Bug 205001

Hi,

can you try to load iwlwifi with:

disable_msix=1

and see what happens?

Thanks.

What happens is:

[ 6315.116362] iwlwifi: unknown parameter 'disable_msix' ignored

modinfo shows that iwlwifi has no parameter containing the string "msix".

Oh well.. This is not upstream, but in our internal backport tree :(

So let's do it manually, can you do this:

diff --git a/drivers/net/wireless/intel/iwlwifi/pcie/trans.c b/drivers/net/wireless/intel/iwlwifi/pcie/trans.c
index eb544811759d..63c997e0b800 100644
--- a/drivers/net/wireless/intel/iwlwifi/pcie/trans.c
+++ b/drivers/net/wireless/intel/iwlwifi/pcie/trans.c
@@ -1706,7 +1706,7 @@ iwl_pcie_set_interrupt_capa(struct pci_dev *pdev,
        int max_irqs, num_irqs, i, ret;
        u16 pci_cmd;
 
-       if (!cfg_trans->mq_rx_supported || iwlwifi_mod_params.disable_msix)
+       if (!cfg_trans->mq_rx_supported || iwlwifi_mod_params.disable_msix || true)
                goto enable_msi;
 
        max_irqs = min_t(u32, num_online_cpus() + 2, IWL_MAX_RX_HW_QUEUES);

Your patch doesn't apply to 5.3.1, but I applied this one:

--- linux-5.3.1/drivers/net/wireless/intel/iwlwifi/pcie/trans.c.iwlwifi-msix-diags      2019-09-26 09:25:43.328456574 +0100
+++ linux-5.3.1/drivers/net/wireless/intel/iwlwifi/pcie/trans.c 2019-09-26 09:29:15.764465124 +0100
@@ -1588,7 +1588,7 @@ static void iwl_pcie_set_interrupt_capa(
        int max_irqs, num_irqs, i, ret;
        u16 pci_cmd;
 
-       if (!trans->cfg->mq_rx_supported)
+       if (!trans->cfg->mq_rx_supported || true)
                goto enable_msi;
 
        max_irqs = min_t(u32, num_online_cpus() + 2, IWL_MAX_RX_HW_QUEUES);


With that patch, I've downloaded the Fedora 30 iso.file eight times and the Microcode SW error has not reoccured.

All right, this is interesting...

Thank you.

Just a quick update. I've been trying to recreate this problem on 4.19.75 but seem unable to. I've downloaded the large .iso file numerous times including a couple of tests where two downloads of the file were running concurrently.

I didn't notice the problem on 5.2 series kernels, so I'll try and bisect 5.3 kernels over the next couple of days.

I've bisected and landed at:
3c514bf831ac12356b695ff054bef641b9e99593
iwlwifi: mvm: add a loose synchronization of the NSSN across Rx queues

A kernel built at the parent commit (521dc6c7c74e88fbd02947e4e50a5cb0d49b4395- iwlwiif: mvm: refactor iwl_mvm_notify_rx_queue) does not exhibit the microcode SW error.

I'll attach the bisect log and 'first bad commit' message in a moment.

Created attachment 285275 [details]
bisect log

Created attachment 285277 [details]
first bad commit message

Ok, I can't see how this commit could be causing this, but we did see that this commit is causing other issues. So this is a sort of "known issue". For sure, your report helps us.

The problem is that this commit is needed for another problem that is described in its commit message.

Your work is very useful for us, thank you very much.
The ball is now in our court. Thanks.

On 15/10/2019 18:29, bugzilla-daemon@bugzilla.kernel.org wrote:
> https://bugzilla.kernel.org/show_bug.cgi?id=205001
> 

What's the latest update on this bug? Will it be fixed in 5.4 timescales. It is a regression after all.

> bugzilla@goodbit.net changed:
> 
>            What    |Removed                     |Added
> ----------------------------------------------------------------------------
>                  CC|                            |bugzilla@goodbit.net
>

It is happening to me too, on 5.3 and 5.4rc3, however not on 5.2

*** This bug has been marked as a duplicate of bug 204873 ***