Miri Korenblit [Mon, 27 May 2024 16:06:14 +0000 (19:06 +0300)]
wifi: iwlwifi: move amsdu_size parsing to iwlwifi
The code that is parsing the amsdu_size module parameter and mapping it
to the corresponding Rx buffer size is common to all opmodes.
Move it into a function in iwlwifi, as preparation to a new op mode we
are working on.
Miri Korenblit [Mon, 27 May 2024 16:06:13 +0000 (19:06 +0300)]
wifi: iwlwifi: remove redundant prints
Upon start, the different opmodes are currently printing:
1. HW rev, which is already print by iwlwifi
2. The HW name (e.g. "Intel(R) Wi-Fi 7 BE201 320MHz")
cleanup things such as the hw rev won't be printed again, the HW name will
be printed by iwlwifi instead of each opmode.
Johannes Berg [Mon, 27 May 2024 16:06:10 +0000 (19:06 +0300)]
wifi: iwlwifi: mvm: show full firmware ID in debugfs
The firmware prefix is unused today, but it might still be
useful to have some information. Since the prefix will get
the intel/ directory in some cases, replace the prefix by
the full FW id which has all the information.
Johannes Berg [Mon, 27 May 2024 16:06:08 +0000 (19:06 +0300)]
wifi: iwlwifi: mvm: use only beacon BSS load for active links
For active links, don't take information that may have been
received in probe responses, as those are not protected. For
inactive links, there may not be a choice.
Modify iwl_mvm_ftm_set_secured_ranging() parameters to support
multiple versions of the target struct.
This is done as preparation for moving to the new range request
version.
Avraham Stern [Mon, 27 May 2024 16:06:04 +0000 (19:06 +0300)]
wifi: iwlwifi: mvm: initiator: move setting the sta_id into a function
Move setting the target's sta_id (and related flags) into a function
to support different versions of the target struct.
This is done as preparation for moving to the new range request
version.
Avraham Stern [Mon, 27 May 2024 16:06:02 +0000 (19:06 +0300)]
wifi: iwlwifi: mvm: initiator: move setting target flags into a function
Move setting the target flags into a dedicated function to support
different versions of the target struct.
This is done as preparation for moving to the new range request
version.
Johannes Berg [Mon, 27 May 2024 16:06:01 +0000 (19:06 +0300)]
wifi: iwlwifi: mvm: handle TPE advertised by AP
6 GHz BSS SP client shall respect TX power limits advertised
by the AP in TPE elements, send the data to the firmware using
the AP_TX_POWER_CONSTRAINTS_CMD command, so do that.
Use the new RX_FLAG_MACTIME_IS_RTAP_TS64 in mac80211 to report
a 64-bit timestamp in the mactime for radiotap only, in case we
report the synchronized PTP clock timestamp (otherwise we really
only have 32 bits anyway).
Signed-off-by: Johannes Berg <johannes.berg@intel.com> Reviewed-by: Gregory Greenman <gregory.greenman@intel.com> Reviewed-by: Benjamin Berg <benjamin.berg@intel.com> Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com> Link: https://msgid.link/20240527190228.4fc44356e7df.Ic8642d35f1090a415e09299a1bd409f51dfb6351@changeid Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Avraham Stern [Mon, 27 May 2024 16:05:56 +0000 (19:05 +0300)]
wifi: iwlwifi: mvm: add an option to use ptp clock for rx timestamp
Add a debugfs option to use the ptp clock time for rx device
timestamps. This can be useful to e.g. synchronize multiple NICs or
reporting the timestamp in the system clock instead of gp2.
Johannes Berg [Fri, 10 May 2024 09:26:02 +0000 (11:26 +0200)]
wifi: mac80211: send DelBA with correct BSSID
In MLO, the deflink BSSID is clearly invalid. Since we fill
the addresses as MLD addresses and translate later, use the
AP address here instead.
This fixes an issue that happens with HW restart, where the
DelBA frame is transmitted, but not processed correctly due
to the wrong BSSID (or even just discarded entirely). As a
result, the BA sessions are kept alive; however, as other
state is reset during HW restart, this then fails (reorder,
etc.) and data doesn't go through until new BA sessions are
established.
Dmitry Antipov [Fri, 17 May 2024 15:33:31 +0000 (18:33 +0300)]
wifi: cfg80211: use __counted_by where appropriate
Annotate 'sub_specs' of 'struct cfg80211_sar_specs', 'channels'
of 'struct cfg80211_sched_scan_request', 'channels' of 'struct
cfg80211_wowlan_nd_match', and 'matches' of 'struct
cfg80211_wowlan_nd_info' with '__counted_by' attribute. Briefly
tested with clang 18.1.1 and CONFIG_UBSAN_BOUNDS running iwlwifi.
Johannes Berg [Mon, 6 May 2024 19:54:52 +0000 (21:54 +0200)]
wifi: mac80211: refactor chanreq.ap setting
There are now three places setting up chanreq.ap which always
depends on the mode (EHT being used or not) and override flag.
Refactor that code into a common function with a comment, to
make that clearer.
Johannes Berg [Mon, 6 May 2024 19:54:50 +0000 (21:54 +0200)]
wifi: mac80211: handle wider bandwidth OFDMA during CSA
During channel switch, track the AP configuration in the
chanreq, so that wider bandwidth OFDMA is taken into
account correctly, since multiple channel contexts may
be needed due to sharing not being possible due to
wider bandwidth OFDMA.
Johannes Berg [Mon, 6 May 2024 19:54:49 +0000 (21:54 +0200)]
wifi: mac80211: collect some CSA data into sub-structs
Collect the CSA data in ieee80211_link_data_managed and
ieee80211_link_data into a csa sub-struct to clean up a
bit and make adding new things more obvious.
Johannes Berg [Mon, 15 Apr 2024 08:27:11 +0000 (11:27 +0300)]
wifi: mac80211: mlme: handle cross-link CSA
If we see a channel switch announcement on one link for
another, handle that case and start the CSA. The driver
can react to this in whatever way it needs. The stack
will have the ability to track it via the RNR/MLE in the
reporting link's beacon if it sees it for inactive links
and adjust everything accordingly.
Note that currently the timings for the CSA aren't set,
the values are only used by the Intel drivers, and they
don't need this for newer devices that support MLO, so
I've left it out for now.
Johannes Berg [Mon, 6 May 2024 19:11:59 +0000 (21:11 +0200)]
wifi: cfg80211: restrict operation during radar detection
Just like it's not currently possible to start radar
detection while already operating, it shouldn't be
possible to start operating while radar detection is
running. Fix that.
Also, improve the check whether operating (carrier
might not be up if e.g. attempting to join IBSS).
Johannes Berg [Mon, 6 May 2024 19:13:12 +0000 (21:13 +0200)]
wifi: mac80211: move radar detect work to sdata
At some point we thought perhaps this could be per link, but
really that didn't happen, and it's confusing. Radar detection
still uses the deflink to allocate the channel, but the work
need not be there. Move it back.
Miri Korenblit [Sun, 12 May 2024 12:24:59 +0000 (15:24 +0300)]
wifi: iwlwifi: mvm: don't skip link selection
If we exit EMLSR due to a IWL_MVM_ESR_EXIT*, a MLO scan followed by a
link selection is scheduled with a delay of 30 seconds.
If during that 30 seconds EMLSR was blocked and unblocked
(IWL_MVM_ESR_BLOCKED*), we would still want to get the needed data from
the MLO scan and select link accordingly, and not return immediately to
EMLSR.
Johannes Berg [Sun, 12 May 2024 12:24:56 +0000 (15:24 +0300)]
wifi: iwlwifi: tracing: fix condition to allocate buf1
There's no need to allocate buf1 if we're not going to
write anything to it, and the condition for writing is
(correctly) "hdr_len > 0 && !iwl_trace_data()". Also
allocate the event space only under the same condition.
Emmanuel Grumbach [Sun, 12 May 2024 04:58:57 +0000 (07:58 +0300)]
wifi: iwlwifi: mvm: leave a print in the logs when we call fw_nmi()
When we crash the firmware, we need to know why we decided to do so.
Almost all the callsites of iwl_force_nmi() print something in the logs
that explain why the driver decided to crash the firmware.
Debugfs doesn't print anything and it is then hard to understand why the
firmware has crashed.
Add a simple print in the debugfs hook to ease the debug.
Miri Korenblit [Sun, 12 May 2024 04:31:10 +0000 (07:31 +0300)]
wifi: iwlwifi: mvm: disable dynamic EMLSR when AUTO_EML is false
When AUTO_EML is set to false, this change ignores the exit criteria,
for testing purposes. Currently, if AUTO_EML is disabled, the
driver will not select a link or enter EMLSR, but will still exit if one
of the criteria is not fulfilled.
Johannes Berg [Sun, 12 May 2024 04:31:07 +0000 (07:31 +0300)]
wifi: iwlwifi: mvm: don't track used links separately
We track which link is using which FW link ID, so there
really isn't a need to separately track which link IDs
are in use. Remove that code and check the table when
looking for a new link ID to use.
Johannes Berg [Sun, 12 May 2024 04:31:03 +0000 (07:31 +0300)]
wifi: iwlwifi: fw: avoid bad FW config on RXQ DMA failure
In practice, iwl_trans_get_rxq_dma_data() will not fail,
it only can do that if called with the wrong arguments.
But it does have an error value and doesn't initialize
the argument when it fails, so don't use the value then.
The two versions of iwl_mei_set_power_limit() are a bit different,
and while really the const isn't all that necessary on the inline,
it's still better to have it be the same.
Johannes Berg [Sun, 12 May 2024 04:30:59 +0000 (07:30 +0300)]
wifi: iwlwifi: mvm: mark bad no-data RX as having bad PLCP
Just like with any other frame, mac80211 will do sanity checks
on no-data RX (representing e.g. sounding PPDUs), and if e.g.
the NSS is wrong, it will warn. This isn't a good idea if we
already know the frame wasn't received well, e.g. has bad PLCP.
Unless the firmware reports "no error", set the bad PLCP flag
to skip checks in mac80211.
Also, since we're now extracting two different values from the
info field, use le32_get_bits() for both.
Shahar S Matityahu [Fri, 10 May 2024 14:06:40 +0000 (17:06 +0300)]
wifi: iwlwifi: remove fw_running op
fw_running assumes that memory can be retrieved only after alive.
This assumption is no longer true as we support dump before alive.
To avoid invalid access to the NIC, check that STATUS_DEVICE_ENABLED
bit in trans status is set before dumping instead of the prior check.
Signed-off-by: Shahar S Matityahu <shahar.s.matityahu@intel.com> Reviewed-by: Luciano Coelho <luciano.coelho@intel.com> Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com> Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com> Link: https://msgid.link/20240510170500.ca07138cedeb.I090e31d3eaeb4ba19f5f84aba997ccd36927e9ac@changeid Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Andrei Otcheretianski [Fri, 10 May 2024 14:06:38 +0000 (17:06 +0300)]
wifi: iwlwifi: mvm: Don't set NO_HT40+/- flags on 6 GHz band
These flags are not needed on 6 GHz channels, and anyway they were set
incorrectly.
This in turn resulted in alternating channel flags, preventing reg domain
rules to be merged together, so cfg80211 couldn't even send them to the
user space, as the regulatory domain was too large.
Fix it.
Emmanuel Grumbach [Fri, 10 May 2024 14:06:30 +0000 (17:06 +0300)]
wifi: iwlwifi: mvm: simplify the uAPSD coexistence limitation code
uAPSD can't be enabled on BSS or P2P Client if there is
another active interface (e.g. associated client). Allow
the code that apply those limitations to run on BSS as well.
Also forbid uAPSD if we have an IBSS or NAN interface in
the system.
Since we are now forbidding uAPSD in any concurrency
scenario, we don't need to check the number of PHY
contexts.
Johannes Berg [Fri, 10 May 2024 14:06:28 +0000 (17:06 +0300)]
wifi: iwlwifi: dvm: clean up rs_get_rate() logic
The logic here is rather convoluted - we cannot get here with
lq_sta being NULL as mac80211 will (no longer) call us like
that, and since I removed the rate_control_send_low() call in
this function there's no longer any point in setting priv_sta
to NULL either.
So the only thing that remains to check is if we have actually
initialized our lq_sta->drv pointer, and exit if we didn't in
which case we'll use the data mac80211 already set up for the
low rate usage.
Johannes Berg [Mon, 6 May 2024 19:37:56 +0000 (21:37 +0200)]
wifi: mac80211: pass parsed TPE data to drivers
Instead of passing the full TPE elements, in all their glory
and mixed up data formats for HE backward compatibility, parse
them fully into the right values, and pass that to the drivers.
Also introduce proper validation already in mac80211, so that
drivers don't need to do it, and parse the EHT portions.
The code now passes the values in the right order according to
the channel used by an interface, which could also be a subset
of the data advertised by the AP, if we couldn't connect with
the full bandwidth (for whatever reason.)
Also add kunit tests for the more complicated bits of it.
Johannes Berg [Mon, 6 May 2024 19:37:55 +0000 (21:37 +0200)]
wifi: ath11k: fix TPE power levels
The chanctx can be shared, but it seems like the TPE values for
this vif should only reflect the channel that the vif uses, which
is captured in bss_conf->chanreq.oper rather than the chanctx.
Use that when calculating defaults, a similar issue is where the
data from the AP is processed, but I'm moving that to mac80211.
The define IEEE80211_MAX_NUM_PWR_LEVEL doesn't make much
sense. Yes, that table has a maximum value of 8, and the
table will actually remain that way, but EHT introduced
a way to encode more levels for 320 MHz channels.
Remove IEEE80211_MAX_NUM_PWR_LEVEL and, for ath11k being
the only user, add ATH11K_NUM_PWR_LEVELS, where it makes
sense since it cannot support 320 MHz channels.
Mukesh Sisodiya [Mon, 6 May 2024 19:37:53 +0000 (21:37 +0200)]
wifi: mac80211: update 6 GHz AP power type before association
6 GHz AP power type details are required to set proper tx power
used to send frames.
Update AP power type received in beacon while preparing
for connection instead of after association so the frames
before association can use the correct tx power.
Johannes Berg [Mon, 6 May 2024 19:00:03 +0000 (21:00 +0200)]
wifi: cfg80211: add tracing for wiphy work
Add trace events to trace when wiphy works are queued (or
delayed ones scheduled), and other APIs are called. Also
add an event when the worker starts, before acquiring the
mutex, to be able to see potential delays due to locking.
Johannes Berg [Wed, 15 May 2024 12:16:00 +0000 (14:16 +0200)]
wifi: nl80211: disallow setting special AP channel widths
Setting the AP channel width is meant for use with the normal
20/40/... MHz channel width progression, and switching around
in S1G or narrow channels isn't supported. Disallow that.
This is actually completely equivalent to the other function
ieee80211_is_s1g_short_beacon(), but open-codes the logic.
Implement the necessary logic in ieee80211_is_s1g_short_beacon()
and remove ieee80211_next_tbtt_present().
Dmitry Antipov [Wed, 15 May 2024 15:28:24 +0000 (18:28 +0300)]
wifi: rt2x00: remove unused delayed work data from link description
Remove unused 'agc_work' and 'vco_work' members of 'struct link'.
This follows commit d96324703ffa ("rt2x00: merge agc and vco works
with link tuner") where AGC and VCO calibration code was completely
redesigned. Compile tested only.
Rafael Beims [Fri, 10 May 2024 11:04:58 +0000 (13:04 +0200)]
wifi: mwifiex: Fix interface type change
When changing the interface type we also need to update the bss_num, the
driver private data is searched based on a unique (bss_type, bss_num)
tuple, therefore every time bss_type changes, bss_num must also change.
This fixes for example an issue in which, after the mode changed, a
wireless scan on the changed interface would not finish, leading to
repeated -EBUSY messages to userspace when other scan requests were
sent.
Fixes: c606008b7062 ("mwifiex: Properly initialize private structure on interface type changes") Cc: stable@vger.kernel.org Signed-off-by: Rafael Beims <rafael.beims@toradex.com> Signed-off-by: Francesco Dolcini <francesco.dolcini@toradex.com> Signed-off-by: Kalle Valo <kvalo@kernel.org> Link: https://msgid.link/20240510110458.15475-1-francesco@dolcini.it
Ajay Singh [Tue, 14 May 2024 15:51:24 +0000 (17:51 +0200)]
wifi: wilc1000: read MAC address from fuse at probe
The default netdev interface exposed by WILC1000 is registered at probe,
but the chip mac address is not known until ndo_open, which will load and
start chip firmware and then retrieve stored MAC address from it. As a
consequence, the interface has uninitialized value (00:00:00:00:00) until a
user brings up the interface.
Fix MAC address at probe by setting the following steps:
- at probe, read MAC address directly from fuse
- whenever a new netdevice is created, apply saved mac address (which can
be a user-provided address, or the eFuse Mac address if no address has
been passed by user)
- whenever an interface is brought up for the first time (and so the
firmware is loaded and started), enforce netdevice mac address to the
chip (in case user has changed it)
Alexis Lothoré [Tue, 14 May 2024 15:51:23 +0000 (17:51 +0200)]
wifi: wilc1000: make sdio deinit function really deinit the sdio card
In order to be able to read raw registers (eg the nv mac address) in
wilc1000 during probe before the firmware is loaded and running, we need to
run the basic sdio functions initialization, but then we also need to
properly deinitialize those right after, to preserve the current driver
behavior (keeping the chip idle/unconfigured until the corresponding
interface is brought up). Calling wilc_sdio_deinit in its current form is
not enough because it merely resets an internal flag.
Implement a deinit sequence which symmetrically reset all steps performed
in wilc_sdio_init (only for parts activating/deactivating features, for the
sake of simplicity, let's ignore blocks size configuration reset)
Alexis Lothoré [Tue, 14 May 2024 15:51:22 +0000 (17:51 +0200)]
wifi: wilc1000: add function to read mac address from eFuse
wilc driver currently reads and sets mac address by firmware calls. It
means that we can not access mac address if no interface has been brought
up (so firmware is up and running). Another way to get mac address is to
read it directly from eFUSE.
Add a function helper to read the mac address written in eFuse, without
firmware assistance
Alexis Lothoré [Tue, 14 May 2024 15:51:21 +0000 (17:51 +0200)]
wifi: wilc1000: set wilc_set_mac_address parameter as const
Any attempt to provide a const mac address to wilc_set_mac_address results
in the following warning:
warning: passing argument 2 of 'wilc_set_mac_address' discards 'const'
qualifier from pointer target type [-Wdiscarded-qualifiers]
[...]
drivers/net/wireless/microchip/wilc1000/hif.h:170:52: note: expected 'u8 *'
{aka 'unsigned char *'} but argument is of type 'const unsigned char *'a
int wilc_set_mac_address(struct wilc_vif *vif, u8 *mac_addr);
Instead of using an explicit cast each time we need provide a MAC address,
set the function parameter as const
Alexis Lothoré [Tue, 14 May 2024 15:51:20 +0000 (17:51 +0200)]
wifi: wilc1000: register net device only after bus being fully initialized
SDIO/SPI probes functions automatically add a default wlan interface on top
of registered wiphy, through wilc_cfg80211_init which in turn calls
wilc_netdev_ifc_init. However, bus is still not fully initialized when we
register corresponding net device (for example we still miss some private
driver data pointers), which for example makes it impossible to
retrieve MAC address from chip (which is supposed to be set on net device
before its registration) before registering net device. More generally, net
device registration should not be done until driver has fully initialized
everything and is ready to handle any operation on the net device.
Prevent net device from being registered so early by doing it at the end of
probe functions. Apply this logic to both sdio and spi buses.
Alexis Lothoré [Tue, 14 May 2024 15:51:19 +0000 (17:51 +0200)]
wifi: wilc1000: set net device registration as last step during interface creation
net device registration is currently done in wilc_netdev_ifc_init but
other initialization operations are still done after this registration.
Since net device is assumed to be usable right after registration, it
should be the very last step of initialization.
Move netdev registration at the very end of wilc_netdev_ifc_init to let
this function completely initialize netdevice before registering it.
wifi: brcmsmac: LCN PHY code is used for BCM4313 2G-only device
The band_idx variable in the function wlc_lcnphy_tx_iqlo_cal() will
never be set to 1 as BCM4313 is the only device for which the LCN PHY
code is used. This is a 2G-only device.
Fixes: 5b435de0d786 ("net: wireless: add brcm80211 drivers") Signed-off-by: Samasth Norway Ananda <samasth.norway.ananda@oracle.com> Acked-by: Arend van Spriel <arend.vanspriel@broadcom.com> Signed-off-by: Kalle Valo <kvalo@kernel.org> Link: https://msgid.link/20240509231037.2014109-1-samasth.norway.ananda@oracle.com
Jakub Kicinski [Thu, 9 May 2024 02:09:38 +0000 (19:09 -0700)]
Merge tag 'wireless-next-2024-05-08' of git://git.kernel.org/pub/scm/linux/kernel/git/wireless/wireless-next
Kalle Valo says:
====================
wireless-next patches for v6.10
The third, and most likely the last, "new features" pull request for
v6.10 with changes both in stack and in drivers. In ath12k and rtw89
we disabled Wireless Extensions just like with iwlwifi earlier. Wi-Fi
7 devices will not support Wireless Extensions (WEXT) anymore so if
someone is still using the legacy WEXT interface it's time to switch
to nl80211 now!
We merged wireless into wireless-next as we decided not to send a
wireless pull request to v6.9 this late in the cycle. Also an
immutable branch with MHI subsystem was merged to get ath11k and
ath12k hibernation working.
Major changes:
mac80211/cfg80211
* handle color change per link
mt76
* mt7921 LED control
* mt7925 EHT radiotap support
* mt7920e PCI support
ath12k
* debugfs support
* dfs_simulate_radar debugfs file
* disable Wireless Extensions
* suspend and hibernation support
* ACPI support
* refactoring in preparation of multi-link support
ath11k
* support hibernation (required changes in qrtr and MHI subsystems)
* ieee80211-freq-limit Device Tree property support
ath10k
* firmware-name Device Tree property support
rtw89
* complete features of new WiFi 7 chip 8922AE including BT-coexistence
and WoWLAN
* use BIOS ACPI settings to set TX power and channels
* disable Wireless Extensios on Wi-Fi 7 devices
iwlwifi
* block_esr debugfs file
* support again firmware API 90 (was reverted earlier)
* provide channel survey information for Automatic Channel Selection (ACS)
* tag 'wireless-next-2024-05-08' of git://git.kernel.org/pub/scm/linux/kernel/git/wireless/wireless-next: (214 commits)
wifi: mwl8k: initialize cmd->addr[] properly
wifi: iwlwifi: Ensure prph_mac dump includes all addresses
wifi: iwlwifi: mvm: don't request statistics in restart
wifi: iwlwifi: mvm: exit EMLSR if secondary link is not used
wifi: iwlwifi: mvm: add beacon template version 14
wifi: iwlwifi: mvm: align UATS naming with firmware
wifi: iwlwifi: Force SCU_ACTIVE for specific platforms
wifi: iwlwifi: mvm: record and return channel survey information
wifi: iwlwifi: mvm: add the firmware API for channel survey
wifi: iwlwifi: mvm: Fix race in scan completion
wifi: iwlwifi: mvm: Add a print for invalid link pair due to bandwidth
wifi: iwlwifi: mvm: add a debugfs for reading EMLSR blocking reasons
wifi: iwlwifi: mvm: Add active EMLSR blocking reasons prints
wifi: iwlwifi: bump FW API to 90 for BZ/SC devices
wifi: iwlwifi: mvm: fix primary link setting
wifi: iwlwifi: mvm: use already determined cmd_id
wifi: iwlwifi: mvm: don't reset link selection during restart
wifi: iwlwifi: Print EMLSR states name
wifi: iwlwifi: mvm: Block EMLSR when a p2p/softAP vif is active
wifi: iwlwifi: mvm: fix typo in debug print
...
====================
Jakub Kicinski [Thu, 9 May 2024 01:59:51 +0000 (18:59 -0700)]
Merge branch 'netdevsim-add-napi-support'
David Wei says:
====================
netdevsim: add NAPI support
Add NAPI support to netdevsim and register its Rx queues with NAPI
instances. Then add a selftest using the new netdev Python selftest
infra to exercise the existing Netdev Netlink API, specifically the
queue-get API.
This expands test coverage and further fleshes out netdevsim as a test
device. It's still my goal to make it useful for testing things like
flow steering and ZC Rx.
====================
David Wei [Tue, 7 May 2024 16:32:27 +0000 (09:32 -0700)]
netdevsim: add NAPI support
Add NAPI support to netdevim, similar to veth.
* Add a nsim_rq rx queue structure to hold a NAPI instance and a skb
queue.
* During xmit, store the skb in the peer skb queue and schedule NAPI.
* During napi_poll(), drain the skb queue and pass up the stack.
* Add assoc between rxq and NAPI instance using netif_queue_set_napi().
The test direction is reversed between receive and transmit tests, so
that the NIC under test is always the local machine.
In total this adds up to 12 testcases, with more to follow. For
conciseness, I replaced individual functions with a function factory.
Also detect hardware offload feature availability using Ethtool
netlink and skip tests when either feature is off. This need may be
common for offload feature tests and eventually deserving of a thin
wrapper in lib.py.
Missing are the PF_PACKET based send tests ('-P'). These use
virtio_net_hdr to program hardware checksum offload. Which requires
looking up the local MAC address and (harder) the MAC of the next hop.
I'll have to give it some though how to do that robustly and where
that code would belong.
Tested:
make -C tools/testing/selftests/ \
TARGETS="drivers/net drivers/net/hw" \
install INSTALL_PATH=/tmp/ksft
cd /tmp/ksft
Eric Dumazet [Tue, 7 May 2024 12:17:48 +0000 (12:17 +0000)]
phonet: no longer hold RTNL in route_dumpit()
route_dumpit() already relies on RCU, RTNL is not needed.
Also change return value at the end of a dump.
This allows NLMSG_DONE to be appended to the current
skb at the end of a dump, saving a couple of recvmsg()
system calls.
projects / users / jedix / linux-maple.git / log