RE: [PATCH net-next 3/3] net: stmmac: Introducing support for Page Pool

[Jose Abreu]

From: Jose Abreu <joabreu@xxxxxxxxxxxx>
Date: Jul/22/2019, 15:04:49 (UTC+00:00)

> From: Jon Hunter <jonathanh@xxxxxxxxxx>
> Date: Jul/22/2019, 13:05:38 (UTC+00:00)
> 
> > 
> > On 22/07/2019 12:39, Jose Abreu wrote:
> > > From: Lars Persson <lists@xxxxxxx>
> > > Date: Jul/22/2019, 12:11:50 (UTC+00:00)
> > > 
> > >> On Mon, Jul 22, 2019 at 12:18 PM Ilias Apalodimas
> > >> <ilias.apalodimas@xxxxxxxxxx> wrote:
> > >>>
> > >>> On Thu, Jul 18, 2019 at 07:48:04AM +0000, Jose Abreu wrote:
> > >>>> From: Jon Hunter <jonathanh@xxxxxxxxxx>
> > >>>> Date: Jul/17/2019, 19:58:53 (UTC+00:00)
> > >>>>
> > >>>>> Let me know if you have any thoughts.
> > >>>>
> > >>>> Can you try attached patch ?
> > >>>>
> > >>>
> > >>> The log says  someone calls panic() right?
> > >>> Can we trye and figure were that happens during the stmmac init phase?
> > >>>
> > >>
> > >> The reason for the panic is hidden in this one line of the kernel logs:
> > >> Kernel panic - not syncing: Attempted to kill init! exitcode=0x0000000b
> > >>
> > >> The init process is killed by SIGSEGV (signal 11 = 0xb).
> > >>
> > >> I would suggest you look for data corruption bugs in the RX path. If
> > >> the code is fetched from the NFS mount then a corrupt RX buffer can
> > >> trigger a crash in userspace.
> > >>
> > >> /Lars
> > > 
> > > 
> > > Jon, I'm not familiar with ARM. Are the buffer addresses being allocated 
> > > in a coherent region ? Can you try attached patch which adds full memory 
> > > barrier before the sync ?
> > 
> > TBH I am not sure about the buffer addresses either. The attached patch
> > did not help. Same problem persists.
> 
> OK. I'm just guessing now at this stage but can you disable SMP ?
> 
> We have to narrow down if this is coherency issue but you said that 
> booting without NFS and then mounting manually the share works ... So, 
> can you share logs with same debug prints in this condition in order to 
> compare ?

Jon, I have one ARM based board and I can't face your issue but I 
noticed that my buffer addresses are being mapped using SWIOTLB. Can you 
disable IOMMU support on your setup and let me know if the problem 
persists ?

---
Thanks,
Jose Miguel Abreu