Re: Today's linux-next iwlwifi broken

[Heiner Kallweit <hkall@kernel.org>]

On 12.02.2021 15:38, Sedat Dilek wrote:
> On Fri, Feb 12, 2021 at 3:15 PM Heiner Kallweit <hkall@kernel.org> wrote:
>>
>> On 12.02.2021 13:21, Heiner Kallweit wrote:
>>> On 11.02.2021 21:23, Luca Coelho wrote:
>>>> On Thu, 2021-02-11 at 17:10 +0100, Heiner Kallweit wrote:
>>>>> Today's linux-next has a bunch of issues, at least with my AX210 (0024) card.
>>>>>
>>>>> First I got a GPF in iwl_request_firmware(), cfg->fw_name_pre was invalid.
>>>>> After some digging into it I found that 6f60fb03c8e7 ("iwlwifi: move SnJ
>>>>> and So rules to the new tables") removed the entries for some cards.
>>>>> Obviously 5febcdef3090 and 9b15596c5006 weren't synced with some parallel
>>>>> development.
>>>>> After adding the entry for 0024 I now get the following lockdep warning.
>>>>> Isn't there any regression testing with lockdep enabled?
>>>>> Also warning "api flags index 2 larger than supported by driver" is still
>>>>> there, IIRC we talked about this weeks ago.
>>>>>
>>>>>
>>>>> [    4.561774] iwlwifi 0000:01:00.0: enabling device (0000 -> 0002)
>>>>> [    4.630849] iwlwifi 0000:01:00.0: Direct firmware load for iwlwifi-ty-a0-gf-a0-61.ucode failed with error -2
>>>>> [    4.639817] iwlwifi 0000:01:00.0: api flags index 2 larger than supported by driver
>>>>> [    4.640062] iwlwifi 0000:01:00.0: TLV_FW_FSEQ_VERSION: FSEQ Version: 93.8.73.28
>>>>> [    4.642138] iwlwifi 0000:01:00.0: loaded firmware version 60.84d9abea.0 ty-a0-gf-a0-60.ucode op_mode iwlmvm
>>>>> [    4.760541] iwlwifi 0000:01:00.0: Detected Intel(R) Wi-Fi 6 AX210 160MHz, REV=0x420
>>>>> [    4.942350] irq/131-iwlwifi/2094 just changed the state of lock:
>>>>> [    4.942358] ffffa3fc07f080b0 (&rxq->lock){+.-.}-{2:2}, at: iwl_pcie_rx_handle+0x7c/0x8b0 [iwlwifi]
>>>>> [    4.942504] 1 lock held by irq/131-iwlwifi/2094:
>>>>> [    4.942511]  #0: ffffa3fc0855c170 (sync_cmd_lockdep_map){+.+.}-{0:0}, at: iwl_pcie_irq_rx_msix_handler+0x47/0x100 [iwlwifi]
>>>>> [    4.943219]    _iwl_pcie_rx_init+0x208/0x6e0 [iwlwifi]
>>>>> [    4.943242]    iwl_pcie_gen2_rx_init+0x20/0x30 [iwlwifi]
>>>>> [    4.943264]    iwl_trans_pcie_gen2_start_fw+0x1e3/0x350 [iwlwifi]
>>>>> [    4.943289]    iwl_mvm_load_ucode_wait_alive+0xfc/0x440 [iwlmvm]
>>>>> [    4.943329]    iwl_run_unified_mvm_ucode+0xa5/0x250 [iwlmvm]
>>>>> [    4.943360]    iwl_run_init_mvm_ucode+0x292/0x370 [iwlmvm]
>>>>> [    4.943391]    iwl_op_mode_mvm_start+0x7f4/0xb60 [iwlmvm]
>>>>> [    4.943424]    _iwl_op_mode_start.isra.0+0x42/0x80 [iwlwifi]
>>>>> [    4.943446]    iwl_opmode_register+0x70/0xe0 [iwlwifi]
>>>>> [    4.943467]    iwl_mvm_init+0x35/0x1000 [iwlmvm]
>>>>> [    4.943588]                     _iwl_pcie_rx_init+0x140/0x6e0 [iwlwifi]
>>>>> [    4.943612]                     iwl_pcie_gen2_rx_init+0x20/0x30 [iwlwifi]
>>>>> [    4.943635]                     iwl_trans_pcie_gen2_start_fw+0x1e3/0x350 [iwlwifi]
>>>>> [    4.943662]                     iwl_mvm_load_ucode_wait_alive+0xfc/0x440 [iwlmvm]
>>>>> [    4.943695]                     iwl_run_unified_mvm_ucode+0xa5/0x250 [iwlmvm]
>>>>> [    4.943728]                     iwl_run_init_mvm_ucode+0x292/0x370 [iwlmvm]
>>>>> [    4.943760]                     iwl_op_mode_mvm_start+0x7f4/0xb60 [iwlmvm]
>>>>> [    4.943793]                     _iwl_op_mode_start.isra.0+0x42/0x80 [iwlwifi]
>>>>> [    4.943816]                     iwl_opmode_register+0x70/0xe0 [iwlwifi]
>>>>> [    4.943838]                     iwl_mvm_init+0x35/0x1000 [iwlmvm]
>>>>> [    4.943954]                     iwl_pcie_rx_handle+0x7c/0x8b0 [iwlwifi]
>>>>> [    4.943978]                     iwl_pcie_napi_poll_msix+0x2a/0x90 [iwlwifi]
>>>>> [    4.944063]                     iwl_pcie_irq_rx_msix_handler+0xbc/0x100 [iwlwifi]
>>>>> [    4.944147]                    _iwl_pcie_rx_init+0x140/0x6e0 [iwlwifi]
>>>>> [    4.944170]                    iwl_pcie_gen2_rx_init+0x20/0x30 [iwlwifi]
>>>>> [    4.944194]                    iwl_trans_pcie_gen2_start_fw+0x1e3/0x350 [iwlwifi]
>>>>> [    4.944220]                    iwl_mvm_load_ucode_wait_alive+0xfc/0x440 [iwlmvm]
>>>>> [    4.944255]                    iwl_run_unified_mvm_ucode+0xa5/0x250 [iwlmvm]
>>>>> [    4.944287]                    iwl_run_init_mvm_ucode+0x292/0x370 [iwlmvm]
>>>>> [    4.944319]                    iwl_op_mode_mvm_start+0x7f4/0xb60 [iwlmvm]
>>>>> [    4.944352]                    _iwl_op_mode_start.isra.0+0x42/0x80 [iwlwifi]
>>>>> [    4.944375]                    iwl_opmode_register+0x70/0xe0 [iwlwifi]
>>>>> [    4.944397]                    iwl_mvm_init+0x35/0x1000 [iwlmvm]
>>>>> [    4.944493]  ... key      at: [<ffffffffc0c2e0c0>] __key.12+0x0/0xffffffffffff0f40 [iwlwifi]
>>>>> [    4.944547]    iwl_pcie_rx_handle+0x7c/0x8b0 [iwlwifi]
>>>>> [    4.944569]    iwl_pcie_napi_poll_msix+0x2a/0x90 [iwlwifi]
>>>>> [    4.944643]    iwl_pcie_irq_rx_msix_handler+0xbc/0x100 [iwlwifi]
>>>>> [    4.944705] CPU: 0 PID: 2094 Comm: irq/131-iwlwifi Not tainted 5.11.0-rc7-next-20210211+ #1
>>>>> [    4.944786]  ? iwl_pcie_rx_handle+0x7c/0x8b0 [iwlwifi]
>>>>> [    4.944825]  ? iwl_pcie_rx_handle+0x7c/0x8b0 [iwlwifi]
>>>>> [    4.944847]  iwl_pcie_rx_handle+0x7c/0x8b0 [iwlwifi]
>>>>> [    4.944878]  iwl_pcie_napi_poll_msix+0x2a/0x90 [iwlwifi]
>>>>> [    4.944949]  ? iwl_pcie_irq_rx_msix_handler+0xab/0x100 [iwlwifi]
>>>>> [    4.944981]  iwl_pcie_irq_rx_msix_handler+0xbc/0x100 [iwlwifi]
>>>>
>>>> Thanks for reporting!
>>>>
>>>> We made a bunch of fixes in this area and Kalle just pulled them to
>>>> wireless-drivers-next.  I believe these changes will be in tomorrow's
>>>> linux-next.  Can you retry it then?
>>>>
>>> I tested today's linux-next:
>>>
>>> The lockdep warning is still there and occurs w/o any traffic on boot.
>>> No CI machine on your side w/ lockdep enabled?
>>>
>>>
>>> [    5.845931] ========================================================
>>> [    5.845938] WARNING: possible irq lock inversion dependency detected
>>> [    5.845946] 5.11.0-rc7-next-20210212+ #1 Not tainted
>>> [    5.845954] --------------------------------------------------------
>>> [    5.845961] irq/131-iwlwifi/2132 just changed the state of lock:
>>> [    5.845969] ffff8ca6c88600b0 (&rxq->lock){+.-.}-{2:2}, at: iwl_pcie_rx_handle+0x7c/0x8b0 [iwlwifi]
>>> [    5.846011] but this lock took another, SOFTIRQ-unsafe lock in the past:
>>> [    5.846019]  (napi_hash_lock){+.+.}-{2:2}
>>> [    5.846023]
>>>
>>>                and interrupts could create inverse lock ordering between them.
>>>
>>> [    5.846039]
>>>                other info that might help us debug this:
>>> [    5.846047]  Possible interrupt unsafe locking scenario:
>>>
>>> [    5.846055]        CPU0                    CPU1
>>> [    5.846061]        ----                    ----
>>> [    5.846066]   lock(napi_hash_lock);
>>> [    5.846074]                                local_irq_disable();
>>> [    5.846081]                                lock(&rxq->lock);
>>> [    5.846090]                                lock(napi_hash_lock);
>>> [    5.846099]   <Interrupt>
>>> [    5.846103]     lock(&rxq->lock);
>>> [    5.846110]
>>>                 *** DEADLOCK ***
>>>
>>> "api flags index 2 larger than supported by driver" is also still there
>>>
>>> 0024 AX210 version is now detected. When I disable the entry I don't get a GPF
>>> any longer but a WARN() is triggered. Not sure why a WARN() was chosen because
>>> the call trace provides no benefit here. More helpful would be a simple
>>> error message stating: "dev id foo: unknown dev sub id bar. Contact maintainers."
>>>
>>>> --
>>>> Cheers,
>>>> Luca.
>>>>
>>> Heiner
>>>
>>
>>
>> One more comment, as I just wanted to upgrade my AX210 firmware version:
>>
>> Latest version in linux-firmware is 59. Seems you're at 62 already.
>> And last but not least: The iwlwifi firmware pollutes the linux-firmware
>> root directory. Wouldn't it be better to move all these firmware files
>> to something like intel/iwlwifi ?
>> Maybe for now you could leave symlinks in the root directory.
>>
> 
> I agree with Heiner, that's something I do/did/will not like: the
> locations for firmware files.
> 
> Here on Debian all firmware files are stored in "/lib/firmware"
> directory - which should be plural "firmwares" IMHO.
> 
> We already have a lot of directories for vendors like "intel",
> "nvidia", "matrox", etc.
> 
> If we move intel stuff around then please do it for intel-gfx stuff
> like "i915" or ucodes like "intel-ucode".
> 
> I vote to move all intel stuff into a single directory called "intel"
> where we have sub-dirs:
> intel > i915
> intel > iwlwifi
> intel > ucode (renamed from "intel-ucode")
> intel > e100
> ...
> 
> But before doing any move - this should be done for all vendors.
> See realtek, broadcom, amd (radeon-gfx), etc.
> Means collect all firmware stuff for bluetooth, wlan, ethernet, gfx
> etc. of a single vendor.
> 
> If we do a cleanup then I vote to do this for all vendors?
> 
> Dunno, who is the maintainer of linux-firmware.
> Guess we need to change code in the Linux sources to say "load
> firmware from </path/to/dirname>".
> ( Should be one-liners. )
> 
> For that request we should open a new thread.
> 
> My €0,99 (if I would be a good guy I have written €1,00).
> 
> - Sedat -
> 

+Josh for linux-firmware