell/mwifiex/util.c | 6 +----- 2 files changed, 2 insertions(+), 10 deletions(-)
drivers/net/wireless/marvell/mwifiex/uap_txrx.c +1 -5
@@ -350,11 +350,7 @@ int mwifiex_uap_recv_packet(struct mwifi skb->truesize += (skb->len -MWIFIEX_RX_DATA_BUF_SIZE );/* Forward multicast/broadcast packet to upper layer*/ -if (in_interrupt ()) -netif_rx (skb); -else -netif_rx_ni (skb); - +netif_rx_any_context (skb);return 0 ; }
drivers/net/wireless/marvell/mwifiex/util.c +1 -5
@@ -488,11 +488,7 @@ int mwifiex_recv_packet(struct mwifiex_p (skb->truesize > MWIFIEX_RX_DATA_BUF_SIZE)) skb->truesize += (skb->len - MWIFIEX_RX_DATA_BUF_SIZE); -if (in_interrupt ()) -netif_rx (skb); -else -netif_rx_ni (skb); - +netif_rx_any_context (skb);return 0 ; }
PATCH 31/35 net: libertas libertas_tf: Remove in_interrupt() from debug macro.
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
The debug macro prints (INT) when in_interrupt() returns true. The value of this information is dubious as it does not distinguish between the various contexts which are covered by in_interrupt().
As the usage of in_interrupt() in drivers is phased out and the same information can be more precisely obtained with tracing, remove the in_interrupt() conditional from this debug printk.
--- drivers/net/wireless/marvell/libertas/defs.h | 3 +-- drivers/net/wireless/marvell/libertas_tf/deb_defs.h | 3 +-- 2 files changed, 2 insertions(+), 4 deletions(-)
drivers/net/wireless/marvell/libertas/defs.h +1 -2
@@ -50,8 +50,7 @@ extern unsigned int lbs_debug;#ifdef DEBUG #define LBS_DEB_LL(grp, grpnam, fmt, args...) \ do { if ((lbs_debug & (grp)) == (grp)) \ - printk(KERN_DEBUG DRV_NAME grpnam "%s: " fmt, \ - in_interrupt() ? " (INT)" : "", ## args); } while (0) + printk(KERN_DEBUG DRV_NAME grpnam ": " fmt, ## args); } while (0)#else #define LBS_DEB_LL(grp, grpnam, fmt, args...) do {} while (0)#endif
drivers/net/wireless/marvell/libertas_tf/deb_defs.h +1 -2
@@ -48,8 +48,7 @@ extern unsigned int lbtf_debug;#ifdef DEBUG #define LBTF_DEB_LL(grp, grpnam, fmt, args...) \ do { if ((lbtf_debug & (grp)) == (grp)) \ - printk(KERN_DEBUG DRV_NAME grpnam "%s: " fmt, \ - in_interrupt() ? " (INT)" : "", ## args); } while (0) + printk(KERN_DEBUG DRV_NAME grpnam ": " fmt, ## args); } while (0)#else #define LBTF_DEB_LL(grp, grpnam, fmt, args...) do {} while (0)#endif
PATCH 32/35 net: libertas: Use netif_rx_any_context()
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
The usage of in_interrupt() in non-core code is phased out. Ideally the information of the calling context should be passed by the callers or the functions be split as appropriate.
libertas uses in_interupt() to select the netif_rx*() variant which matches the calling context. The attempt to consolidate the code by passing an arguemnt or by distangling it failed due lack of knowledge about this driver and because the call chains are hard to follow.
As a stop gap tse netif_rx_any_context() which invokes the correct code path depending on context and confines the in_interrupt() usage to core code.
--- drivers/net/wireless/marvell/libertas/rx.c | 11 ++--------- 1 file changed, 2 insertions(+), 9 deletions(-)
drivers/net/wireless/marvell/libertas/rx.c +2 -9
@@ -147,10 +147,7 @@ int lbs_process_rxed_packet(struct lbs_p dev->stats.rx_packets++; skb->protocol =eth_type_trans (skb, dev); -if (in_interrupt ()) -netif_rx (skb); -else -netif_rx_ni (skb); +netif_rx_any_context (skb); ret =0 ; done:
@@ -265,11 +262,7 @@ static int process_rxed_802_11_packet(st dev->stats.rx_packets++; skb->protocol =eth_type_trans (skb, priv->dev); - -if (in_interrupt ()) -netif_rx (skb); -else -netif_rx_ni (skb); +netif_rx_any_context (skb); ret =0 ;
PATCH 33/35 net: rtlwifi: Remove void* casts related to delayed work
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
INIT_DELAYED_WORK() takes two arguments: A pointer to the delayed work and a function reference for the callback.
The rtl code casts all function references to (void *) because the callbacks in use are not matching the required function signature. That's error prone and bad pratice.
Some of the callback functions are also global, but only used in a single file.
Clean the mess up by:
- Adding the proper arguments to the callback functions and using them in the container_of() constructs correctly which removes the hideous container_of_dwork_rtl() macro as well.
- Removing the type cast at the initializers
- Making the unnecessary global functions static
--- drivers/net/wireless/realtek/rtlwifi/base.c | 39 +++++++++++++--------------- drivers/net/wireless/realtek/rtlwifi/base.h | 3 -- drivers/net/wireless/realtek/rtlwifi/ps.c | 19 ++++++------- drivers/net/wireless/realtek/rtlwifi/ps.h | 6 ++-- drivers/net/wireless/realtek/rtlwifi/wifi.h | 3 -- 5 files changed, 31 insertions(+), 39 deletions(-)
drivers/net/wireless/realtek/rtlwifi/base.c +19 -20
@@ -436,6 +436,10 @@ static void _rtl_init_mac80211(struct ie } } +static void rtl_watchdog_wq_callback (struct work_struct *work); +static void rtl_fwevt_wq_callback (struct work_struct *work); +static void rtl_c2hcmd_wq_callback (struct work_struct *work); +static void _rtl_init_deferred_work (struct ieee80211_hw *hw) {struct rtl_priv *rtlpriv =rtl_priv (hw);
@@ -454,17 +458,14 @@ static void _rtl_init_deferred_work(stru }INIT_DELAYED_WORK (&rtlpriv->works.watchdog_wq, - (void *)rtl_watchdog_wq_callback); + rtl_watchdog_wq_callback);INIT_DELAYED_WORK (&rtlpriv->works.ips_nic_off_wq, - (void *)rtl_ips_nic_off_wq_callback); -INIT_DELAYED_WORK (&rtlpriv->works.ps_work, - (void *)rtl_swlps_wq_callback); + rtl_ips_nic_off_wq_callback); +INIT_DELAYED_WORK (&rtlpriv->works.ps_work, rtl_swlps_wq_callback);INIT_DELAYED_WORK (&rtlpriv->works.ps_rfon_wq, - (void *)rtl_swlps_rfon_wq_callback); -INIT_DELAYED_WORK (&rtlpriv->works.fwevt_wq, - (void *)rtl_fwevt_wq_callback); -INIT_DELAYED_WORK (&rtlpriv->works.c2hcmd_wq, - (void *)rtl_c2hcmd_wq_callback); + rtl_swlps_rfon_wq_callback); +INIT_DELAYED_WORK (&rtlpriv->works.fwevt_wq, rtl_fwevt_wq_callback); +INIT_DELAYED_WORK (&rtlpriv->works.c2hcmd_wq, rtl_c2hcmd_wq_callback); }void rtl_deinit_deferred_work (struct ieee80211_hw *hw,bool ips_wq)
@@ -2042,11 +2043,10 @@ void rtl_collect_scan_list(struct ieee80 }EXPORT_SYMBOL (rtl_collect_scan_list); -void rtl_watchdog_wq_callback (void *data) +static void rtl_watchdog_wq_callback (struct work_struct *work) { -struct rtl_works *rtlworks =container_of_dwork_rtl (data, - struct rtl_works, - watchdog_wq); + struct rtl_works *rtlworks = container_of(work, struct rtl_works, + watchdog_wq.work);struct ieee80211_hw *hw = rtlworks->hw;struct rtl_priv *rtlpriv =rtl_priv (hw);struct rtl_hal *rtlhal =rtl_hal (rtl_priv (hw));
@@ -2239,10 +2239,10 @@ void rtl_watch_dog_timer_callback(struct jiffies + MSECS(RTL_WATCH_DOG_TIME)); } -void rtl_fwevt_wq_callback (void *data) +static void rtl_fwevt_wq_callback (struct work_struct *work) { -struct rtl_works *rtlworks = - container_of_dwork_rtl(data, struct rtl_works, fwevt_wq); + struct rtl_works *rtlworks = container_of(work, struct rtl_works, + fwevt_wq.work);struct ieee80211_hw *hw = rtlworks->hw;struct rtl_priv *rtlpriv =rtl_priv (hw);
@@ -2368,11 +2368,10 @@ void rtl_c2hcmd_launcher(struct ieee8021 } } -void rtl_c2hcmd_wq_callback (void *data) +static void rtl_c2hcmd_wq_callback (struct work_struct *work) { -struct rtl_works *rtlworks =container_of_dwork_rtl (data, - struct rtl_works, - c2hcmd_wq); + struct rtl_works *rtlworks = container_of(work, struct rtl_works, + c2hcmd_wq.work);struct ieee80211_hw *hw = rtlworks->hw;rtl_c2hcmd_launcher (hw,1 );
drivers/net/wireless/realtek/rtlwifi/base.h +0 -3
@@ -108,9 +108,6 @@ int rtl_rx_agg_start(struct ieee80211_hwint rtl_rx_agg_stop (struct ieee80211_hw *hw,struct ieee80211_sta *sta,u16 tid);void rtl_rx_ampdu_apply (struct rtl_priv *rtlpriv); -void rtl_watchdog_wq_callback (void *data); -void rtl_fwevt_wq_callback (void *data); -void rtl_c2hcmd_wq_callback (void *data);void rtl_c2hcmd_launcher (struct ieee80211_hw *hw,int exec);void rtl_c2hcmd_enqueue (struct ieee80211_hw *hw,struct sk_buff *skb);
drivers/net/wireless/realtek/rtlwifi/ps.c +9 -10
@@ -179,10 +179,10 @@ static void _rtl_ps_inactive_ps(struct i ppsc->swrf_processing = false; } -void rtl_ips_nic_off_wq_callback (void *data) +void rtl_ips_nic_off_wq_callback (struct work_struct *work) { -struct rtl_works *rtlworks = - container_of_dwork_rtl(data, struct rtl_works, ips_nic_off_wq); + struct rtl_works *rtlworks = container_of(work, struct rtl_works, + ips_nic_off_wq.work);struct ieee80211_hw *hw = rtlworks->hw;struct rtl_priv *rtlpriv =rtl_priv (hw);struct rtl_hal *rtlhal =rtl_hal (rtl_priv (hw));
@@ -562,10 +562,10 @@ void rtl_swlps_rf_awake(struct ieee80211mutex_unlock (&rtlpriv->locks.lps_mutex); } -void rtl_swlps_rfon_wq_callback (void *data) +void rtl_swlps_rfon_wq_callback (struct work_struct *work) { -struct rtl_works *rtlworks = - container_of_dwork_rtl(data, struct rtl_works, ps_rfon_wq); + struct rtl_works *rtlworks = container_of(work, struct rtl_works, + ps_rfon_wq.work);struct ieee80211_hw *hw = rtlworks->hw;rtl_swlps_rf_awake (hw);
@@ -675,11 +675,10 @@ void rtl_lps_leave(struct ieee80211_hw * }EXPORT_SYMBOL_GPL (rtl_lps_leave); -void rtl_swlps_wq_callback (void *data) +void rtl_swlps_wq_callback (struct work_struct *work) { -struct rtl_works *rtlworks =container_of_dwork_rtl (data, - struct rtl_works, - ps_work); + struct rtl_works *rtlworks = container_of(work, struct rtl_works, + ps_work.work);struct ieee80211_hw *hw = rtlworks->hw;struct rtl_priv *rtlpriv =rtl_priv (hw);bool ps = false;
drivers/net/wireless/realtek/rtlwifi/ps.h +3 -3
@@ -10,15 +10,15 @@ bool rtl_ps_enable_nic(struct ieee80211_bool rtl_ps_disable_nic (struct ieee80211_hw *hw);void rtl_ips_nic_off (struct ieee80211_hw *hw);void rtl_ips_nic_on (struct ieee80211_hw *hw); -void rtl_ips_nic_off_wq_callback (void *data); +void rtl_ips_nic_off_wq_callback (struct work_struct *work);void rtl_lps_enter (struct ieee80211_hw *hw);void rtl_lps_leave (struct ieee80211_hw *hw);void rtl_lps_set_psmode (struct ieee80211_hw *hw,u8 rt_psmode);void rtl_swlps_beacon (struct ieee80211_hw *hw,void *data,unsigned int len); -void rtl_swlps_wq_callback (void *data); -void rtl_swlps_rfon_wq_callback (void *data); +void rtl_swlps_wq_callback (struct work_struct *work); +void rtl_swlps_rfon_wq_callback (struct work_struct *work);void rtl_swlps_rf_awake (struct ieee80211_hw *hw);void rtl_swlps_rf_sleep (struct ieee80211_hw *hw);void rtl_p2p_ps_cmd (struct ieee80211_hw *hw ,u8 p2p_ps_state);
drivers/net/wireless/realtek/rtlwifi/wifi.h +0 -3
@@ -2936,9 +2936,6 @@ enum bt_radio_shared {#define RT_SET_PS_LEVEL (ppsc, _ps_flg) \ (ppsc->cur_ps_level |= _ps_flg) -#define container_of_dwork_rtl (x, y, z) \ - container_of(to_delayed_work(x), y, z) -#define FILL_OCTET_STRING (_os, _octet, _len) \ (_os).octet = (u8 *)(_octet); \ (_os).length = (_len);
PATCH 34/35 net: rtlwifi: Remove in_interrupt() from debug macro
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
The usage of in_interrupt() in drivers in is phased out.
rtlwifi uses in_interrupt() in the RT_TRACE() debug macro which is sprinkled all over the driver. RT_TRACE() is almost identical to RTPRINT() which another hideous debug printk wrapper. The only difference is the printout of in_interrupt().
The decoding of in_interrupt() as hexvalue is non-trivial and aside of being phased out for driver use the return value is just by chance the masked preempt count value and not a boolean.
These home brewn printk debug aids are tedious to work with and provide only minimal context. They should be replaced by trace_printk() or a debug tracepoint which automatically records all context information.
To make progress on the in_interrupt() cleanup, make RT_TRACE() use the RTPRINT() debug function and remove _rtl_dbg_trace().
--- drivers/net/wireless/realtek/rtlwifi/debug.c | 20 -------------------- drivers/net/wireless/realtek/rtlwifi/debug.h | 6 +----- 2 files changed, 1 insertion(+), 25 deletions(-)
drivers/net/wireless/realtek/rtlwifi/debug.c +0 -20
@@ -8,26 +8,6 @@#include <linux/vmalloc.h> #ifdef CONFIG_RTLWIFI_DEBUG -void _rtl_dbg_trace (struct rtl_priv *rtlpriv,u64 comp,int level, -const char *fmt, ...) -{ -if (unlikely ((comp & rtlpriv->cfg->mod_params->debug_mask) && - level <= rtlpriv->cfg->mod_params->debug_level)) { -struct va_format vaf; -va_list args; - -va_start (args, fmt); - - vaf.fmt = fmt; - vaf.va = &args; - -pr_info (":<%lx> %pV" ,in_interrupt (), &vaf); - -va_end (args); - } -} -EXPORT_SYMBOL_GPL (_rtl_dbg_trace); -void _rtl_dbg_print (struct rtl_priv *rtlpriv,u64 comp,int level,const char *fmt, ...) {
drivers/net/wireless/realtek/rtlwifi/debug.h +1 -5
@@ -149,10 +149,6 @@ enum dbgp_flag_e {struct rtl_priv ; __printf(4, 5) -void _rtl_dbg_trace (struct rtl_priv *rtlpriv,u64 comp,int level, -const char *fmt, ...); - -__printf(4, 5)void _rtl_dbg_print (struct rtl_priv *rtlpriv,u64 comp,int level,const char *fmt, ...);
@@ -161,7 +157,7 @@ void _rtl_dbg_print_data(struct rtl_priv const void *hexdata, int hexdatalen);#define RT_TRACE (rtlpriv, comp, level, fmt, ...) \ - _rtl_dbg_trace(rtlpriv, comp, level, \ + _rtl_dbg_print(rtlpriv, comp, level, \ fmt, ##__VA_ARGS__)#define RTPRINT (rtlpriv, dbgtype, dbgflag, fmt, ...) \
PATCH 35/35 net: rtlwifi: Replace in_interrupt() for context detection
In reply to Thomas GleixnerFrom: Sebastian Andrzej Siewior <[email protected]>
rtl_lps_enter() and rtl_lps_leave() are using in_interrupt() to detect whether it is safe to acquire a mutex or if it is required to defer to a workqueue.
The usage of in_interrupt() in drivers is phased out and Linus clearly requested that code which changes behaviour depending on context should either be seperated or the context be conveyed in an argument passed by the caller, which usually knows the context.
in_interrupt() also is only partially correct because it fails to chose the correct code path when just preemption or interrupts are disabled.
Add an argument 'may_block' to both functions and adjust the callers to pass the context information.
The following call chains were analyzed to be safe to block:
rtl_watchdog_wq_callback()
rlf_lps_leave/enter()
rtl_op_suspend()
rtl_lps_leave()
rtl_op_bss_info_changed()
rtl_lps_leave()
rtl_op_sw_scan_start()
rtl_lps_leave()The following call chains were analyzed to be unsafe to block:
_rtl_pci_interrupt()
_rtl_pci_rx_interrupt()
rtl_lps_leave()
_rtl_pci_interrupt()
_rtl_pci_rx_interrupt()
rtl_is_special_data()
rtl_lps_leave()
_rtl_pci_interrupt()
_rtl_pci_rx_interrupt()
rtl_is_special_data()
setup_special_tx()
rtl_lps_leave()
_rtl_pci_interrupt()
_rtl_pci_tx_isr
rtl_lps_leave()
halbtc_leave_lps()
rtl_lps_leave()This leaves four callers of rtl_lps_enter/leave() where the analyzis stopped dead in the maze of several nested pointer based callchains and lack of rtlwifi hardware to debug this via tracing:
halbtc_leave_lps(), halbtc_enter_lps(), halbtc_normal_lps(), halbtc_pre_normal_lps()
These four have been cautionally marked to be unable to block which is the safe option, but the rtwifi wizards should be able to clarify that.
--- drivers/net/wireless/realtek/rtlwifi/base.c | 8 +++--- drivers/net/wireless/realtek/rtlwifi/btcoexist/halbtcoutsrc.c | 12 ++++++---- drivers/net/wireless/realtek/rtlwifi/core.c | 6 ++--- drivers/net/wireless/realtek/rtlwifi/pci.c | 4 +-- drivers/net/wireless/realtek/rtlwifi/ps.c | 8 +++--- drivers/net/wireless/realtek/rtlwifi/ps.h | 4 +-- 6 files changed, 23 insertions(+), 19 deletions(-)
drivers/net/wireless/realtek/rtlwifi/base.c +4 -4
@@ -1456,7 +1456,7 @@ static void setup_special_tx(struct rtl_if (rtlpriv->cfg->ops->get_btc_status ()) rtlpriv->btcoexist.btc_ops->btc_special_packet_notify ( rtlpriv, type); -rtl_lps_leave (hw); +rtl_lps_leave (hw, false); ppsc->last_delaylps_stamp_jiffies = jiffies; }
@@ -1546,7 +1546,7 @@ u8 rtl_is_special_data(struct ieee80211_if (is_tx) { rtlpriv->ra.is_special_data = true; -rtl_lps_leave (hw); +rtl_lps_leave (hw, false); ppsc->last_delaylps_stamp_jiffies = jiffies;setup_special_tx (rtlpriv, ppsc,PACKET_EAPOL );
@@ -2147,9 +2147,9 @@ static void rtl_watchdog_wq_callback(strif (rtlpriv->link_info.num_rx_inperiod + rtlpriv->link_info.num_tx_inperiod >8 || rtlpriv->link_info.num_rx_inperiod >2 ) -rtl_lps_leave (hw); +rtl_lps_leave (hw, true);else -rtl_lps_enter (hw); +rtl_lps_enter (hw, true); label_lps_done: ;
drivers/net/wireless/realtek/rtlwifi/btcoexist/halbtcoutsrc.c +8 -4
@@ -285,7 +285,8 @@ static void halbtc_leave_lps(struct btc_ btcoexist->bt_info.bt_ctrl_lps = true; btcoexist->bt_info.bt_lps_on = false; -rtl_lps_leave (rtlpriv->mac80211.hw); +/* FIXME: Context is unclear. Is it allowed to block? */ +rtl_lps_leave (rtlpriv->mac80211.hw, false); }static void halbtc_enter_lps (struct btc_coexist *btcoexist)
@@ -306,7 +307,8 @@ static void halbtc_enter_lps(struct btc_ btcoexist->bt_info.bt_ctrl_lps = true; btcoexist->bt_info.bt_lps_on = true; -rtl_lps_enter (rtlpriv->mac80211.hw); +/* FIXME: Context is unclear. Is it allowed to block? */ +rtl_lps_enter (rtlpriv->mac80211.hw, false); }static void halbtc_normal_lps (struct btc_coexist *btcoexist)
@@ -317,7 +319,8 @@ static void halbtc_normal_lps(struct btcif (btcoexist->bt_info.bt_ctrl_lps) { btcoexist->bt_info.bt_lps_on = false; -rtl_lps_leave (rtlpriv->mac80211.hw); +/* FIXME: Context is unclear. Is it allowed to block? */ +rtl_lps_leave (rtlpriv->mac80211.hw, false); btcoexist->bt_info.bt_ctrl_lps = false; } }
@@ -328,7 +331,8 @@ static void halbtc_pre_normal_lps(structif (btcoexist->bt_info.bt_ctrl_lps) { btcoexist->bt_info.bt_lps_on = false; -rtl_lps_leave (rtlpriv->mac80211.hw); +/* FIXME: Context is unclear. Is it allowed to block? */ +rtl_lps_leave (rtlpriv->mac80211.hw, false); } }
drivers/net/wireless/realtek/rtlwifi/core.c +3 -3
@@ -544,7 +544,7 @@ static int rtl_op_suspend(struct ieee802 rtlhal->driver_is_goingto_unload = true; rtlhal->enter_pnp_sleep = true; -rtl_lps_leave (hw); +rtl_lps_leave (hw, true);rtl_op_stop (hw);device_set_wakeup_enable (wiphy_dev (hw->wiphy), true);return 0 ;
@@ -1151,7 +1151,7 @@ static void rtl_op_bss_info_changed(stru mstatus =RT_MEDIA_DISCONNECT ;if (mac->link_state ==MAC80211_LINKED ) -rtl_lps_leave (hw); +rtl_lps_leave (hw, true);if (ppsc->p2p_ps_info.p2p_ps_mode >P2P_PS_NONE )rtl_p2p_ps_cmd (hw,P2P_PS_DISABLE ); mac->link_state =MAC80211_NOLINK ;
@@ -1448,7 +1448,7 @@ static void rtl_op_sw_scan_start(struct }if (mac->link_state ==MAC80211_LINKED ) { -rtl_lps_leave (hw); +rtl_lps_leave (hw, true); mac->link_state =MAC80211_LINKED_SCANNING ; }else {rtl_ips_nic_on (hw);
drivers/net/wireless/realtek/rtlwifi/pci.c +2 -2
@@ -622,7 +622,7 @@ static void _rtl_pci_tx_isr(struct ieee8if (((rtlpriv->link_info.num_rx_inperiod + rtlpriv->link_info.num_tx_inperiod) >8 ) || rtlpriv->link_info.num_rx_inperiod >2 ) -rtl_lps_leave (hw); +rtl_lps_leave (hw, false); } static int _rtl_pci_init_one_rxdesc(struct ieee80211_hw *hw,
@@ -875,7 +875,7 @@ static void _rtl_pci_rx_interrupt(structif (((rtlpriv->link_info.num_rx_inperiod + rtlpriv->link_info.num_tx_inperiod) >8 ) || rtlpriv->link_info.num_rx_inperiod >2 ) -rtl_lps_leave (hw); +rtl_lps_leave (hw, false); skb = new_skb; no_new:if (rtlpriv->use_new_trx_flow) {
drivers/net/wireless/realtek/rtlwifi/ps.c +4 -4
@@ -653,22 +653,22 @@ void rtl_lps_change_work_callback(struct }EXPORT_SYMBOL_GPL (rtl_lps_change_work_callback); -void rtl_lps_enter (struct ieee80211_hw *hw) +void rtl_lps_enter (struct ieee80211_hw *hw,bool may_block) {struct rtl_priv *rtlpriv =rtl_priv (hw); -if (!in_interrupt ()) +if (may_block)return rtl_lps_enter_core (hw); rtlpriv->enter_ps = true;schedule_work (&rtlpriv->works.lps_change_work); }EXPORT_SYMBOL_GPL (rtl_lps_enter); -void rtl_lps_leave (struct ieee80211_hw *hw) +void rtl_lps_leave (struct ieee80211_hw *hw,bool may_block) {struct rtl_priv *rtlpriv =rtl_priv (hw); -if (!in_interrupt ()) +if (may_block)return rtl_lps_leave_core (hw); rtlpriv->enter_ps = false;schedule_work (&rtlpriv->works.lps_change_work);
drivers/net/wireless/realtek/rtlwifi/ps.h +2 -2
@@ -11,8 +11,8 @@ bool rtl_ps_disable_nic(struct ieee80211void rtl_ips_nic_off (struct ieee80211_hw *hw);void rtl_ips_nic_on (struct ieee80211_hw *hw);void rtl_ips_nic_off_wq_callback (struct work_struct *work); -void rtl_lps_enter (struct ieee80211_hw *hw); -void rtl_lps_leave (struct ieee80211_hw *hw); +void rtl_lps_enter (struct ieee80211_hw *hw,bool may_block); +void rtl_lps_leave (struct ieee80211_hw *hw,bool may_block);void rtl_lps_set_psmode (struct ieee80211_hw *hw,u8 rt_psmode);
Quoted 6 lines · Thomas Gleixner
Folks,
in the discussion about preempt count consistency accross kernel configurations:
https://lore.kernel.org/r/20200914204209.256266093 at linutronix.de/
Quoted 8 lines
Linus clearly requested that code in drivers and libraries which changes behaviour based on execution context should either be split up so that e.g. task context invocations and BH invocations have different interfaces or if that's not possible the context information has to be provided by the caller which knows in which context it is executing.
This includes conditional locking, allocation mode (GFP_*) decisions and avoidance of code paths which might sleep.
Are these patches intended to be applied to Linus' tree before v5.9 is released?
This patchset will cause conflicts against -next.
For instance, in patch 34, RT_TRACE has already
been removed in -next.
David Miller
maintainerFrom: Thomas Gleixner <[email protected]>
Date: Sun, 27 Sep 2020 21:48:46 +0200
Quoted 1 line
in the discussion about preempt count consistency accross kernel configurations:
Please respin this against net-next, some of the patches in here are already in net-next (the wireless debug macro one) and even after that the series doesn't build:
drivers/net/ethernet/cisco/enic/enic_main.c: In function ?enic_reset?:drivers/net/ethernet/cisco/enic/enic_main.c:2315:2: error: implicit declaration of function ?enic_set_api_state?; did you mean ?enic_set_api_busy?? [-Werror=implicit-function-declaration]
2315 | enic_set_api_state(enic, true);
| ^~~~~~~~~~~~~~~~~~
| enic_set_api_busy
At top level:
drivers/net/ethernet/cisco/enic/enic_main.c:2298:13: warning: ?enic_set_api_busy? defined but not used [-Wunused-function]
2298 | static void enic_set_api_busy(struct enic *enic, bool busy)
| ^~~~~~~~~~~~~~~~~Quoted 8 lines · David Miller
From: Thomas Gleixner <[email protected]>
Date: Sun, 27 Sep 2020 21:48:46 +0200Quoted 1 line
in the discussion about preempt count consistency accross kernel configurations:
Please respin this against net-next, some of the patches in here are already in net-next (the wireless debug macro one) and even after that the series doesn't build:
Will do.
Quoted 9 lines
drivers/net/ethernet/cisco/enic/enic_main.c: In function ?enic_reset?:drivers/net/ethernet/cisco/enic/enic_main.c:2315:2: error: implicit declaration of function ?enic_set_api_state?; did you mean ?enic_set_api_busy?? [-Werror=implicit-function-declaration]2315 | enic_set_api_state(enic, true); | ^~~~~~~~~~~~~~~~~~ | enic_set_api_busy At top level: drivers/net/ethernet/cisco/enic/enic_main.c:2298:13: warning: ?enic_set_api_busy? defined but not used [-Wunused-function] 2298 | static void enic_set_api_busy(struct enic *enic, bool busy) | ^~~~~~~~~~~~~~~~~
Duh, not sure how I managed that. Sorry. will fix and rebase.
Thanks,
tglx
cover net: in_interrupt() cleanup and fixes
In reply to Thomas GleixnerQuoted 81 lines · Thomas Gleixner
Folks,
in the discussion about preempt count consistency accross kernel configurations:
https://lore.kernel.org/r/20200914204209.256266093 at linutronix.de/
Linus clearly requested that code in drivers and libraries which changes behaviour based on execution context should either be split up so that e.g. task context invocations and BH invocations have different interfaces or if that's not possible the context information has to be provided by the caller which knows in which context it is executing.
This includes conditional locking, allocation mode (GFP_*) decisions and avoidance of code paths which might sleep.
In the long run, usage of 'preemptible, in_*irq etc.' should be banned from driver code completely.
Our team started to dig through drivers and this it the first batch of cleanups in
drivers/net/. It's not yet complete, so expect further patches in the next days.The series contains:
- A couple of bug fixes
- Removal of the bitrotting CAIF SPI driver which has never had a matching driver providing the necessary platform device support.
- Removal of WARN/BUG(in_interrupt()) en masse as most of them are incomplete because they won't detect other non-preemptible context. All of the functions which have these WARN/BUG invoke core code functions which can sleep. These have plenty of checks to catch _all_ invalid contexts. So it's pointless to have incomplete WARN/BUG in the drivers.
If a driver wants to have such a check for paranoia reasons, then e.g.
lockdep_assert_preemtion_enabled()is the right mechanism to chose because lockdep guarantees to catch all invalid contexts independent of kernel configuration while e.g.preemptible()does not.- Conversion of
in_interrupt()checks to use either different functions or to hand the context information in from the caller.- For some drivers handing the context into functions which decided between
netif_rx()andnetif_rx_ni()turned out to be impossible due to lack of driver knowledge and convoluted code pathes with multiple indirections. For those a core code functionnetif_rx_any_context()is provided which contains anin_interrupt()check as a stop gap. This allows to make progess on the driver side cleanup and the function should go away once the driver wizards have fixed it up proper.- Simplifcation and cleanups in various places where code pointlessly contains
in_interrupt()conditionals which are mostly leftovers from calling conventions in older kernels and have never been cleaned up.Along with removing if from the horrible
DBG_FOO()macro mess which probably should be removed completely as the kernel today provides way more sensible mechanisms to do function tracing and similar.- A few other cleanups which were obvious when chasing the
in_interrupt()usage.The pile is also available from:
git://git.kernel.org/pub/scm/linux/kernel/git/tglx/devel.gitsoftirqThe diffstat summary is:
86 files changed, 300 insertions(+), 2233 deletions(-)which is biased by the CAIF SPI removal. Without that it is:
79 files changed, 300 insertions(+), 697 deletions(-)Thanks,
tglx
---
[...]
Quoted 24 lines
b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/bcmsdh.c | 4 b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/bus.h | 5 b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/core.c | 20 b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/fweh.c | 8 b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/fweh.h | 7 b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/msgbuf.c | 2 b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/sdio.c | 12 b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/sdio.h | 2 b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/usb.c | 2 b/drivers/net/wireless/intel/ipw2x00/ipw2100.c | 3 b/drivers/net/wireless/intel/ipw2x00/ipw2200.h | 6 b/drivers/net/wireless/intel/ipw2x00/libipw.h | 3 b/drivers/net/wireless/intel/iwlegacy/common.h | 4 b/drivers/net/wireless/intel/iwlwifi/iwl-debug.c | 5 b/drivers/net/wireless/intel/iwlwifi/iwl-devtrace-msg.h | 6 b/drivers/net/wireless/intersil/hostap/hostap_hw.c | 12 b/drivers/net/wireless/marvell/libertas/defs.h | 3 b/drivers/net/wireless/marvell/libertas/rx.c | 11 b/drivers/net/wireless/marvell/libertas_tf/deb_defs.h | 3 b/drivers/net/wireless/marvell/mwifiex/uap_txrx.c | 6 b/drivers/net/wireless/marvell/mwifiex/util.c | 6 b/drivers/net/wireless/realtek/rtlwifi/base.c | 47 b/drivers/net/wireless/realtek/rtlwifi/base.h | 3 b/drivers/net/wireless/realtek/rtlwifi/btcoexist/halbtcoutsrc.c | 12
Quoted 8 lines
b/drivers/net/wireless/realtek/rtlwifi/core.c | 6 b/drivers/net/wireless/realtek/rtlwifi/debug.c | 20 b/drivers/net/wireless/realtek/rtlwifi/debug.h | 6 b/drivers/net/wireless/realtek/rtlwifi/pci.c | 4 b/drivers/net/wireless/realtek/rtlwifi/ps.c | 27 b/drivers/net/wireless/realtek/rtlwifi/ps.h | 10 b/drivers/net/wireless/realtek/rtlwifi/wifi.h | 3 b/drivers/net/wireless/zydas/zd1211rw/zd_usb.c | 1
For the wireless patches:
I assume Dave will take them, but just let me know if I should take them instead.
Signature · 3 lines
https://patchwork.kernel.org/project/linux-wireless/list/ https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches