To protect firmware not downloaded in system suspend, acquire/release
the cnss semaphore cnss_pm_sem. It is complex, and cnss function
cnss_pm_notify may trigger dead lock issue once PM_POST_SUSPEND is
ahead of PM_SUSPEND_PREPARE.
Optimize code to use PM wakelock API to protect firmware downloading.
Change-Id: I533c373b85f554fbcceb562d9f56c6b88e5155bb
CRs-Fixed: 3280247
-Wimplicit-fallthrough is being enabled by default. Some compilers
such as clang require the attribute instead of just a fallthrough comment.
Change-Id: I220132ef01f4bf300bbd1f54e5f9dc5208787603
CRs-Fixed: 3217416
For platform without pld support, SSR would do rmmod operation, which tries
to do suspend target, there is no ACK from crashed firmware during SSR.
Keep the recovery in progress flag to skip suspend target command.
Change-Id: Ic6b8f4ebacbf9dc5798730b253eae239fe30c159
CRs-Fixed: 3005884
From ROME, when FW assert, need to wait for ram dump via PCIe completed
before calling QDF_BUG.
Change-Id: I3d76f34c5eacccae07ed485b679a4177be12b15e
CRs-Fixed: 2818715
cds_get_context() function already takes care of logging the
caller function name in case of any error.
Hence there is no need to add the error log again.
Getting rid of these unnecessary logs reduces driver memory footprint.
CRs-Fixed: 2816598
Change-Id: I16fc2a3cdf675d35fb297ec62f4d3f5fb2cd4dd3
qdf_mem_malloc() function already takes care of logging the
caller function name and line number in case of any allocation error.
Hence there is no need to add the error log again.
Getting rid of these unnecessary logs reduces driver memory footprint.
Change-Id: If24f25fb61bda6f9cd18ec45c5ded2eb53748736
CRs-Fixed: 2781932
In cds_trigger_recovery_handler, Host driver first set flag recovering
before cds_force_assert_target. For Rome, ol_target_failure run when
FW crash, Self-Recovery stop when detect flag recoverying be set
Add CDS_DRIVER_STATE_ASSERTING_TARGET to record Host asserting in
cds_force_assert_target. Skip recovering check in ol_target_failure
if detect this flag.
Change-Id: I290e2f21bd0bca6dc6d76de43de4f8dd801d7ff5
When kernel panic, irq will be disabled, when dump Rome
firmware, dma buffer is malloced and freed every time
hif_diag_read_mem is called to dump 20K mem, dma_free_attrs is
called 100 times for 2M mem dump, WARN_ON(irqs_disabled())
will printk stack about 100 times in short time, watchdog bite.
Fix: Change dma mem malloc size from 20KB to 256KB, then only
malloc/free less than 10 times totally when dump Rome firmware.
Change-Id: I803b2cd30a1c6cdc1ce8f8f1073cdb6c17b650af
CRs-Fixed: 2693690
There is no sync between driver unloading & SSR recovering.
If SSR is in progress, need to wait when driver unloading.
Change-Id: I81b935f9d8d6336b68abd6aecbdc7d28f4762be9
CRs-Fixed: 2534890
This reverts change-id Ie40db69a24dc1e0b86f140ca1aad91bd3b10e95e.
The reverted changes was causing regression, that optional
firmware not loadable even it exists.
Change-Id: I02a45e9be635e4ce0787f0a24480e42129c59874
CRs-Fixed: 2516015
In function ol_target_failure(), if enable_self_recovery is true,
fw_indication_work_handler() will be called. Ramdump collection
will not be executed. But for no CNSS module platform, recovery flag
is neither cleaned, no any FW indication is sent to user space.
Even enable_self_recovery is false, no FW crashed indication sent
to user space.
So, call function ol_check_clean_recovery_flag() to clean recovery
flag, and send FW CRASHED indication to user space by function
ol_target_failure().
Change-Id: I44396caf4972dc267d60757c479e4681f8885de5
CRs-Fixed: 2482596
The request_firmware() will take too long time(60s) when the
requested fw file do not exist. Replace this API with
request_firmware_direct.
Change-Id: Ie40db69a24dc1e0b86f140ca1aad91bd3b10e95e
CRs-Fixed: 2435125
Per the Linux Kernel coding style, as enforced by the kernel
checkpatch script, pointers should not be explicitly compared to
NULL. Therefore within bmi replace any such comparisons with logical
operations performed on the pointer itself.
Change-Id: I85578cf7e40e7d6c81246b423d5b626a9570c8b4
CRs-Fixed: 2418396
Peer and vdev leaked during SSR, Helium/Rome SSR is different.
Helium:
FW assert will indicate to CNSS driver, then PLD_FW_DOWN
is sent, in wlan_hdd_pld_uevent, target ready is cleared and
RECOVERING is set, so wma_force_objmgr_vdev_peer_cleanup can be
called, no such issue.
Rome:
FW assert will indicate to wlan driver directly, in
ol_target_failure, only RECOVERING is set, target ready isn't
cleared, wma_force_objmgr_vdev_peer_cleanup can't be called,
so peer and vdev leaked.
Change-Id: I6ecf6cc439f931d8da67972c543376b2148982af
CRs-Fixed: 2403501
There are a few consumers of the legacy OS_* APIs. Directly use
appropriate QDF APIs instead.
Change-Id: Ibe46172200119c6d293acb301baf796276c5949c
CRs-Fixed: 2403793
EOK is a legacy definition and frequently misused. Remove references to
it from BMI.
Change-Id: I38455eca00a6c1f3ffe25635019677708e312f92
CRs-Fixed: 2404901
The header file ol_if_athvar.h is strange in that it is a qcacld-3.0
file that is included by files in the qca-wifi-host-cmn project. That
model does not conform to the software architecture so there is a plan
to correctly refactor the file. As a preliminary step remove all
include directives that are not actually required by the files that
include this header file.
Change-Id: Ifbb38adba434aa19fade4838e30d39884b6e4615
CRs-Fixed: 2366965
Currently ol_if_athvar.h defines the following:
- struct ol_version
- enum ol_ath_tx_ecodes
- struct ol_ath_stats
None of these are used, so remove them. This leaves behind just a set
of #include directives.
Change-Id: Ie8c2ee1fa1b3c34f37430633c28987651e60434b
CRs-Fixed: 2366964
Header file i_bmi.h utilizes several QDF data structures, but it does
not explicitly include the header files which define them. This must
mean that the data structures are being included as a side-effect of
including an unrelated header file.
In an upcoming change the header file ol_if_athvar.h is being
refactored, and that change somehow interferes with the side-effect
that causes the QDF header files to be included by i_bmi.h. So rather
than rely upon the side-effect, explicitly include the necessary
header files.
Change-Id: Icf7d7e4babb14faa41b2ac18c6181a206da9b858
CRs-Fixed: 2366963
In i_bmi.h the hif_reg_based_get_target_info() conditional
compilation references SNOC but this is not a macro defined
by the driver. Replace this with the correct macro, HIF_SNOC.
Change-Id: I52969426c58c078b3dd558e825e2a9fb68394536
CRs-Fixed: 2366958
Change I2c217891d0ca4b503e7388b3ebe2f787e8325af5 ("qcacld-3.0: Add
firmware download support for sdio bus (Part 5 - HIF SDIO)") added a
prototype for ol_get_fw_files() but did not add an implementation.
Since nothing uses this prototype, remove it.
Change-Id: Ie3b9aee767e14acc425ca7c365d10cbb73682fb5
CRs-Fixed: 2366957
The definitions in file ol_ctrl_addba_api.h are unused, so remove the
file and remove the #include directive in the one place it appears.
Change-Id: I9056791b1a21dc8c8e113db2d49468e3abafab34
CRs-Fixed: 2364105
Currently, driver without CNSS module can't be unloaded when firmware
crashed. Without CNSS module, driver should set recovery in progress
flag to false when fw dump finished or failed. Driver should add ramdump
memory address and length too.
Change-Id: I672c50bd90ada7ff47ddff64ff3b313a75faff13
CRs-Fixed: 2343124
As part of WIN BMI modularization, hif_bmi_register_callbacks
are removed from if_pci.c in qcacmn project. As a result,
MCL needs to invoke hif_bmi_register_callbacks in qcacld project.
Fix is to call bus agnostic hif_register_bmi_callbacks method
in bmi_download_firmware.
Change-Id: I72febb02150cf9c672bea4949c53f1b87fa69cdf
CRs-Fixed: 2332266
When downloading a corrupted firmware file that has chunk length in
header which doesn't match the contents, buffer over-read may occur.
To fix it, before downloading the data, ensure the length is equal or
smaller than the left size of the firmware file.
Change-Id: I4e0c6c4423f94f26a8c4573b5d234296890f4ecf
CRs-Fixed: 2314182
Logging macros already print function name by default,
there is no need to print the function name again as part of the log.
Hence, remove the function name or
wherever required use '"%s...", __func__'
Change-Id: Ib6dc6cc03ef148146b1c626271d5ac3b7345c4bb
CRs-Fixed: 2312593
Remaining may be greater than its original value when
(remaining & 0x3)!=0, then OOB occurs when memcpy.To address
this, align the remaining after doing memcpy.
Change-Id: I3e39a791a76a272e82beb6561375e26ca84ec0f4
CRs-Fixed: 2259721
Do not tie up names with mailbox.
This results from SDIO HIF refactoring.
See qcacmn: I12bdf8b07350411093ba35071411525a8333cf93
Change-Id: I99184fbfc95a601f09dba5891098c35d5aa794c5
CRs-Fixed: 2252448
Address the following issues in the core/bmi folder:
CHECK: 'Defintions' may be misspelled - perhaps 'Definitions'?
CHECK: 'initilization' may be misspelled - perhaps 'initialization'?
Change-Id: I649b42e30e10e51c2c734d909a8c9ab2811b9421
CRs-Fixed: 2241939
This reverts Change-Id: I2a4980b6cd4b689da07c34f2acdfaafbbb0715bb.
This change set driver unload state before unload work is scheduled,
which causes async between platform driver and wlan host driver.
Change-Id: Ieaf1cf0a257f69bf8958085c48d9020f1b5ec610
CRs-Fixed: 2124953
Fw assert and unloading can occur at same moment, there is no need
to recover if wlan unloading is ongoing, so skip notifying SSR framework
in such case.
Change-Id: I2176d1505e8e358f4436277e4d0b706923596f27
CRs-Fixed: 2087634
Add ini param 'gtsf_ptp_options' to control
PTP options, it's a bitmap:
bit0 - PTP_OPT_RX(0x1)
set this bit to enable RX time stamping
bit1 - PTP_OPT_TX(0x2)
set this bit to enable TX time stamping
bit2 - PTP_OPT_RAW(0x4)
set this bit to use raw time as timestamp
bit3 - TSF_DBG_FS(0x8)
set this bit to add device attriubte 'tsf' for iface
The default value of gtsf_ptp_options is 0xf
Propagated from qcacld-2.0.
Change-Id: Ie53d503bdd2e85790502583a238ee138f4bcf6c6
CRs-Fixed: 2079466
qcacld-2.0 to qcacld-3.0 propagation
The board data file is selected based on the board id returned
by executing otp binary.
Based on the chip id and the board id, select the appropriate
board data file and download.
The board data file selection is as follows:
Mission mode: bdwlan<chip_id>.b<board_id>
Factory mode: utfbd<chip_id>.b<board_id>
If the auto board data files are not present in the filesystem,
then download the default board data file i.e bdwlan<chip_id>.bin
Change-Id: I26c3eb377513d2c4439eddc4e5dc75eba66c78f4
CRs-Fixed: 805800
qcacld-2.0 to qcacld-3.0 propagation
Cleanup target dump code and add support to dump IRAM region
for different platforms. Update the Target Memory Region for
different platforms.
Change-Id: Ie29fb62e0f1bc279311f77400e8be490ccf987a4
CRs-Fixed: 2088394
In function ol_configure_target cap is used uninitialized when call
pld_get_platform_cap fail, although pld_get_bus_type and
hif_get_bus_type return same bus type, so actually it has no issue here,
but it look not clear here, init it explicitly.
Change-Id: I001956521c62c14b3537b399bbca1791b274a4c7
CRs-Fixed: 2115553
