4.1: XPA breaks Alchemy

4.1: XPA breaks Alchemy

[Manuel Lauss]

Hi Steve,

your patch commit c5b367835cfc7a8ef53b9670a409ffcc95194344
("MIPS: Add support for XPA") breaks userspace on my Alchemy systems,
everything is killed with error -14:

Kernel panic - not syncing: Requested init /bin/sh failed (error -14).

Reverting that patch fixes everything.  I bet it's related to the 36bit
address space in some way.

Thanks!
     Manuel

Re: 4.1: XPA breaks Alchemy

[Manuel Lauss]

Ping

This issue still exists in latest -git.  It goes away when
CONFIG_PHYS_ADDR_T_64BIT
is disabled, but then we lose PCI and PCMCIA (and other chipselect signals which
are mapped beyond 4G).

I'm going to look into this today if nobody beats me to it.

Manuel


On Thu, Apr 30, 2015 at 7:34 PM, Manuel Lauss <manuel.lauss@gmail.com> wrote:
> your patch commit c5b367835cfc7a8ef53b9670a409ffcc95194344
> ("MIPS: Add support for XPA") breaks userspace on my Alchemy systems,
> everything is killed with error -14:
>
> Kernel panic - not syncing: Requested init /bin/sh failed (error -14).
>
> Reverting that patch fixes everything.  I bet it's related to the 36bit
> address space in some way.

Re: 4.1: XPA breaks Alchemy

[Steven J. Hill]

On 07/14/2015 01:28 AM, Manuel Lauss wrote:
> Ping
>
> This issue still exists in latest -git.  It goes away when
> CONFIG_PHYS_ADDR_T_64BIT
> is disabled, but then we lose PCI and PCMCIA (and other chipselect signals which
> are mapped beyond 4G).
>
> I'm going to look into this today if nobody beats me to it.
>
> Manuel
>
Hey Manuel.

I'm sorry that I did not pay attention and see this on the mailing list. 
I will get a fix out today (tomorrow your time).

Steve

Re: 4.1: XPA breaks Alchemy

[Paul Burton]

On Tue, Nov 03, 2015 at 07:12:55AM -0600, Steven J. Hill wrote:
> On 07/14/2015 01:28 AM, Manuel Lauss wrote:
> >Ping
> >
> >This issue still exists in latest -git.  It goes away when
> >CONFIG_PHYS_ADDR_T_64BIT
> >is disabled, but then we lose PCI and PCMCIA (and other chipselect signals which
> >are mapped beyond 4G).
> >
> >I'm going to look into this today if nobody beats me to it.
> >
> >Manuel
> >
> Hey Manuel.
> 
> I'm sorry that I did not pay attention and see this on the mailing list. I
> will get a fix out today (tomorrow your time).
> 
> Steve

Hi Steve,

Did you ever fix this? I don't see anything in patchwork from you since
a long time before this email.

I believe the same issue occurs if you take any system with a 32 bit
kernel (eg. malta or boston) & select ARCH_PHYS_ADDR_T_64BIT.

One issue is that iPTE_SW tries to ori a mode with bits beyond bit 15
set if you are running that 32 bit kernel on a 64 bit CPU, leading to
uasm overflow errors & garbage TLB handlers. Just having it or in all of
mode isn't sufficient to get things working though.

Thanks,
    Paul

Re: 4.1: XPA breaks Alchemy

[Paul Burton]

On Thu, Apr 30, 2015 at 07:34:06PM +0200, Manuel Lauss wrote:
> Hi Steve,
> 
> your patch commit c5b367835cfc7a8ef53b9670a409ffcc95194344
> ("MIPS: Add support for XPA") breaks userspace on my Alchemy systems,
> everything is killed with error -14:
> 
> Kernel panic - not syncing: Requested init /bin/sh failed (error -14).
> 
> Reverting that patch fixes everything.  I bet it's related to the 36bit
> address space in some way.
> 
> Thanks!
>      Manuel

Hi Manuel,

I don't suppose you'd be able to try this kernel branch?

    https://git.linux-mips.org/cgit/paul/linux.git/log/?h=v4.6-tlb

    git://git.linux-mips.org/pub/scm/paul/linux.git -b v4.6-tlb

I'm working on fixing up a number of issues with commit c5b367835cfc
("MIPS: Add support for XPA.") but unfortunately don't have access to
any Alchemy systems to test it myself.

Thanks,
    Paul

Re: 4.1: XPA breaks Alchemy

[Ralf Baechle]

On Thu, Apr 07, 2016 at 01:06:58AM +0100, Paul Burton wrote:

> 
> I don't suppose you'd be able to try this kernel branch?
> 
>     https://git.linux-mips.org/cgit/paul/linux.git/log/?h=v4.6-tlb
> 
>     git://git.linux-mips.org/pub/scm/paul/linux.git -b v4.6-tlb
> 
> I'm working on fixing up a number of issues with commit c5b367835cfc
> ("MIPS: Add support for XPA.") but unfortunately don't have access to
> any Alchemy systems to test it myself.

The unique architecural feature of Alchemy is that it has devices such as
the PCI bus outside the low 4GB of physical address space.  So I'd
suspect something is wrong there.

Everybody is running Sibyte 64 bit; I wonder if highmem with Sibyte is
also affected.

  Ralf

Re: 4.1: XPA breaks Alchemy

[Manuel Lauss]

On Thu, Apr 7, 2016 at 2:06 AM, Paul Burton <paul.burton@imgtec.com> wrote:
> git://git.linux-mips.org/pub/scm/paul/linux.git -b v4.6-tlb


Hi Paul,

This branch does indeed boot again, and PCI/PCMCIA work as well.

Thank you for looking at this!
      Manuel

Re: 4.1: XPA breaks Alchemy

[Paul Burton]

On Thu, Apr 07, 2016 at 08:40:17AM +0200, Manuel Lauss wrote:
> On Thu, Apr 7, 2016 at 2:06 AM, Paul Burton <paul.burton@imgtec.com> wrote:
> > git://git.linux-mips.org/pub/scm/paul/linux.git -b v4.6-tlb
> 
> 
> Hi Paul,
> 
> This branch does indeed boot again, and PCI/PCMCIA work as well.
> 
> Thank you for looking at this!
>       Manuel

Thanks for that Manuel :)

Am I OK to add your Tested-by to the current last patch in that branch
("MIPS: Fix MIPS32 36b physical addressing (alchemy, netlogic)")?

Paul

Re: 4.1: XPA breaks Alchemy

[Manuel Lauss]

On Thu, Apr 7, 2016 at 8:46 AM, Paul Burton <paul.burton@imgtec.com> wrote:
> On Thu, Apr 07, 2016 at 08:40:17AM +0200, Manuel Lauss wrote:
>> On Thu, Apr 7, 2016 at 2:06 AM, Paul Burton <paul.burton@imgtec.com> wrote:
>> > git://git.linux-mips.org/pub/scm/paul/linux.git -b v4.6-tlb
>>
>>
>> Hi Paul,
>>
>> This branch does indeed boot again, and PCI/PCMCIA work as well.
>>
>> Thank you for looking at this!
>>       Manuel
>
> Thanks for that Manuel :)
>
> Am I OK to add your Tested-by to the current last patch in that branch
> ("MIPS: Fix MIPS32 36b physical addressing (alchemy, netlogic)")?

Of course, any time.

> Paul

Re: 4.1: XPA breaks Alchemy

[Paul Burton]

On Thu, Apr 07, 2016 at 07:58:13AM +0200, Ralf Baechle wrote:
> On Thu, Apr 07, 2016 at 01:06:58AM +0100, Paul Burton wrote:
> 
> > 
> > I don't suppose you'd be able to try this kernel branch?
> > 
> >     https://git.linux-mips.org/cgit/paul/linux.git/log/?h=v4.6-tlb
> > 
> >     git://git.linux-mips.org/pub/scm/paul/linux.git -b v4.6-tlb
> > 
> > I'm working on fixing up a number of issues with commit c5b367835cfc
> > ("MIPS: Add support for XPA.") but unfortunately don't have access to
> > any Alchemy systems to test it myself.
> 
> The unique architecural feature of Alchemy is that it has devices such as
> the PCI bus outside the low 4GB of physical address space.  So I'd
> suspect something is wrong there.
> 
> Everybody is running Sibyte 64 bit; I wonder if highmem with Sibyte is
> also affected.
> 
>   Ralf

Hi Ralf,

The problem is anywhere that formerly used 64 bit physical addresses
(CONFIG_PHYS_ADDR_T_64BIT) with a MIPS32 CPU (CONFIG_CPU_MIPS32), since
the XPA support essentially clobbered all that code with the XPA
implementation rather than treating them as the distinct cases that they
are. My fix is over here, and I'll submit it for merging in the v4.7
cycle (I guess I could submit now as it's a fix, but it's built atop
some rework of pgtable-bits.h to make it more readable so would need
that to go in too):

    https://git.linux-mips.org/cgit/paul/linux.git/commit/?h=v4.6-tlb&id=3a74e3b7bcb1b392da2400ff27ee4e41989dd54f

Thanks,
    Paul

Re: 4.1: XPA breaks Alchemy

[Maciej W. Rozycki]

On Thu, 7 Apr 2016, Ralf Baechle wrote:

> Everybody is running Sibyte 64 bit; I wonder if highmem with Sibyte is
> also affected.

 Last time I tried SiByte with a 32-bit kernel (a couple years ago) it 
didn't boot at all in the first place.

 FWIW,

  Maciej