Re: [BUG] iwlwifi: card unusable after firmware crash

[Rui Salvaterra]

Hi again, Emmanuel,

On Wed, 9 Dec 2020 at 21:07, Emmanuel Grumbach <egrumbach@xxxxxxxxx> wrote:
>
> Indeed, the bit is reverse logic. So we can put that aside.
> Frankly, I have no clue. You can try our backport tree to bisect,
> should be easier..
> What I see here is that your GP_CTRL value is 080003d8
>
> #define CSR_GP_CNTRL_REG_FLAG_HW_RF_KILL_SW          (0x08000000)
> which means sense since apparently, HW RF-Kill was asserted.
> #define CSR_GP_CNTRL_REG_FLAG_GOING_TO_SLEEP         (0x00000010)
> Which means that the device is going to sleep... And that's the problem:
>
> iwl_trans_pcie_grab_nic_access:
>         ret = iwl_poll_bit(trans, CSR_GP_CNTRL,
>                            CSR_GP_CNTRL_REG_VAL_MAC_ACCESS_EN,
>                            (CSR_GP_CNTRL_REG_FLAG_MAC_CLOCK_READY |
>                             CSR_GP_CNTRL_REG_FLAG_GOING_TO_SLEEP), 15000);
>         if (unlikely(ret < 0)) {
>                 u32 cntrl = iwl_read32(trans, CSR_GP_CNTRL);
>
>                 WARN_ONCE(1,
>                           "Timeout waiting for hardware access
> (CSR_GP_CNTRL 0x%08x)\n",
>                           cntrl);
>
> but I'd expect the splat in your log...
> Or maybe you can't load the firmware?

Well, my kernel doesn't have any modules, it's all built-in. The
firmware is obviously loading fine, otherwise the card wouldn't work,
but yeah, that WARN_ONCE hasn't triggered at all.

> Can you try this:
> diff --git a/drivers/net/wireless/intel/iwlwifi/pcie/trans.c
> b/drivers/net/wireless/intel/iwlwifi/pcie/trans.c
> index 2fffbbc8462f..748300752630 100644
> --- a/drivers/net/wireless/intel/iwlwifi/pcie/trans.c
> +++ b/drivers/net/wireless/intel/iwlwifi/pcie/trans.c
> @@ -2121,6 +2121,7 @@ static bool
> iwl_trans_pcie_grab_nic_access(struct iwl_trans *trans,
>          * track nic_access anyway.
>          */
>         __release(&trans_pcie->reg_lock);
> +       mdelay(1);
>         return true;
>  }
>
> If that helps, then... I'd have no clue why it helps, but this
> specific device caused us trouble like bad timing after
> grab_nic_access..

I'll give it a spin. Nasty hack, but if it works, it works. :)

Thanks,
Rui