06 Apr, 2016
21 commits
-
If the driver advertises the new HW flag USE_RSS, make the
station statistics on the fast-rx path per-CPU. This will
enable calling the RX in parallel, only hitting locking or
shared cachelines when the fast-RX path isn't available.Signed-off-by: Johannes Berg
-
The regular RX path has a lot of code, but with a few
assumptions on the hardware it's possible to reduce the
amount of code significantly. Currently the assumptions
on the driver are the following:
* hardware/driver reordering buffer (if supporting aggregation)
* hardware/driver decryption & PN checking (if using encryption)
* hardware/driver did de-duplication
* hardware/driver did A-MSDU deaggregation
* AP_LINK_PS is used (in AP mode)
* no client powersave handling in mac80211 (in client mode)of which some are actually checked per packet:
* de-duplication
* PN checking
* decryption
and additionally packets must
* not be A-MSDU (have been deaggregated by driver/device)
* be data packets
* not be fragmented
* be unicast
* have RFC 1042 headerAdditionally dynamically we assume:
* no encryption or CCMP/GCMP, TKIP/WEP/other not allowed
* station must be authorized
* 4-addr format not enabledSome data needed for the RX path is cached in a new per-station
"fast_rx" structure, so that we only need to look at this and
the packet, no other memory when processing packets on the fast
RX path.After doing the above per-packet checks, the data path collapses
down to a pretty simple conversion function taking advantage of
the data cached in the small fast_rx struct.This should speed up the RX processing, and will make it easier
to reason about parallelizing RX (for which statistics will need
to be per-CPU still.)Signed-off-by: Johannes Berg
-
On 32-bit platforms, the 64-bit counters we keep need to be protected
to be consistently read. Use the u64_stats_sync mechanism to do that.In order to not end up with overly long lines, refactor the tidstats
assignments a bit.Signed-off-by: Johannes Berg
-
When storing the last_rate_* values in the RX code, there's nothing
to guarantee consistency, so a concurrent reader could see, e.g.
last_rate_idx on the new value, but last_rate_flag still on the old,
getting completely bogus values in the end.To fix this, I lifted the sta_stats_encode_rate() function from my
old rate statistics code, which encodes the entire rate data into a
single 16-bit value, avoiding the consistency issue.Signed-off-by: Johannes Berg
-
Instead of touching the rx_stats.last_rx from the status path, introduce
and use a status_stats.last_ack variable. This will make rx_stats.last_rx
indicate when the last frame was received, making it available for real
"last_rx" and statistics gathering; statistics, when done per-CPU, will
need to figure out which place was updated last for those items where the
"last" value is exposed.Signed-off-by: Johannes Berg
-
There's no need to update rx_stats.last_rx after allocating
a station since it's already updated during allocation.Signed-off-by: Johannes Berg
-
Move the averaged values out of rx_stats and into rx_stats_avg,
to cleanly split them out. The averaged ones cannot be supported
for parallel RX in a per-CPU fashion, while the other values can
be collected per CPU and then combined/selected when needed.Signed-off-by: Johannes Berg
-
Move the semicolon, people typically assume that and
once line already put a semicolon behind the "call".Signed-off-by: Johannes Berg
-
For the RX MSDU statistics, we need to count the number of
MSDUs created and accepted from an A-MSDU. Right now, all
frames in any A-MSDUs were completely ignored. Fix this by
moving the RX MSDU statistics accounting into the deliver
function.Signed-off-by: Johannes Berg
-
Sometimes drivers already looked up, or know out-of-band
from their device, which station transmitted a given RX
frame. Allow them to pass the station pointer to mac80211
to save the extra lookup.Signed-off-by: Johannes Berg
-
Lockdep warned of a lock dependency between the mesh_plink lock
and the internal lock for the rhashtable. The problem is that
the rhashtable code uses a spin lock with softirqs enabled, while
mesh_plink_timer executes a walk (to flush paths on a state change)
inside a softirq with the plink lock held.This leads to the following deadlock if the timer fires while rht
lock is held on this CPU, and plink lock is held on another CPU:CPU0 CPU1
---- ----
lock(&(&ht->lock)->rlock);
local_irq_disable();
lock(&(&sta->mesh->plink_lock)->rlock);
lock(&(&ht->lock)->rlock);
lock(&(&sta->mesh->plink_lock)->rlock);
*** DEADLOCK ***Fix by waiting until we drop the plink lock to flush paths.
Fixes: d48a1b7cd439 ("mac80211: mesh: convert path table to rhashtable")
Signed-off-by: Bob Copeland
Signed-off-by: Johannes Berg -
The mesh path table needs to be around for the entire time the
interface is in mesh mode, as users can perform an mpath dump
at any time. The existing path table lifetime is instead tied
to the mesh BSS which can cause crashes when different MBSSes
are joined in the context of a single interface, or when the
path table is dumped when no MBSS is joined.Introduce a new function to perform the final teardown of the
interface and perform path table cleanup there. We already
free the individual path elements when the leaving the mesh
so no additional cleanup is needed there. This fixes the
following crash:[ 47.753026] BUG: unable to handle kernel paging request at fffffff0
[ 47.753026] IP: [] kthread_data+0xa/0xe
[ 47.753026] *pde = 00741067 *pte = 00000000
[ 47.753026] Oops: 0000 [#4] PREEMPT
[ 47.753026] Modules linked in: ppp_generic slhc 8021q garp mrp sch_fq_codel iptable_mangle ipt_MASQUERADE nf_nat_masquerade_ipv4 iptable_nat nf_conntrack_ipv4 nf_defrag_ipv4 nf_nat_ipv4 nf_nat ip_tables ath9k_htc ath5k 8139too ath10k_pci ath10k_core arc4 ath9k ath9k_common ath9k_hw mac80211 ath cfg80211 cpufreq_powersave br_netfilter bridge stp llc ipw usb_wwan sierra_net usbnet af_alg natsemi via_rhine mii iTCO_wdt iTCO_vendor_support gpio_ich sierra coretemp pcspkr i2c_i801 lpc_ich ata_generic ata_piix libata ide_pci_generic piix e1000e igb i2c_algo_bit ptp pps_core [last unloaded: 8139too]
[ 47.753026] CPU: 0 PID: 12 Comm: kworker/u2:1 Tainted: G D W 4.5.0-wt-V3 #6
[ 47.753026] Hardware name: To Be Filled By O.E.M./To be filled by O.E.M., BIOS 080016 11/07/2014
[ 47.753026] task: f645a0c0 ti: f6462000 task.ti: f6462000
[ 47.753026] EIP: 0060:[] EFLAGS: 00010002 CPU: 0
[ 47.753026] EIP is at kthread_data+0xa/0xe
[ 47.753026] EAX: 00000000 EBX: 00000000 ECX: 00000000 EDX: 00000000
[ 47.753026] ESI: f645a0c0 EDI: f645a2fc EBP: f6463a80 ESP: f6463a78
[ 47.753026] DS: 007b ES: 007b FS: 0000 GS: 0000 SS: 0068
[ 47.753026] CR0: 8005003b CR2: 00000014 CR3: 353e5000 CR4: 00000690
[ 47.753026] Stack:
[ 47.753026] c0236866 00000000 f6463aac c05768b4 00000009 f6463ba8 f6463ab0 c0247010
[ 47.753026] 00000000 f645a0c0 f6464000 00000009 f6463ba8 f6463ab8 c0576eb2 f645a0c0
[ 47.753026] f6463aec c0228be4 c06335a4 f6463adc f6463ad0 c06c06d4 f6463ae4 c02471b0
[ 47.753026] Call Trace:
[ 47.753026] [] ? wq_worker_sleeping+0xb/0x78
[ 47.753026] [] __schedule+0xda/0x587
[ 47.753026] [] ? vprintk_default+0x12/0x14
[ 47.753026] [] schedule+0x72/0x89
[ 47.753026] [] do_exit+0xb8/0x71d
[ 47.753026] [] ? kmsg_dump+0xa9/0xae
[ 47.753026] [] oops_end+0x69/0x70
[ 47.753026] [] no_context+0x1bb/0x1c5
[ 47.753026] [] __bad_area_nosemaphore+0x136/0x140
[ 47.753026] [] ? vmalloc_sync_all+0x19a/0x19a
[ 47.753026] [] bad_area_nosemaphore+0xd/0x10
[ 47.753026] [] __do_page_fault+0x26c/0x320
[ 47.753026] [] ? vmalloc_sync_all+0x19a/0x19a
[ 47.753026] [] do_page_fault+0xb/0xd
[ 47.753026] [] error_code+0x58/0x60
[ 47.753026] [] ? vmalloc_sync_all+0x19a/0x19a
[ 47.753026] [] ? kthread_data+0xa/0xe
[ 47.753026] [] ? wq_worker_sleeping+0xb/0x78
[ 47.753026] [] __schedule+0xda/0x587
[ 47.753026] [] ? vprintk_default+0x12/0x14
[ 47.753026] [] schedule+0x72/0x89
[ 47.753026] [] do_exit+0xb8/0x71d
[ 47.753026] [] ? kmsg_dump+0xa9/0xae
[ 47.753026] [] oops_end+0x69/0x70
[ 47.753026] [] no_context+0x1bb/0x1c5
[ 47.753026] [] __bad_area_nosemaphore+0x136/0x140
[ 47.753026] [] ? vmalloc_sync_all+0x19a/0x19a
[ 47.753026] [] bad_area_nosemaphore+0xd/0x10
[ 47.753026] [] __do_page_fault+0x26c/0x320
[ 47.753026] [] ? vmalloc_sync_all+0x19a/0x19a
[ 47.753026] [] do_page_fault+0xb/0xd
[ 47.753026] [] error_code+0x58/0x60
[ 47.753026] [] ? vmalloc_sync_all+0x19a/0x19a
[ 47.753026] [] ? kthread_data+0xa/0xe
[ 47.753026] [] ? wq_worker_sleeping+0xb/0x78
[ 47.753026] [] __schedule+0xda/0x587
[ 47.753026] [] ? put_io_context_active+0x6d/0x95
[ 47.753026] [] schedule+0x72/0x89
[ 47.753026] [] do_exit+0x6cc/0x71d
[ 47.753026] [] oops_end+0x69/0x70
[ 47.753026] [] no_context+0x1bb/0x1c5
[ 47.753026] [] __bad_area_nosemaphore+0x136/0x140
[ 47.753026] [] ? vmalloc_sync_all+0x19a/0x19a
[ 47.753026] [] bad_area_nosemaphore+0xd/0x10
[ 47.753026] [] __do_page_fault+0x26c/0x320
[ 47.753026] [] ? debug_smp_processor_id+0x12/0x16
[ 47.753026] [] ? __switch_to+0x24/0x40e
[ 47.753026] [] ? vmalloc_sync_all+0x19a/0x19a
[ 47.753026] [] do_page_fault+0xb/0xd
[ 47.753026] [] error_code+0x58/0x60
[ 47.753026] [] ? vmalloc_sync_all+0x19a/0x19a
[ 47.753026] [] ? rhashtable_walk_init+0x5c/0x93
[ 47.753026] [] mesh_path_tbl_expire.isra.24+0x19/0x82 [mac80211]
[ 47.753026] [] mesh_path_expire+0x11/0x1f [mac80211]
[ 47.753026] [] ieee80211_mesh_work+0x73/0x1a9 [mac80211]
[ 47.753026] [] ieee80211_iface_work+0x2ff/0x311 [mac80211]
[ 47.753026] [] process_one_work+0x14b/0x24e
[ 47.753026] [] worker_thread+0x249/0x343
[ 47.753026] [] ? process_scheduled_works+0x24/0x24
[ 47.753026] [] kthread+0x9e/0xa3
[ 47.753026] [] ret_from_kernel_thread+0x20/0x40
[ 47.753026] [] ? kthread_parkme+0x18/0x18
[ 47.753026] Code: 6b c0 85 c0 75 05 e8 fb 74 fc ff 89 f8 84 c0 75 08 8d 45 e8 e8 34 dd 33 00 83 c4 28 5b 5e 5f 5d c3 55 8b 80 10 02 00 00 89 e5 5d 40 f0 c3 55 b9 04 00 00 00 89 e5 52 8b 90 10 02 00 00 8d 45
[ 47.753026] EIP: [] kthread_data+0xa/0xe SS:ESP 0068:f6463a78
[ 47.753026] CR2: 00000000fffffff0
[ 47.753026] ---[ end trace 867ca0bdd0767790 ]---Fixes: 3b302ada7f0a ("mac80211: mesh: move path tables into if_mesh")
Reported-by: Fred Veldini
Signed-off-by: Bob Copeland
Signed-off-by: Johannes Berg -
Several of the mesh path fields are undocumented and some
of the documentation is no longer correct or relevant after
the switch to rhashtable. Clean up the kernel doc
accordingly and reorder some fields to match the structure
layout.Signed-off-by: Bob Copeland
Signed-off-by: Johannes Berg -
Reduce padding waste in struct mesh_table and struct rmc_entry by
moving the smaller fields to the end.Signed-off-by: Bob Copeland
Signed-off-by: Johannes Berg -
Since we have converted the mesh path tables to rhashtable, we are
no longer swapping out the entire mesh_pathtbl pointer with RCU.
As a result, we no longer need indirection to the hlist head for
the gates list and can simply embed it, saving a pair of
pointer-sized allocations.Signed-off-by: Bob Copeland
Signed-off-by: Johannes Berg -
The RMC cache has 256 list heads plus a u32, which puts it at the
unfortunate size of 4104 bytes with padding. kmalloc() will then
round this up to the next power-of-two, so we wind up actually
using two pages here where most of the second is wasted.Switch to hlist heads here to reduce the structure size down to
fit within a page.Signed-off-by: Bob Copeland
Signed-off-by: Johannes Berg -
In the unlikely case that mesh_rmc_init() fails with -ENOMEM,
the rmc pointer will be left as NULL but the interface is still
operational because ieee80211_mesh_init_sdata() is not allowed
to fail.If this happens, we would blindly dereference rmc when checking
whether a multicast frame is in the cache. Instead just drop the
frames in the forwarding path.Signed-off-by: Bob Copeland
Signed-off-by: Johannes Berg -
The mesh_path_reclaim() function, called from an rcu callback, cancels
the mesh_path_timer associated with a mesh path. Unfortunately, this
call can happen much later, perhaps after the hash table itself is
destroyed.Such a situation led to the following crash in mesh_path_send_to_gates()
when dereferencing the tbl pointer:[ 23.901661] BUG: unable to handle kernel NULL pointer dereference at 0000000000000008
[ 23.905516] IP: [] mesh_path_send_to_gates+0x2b/0x740
[ 23.908757] PGD 99ca067 PUD 99c4067 PMD 0
[ 23.910789] Oops: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
[ 23.913485] CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.5.0-rc6-wt+ #43
[ 23.916675] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Debian-1.8.2-1 04/01/2014
[ 23.920471] task: ffffffff81685500 ti: ffffffff81678000 task.ti: ffffffff81678000
[ 23.922619] RIP: 0010:[] [] mesh_path_send_to_gates+0x2b/0x740
[ 23.925237] RSP: 0018:ffff88000b403d30 EFLAGS: 00010286
[ 23.926739] RAX: 0000000000000000 RBX: ffff880009bc0d20 RCX: 0000000000000102
[ 23.928796] RDX: 000000000000002e RSI: 0000000000000001 RDI: ffff880009bc0d20
[ 23.930895] RBP: ffff88000b403e18 R08: 0000000000000001 R09: 0000000000000001
[ 23.932917] R10: 0000000000000000 R11: 0000000000000001 R12: ffff880009c20940
[ 23.936370] R13: ffff880009bc0e70 R14: ffff880009c21c40 R15: ffff880009bc0d20
[ 23.939823] FS: 0000000000000000(0000) GS:ffff88000b400000(0000) knlGS:0000000000000000
[ 23.943688] CS: 0010 DS: 0000 ES: 0000 CR0: 000000008005003b
[ 23.946429] CR2: 0000000000000008 CR3: 00000000099c5000 CR4: 00000000000006b0
[ 23.949861] Stack:
[ 23.950840] 000000000000002e ffff880009c20940 ffff88000b403da8 ffffffff8109e551
[ 23.954467] ffffffff82711be2 000000000000002e 0000000000000000 ffffffff8166a5f5
[ 23.958141] 0000000000685ce8 0000000000000246 ffff880009bc0d20 ffff880009c20940
[ 23.961801] Call Trace:
[ 23.962987]
[ 23.963963] [] ? vprintk_emit+0x351/0x5e0
[ 23.966782] [] ? vprintk_default+0x1f/0x30
[ 23.969529] [] ? printk+0x48/0x50
[ 23.971956] [] mesh_path_timer+0x133/0x160
[ 23.974707] [] ? mesh_nexthop_resolve+0x230/0x230
[ 23.977775] [] call_timer_fn+0xce/0x330
[ 23.980448] [] ? call_timer_fn+0x5/0x330
[ 23.983126] [] ? mesh_nexthop_resolve+0x230/0x230
[ 23.986091] [] run_timer_softirq+0x22c/0x390Instead of cancelling in the RCU callback, set a new flag to prevent the
timer from being rearmed, and then cancel the timer synchronously when
freeing the mesh path. This leaves mesh_path_reclaim() doing nothing
but kfree, so switch to kfree_rcu().Fixes: 3b302ada7f0a ("mac80211: mesh: move path tables into if_mesh")
Signed-off-by: Bob Copeland
Signed-off-by: Johannes Berg -
Legacy clients don't support P2P power save mechanism, and thus if a P2P GO
has a legacy client connected to it, it should disable P2P PS mechanisms.
Let the driver know about this with a new bss_conf parameter.Signed-off-by: Ayala Beker
Signed-off-by: Emmanuel Grumbach
Signed-off-by: Johannes Berg -
Legacy clients don't support P2P power save mechanisms, and thus
if a P2P GO has a legacy client connected to it, it has to make
some changes in the PS behavior.To handle this, add an attribute to specify whether a station supports
P2P PS or not. If the attribute was not specified cfg80211 will assume
that station supports it for P2P GO interface, and does NOT support it
for AP interface, matching the current assumptions in the code.Signed-off-by: Ayala Beker
Signed-off-by: Emmanuel Grumbach
Signed-off-by: Johannes Berg -
In the likely case that probe_count is 0, don't write to the
memory there.Also use ifmgd consistently in the function, instead of using
sdata->u.mgd as well.Signed-off-by: Johannes Berg
Signed-off-by: Emmanuel Grumbach
Signed-off-by: Johannes Berg
05 Apr, 2016
19 commits
-
The code is only used with iwlwifi, but still should have proper
mac80211 naming scheme; fix that.Signed-off-by: Johannes Berg
-
Avoid the really strange %s%s%s expression, use an array
of flag names and check that all flags are present.Signed-off-by: Johannes Berg
-
If the device implements dynamic PS itself, there's no need
to ever start the dynamic powersave timer on RX.While at it, fix up some indentation in this code.
Signed-off-by: Johannes Berg
-
Since the previous patch, the struct only has a single member,
so remove the struct and leave just the single member.Signed-off-by: Johannes Berg
-
Remove unused variable in per STA debugfs structure, 'commit 34e895075e21
("mac80211: allow station add/remove to sleep")' removed the only user of
'add_has_run'.Signed-off-by: Mohammed Shafi Shajakhan
Signed-off-by: Johannes Berg -
Currently the debugfs entry for starting aggregation session
starts it with timeout of 5 seconds. Allow opening a session
with a custom timeout (according to spec 0 is no timeout).
while at it, refactor the function and remove the magic numbers.Signed-off-by: Sara Sharon
Signed-off-by: Emmanuel Grumbach
Signed-off-by: Johannes Berg -
NETIF_F_RXCSUM is not in the white list, though some
drivers may want to set it in order to enable seeing the
actual RX checksum status in ethtool.Signed-off-by: Sara Sharon
Signed-off-by: Emmanuel Grumbach
Signed-off-by: Johannes Berg -
Allow publishing RRM capabilities for features that are not
HW dependent.Signed-off-by: Emmanuel Grumbach
Signed-off-by: Johannes Berg -
There were a few issues that were slowing down the process of finding
the optimal rate, especially on devices with multi-rate retry
limitations:When max_tp_rate[0] was slower than max_tp_rate[1], the code did not
sample max_tp_rate[1], which would often allow it to switch places with
max_tp_rate[0] (e.g. if only the first sampling attempts were bad, but the
rate is otherwise good).Also, sample attempts of rates between max_tp_rate[0] and [1] were being
ignored in this case, because the code only checked if the rate was
slower than [1].Fix this by checking against the fastest / second fastest max_tp_rate
instead of assuming a specific order between the two.In my tests this patch significantly reduces the time until minstrel_ht
finds the optimal rate right after assocSigned-off-by: Felix Fietkau
Signed-off-by: Johannes Berg -
Fall back to rate control if the requested bitrate was not found.
Fixes: dfdfc2beb0dd ("mac80211: Parse legacy and HT rate in injected frames")
Signed-off-by: Felix Fietkau
Signed-off-by: Johannes Berg -
The MCS bandwidth part of the radiotap header is 2 bits wide. The full 2
bit have to compared against IEEE80211_RADIOTAP_MCS_BW_40 and not only if
the first bit is set. Otherwise IEEE80211_RADIOTAP_MCS_BW_40 can be
confused with IEEE80211_RADIOTAP_MCS_BW_20U.Fixes: dfdfc2beb0dd ("mac80211: Parse legacy and HT rate in injected frames")
Signed-off-by: Sven Eckelmann
Signed-off-by: Johannes Berg -
Introducing a new feature that the driver can use to
indicate the driver/firmware supports configuration of BSS
selection criteria upon CONNECT command. This can be useful
when multiple BSS-es are found belonging to the same ESS,
ie. Infra-BSS with same SSID. The criteria can then be used to
offload selection of a preferred BSS.Reviewed-by: Hante Meuleman
Reviewed-by: Franky (Zhenhui) Lin
Reviewed-by: Pieter-Paul Giesberts
Reviewed-by: Lei Zhang
Signed-off-by: Arend van Spriel
[move wiphy support check into parse_bss_select()]
Signed-off-by: Johannes Berg -
Some devices, like iwlwifi, have RSS queues. This may cause a
situation where a disassociation is handled in control path and
results in station removal while there are prior RX frames
that were still not processed in other queues. When they will
be processed the station will be gone, and the frames will be
dropped.
Add a synchronization interface to avoid that. When driver returns
from the synchronization mac80211 may remove the station.Signed-off-by: Sara Sharon
Signed-off-by: Emmanuel Grumbach
Signed-off-by: Johannes Berg -
In the time since the mesh path table was implemented as an
RCU-traversable, dynamically growing hash table, a generic RCU
hashtable implementation was added to the kernel.Switch the mesh path table over to rhashtable to remove some code
and also gain some features like automatic shrinking.Cc: Thomas Graf
Cc: netdev@vger.kernel.org
Signed-off-by: Bob Copeland
Signed-off-by: Johannes Berg -
In certain cases, the 802.11 mesh pathtable code wants to
iterate over all of the entries in the forwarding table from
the receive path, which is inside an RCU read-side critical
section. Enable walks inside atomic sections by allowing
GFP_ATOMIC allocations for the walker state.Change all existing callsites to pass in GFP_KERNEL.
Acked-by: Thomas Graf
Signed-off-by: Bob Copeland
[also adjust gfs2/glock.c and rhashtable tests]
Signed-off-by: Johannes Berg -
The mesh path table uses a struct mesh_node in its hlists in
order to support a resizable hash table: the mesh_node provides
an indirection to the actual mesh path so that two different
bucket lists can point to the same path entry.However, for the known gates list, we don't need this indirection
because there is ever only one list. So we can just embed the
hlist_node in the mesh path itself, which simplifies things a bit
and saves a linear search whenever we need to find an item in
the list.Signed-off-by: Bob Copeland
Signed-off-by: Johannes Berg -
Remove duplicate code to allocate and initialize a mesh
path or mesh proxy path.Signed-off-by: Bob Copeland
Signed-off-by: Johannes Berg -
Now that the sdata pointer is the same for all entries of a
path table, hashing it is pointless, so hash only the address.Signed-off-by: Bob Copeland
Signed-off-by: Johannes Berg -
The mesh path and mesh gate hashtables are global, containing
all of the mpaths for every mesh interface, but the paths are
all tied logically to a single interface. The common case is
just a single mesh interface, so optimize for that by moving
the global hashtable into the per-interface struct.Doing so allows us to drop sdata pointer comparisons inside
the lookups and also saves a few bytes of BSS and data.Signed-off-by: Bob Copeland
Signed-off-by: Johannes Berg
