| Age | Commit message (Collapse) | Author |
|
Newer firmware versions will start reporting a separate RX time value
that represents for how long the channel was busy with Wi-Fi frames.
Previously the reported value for RX time was equal to the busy time.
As such, report the information as it will be useful in the future.
Signed-off-by: Benjamin Berg <benjamin.berg@intel.com>
Link: https://patch.msgid.link/20260930121740.727c05ab3c58.I9509587622fa7c6493b6baee53372c548bf60ed8@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
TX command v12 adds a driver timestamp in the gp2 clock domain.
Firmware uses it to age out frames deferred by CCA for more than
four seconds. Without it, a blocked queue can stall until the
datapath monitor fires LMAC assert 0x4433.
Reading the gp2 register per frame is too expensive for the TX path.
Instead, use a worker to measure its offset from host boottime and
publish the offset and validity deadline using atomic stores. TX
then estimates gp2 without allocation or locking for timestamp reads.
TX requests a refresh after five seconds, while the estimate remains
valid for ten seconds to tolerate worker delays and bound clock drift.
Successful refreshes require no periodic work while traffic is idle.
A failed read invalidates the estimate and schedules a retry after
roughly one second, rounded for timer coalescing, even without traffic.
Invalidate the estimate before D3 entry and firmware shutdown, and
refresh it on resume or reload because gp2 restarts with firmware.
Frames without a valid estimate carry a zero timestamp, which firmware
ignores.
Signed-off-by: Daniel Gabay <daniel.gabay@intel.com>
Assisted-by: LLM
Link: https://patch.msgid.link/20260930121740.a5016741a467.Ie4d4af8d7100a656a8469edc0c38ec81b0b58860@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
We hit a WARNING in mac80211 in sniffer that complains that
rx_status->nss is 0. This happens on frames that the firmware doesn't
really understand and when that happens, somehow the STBC bit is set.
The driver divides the number of antennas by 2 to get the NSS with STBC
set because each stream requires 2 antennas. Since NSTS=1, NSS end up
being 0.
Since we don't have devices with more than 2 antennas, just consider
that when we receive STBC traffic, we have one spatial stream.
Since this problem has been seen on HE traffic only, don't apply this
logic on VHT / HT code paths.
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Reviewed-by: Johannes Berg <johannes.berg@intel.com>
Link: https://patch.msgid.link/20260930121740.2589668e377b.Ib41003bd4e10f0fe6d7f4aa1038291be330c3175@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
There are several helpers that exported to the opmodes but are really only
called from the pcie transport. Define those functions there.
No functional change intended.
Assisted-by: GitHubCopilot:claude-opus-4.8
Reviewed-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Link: https://patch.msgid.link/20260930121740.480246f9b623.I88de3ad2bbd8a8444ff2a154f7eb1f08ecb7b7ea@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
A function that accesses the nic memory without grabbing nic access
first, needs to have a _no_grab postfix. It makes it clear to the caller
that he is responsible for grabbing it.
Add it to iwl_trans_write_prph and iwl_trans_pcie_write_prph as well.
Reviewed-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Link: https://patch.msgid.link/20260930121740.68337f23f322.I38ed9c92bf6358158974b0af025d8bf8fdd7e4f5@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
HE ER mode (~3 Mbps) provides marginal range benefit over legacy rates,
while effective coverage remains constrained by beacon reception. In
addition, ~3 Mbps is insufficient for laptop connectivity.
Furthermore, any range benefit in 5 and 6 GHz is more easily achieved
by roaming to 2.4 GHz, while avoiding degrading 5 and 6 GHz aggregate
network capacity with low-rate links. Therefore, disable HE ER mode.
Signed-off-by: Shahar Tzarfati <shahar.tzarfati@intel.com>
Link: https://patch.msgid.link/20260926201527.aac48f46b60e.I5dcf9ed88b2b73bc09c81c0d9c841dc624447faf@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
commit 5b63d0ae94cc ("compiler-context-analysis: Remove Sparse support")
deleted the Sparse implementation of the lock annotations.
__acquires()/__releases()/__acquire()/__release() are gone; the macros
now only map onto clang's context analysis and expand to nothing for gcc.
The nic_access annotations describe nothing anymore:
- Every one of them is a __releases() with no matching __acquires().
The acquire side was the __cond_lock() wrapper around
_iwl_trans_grab_nic_access(), which went away together with
__cond_lock() itself. The names don't even agree - PCIe releases
nic_access_nobh while the generic wrapper calling it claims
nic_access.
- The __acquire()/__release(reg_lock) pairs only existed
to silence Sparse
Use lockdep_assert_held() instead, which unlike the annotations actually
runs, and add the missing assertion to
iwl_trans_pcie_resched_with_nic_access().
Reviewed-by: Johannes Berg <johannes.berg@intel.com>
Link: https://patch.msgid.link/20260926201527.f818a7e09def.Iccecaabe0f37525430377ae1c0db6fd7bbf7625a@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
Version 3 of the RX BAID allocation modify command shrinks the tid
field to a u8 and adds a window size field, so that the window size
can be updated while moving a BAID to a new AP. A window size of 0
tells the firmware to keep the current one, which is what we need
for now.
Add the new layout and pick it based on the command version.
Assisted-by: LLM
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Link: https://patch.msgid.link/20260926201527.46c77de75c0b.Ie41b1d43f945a0c5e2bb68981a0cec0c62111233@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
The firmware now reports in DW-12 bits 31:24 the RU/MRU/DRU that was
used for the received frame, in trigger-like format.
The MAC context byte moved to bits 31:24 of DW-10.
It isn't used by the driver, so just follow the new layout.
Assisted-by: LLM
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Link: https://patch.msgid.link/20260926201527.cf236c23d1aa.Ie2d919c884f7db8cd76b4cb6a9f4584eb285eea5@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
iwl_force_nmi() forces a firmware NMI and has nothing to do with the
register I/O helpers in iwl-io.c. Move it to iwl-trans.c and rename it
to iwl_trans_force_nmi() to match the transport API naming.
Reviewed-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Link: https://patch.msgid.link/20260926201527.c3312fb2f2cc.Ie0d3ea9b23e35d7ff4b67e129c7b97ccb7c6c453@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
iwl_trans_sync_nmi_with_addr() had a single caller,
iwl_trans_pcie_sync_nmi(). There is no reason to keep it as a separate
exported function in iwl-io.c, so embed its body directly in the caller
and drop the now-unused declaration.
Link: https://patch.msgid.link/20260926201527.a2048732f45f.I23611eb40d875c6a69df0c08add3834f6f7d8528@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
This is no longer needed, and adds lots of parsing code
Reviewed-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Link: https://patch.msgid.link/20260926201527.3e2f229ac9f6.I122eaf364743ffc42bd18a84a766d7920ec5700d@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
iwl_pcie_rx_handle() reads the last closed RB index from rb_stts and
then reads the completion descriptor and the RB contents that this
index makes visible. Nothing orders those two reads: there is no
barrier, the descriptor address derives from rxq->read rather than
from the index just read, so there is no address dependency, and the
"while (i != r)" test is only a control dependency, which does not
order loads on arm64.
The CPU can therefore speculate past the loop test, perform the
completion descriptor load early and sample the value from before
the device's DMA, then resolve the rb_stts load to the value from
after it. On arm64 (Jetson AGX Orin) this showed up as recurring
"Invalid rxb from HW <vid>" naming an in-range rbid the driver still
owned, and "frame on invalid queue" for an RB tagged with the queue
that used it before the wrap. Both force an NMI and a firmware
restart. x86 never shows it because loads are not reordered with
loads there.
Add a dma_rmb() after reading the write pointer, as other NIC drivers
do when consuming a DMA descriptor ring. It is a no-op on x86 and a
dmb on arm64. Verified to stop the warnings on Orin.
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Assisted-by: GitHub-Copilot:claude-opus-5
Link: https://patch.msgid.link/20260926201527.1bff8cebef69.I24dec81516e37a33674e87ed3359c41e59d12e5e@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
Allow platforms reporting the exact manufacturer string
"Samsung" to use TAS and PPAG, while retaining the existing
Samsung Electronics entry.
Signed-off-by: Pagadala Yesu Anjaneyulu <pagadala.yesu.anjaneyulu@intel.com>
Link: https://patch.msgid.link/20260926201527.e2e2e1f720af.I18a7292b5a8bd878d074b114b45ce99e1309897e@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
UEFI reduced-power parsing advanced by ALIGN(tlv_len, 4) but
validated only tlv_len. When remaining bytes were between these
values, len could underflow and parsing could continue past the
buffer boundary.
Validate remaining bytes against the aligned length before
advancing in both TLV walkers.
Signed-off-by: Pagadala Yesu Anjaneyulu <pagadala.yesu.anjaneyulu@intel.com>
Assisted-by: GitHubCopilot:GPT-5.3-Codex
Link: https://patch.msgid.link/20260926201527.2552432e98e8.I570750f27d9530d4c76922d2699a272d9c60a0ab@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
The iwl_uefi_set_sar_profile() function was receiving a bool parameter,
which would treat any non-zero value as enabled. This is incorrect because
the SAR mode value contains additional bits beyond the enable flag.
Change the parameter from bool enabled to u32 mode and properly extract
the enable bit using IWL_SAR_ENABLE_MSK bitmask. This ensures that only
the designated enable bit determines whether SAR is enabled, not arbitrary
non-zero values.
Signed-off-by: Pagadala Yesu Anjaneyulu <pagadala.yesu.anjaneyulu@intel.com>
Link: https://patch.msgid.link/20260926201527.60160602581b.Ib184c201e677cb6d95f64e725a1e39148234fa41@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
Plumb the standalone WRSS/EWSS SAR data and
send the standalone profile to firmware.
Add REDUCE_TX_POWER_CMD version 12 support with the
IWL_TX_POWER_MODE_SET_STANDALONE_CHAINS set_mode, and rename
iwl_dev_tx_power_cmd_v11 to iwl_dev_tx_power_cmd_v12 to match the
updated command layout.
This enables higher transmit power limits for standalone firmware
operation while preserving the existing concurrent SAR handling.
Signed-off-by: Pagadala Yesu Anjaneyulu <pagadala.yesu.anjaneyulu@intel.com>
Link: https://patch.msgid.link/20260926201527.d5315e93c1ce.Ic02190e91e795dcfe660710721b47de71f2b2c86@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
to a folder"
This reverts commit a6fafaedb5d6 ("wifi: iwlwifi: pcie: move generation
specific files to a folder").
The pcie/gen1_2/ subfolder was introduced to prepare for a future
pcie-transport, which will not happen eventually.
Move all the files back to the pcie folder and remove the gen1_2 folder.
Link: https://patch.msgid.link/20260926201527.91ac209ad70d.I25521996309a25ac779cb31f2c93477b25be9ac3@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
function"
This reverts commit df70a9a86eee ("wifi: iwlwifi: gen1_2: move gen
specific code to a function").
The remove flow body was factored out into into a new function to
prepare for a new pcie-transport, which will not happen eventually.
Inline the code back.
Link: https://patch.msgid.link/20260926201527.44edf3234ede.Ic19d96486253518e2107d91e512654596689a426@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
As a preparation of a future pcie-transport, we added a gen1_2 prefix to
some functions to distinguish them from the new implementation.
Since the new pcie-transport will not happen eventaully, remove this
prefix from those functions.
Link: https://patch.msgid.link/20260926201527.dc1a22771b0e.I20e77f228e1536c6eb327dc077b030dbea889ac2@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
Add support for reading WRSS and EWSS standalone SAR tables
from the BIOS in the ACPI and UEFI paths.
WRDS/EWRD tables describe SAR limits for concurrent Wi-Fi
and Bluetooth operation. When Bluetooth is not active,
those limits can unnecessarily restrict Wi-Fi transmit power.
WRSS/EWSS provide separate standalone SAR profiles for Wi-Fi-only
operation, allowing the firmware to select the appropriate
higher-power limits while still respecting the platform-specific
regulatory constraints.
Signed-off-by: Pagadala Yesu Anjaneyulu <pagadala.yesu.anjaneyulu@intel.com>
Link: https://patch.msgid.link/20260923160318.9a8eb3ff9512.Id79836e772136e1e38f107b578007faf39b13b4e@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
The BAID from the ADD_STA response is used to access mvm->baid_map[]
without an upper bound check.
Check and reject an invalid value.
Signed-off-by: Shahar Tzarfati <shahar.tzarfati@intel.com>
Link: https://patch.msgid.link/20260923160318.22624c6dfa10.I9a99a140ab72cb1790a03a130ae7c8aa93cfac01@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
Add support of another flavor of AX231.
Link: https://patch.msgid.link/20260923160318.af9e8f9c1548.I1fd3aa97da28065966952c36edcdb99acd5313cb@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
The BIOS tables can be sourced from either UEFI or ACPI. The UEFI
copies are writable, so they may be trusted only when the platform
attests they are locked. Resolve this trust from a single source,
preferring the UEFI GUID Lock Indicator (GLUI) and consulting the ACPI
GLAI whenever GLUI does not attest the tables as locked or in test mode.
Add iwl_bios_get_guid_lock_status() as that entry point and have the
mvm and mld op-modes use it instead of the ACPI-only helper, so trust is
resolved consistently regardless of whether a platform ships GLUI or
GLAI.
Signed-off-by: Shahar Tzarfati <shahar.tzarfati@intel.com>
Assisted-by: GitHub-Copilot:claude-opus-4-8
Link: https://patch.msgid.link/20260923160318.c6a2252dafb9.I46f8c4d55bd22d98a7200cce8f528c5e182f443c@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
The connectivity UEFI variables are only trusted when the platform
attests they are locked. When the lock status is unknown, the trust of
each variable has to be discovered on its own by checking whether it is
write protected. Probing depends on trying to rewrite UEFI variables and
it is only attempted on non-x86 platforms. If the write is rejected the
variable is considered locked and therefore trusted, otherwise it is
treated as not trusted.
Separate reading such a variable into two steps:
iwl_uefi_get_variable_guid() only fetches and validates the variable,
while iwl_uefi_get_verified_variable_guid() layers the trust decision on
top of it. The latter now takes an iwl_fw_runtime so it can consult the
lock status. Rename its WiFi-GUID wrapper from
iwl_uefi_get_verified_variable() to iwl_uefi_get_verified_wifi_var().
With the trust check now performed by every iwl_uefi_get_*() reached
through the GET_BIOS_TABLE() macro, the per-table lock-status gate in
that macro is redundant, so drop it here.
This move, and the new unknown-status handling, affect only the tables
reached through GET_BIOS_TABLE(), i.e. those that have both a UEFI and
an ACPI source.
Assisted-by: GitHub-Copilot:claude-opus-4-8
Signed-off-by: Shahar Tzarfati <shahar.tzarfati@intel.com>
Link: https://patch.msgid.link/20260923160318.6caa9ca17ffe.I877c0d6ab5126edb3e4f1f187ca1e20e563b2ec4@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
The platform provides a root-of-trust mechanism for connectivity UEFI
variables via a dedicated UEFI GUID Lock Indicator (GLUI) variable. The
GLUI may remain locked in both production and testing modes. A UEFI
indicator is used rather than relying on the ACPI GLAI alone, so the
mechanism also works on platforms that do not expose ACPI.
Add the building blocks for consuming this variable and
iwl_uefi_get_guid_lock_status() to read its value and store it, so the
driver can use its lock state to decide whether the connectivity UEFI
variables are trusted.
Signed-off-by: Shahar Tzarfati <shahar.tzarfati@intel.com>
Assisted-by: GitHub-Copilot:claude-opus-4-8
Link: https://patch.msgid.link/20260923160318.55d7dc9c9531.If4c9949de70fe78951464b983caf2c9d974061b3@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
iwl_mvm_rx_mpdu_mq path only checks that the mpdu_len fits into the pkt
len, but it doesn't validate len itself before dereferencing parts of
the mpdu: the header, the padding, and the mic_crc_len.
Check that the mpdu len covers the hdrlen, the padding (if exists) and
the mic_crc_len.
Reviewed-by: Johannes Berg <johannes.berg@intel.com>
Link: https://patch.msgid.link/20260923160318.fbf29ec27bc0.I4bb0dbd739ff1ddf969e51bbd68c42ca9e4e0634@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
Start supporting Core 108 FW on these devices.
The BZ/FM combination is frozen on Core107 (while each of BZ, FM is not
yet frozen and is still supported in other combinations).
Indicate this in the relevant FW file mames, for modinfo.
While at it, remove names that are no longer needed.
Link: https://patch.msgid.link/20260923160318.61c292103e38.I0e2918e21f8a20d200d371c1861ed2c5deac829d@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
Since API 77, firmware reports scan offload profile matches using
SCAN_OFFLOAD_PROFILE_MATCH_RESULTS_S_VER_3, which has more channels (16
byte instead of 7).
Somehow we missed this change, and the driver only handles the 5- and
7-byte layouts, so on such firmware it misparsed the match (reading band
as energy) and could not recover matches beyond the first 56 channels.
Add the version 3 match structure and select it based on the
SCAN_OFFLOAD_PROFILES_QUERY_CMD notification version (in MVM, in MLD,
this is the only version supported).
In MVM, while walking the matched-channels bitmap, bound the channel
index with IWL_FW_CHECK so firmware that sets a bit beyond n_nd_channels
can no longer trigger an out-of-bounds read of nd_channels[].
Reviewed-by: Ilan Peer <ilan.peer@intel.com>
Link: https://patch.msgid.link/20260923160318.07738ca44b54.I58162891250af9f46220fd01db667460e6dd621b@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
This needs to be combined from two places, fix that.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Link: https://patch.msgid.link/20260923160318.e65cdac4424f.I23b05c26fc788081ad8aa2673e104e2215697416@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
These were wrong in the driver based on the hardware
definitions, fix them.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Link: https://patch.msgid.link/20260923160318.3fa552b3d795.I8098855c931e1f9df16951e2d527bb079eca0549@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
In UHR, currently supported devices have 2x LDPC support.
Enable the bit.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Link: https://patch.msgid.link/20260923160318.9771e88dbc3b.I22693891ed85fd0449ae17106e2502ec4ee2e086@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
The version guard used `< 1` instead of `!= 1`. With `< 1` the
command would be sent for any firmware version >= 1, including
future versions that may have a different structure, which could
trigger a firmware assert due to a size/format mismatch.
Fix the check to `!= 1` so the command is only sent when the
firmware supports exactly version 1.
Signed-off-by: Pagadala Yesu Anjaneyulu <pagadala.yesu.anjaneyulu@intel.com>
Link: https://patch.msgid.link/20260923160318.27693f88a8d7.Id6a4b91938cd7758c2be3b98379f1057e574cdad@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
iwl_dbgfs_tas_get_status_read() trusted the FW response layout and
iterated with rsp->in_dual_radio + 1 without capping to the
tas_status_mac array size. A malformed or inconsistent response could
lead to out-of-bounds access while printing TAS status.
Validate the GET_TAS_STATUS payload length before dereferencing rsp,
free the response on early allocation failure, and cap both TAS loops
with array size.
This keeps the existing behavior for valid responses and hardens the
debugfs path against invalid firmware data.
Signed-off-by: Pagadala Yesu Anjaneyulu <pagadala.yesu.anjaneyulu@intel.com>
Link: https://patch.msgid.link/20260923160318.f03b511b938e.Ib5069ecf15340506498a28d965c49387cab609b1@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
The RSS indirection table debugfs knob is not used, and its pattern
repetition logic triggered a static analysis out-of-bounds report on
the trailing copy that forms a one-past-the-end pointer. Rather than
reworking unused code, drop the entry entirely.
Assisted-by: GitHub-Copilot:claude-opus-5
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Link: https://patch.msgid.link/20260923160318.bc19cf5a20b9.I04cbf6a929070775a7e0cb04c7330421e6777f4e@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
It appears that there are some APs that advertise STBC
for HT and VHT, but don't actually support it for HE.
Since the FW TLC will only use the mode we set (or fall
back to legacy rates), we can set the bits according to
the mode for STBC and LDPC. For EHT, there's no STBC at
all, but LDPC support is more complex to determine.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Link: https://patch.msgid.link/20260922144819.cf235dddcbb3.I315c8089e341c6f7b2108276a32560595a8571fd@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
Validate FW_DBG_DEST TLV size before using it:
Validate the version byte first, then check the length of each version
of the payload.
Signed-off-by: Pagadala Yesu Anjaneyulu <pagadala.yesu.anjaneyulu@intel.com>
Link: https://patch.msgid.link/20260922144819.7eb8d71ff962.I81295f046c785da613331c6a619897ccf66007d6@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
Some platforms expose a UEFI DSM table with the expected revision
but with an empty QUERY bitmap, which means no DSM function is
advertised as valid.
Signed-off-by: Pagadala Yesu Anjaneyulu <pagadala.yesu.anjaneyulu@intel.com>
Link: https://patch.msgid.link/20260922144819.fe9c61c887e1.I735e6a67f7fb4903ddeca84f0a2ca5cd062ba1df@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
If we allocated a recovery buffer, but didn't eventually send the
recovery command (for example because we failed to load the FW),
We won't free the buffer until the module is unloaded.
If in between another FW error occured, we will override the pointer
with a new allocated buffer without freeing the previous one.
Fix this by freeing a previous buffer before allocating a new one.
Reviewed-by: Johannes Berg <johannes.berg@intel.com>
Link: https://patch.msgid.link/20260922144819.36e637b8162a.Ida6e834d7f5d1e69d46c47e2e46db1e87a57ac71@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
If we are recovering from FW error but we don't have a reocvery buffer
(for example due to a driver bug), it means that the FW might not have all
the data it needs and we can therefore have bugs after the recovery.
Those bugs can be difficult to debug. Log a clear error in such case.
Reviewed-by: Johannes Berg <johannes.berg@intel.com>
Link: https://patch.msgid.link/20260922144819.afd6c953d680.Ibbf29ef47b99a9037a56b3342883376e9f7749dd@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
Add support for scan iteration complete notification v3. This
version is a superset of the previous version and adds information
about the EBS corresponding with this scan iteration.
Assisted-by: GitHub-Copilot:claude-opus-5
Signed-off-by: Ilan Peer <ilan.peer@intel.com>
Link: https://patch.msgid.link/20260922144819.daf57b66d8a9.I1ff60b070464ea5cc1f402293c2342157d29ea79@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
In some cases the frame status is changed to AID_NOT_FOR_US if no data
is received. Pass the status explicitly so that the PHY data remains
intact as that simplifies debugging.
Signed-off-by: Benjamin Berg <benjamin.berg@intel.com>
Reviewed-by: Johannes Berg <johannes.berg@intel.com>
Link: https://patch.msgid.link/20260922144819.86c1201ae2f6.I4b10b530325df711c7d229f1c487285a4dfcd294@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
A failed ctxt-info-v2 allocation can free prph_scratch/prph_info
without clearing the stored transport pointers.
Clear stale DMA address/pointer fields during unwind.
Also add an alloc-entry WARN_ON_ONCE/-EBUSY invariant to reject re-entry
when ctxt/prph/iml state is already populated.
Also, assign iml_len only the allocation.
Assisted-by: GitHubCopilot:gpt-5.3-codex
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Link: https://patch.msgid.link/20260922144819.be2c3cf96a4e.I207172d76a22c0302575f3383a044700f97a9980@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
iwl_pcie_prep_tso() obtains a TSO header page via
iwl_pcie_get_page_hdr(), which stores the page in skb->cb
and increments its internal reference count.
If skb_to_sgvec() or dma_map_sgtable() then fails, the
function returns NULL without calling iwl_pcie_free_tso_pages()
or otherwise dropping the page reference and/or DMA mapping.
Fix this page leak.
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Link: https://patch.msgid.link/20260922144819.9f5a0b76e75e.Ic7d2303cb85a6de5158a713f57d4366c7b0d93b0@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
Add support for version 9 of the LINK context config command, which
carries the max BSSID indicator N, where 2**N is the maximal number of
BSSIDs the physical AP can support.
The firmware uses this to mask N bits of the transmitter address for
NPCA OBSS and inter/intra BSS NAV decisions. N=0 disables the feature.
Assisted-by: GithubCopilot:claude-opus-4.8
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Link: https://patch.msgid.link/20260922144819.fafac4d1052d.I91be3bcfcd25d23b5c10c043d6cd639de4efa214@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
The htc_trig_based_pkt_ext, rand_alloc_ecwmin, rand_alloc_ecwmax and
frame_time_rts_th fields of the LINK context config command are not
used by the firmware. Stop filling them.
Assisted-by: GithubCopilot:claude-opus-4.8
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Link: https://patch.msgid.link/20260922144819.887695af3d15.Ie0f0804a28ec4e1e8e41d0eb11615ed2a0f2823e@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
Treat firmware-provided NVM response length as untrusted before copying
data into the host buffer.
Reject responses that are shorter than the fixed response header, where
reported bytes exceed the requested chunk length, or where payload size
does not contain the reported number of bytes.
This prevents heap overwrite and RX payload over-read in malformed or
malicious NVM_ACCESS responses.
Signed-off-by: Pagadala Yesu Anjaneyulu <pagadala.yesu.anjaneyulu@intel.com>
Reviewed-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Link: https://patch.msgid.link/20260922144819.a9d384de40a0.Id6028e4753a2af5638bed0279bbf55842845df3b@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
The monitor_data debugfs file exposed the DRAM firmware monitor buffer
to userspace. Remove it along with all the code and data that only
existed to support it: the read/open/release handlers and the
iwl_write_to_user_buf() helper, the cont_rec state tracking structure
and iwl_fw_mon_dbgfs_state enum, the fw_mon_data field and its
initialization, and the now-unused debugfs cleanup path
(iwl_trans_pcie_debugfs_cleanup() and iwl_trans_debugfs_cleanup()).
This relied on old firmware monitor format.
Assisted-By: GitHubCopilot:Claude Opus 4.8
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Link: https://patch.msgid.link/20260922144819.9e5d30a9caa4.Idf62277c82970daf49d3d4d7b7a021595672eeed@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
There is a race between the PCI link power save flow and the
notification flow from the firmware. The firmware prepares the
notification in DRAM and then sends an MSI-X message to interrupt the
CPU and let it know it can handle the notification. In case the PCI
link goes L1.2 between the access to the DRAM to copy the notification
and the MSI-X interrupt, the host won't be interrupted.
Add a work-around to cope with this situation. If the command response
was not received after 200ms, read the config space of the device just
to wake up the link. This will allow the MSI-X interrupt to go through
and the CPU will get the interrupt.
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=221501
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Link: https://patch.msgid.link/20260922144819.ef433f3da32b.I017c1c7a1dd3ac6f111d4640966968b82f37e291@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|
|
Add STATISTICS_FW_NTFY_OPERATIONAL_API_S_VER_5, which uses a new
per-PHY struct (VER_3) that appends the NPCA channel load fields.
Assisted-by: GitHubCopilot:claude-opus-4.8
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Reviewed-by: Johannes Berg <johannes.berg@intel.com>
Link: https://patch.msgid.link/20260922144819.971d1a747698.I0792723d62dca7cf68056f13c8f733084ce6445f@changeid
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
|