P2020 boot

[tonykleman]

Booting from GPCM, local bus.  Not completing boot, it hangs when I try to write the CCSRBAR register.  In fact it hangs when I do any STW command. 
Im looking at this on an ATE so I control the content of the RAM.  The part goes to rest vector, I branch and write th MASx registers, run tlbwe and isynce.  Next I try to set the CCSRBAR and that is when I have problems.  This is after power on reset and so I should have all necessary permissions.   I believe I'm missing something basic, an ideas?

ba  0xFFFFF000

lis  r5, 0x0000
ori  r5, r5, 0x0000
mtspr 624, r5

lis  r5, 0xC000
ori  r5, r5, 0x0500
mtspr 625, r5

lis  r5, 0xe000
ori  r5, r5, 0x000a
mtspr 626, r5

lis  r5, 0xe000
ori  r5, r5, 0x003f
mtspr 627, r5

tlbwe 0,0,0
msync
isync

//////////////////////////////////
// relocate CCSRBAR at 0xe0000000
//////////////////////////////////
// read current value (forces all accesses to configuration space to complete)
lis  r5, 0xff70
ori  r5, r5, 0x0000
lwz  r4, 0(r5)
isync

//Write the new value to CCSRBAR
lis  r5, 0x000e
ori  r5, r5, 0x0000
lis  r4, 0xFF700000@ha
stw  r5, 0xFF700000@l(r4)

nop

nop

......

[scottwood]

Please double-check your TLB setup.  You put 0 into MAS0, selecting TLB0.  However, TLB0 does not support IPROT nor page sizes other than 4K, both of which you've requested in MAS1.  Also, where did you set up the TLB entry pointing at 0xff700000?

[tonykleman]

Hi Scott,

I'm realizing I'm not setting up the TLB correctly. I've modified code to this, but get the same results. I'm still missing something.

ba 0xFFFFF000

lis r5, 0x1001

ori r5, r5, 0x0000

mtspr 624, r5 //MAS0

lis r5, 0xC000

ori r5, r5, 0x0500

mtspr 625, r5 //MAS1

lis r5, 0xe000

ori r5, r5, 0x000a

mtspr 626, r5 //MAS2

lis r5, 0xe000

ori r5, r5, 0x003f

mtspr 627, r5 //MAS3

tlbwe 1,1,0 //TLB1

msync

isync

lis r5, 0x1002

ori r5, r5, 0x0000

mtspr 624, r5 // MAS0

lis r5, 0xC000

ori r5, r5, 0x0900

mtspr 625, r5 //MAS1

lis r5, 0xf000

ori r5, r5, 0x000a

mtspr 626, r5 //MAS2

lis r5, 0xf000

ori r5, r5, 0x003f

mtspr 627, r5 //MAS3

tlbwe 1,2,0 /TLB2 to set up eLBC and CCSR

msync

isync

//////////////////////////////////

// relocate CCSRBAR at 0xe0000000

//////////////////////////////////

// read current value (forces all accesses to configuration space to complete)

lis r5, 0xff70

ori r5, r5, 0x0000

lwz r4, 0(r5)

isync

//Write the new value to CCSRBAR

lis r5, 0x000e

ori r5, r5, 0x0000

lis r4, 0xFF700000@ha<mailto:0xFF700000@ha>

stw r5, 0xFF700000@l(r4<mailto:0xFF700000@l(r4>) //hangs up on this command

Tony

[scottwood]

The entry you're putting in ESEL 2 is overlapping the boot TLB entry, which is illegal.  Not sure if that's what's causing your problem, though.

You shouldn't enable execute permission on CCSR or other I/O -- the guarded bit doesn't prevent speculative execution.  This is very unlikely to be your problem, but it's something you should be aware of.

Why are you specifying operands to tlbwe?  It doesn't take any on e500 -- MAS0 is used to identify the entry to write.

Are you sure that the stw is what hangs?  What code is after the stw?  Is there any chance your debugger is depending on the location of CCSR not changing?

[tonykleman]

I change the tlbe2 to be 1MB so it shouldn't be overlapping anymore. And I dropped the operands for tlbwe. I know it is the stw command because if I replace with a nop it continues on. All I have after the stw right now is nop's for debug. I found that this also fails.

lis r7,0xff70

ori r7,r7,0000

lwz r4, 0(r7)

So not just writing to CCSBAR, but reading from CCSBAR. Somehow I must not be defining memory correctly.

Tony Kleman

Test Engineer

Integra Technologies LLC

3450 N. Rock Rd. bld 100

Wichita, KS 67226

316-630-6809 phone/fax

[scottwood]

So the first lwz doesn't hang but a second one does?  What happens if you just do the store without an lwz first?

Just to be sure, when you adjusted the size of tlbe2, you adjusted the base address (in both MAS2 and MAS3) to 0xff700000, right?

[tonykleman]

I supposed this makes some sort of sense. FF70_0000 is not in the FFFF_F000 4k space defined in boot. But I'm supposed to write that address to define the CCSRBAR. Am I missing an indirect addressing somewhere?

Tony

[scottwood]

There shouldn't be anything required other than setting up the TLB entry.

Has your debug environment set up any LAWs, BTW?  While CCSR normally takes precedence over LAWs, there is a note in the manual that says that undefined behavior can occur if a DDR LAW overlaps CCSR.

[tonykleman]

It never gets to the LAW section before hanging. I must be setting up the TLB wrong then if nothing else is required.

ba 0xFFFFF000

lis r5, 0x1001

ori r5, r5, 0x0000

mtspr 624, r5

lis r5, 0xC000

ori r5, r5, 0x0500

mtspr 625, r5

lis r5, 0xe000

ori r5, r5, 0x000a

mtspr 626, r5

lis r5, 0xe000

ori r5, r5, 0x003f

mtspr 627, r5

tlbwe

isync

lis r5, 0x1002

ori r5, r5, 0x0000

mtspr 624, r5

lis r5, 0xC000

ori r5, r5, 0x0500

mtspr 625, r5

lis r5, 0xf000

ori r5, r5, 0x000a

mtspr 626, r5

lis r5, 0xf000

ori r5, r5, 0x003f

mtspr 627, r5

tlbwe

isync

//////////////////////////////////

// relocate CCSRBAR at 0xe0000000

//////////////////////////////////

// read current value (forces all accesses to configuration space to complete)

lis r5, 0xff70

ori r5, r5, 0x0000

lwz r4, 0(r5) /// yes this is where it hangs

isync

//Write the new value to CCSRBAR

lis r5, 0x000e

ori r5, r5, 0x0000

lis r4, 0xFF700000@ha

stw r5, 0xFF700000@l(r4)

Tony Kleman

Test Engineer

Integra Technologies LLC

3450 N. Rock Rd. bld 100

Wichita, KS 67226

316-630-6809 phone/fax

[scottwood]

From a few comments back: "Just to be sure, when you adjusted the size of tlbe2, you adjusted the base address (in both MAS2 and MAS3) to 0xff700000, right?"

You changed the size of the entry to 1 MiB, but you left the base at 0xf0000000.  So the entry covers from 0xf0000000 to 0xf00fffff, and nothing covers 0xff700000.

[tonykleman]

No I didn't. This is the code snippet I got from freescale

I get confused about the location of CCSRBAR. From the P2020 reference manual, but all code snippets I see is FF70_0000. I've tried FFE0.

2.2.4 CCSR Address Map

The full register address of any CCSR is comprised of the CCSR window base address, specified in

CCSRBAR (default address 0x0_FFE0_0000), plus the functional block base address, plus the specific

register's offset within that block.

4.3.1.1 Accessing Configuration, Control, and Status Registers

The configuration, control, and status registers are memory mapped. The set of configuration, control, and

status registers occupies a 1-Mbyte region of memory. Their location is programmable using the CCSR

base address register (CCSRBAR). The default base address for the configuration, control, and status

registers is 0xFFE0_0000 (CCSRBAR = 0x000F_FE00). CCSRBAR itself is part of the local access block

of CCSR memory, which begins at offset 0x0 from CCSRBAR. Because CCSRBAR is at offset 0x0 from

the beginning of the local access registers, CCSRBAR always points to itself. The contents of CCSRBAR

are broadcast internally in the P2020 to all functional units that need to be able to identify or create

configuration transactions.

// 1MB TLB1 entry 1: 0xE0000000 - 0xE00FFFFF; for CCSR cache inhibited

lis r5, 0x1001

ori r5, r5, 0x0000

mtspr 624, r5

lis r5, 0xC000

ori r5, r5, 0x0500

mtspr 625, r5

lis r5, 0xe000

ori r5, r5, 0x000a

mtspr 626, r5

lis r5, 0xe000

ori r5, r5, 0x003f

mtspr 627, r5

tlbwe

msync

isync

// 256MB TLB1 entry 2 - Local Bus : 0xF0000000 - 0xFFFFFFFF

lis r5, 0x1002

ori r5, r5, 0x0000

mtspr 624, r5

lis r5, 0xC000

ori r5, r5, 0x0900

mtspr 625, r5

lis r5, 0xf000

ori r5, r5, 0x000a

mtspr 626, r5

lis r5, 0xf000

ori r5, r5, 0x003f

mtspr 627, r5

tlbwe

msync

isync

Tony Kleman

Test Engineer

Integra Technologies LLC

3450 N. Rock Rd. bld 100

Wichita, KS 67226

316-630-6809 phone/fax

[scottwood]

I'm not sure what you mean by, "No I didn't."  In the code snippet posted on  Mar 15, 2013 10:51 AM, for the second TLB entry, please change SPR 626 from 0xf000000a to 0xff70000a, and SPR 627 from 0xf000003f to 0xff700005.

The default CCSRBAR on P2020 is 0xff700000.  It is U-Boot that relocates it to 0xffe00000, but since you're running straight from the reset vector U-Boot isn't involved.  What revision of what document do you see 0xffe00000 in?  I see similar language in an older version of the P2020 reference manual, but it says 0xff700000.

Where did you get the code snippet from Freescale?

[tonykleman]

I have changed to this. It no longer hangs up and I'm trying to determine if it is doing what I want. The manual I have is 03/2011

ba 0xFFFFF000

lis r5, 0x1001

ori r5, r5, 0x0000

mtspr 624, r5

lis r5, 0xC000

ori r5, r5, 0x0500

mtspr 625, r5

• lis r5, 0xFF70

ori r5, r5, 0x000a

mtspr 626, r5

• lis r5, 0xFF70

ori r5, r5, 0x003f

mtspr 627, r5

tlbwe

isync

lis r5, 0x1002

ori r5, r5, 0x0000

mtspr 624, r5

lis r5, 0xC000

ori r5, r5, 0x0500

mtspr 625, r5

• lis r5, 0xe000

ori r5, r5, 0x000a

mtspr 626, r5

• lis r5, 0xe000

ori r5, r5, 0x003f

mtspr 627, r5

tlbwe

isync

//////////////////////////////////

// relocate CCSRBAR at 0xe0000000

//////////////////////////////////

// read current value (forces all accesses to configuration space to complete)

lis r5, 0xff70

ori r5, r5, 0x0000

lwz r4, 0(r5)

isync

//Write the new value to CCSRBAR

lis r5, 0x000e

ori r5, r5, 0x0000

lis r4, 0xFF700000@ha

stw r5, 0xFF700000@l(r4)

Tony Kleman

Test Engineer

Integra Technologies LLC

3450 N. Rock Rd. bld 100

Wichita, KS 67226

316-630-6809 phone/fax

[scottwood]

The latest P2020RM is Rev 2, dated 12/2012, and is available from our website.  That said, the older version I was talking about was from 2009, so the error must have been introduced and then fixed.

Again, I recommend clearing the execute bits from MAS3 on these TLB entries -- you're not actually going to execute code from CCSR, and it leaves you open to speculative fetches that could have side effects.  It's unlikely, but possible.

[tonykleman]

Hi Scott

It turns out I can do a lwz and stw as long as it is to the 4K boot space. I can do a mfspr and verify MASx are set correctly. It only hangs when I try to write to FF70_0000

Tony

[scottwood]

So it was hanging on the lwz and not the stw in the previous code snippet?  Can you show what the code looks like now?

[tonykleman]

I'm realizing I'm not setting up the TLB correctly.  I've modified code to this, but get the same results.  I'm still missing something.

ba  0xFFFFF000

lis  r5, 0x1001
ori  r5, r5, 0x0000
mtspr 624, r5   //MAS0

lis  r5, 0xC000
ori  r5, r5, 0x0500
mtspr 625, r5  //MAS1

lis  r5, 0xe000
ori  r5, r5, 0x000a
mtspr 626, r5  //MAS2

lis  r5, 0xe000
ori  r5, r5, 0x003f
mtspr 627, r5  //MAS3

tlbwe 1,1,0  //TLB1
msync
isync

lis  r5, 0x1002
ori  r5, r5, 0x0000
mtspr 624, r5 // MAS0

lis  r5, 0xC000
ori  r5, r5, 0x0900
mtspr 625, r5 //MAS1

lis  r5, 0xf000
ori  r5, r5, 0x000a
mtspr 626, r5 //MAS2

lis  r5, 0xf000
ori  r5, r5, 0x003f
mtspr 627, r5 //MAS3

tlbwe 1,2,0  /TLB2 to set up eLBC and CCSRBAR
msync
isync

//////////////////////////////////
// relocate CCSRBAR at 0xe0000000
//////////////////////////////////
// read current value (forces all accesses to configuration space to complete)
lis  r5, 0xff70
ori  r5, r5, 0x0000
lwz  r4, 0(r5)
isync

//Write the new value to CCSRBAR
lis  r5, 0x000e
ori  r5, r5, 0x0000
lis  r4, 0xFF700000@ha
stw  r5, 0xFF700000@l(r4)    //hangs up on this command