- 01 May, 2019 40 commits
-
-
Felix Fietkau authored
It is simpler in this case Signed-off-by:Felix Fietkau <nbd@nbd.name>
-
Felix Fietkau authored
The previous version only accidentally disabled A-MSDU deaggregation by using the wrong mask for rx length configuration, which left previous length value in place. Fix the length and initialize the register completely to keep A-MSDU de-aggregation remaining disabled Signed-off-by: -
Felix Fietkau authored
Use a separate lock and spin_trylock to avoid disabling interrupts. Should improve performance and latency Signed-off-by: -
Felix Fietkau authored
This allows tx scheduling and tx cleanup to run concurrently Signed-off-by: -
Stanislaw Gruszka authored
mac80211 .start(), .stop() callbacks are never called concurrently with other callbacks. The only concurencly is with mt76 works which we cancel on stop() and schedule on start(). This fixes possible deadlock on cancel_delayed_work_sync(&dev->mac_work) as mac_work also take mutex. Signed-off-by:
Stanislaw Gruszka <sgruszka@redhat.com> Signed-off-by:
-
Stanislaw Gruszka authored
Disabling tasklets on stopping rx/tx is wrong. If blocked tasklet is scheduled and we remove device we will get 100% cpu usage: PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND 9 root 20 0 0 0 0 R 93.8 0.0 1:47.19 ksoftirqd/0 by infinite loop in tasklet_action_common() and eventuall crash on next mt76usb module load: [ 2068.591964] RIP: 0010:tasklet_action_common.isra.17+0x66/0x100 [ 2068.591966] Code: 41 89 f5 eb 25 f0 48 0f ba 33 00 0f 83 b1 00 00 00 48 8b 7a 20 48 8b 42 18 e8 56 a3 b5 00 f0 80 23 fd 48 89 ea 48 85 ed 74 53 <48> 8b 2a 48 8d 5a 08 f0 48 0f ba 6a 08 01 72 0b 8b 42 10 85 c0 74 [ 2068.591968] RSP: 0018:ffff98758c34be58 EFLAGS: 00010206 [ 2068.591969] RAX: ffff98758e6966d0 RBX: ffff98756e69aef8 RCX: 0000000000000006 [ 2068.591970] RDX: 01060a053d060305 RSI: 0000000000000006 RDI: ffff98758e6966d0 [ 2068.591971] RBP: 01060a053d060305 R08: 0000000000000000 R09: 00000000000203c0 [ 2068.591971] R10: 000003ff65b34f08 R11: 0000000000000001 R12: ffff98758e6966d0 [ 2068.591972] R13: 0000000000000006 R14: 0000000000000040 R15: 0000000000000006 [ 2068.591974] FS: 0000000000000000(0000) GS:ffff98758e680000(0000) knlGS:0000000000000000 [ 2068.591975] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 2068.591975] CR2: 00002c5f73a6cc20 CR3: 00000002f920a001 CR4: 00000000003606e0 [ 2068.591977] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 [ 2068.591978] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 [ 2068.591978] Call Trace: [ 2068.591985] __do_softirq+0xe3/0x30a [ 2068.591989] ? sort_range+0x20/0x20 [ 2068.591990] run_ksoftirqd+0x26/0x40 [ 2068.591992] smpboot_thread_fn+0xc5/0x160 [ 2068.591995] kthread+0x112/0x130 [ 2068.591997] ? kthread_create_on_node+0x40/0x40 [ 2068.591998] ret_from_fork+0x35/0x40 [ 2068.591999] Modules linked in: ccm arc4 fuse rfcomm cmac bnep sunrpc snd_hda_codec_hdmi snd_soc_skl snd_soc_core snd_soc_acpi_intel_match snd_hda_codec_realtek snd_soc_acpi snd_hda_codec_generic snd_soc_skl_ipc snd_soc_sst_ipc snd_soc_sst_dsp snd_hda_ext_core iTCO_wdt snd_hda_intel intel_rapl iTCO_vendor_support x86_pkg_temp_thermal intel_powerclamp btusb mei_wdt coretemp btrtl snd_hda_codec btbcm btintel intel_cstate snd_hwdep intel_uncore uvcvideo snd_hda_core videobuf2_vmalloc videobuf2_memops intel_rapl_perf wmi_bmof videobuf2_v4l2 intel_wmi_thunderbolt snd_seq bluetooth joydev videobuf2_common snd_seq_device snd_pcm videodev media i2c_i801 snd_timer idma64 ecdh_generic intel_lpss_pci intel_lpss mei_me mei ucsi_acpi typec_ucsi processor_thermal_device intel_soc_dts_iosf intel_pch_thermal typec thinkpad_acpi wmi snd soundcore rfkill int3403_thermal int340x_thermal_zone int3400_thermal acpi_thermal_rel acpi_pad pcc_cpufreq uas usb_storage crc32c_intel i915 i2c_algo_bit nvme serio_raw [ 2068.592033] drm_kms_helper e1000e nvme_core drm video ipv6 [last unloaded: cfg80211] Fortunate thing is that this not happen frequently, as scheduling tasklet on blocked state is very exceptional, though might happen. Due to different RX/TX tasklet processing fix is different for those. For RX we have to assure rx_tasklet do fail to resubmit buffers by poisoning urb's and kill the tasklet. For TX we need to handle all stop cases properly (suspend, module unload, device removal). Signed-off-by: -
Stanislaw Gruszka authored
On suspend mac80211 .stop callback is called before .suspend(), so hw mac is already stopped and we do not have to do this again. Signed-off-by:
-
Ryder Lee authored
Roy and I actively join the development and review. Signed-off-by:
Ryder Lee <ryder.lee@mediatek.com> Signed-off-by:
-
Ryder Lee authored
Use macro to convert sn and seq_ctrl for better readability. Signed-off-by:
Roy Luo <royluo@google.com> Signed-off-by:
-
Ryder Lee authored
In the tx/rx fastpath, the funciton dma_map_single() rarely fails. This adds unlikely() optimization to this error check conditional. Signed-off-by:
-
Ryder Lee authored
This driver is for a newer generation of MediaTek MT7615 4x4 802.11ac PCIe-based chipsets, which support wave2 MU-MIMO up to 4 users/group and also support up to 160MHz bandwidth. The driver fully supports AP, station and monitor mode. Signed-off-by:
Lorenzo Bianconi <lorenzo@kernel.org> Signed-off-by:
-
Lorenzo Bianconi authored
Introduce skb check for dummy address in mt76_dma_tx_cleanup_idx. This is a preliminary patch to add support for new chipsets (e.g. 7615) Signed-off-by:
-
Lorenzo Bianconi authored
Introduce skb field in mt76_txwi_cache. Moreover add txwi_flags to mt76_driver_ops since new chipsets will release mt76_txwi_cache/skbs at tx completion instead of dma one. This is a preliminary patch to add mt7615 support Signed-off-by:
-
Lorenzo Bianconi authored
Pass skb pointer to tx_prepare_skb through mt76_tx_info data structure. This is a preliminary patch to properly support dma error path for new chipsets (e.g. 7615) Signed-off-by:
-
Lorenzo Bianconi authored
Dynamically allocate txwi since new chipsets will use longer txwi descriptors Signed-off-by:
-
Lorenzo Bianconi authored
Similar to pci counterpart, reduce locking in mt76u_tx_tasklet since q->head is managed just in mt76u_tx_tasklet and q->queued is updated holding q->lock Signed-off-by:
-
Lorenzo Bianconi authored
Move mac_work delayed work in mt76_dev data structure since it is used by all drivers and it will be reused adding mac work to mt7615 Signed-off-by:
-
Felix Fietkau authored
Reduces lock contention from the tx path and improves performance Signed-off-by: -
Felix Fietkau authored
Allows it to be scheduled from core code Signed-off-by: -
Felix Fietkau authored
Signed-off-by: -
Felix Fietkau authored
q->tail can be safely updated without locking, because there is no concurrent access. If called from outside of the tasklet (for flushing), the tasklet is always disabled. q->queued can be safely read without locking, as long as the decrement happens within the locked section. This patch allows cleaning up tx packets outside of the section that holds the queue lock for improved performance Signed-off-by: -
Felix Fietkau authored
Performance improvement and preparation for adding airtime fairness support Signed-off-by: -
Felix Fietkau authored
Switching to readl/writel is faster because it gets rid of an unnecessary wrapper with extra checks. Signed-off-by: -
Felix Fietkau authored
- tx power is stored in the channels after ieee80211_register_hw, so chan->orig_mpwr needs to be updated as well - for non-TSSI devices, mt76x2e needs to use a different target power value from the EEPROM - fix a rounding error in a few places (need to round up, not down) Signed-off-by: -
Lorenzo Bianconi authored
Move mt76_tx_free in mt76_free_device routine in order to unmap all txwi descriptors at module unload Signed-off-by:
-
Stanislaw Gruszka authored
Since now only mt76u_get_next_rx_entry use queue argument move it to this function. Signed-off-by:
-
Stanislaw Gruszka authored
Alloc sg table at the end of urb structure. This will increase cache usage. Signed-off-by:
-
Stanislaw Gruszka authored
We already allocate with GFP_ZERO and sg marker is set later for both RX and TX. Signed-off-by:
-
Stanislaw Gruszka authored
Add new rx_urb_alloc routine and reuse common urb_alloc for tx allocations. Signed-off-by:
-
Stanislaw Gruszka authored
Get the RX queue inside mt76u_refill_rx. This will allow to reuse mt76u_urb_alloc for TX allocations. Signed-off-by:
-
Stanislaw Gruszka authored
Put urb pointer in mt76_queue_entry directly instead of mt76u_buf structure. Signed-off-by:
-
Stanislaw Gruszka authored
mt76_queue_entry has alreay one bool variable, adding new one will not increase it's size. Removing ->done filed from mt76u_buf will allow to use urb directly in mt76usb code. Signed-off-by:
-
Stanislaw Gruszka authored
Remove mt76u_buf->{len, buf} fields and operate on corresponding urb fields directly. Signed-off-by: -
Stanislaw Gruszka authored
Setup usb device private data. This allows to remove mt76u_buf->dev and simplify some routines as no longer we need to get usb device through usb interface. Signed-off-by:
-
Stanislaw Gruszka authored
We do not need to pass len and sglen to the function. Additionally pass gfp to control allocation context. Signed-off-by:
-
Stanislaw Gruszka authored
We can not run mt76u_alloc_buf() concurently, rx_tasklet is stooped when mt76u_submit_rx_buffers(). We can remove rx_page_lock. Signed-off-by:
-
Stanislaw Gruszka authored
Remove unnecessery arguments and change the function name since is now used only for RX. Signed-off-by:
-
Stanislaw Gruszka authored
Enable AP mode. For now without multi-vif support, this will require more testing and investigation. Signed-off-by:
-
Stanislaw Gruszka authored
All mt76 drivers (now also USB drivers) require empty .set_tim callback. Add it to common mt76 module and use on all drivers. Signed-off-by:
-
Stanislaw Gruszka authored
Create software MT_TXQ_PSD queue for USB and map it to MT_TXQ_VO since we do not have USB endpoint for PSD. This should make mt76_release_buffered_frames() work by sending released frames via MT_TXQ_VO. Signed-off-by:
-
