Booting without Uboot on 8548E

Booting without Uboot on 8548E

[M Ptich]

Documentation mentions that the only dependencies between 2.6 Kernel and 
bootloader are:

1. 5 input parameters (bd_t pointer, start and end of command line, start 
and end of ram disk)
2. bd_t pointer and command line must with within 16MB of Kernel
3. Initial TLB1 entry must have IPROT=1

Slowly, we are finding that this list is not complete. For example, address 
calculation at the start of platform_init (r3 + ...) assumes that Kernel has 
been initially loaded at 0-base virtual address.

And now we are stuck with apparently another dependency: our Kernel fails to 
advance beyond time calibration because jiffies are not getting incremented, 
so we are not getting timer interrupts. Is it true that Kernel relies on 
EPIC timer set in Uboot ? If not - where the timer insterrupts are 
configured ?

_________________________________________________________________
Try the next generation of search with Windows Live Search today!  
http://imagine-windowslive.com/minisites/searchlaunch/?locale=en-us&source=hmtagline

Re: Booting without Uboot on 8548E

[Benjamin Herrenschmidt]

On Sat, 2006-11-04 at 08:52 +0000, M Ptich wrote:
> Documentation mentions that the only dependencies between 2.6 Kernel and 
> bootloader are:
> 
> 1. 5 input parameters (bd_t pointer, start and end of command line, start 
> and end of ram disk)
> 2. bd_t pointer and command line must with within 16MB of Kernel
> 3. Initial TLB1 entry must have IPROT=1

This is for arch/ppc, which is sort-of deprecated. You should seriously
think about porting over to arch/powerpc instead.

> Slowly, we are finding that this list is not complete. For example, address 
> calculation at the start of platform_init (r3 + ...) assumes that Kernel has 
> been initially loaded at 0-base virtual address.

For which CPU core ? 6xx/7xx/7xxx/G2 cores don't assume such a thing,
the kernel can be loaded anywhere as long as the virtual and physical
addresses are the same.

> And now we are stuck with apparently another dependency: our Kernel fails to 
> advance beyond time calibration because jiffies are not getting incremented, 
> so we are not getting timer interrupts. Is it true that Kernel relies on 
> EPIC timer set in Uboot ? If not - where the timer insterrupts are 
> configured ?

I'm not familiar about which core family the 85xx is. Doesn't it use the
decrementer ?

Ben.

Re: Booting without Uboot on 8548E

[Benjamin Herrenschmidt]

> > Slowly, we are finding that this list is not complete. For example, address 
> > calculation at the start of platform_init (r3 + ...) assumes that Kernel has 
> > been initially loaded at 0-base virtual address.
> 
> For which CPU core ? 6xx/7xx/7xxx/G2 cores don't assume such a thing,
> the kernel can be loaded anywhere as long as the virtual and physical
> addresses are the same.

I've checked and 85xx is some sort of BookE, so you'll have to wait for
the freescale folks here to help you..

(Note that it's unlikely that people will be motivated to help you
considering the rude tone of your initial email)

Ben.

Low memory not pageable on Freescale BookE

[M Ptich]

For our application, we need low memory to be mapped with pages, but it 
turned out that Freescale BookE maps it with TLB1. Was there any particular 
reason why ? I am thinking of changing it, but wondering if there was 
something fundamental behind it.

Regards
Ptich

_________________________________________________________________
All-in-one security and maintenance for your PC.  Get a free 90-day trial! 
http://clk.atdmt.com/MSN/go/msnnkwlo0050000002msn/direct/01/?href=http://www.windowsonecare.com/?sc_cid=msn_hotmail

Re: Low memory not pageable on Freescale BookE

[Kumar Gala]

On Nov 10, 2006, at 2:14 AM, M Ptich wrote:

> For our application, we need low memory to be mapped with pages,  
> but it
> turned out that Freescale BookE maps it with TLB1. Was there any  
> particular
> reason why ? I am thinking of changing it, but wondering if there was
> something fundamental behind it.

Why do you need low kernel memory to be pageable?

- kumar

KGDB support in 2.6.18 for 85xx platforms

[M Ptich]

PPC_GEN550 config does not seem to be supported for 85xx in arch/powepc of 
2.6.18 kernel. Does it mean that KGDB and early blocking output will not 
work there ?

Ptich

_________________________________________________________________
Get FREE company branded e-mail accounts and business Web site from 
Microsoft Office Live 
http://clk.atdmt.com/MRT/go/mcrssaub0050001411mrt/direct/01/

Re: Low memory not pageable on Freescale BookE

[M Ptich]

>Why do you need low kernel memory to be pageable?

Mainly for debugging purposes - this way we can have guardbands for our 
application stacks, and also for proprietary software that tracks memory 
accesses (for corruption detection).

-Ptich

>>For our application, we need low memory to be mapped with pages,  but it
>>turned out that Freescale BookE maps it with TLB1. Was there any  
>>particular
>>reason why ? I am thinking of changing it, but wondering if there was
>>something fundamental behind it.
>
>Why do you need low kernel memory to be pageable?
>
>- kumar

_________________________________________________________________
All-in-one security and maintenance for your PC.  Get a free 90-day trial! 
http://clk.atdmt.com/MSN/go/msnnkwlo0050000002msn/direct/01/?href=http://www.windowsonecare.com/?sc_cid=msn_hotmail

Re: Low memory not pageable on Freescale BookE

[Kumar Gala]

On Nov 14, 2006, at 6:38 PM, M Ptich wrote:

>> Why do you need low kernel memory to be pageable?
>
> Mainly for debugging purposes - this way we can have guardbands for  
> our application stacks, and also for proprietary software that  
> tracks memory accesses (for corruption detection).

If you are doing at the application level then its already paged.   
The TLB1 stuff is just for kernel memory.

- k

Re: Booting without Uboot on 8548E

[M Ptich]

>You should seriously think about porting
>over to arch/powerpc instead.

8548E is E500v2 core. Our Kernel is from Freescale, it is using arch/ppc. I 
am not sure how well (if at all) E500 is supported in the arch/powerpc from 
the FSF ditribution.

>I'm not familiar about which core family the 85xx is. Doesn't it use the
>decrementer ?

Actually E500 could use Decrementer, with its Autoreload feature to get an 
exact time tick. I did not think about it, will investigate.

M Ptich



>From: Benjamin Herrenschmidt <benh@kernel.crashing.org>
>To: M Ptich <ptich@hotmail.com>
>CC: linuxppc-dev@ozlabs.org
>Subject: Re: Booting without Uboot on 8548E
>Date: Sat, 04 Nov 2006 20:13:40 +1100
>MIME-Version: 1.0
>Received: from gate.crashing.org ([63.228.1.57]) by 
>bay0-mc7-f7.bay0.hotmail.com with Microsoft SMTPSVC(6.0.3790.2444); Sat, 4 
>Nov 2006 01:13:43 -0800
>Received: from [127.0.0.1] (localhost.localdomain [127.0.0.1])by 
>gate.crashing.org (8.13.8/8.13.8) with ESMTP id kA49DeGf013357;Sat, 4 Nov 
>2006 03:13:41 -0600
>X-Message-Info: LsUYwwHHNt3660MmjhEvYg2f34OAemlKtU9j2Z7TuGo=
>References: <BAY102-F21BB220054F2541A7614E4A9FD0@phx.gbl>
>X-Mailer: Evolution 2.8.1 Return-Path: benh@kernel.crashing.org
>X-OriginalArrivalTime: 04 Nov 2006 09:13:43.0980 (UTC) 
>FILETIME=[86C386C0:01C6FFF1]
>
>On Sat, 2006-11-04 at 08:52 +0000, M Ptich wrote:
> > Documentation mentions that the only dependencies between 2.6 Kernel and
> > bootloader are:
> >
> > 1. 5 input parameters (bd_t pointer, start and end of command line, 
>start
> > and end of ram disk)
> > 2. bd_t pointer and command line must with within 16MB of Kernel
> > 3. Initial TLB1 entry must have IPROT=1
>
>This is for arch/ppc, which is sort-of deprecated. You should seriously
>think about porting over to arch/powerpc instead.
>
> > Slowly, we are finding that this list is not complete. For example, 
>address
> > calculation at the start of platform_init (r3 + ...) assumes that Kernel 
>has
> > been initially loaded at 0-base virtual address.
>
>For which CPU core ? 6xx/7xx/7xxx/G2 cores don't assume such a thing,
>the kernel can be loaded anywhere as long as the virtual and physical
>addresses are the same.
>
> > And now we are stuck with apparently another dependency: our Kernel 
>fails to
> > advance beyond time calibration because jiffies are not getting 
>incremented,
> > so we are not getting timer interrupts. Is it true that Kernel relies on
> > EPIC timer set in Uboot ? If not - where the timer insterrupts are
> > configured ?
>
>I'm not familiar about which core family the 85xx is. Doesn't it use the
>decrementer ?
>
>Ben.
>
>

_________________________________________________________________
Get FREE company branded e-mail accounts and business Web site from 
Microsoft Office Live 
http://clk.atdmt.com/MRT/go/mcrssaub0050001411mrt/direct/01/

Re: Booting without Uboot on 8548E

[Andy Fleming]

On Nov 4, 2006, at 03:28, M Ptich wrote:

>> You should seriously think about porting
>> over to arch/powerpc instead.
>
> 8548E is E500v2 core. Our Kernel is from Freescale, it is using  
> arch/ppc. I
> am not sure how well (if at all) E500 is supported in the arch/ 
> powerpc from
> the FSF ditribution.


The 8548 is supported in arch/powerpc as of 2.6.18.

But moving to arch/powerpc (while a good idea) is not going to solve  
your problems.  There are bound to be a large number of things the  
kernel expects to have done for it by the firmware.  In arch/powerpc,  
the 8548 kernel expects PCI to be set up, for instance (We plan to  
fix that, but it involves porting some code that only exists in arch/ 
ppc).

Andy

Re: Booting without Uboot on 8548E

[Benjamin Herrenschmidt]

> But moving to arch/powerpc (while a good idea) is not going to solve  
> your problems.  There are bound to be a large number of things the  
> kernel expects to have done for it by the firmware.  In arch/powerpc,  
> the 8548 kernel expects PCI to be set up, for instance (We plan to  
> fix that, but it involves porting some code that only exists in arch/ 
> ppc).

I hope you don't mean porting pci-auto over :)

While It was good to have back then, we should really fix out PCI layer
to properly assign unassigned devices itself.

There is much work to do to cleanup ppc32, ppc64 and core PCI so they
can deal with all the various setups around, but it's probably a
worthwhile goal.

Ben.

Re: Booting without Uboot on 8548E

[Dan Malek]

On Nov 4, 2006, at 3:52 AM, M Ptich wrote:

> Documentation mentions that the only dependencies between 2.6  
> Kernel and
> bootloader are:
>
> 1. 5 input parameters (bd_t pointer, start and end of command line,  
> start
> and end of ram disk)
> 2. bd_t pointer and command line must with within 16MB of Kernel
> 3. Initial TLB1 entry must have IPROT=1

I don't know where these are "documented", but there are
many more dependencies. :-)  Take a look at the cpu initialization
code from u-boot and you will notice how the TLB CAMs need
to be initialized with the various memory maps.  All of the local
bus windows likewise need to be properly set up and match
the kernel configuration.

> And now we are stuck with apparently another dependency: our Kernel  
> fails to
> advance beyond time calibration because jiffies are not getting  
> incremented,

The time base needs to be enabled in HID0 prior to calling
the kernel.

All of these "dependencies" are typically things a boot rom is
going to also need for proper functioning.  We just chose not
to do it again in the kernel.  You are going to find more.

Thanks.


	-- Dan