[RFC] powerpc/boot: add kernel,end node to the cuboot target

this could be used by the kexec userland code.

Signed-off-by: Sebastian Siewior <bigeasy@linutronix.de>
---
This is done by the 64bit kexec code allready. The 32bit doesn't use the
device tree at all. I'm not sure whether the node has to be a 32bit or
64bit value.

 arch/powerpc/boot/main.c |    6 +++---
 1 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/arch/powerpc/boot/main.c b/arch/powerpc/boot/main.c
index 27f6af1..a8714cf 100644
--- a/arch/powerpc/boot/main.c
+++ b/arch/powerpc/boot/main.c

@@ -27,7 +27,7 @@ struct addr_range {
 
 #undef DEBUG
 
-static struct addr_range prep_kernel(void)
+static struct addr_range prep_kernel(void *chosen)
 {
 	char elfheader[256];
 	void *vmlinuz_addr = _vmlinux_start;

@@ -72,7 +72,7 @@ static struct addr_range prep_kernel(void)
 		fatal("ran out of data!  only got 0x%x of 0x%lx bytes.\n\r",
 				len, ei.loadsize);
 	printf("done 0x%x bytes\n\r", len);
-
+	setprop_val(chosen, "linux,kernel-end", (u32)(ei.memsize));
 	flush_cache(addr, ei.loadsize);
 
 	return (struct addr_range){addr, ei.memsize};

@@ -180,7 +180,7 @@ void start(void)
 	if (!chosen)
 		chosen = create_node(NULL, "chosen");
 
-	vmlinux = prep_kernel();
+	vmlinux = prep_kernel(chosen);
 	initrd = prep_initrd(vmlinux, chosen,
 			     loader_info.initrd_addr, loader_info.initrd_size);
 	prep_cmdline(chosen);

-- 
1.5.6.5

On Sep 23, 2008, at 2:54 PM, Sebastian Siewior wrote:

this could be used by the kexec userland code.

Signed-off-by: Sebastian Siewior <bigeasy@linutronix.de>
---
This is done by the 64bit kexec code allready. The 32bit doesn't use the
device tree at all. I'm not sure whether the node has to be a 32bit or
64bit value.

arch/powerpc/boot/main.c |    6 +++---
1 files changed, 3 insertions(+), 3 deletions(-)

I missing why you are doing this?  Do you plan on using it w/32-bit kexec?

- k

On Wed Sep 24 at about 05:54:04 EST 2008, Sebastian Siewior wrote:

this could be used by the kexec userland code.

NACK.

First of all, you fail to justify why it is needed.

Second of all, its in the wrong place, if it were needed.  The code
in arch/powerpc/boot does not get executed on all platforms, it is
just an assistance and conversion wrapper for those platforms that
need it.

Third, it will break the 64 bit code worse than it is.  (The kernel
does not delete the property before attemtping to add a new one. The
patch got lost in the maintainer then patchwork shuffeles and will
be resent.)

On the first point:  The only reason the kernel exports it for the
64 bit code is because of limitations of the kernel kexec code
unique to the 64 bit powerpc port.  Specifically, the kernel does
not support the next image destination address being on top of
itself.   This this limitation is unique to 64-bit powerpc kexec
code and shold NOT be propagated to 32-bit powerpc support.  The
reason for it is due the combination of the hypervisor architecture
that only gives us access to a small portion of memory in real mode
(MB out of GB of main memory), a single zone memory management
policy for linux alocation of main memory (which is good from a
memory management perspective but combined with the former means
we can not reasonably permit the generic code to just keep allocating
memory until the chosen pages are in the real mode accesble region),
and the many specific hypervisor interfaces to fill in the page
table required to do a virtural mode access.  The three factors
combine to say the only sane answer is to do the copy under the
kernels virtural mode fault handlers, and therefore the new image
must be loaded using memory other than that occupied by the current
kernel text, its static data and bss (and other mmu related tables).
The reason this limiation is to not too severe is the kernel currently
does not care where it is loaded as it will copy itself to the
beginning of memory.  Note that this does not apply to kdump -- the
panic kernel is loaded into memory reserverd for the purpose by the
first kernel, and that memory never overlaps the kernel, page table,
iommu tables, or other simiar data).  Also, this easing only applies

43 more lines

Kumar Gala wrote:

I missing why you are doing this?  Do you plan on using it w/32-bit kexec?

Yes. I plan to kexec my mpc8544 board.

- k

Sebastian

Milton Miller wrote:

On Wed Sep 24 at about 05:54:04 EST 2008, Sebastian Siewior wrote:

quoted

this could be used by the kexec userland code.

NACK.

[...]
Thanks for explanation. I just went through the kexec user land code and
saw that the 64bit code uses the device tree to obtain some address and
32bit returns a hard coded struct on game cube (unless it was another
console I mixed it up) and nothing on every other platform. Therefore I
tried to add the missing nodes.

If you have any questions about kdump or what needs to happen,
please feel free to contact me either by email or on irc (sometimes
I use mdm other times the email login as my nick, and when connected
I tend to leave it well past the hours I am at the computer).

Thanks for the offer.

milton

Sebastian

* Milton Miller | 2008-09-23 20:24:02 [-0500]:

If you have any questions about kdump or what needs to happen,
please feel free to contact me either by email or on irc (sometimes
I use mdm other times the email login as my nick, and when connected
I tend to leave it well past the hours I am at the computer).

I copied most of the 64bit code to parse the device tree without the pci
nodes & moved it to 32. The userland *could* work, I'm not sure. My
outout is:

|load: entry = 0x80053c flags = 0
|nr_segments = 2
|segment[0].buf   = 0x1002b8f0
|segment[0].bufsz = 80
|segment[0].mem   = (nil)
|segment[0].memsz = 1000
|segment[1].buf   = 0x4803f008
|segment[1].bufsz = 3a3138
|segment[1].mem   = 0x800000
|segment[1].memsz = 3b0000

Now. The entry address in image->start is valid and is the entrypoint of
the "custom" cuImage. Custom means that it does not depend any register
values passed from u-boot (the original one needs a pointer to bd_t).
The only requirement is a valid 1:1 memory mapping.

I learned, that I can not disable the MMU on Book-E so I have to create
a new temporary mapping in my relocate_new_kernel routine. _start is
doing the same thing what I am trying to accomplish: create a new
mapping and don't kill the current one and switch over. This is done by
disabling all mappings but the current, creating a new mapping with
EFN/RPN = 0 and swapping the TS bit in MAS1. This is my current patch
which is not really working:

diff --git a/arch/powerpc/kernel/misc_32.S b/arch/powerpc/kernel/misc_32.S
index 7a6dfbc..49c9c2a 100644
--- a/arch/powerpc/kernel/misc_32.S
+++ b/arch/powerpc/kernel/misc_32.S

@@ -878,22 +878,142 @@ relocate_new_kernel:
 	/* r4 = reboot_code_buffer */
 	/* r5 = start_address      */
 
-	li	r0, 0
+	mr r27, r4
+	mr r28, r5
+	mr r29, r6
+
+	li      r25,0           /* phys kernel start (low) */
+	li      r24,0           /* CPU number */
+	li      r23,0           /* phys kernel start (high) */
+
+
+/* 1. Find the index of the entry we're executing in */
+	bl      invstr                          /* Find our address */
+invstr: mflr    r6                              /* Make it accessible */
+	mfmsr   r7
+	rlwinm  r4,r7,27,31,31                  /* extract MSR[IS] */
+	mfspr   r7, SPRN_PID0
+	slwi    r7,r7,16
+	or      r7,r7,r4
+	mtspr   SPRN_MAS6,r7
+	tlbsx   0,r6                            /* search MSR[IS], SPID=PID0 */
+#ifndef CONFIG_E200
+	mfspr   r7,SPRN_MAS1
+	andis.  r7,r7,MAS1_VALID@h
+	bne     match_TLB

The branch above is taken, so I've found my current mapping

+	mfspr   r7,SPRN_PID1
+	slwi    r7,r7,16
+	or      r7,r7,r4
+	mtspr   SPRN_MAS6,r7
+	tlbsx   0,r6                            /* search MSR[IS], SPID=PID1 */
+	mfspr   r7,SPRN_MAS1
+	andis.  r7,r7,MAS1_VALID@h
+	bne     match_TLB
+	mfspr   r7, SPRN_PID2
+	slwi    r7,r7,16
+	or      r7,r7,r4
+	mtspr   SPRN_MAS6,r7
+	tlbsx   0,r6                            /* Fall through, we had to match */
+#endif
+match_TLB:
+
+	rlwinm  r3,r7,16,20,31                  /* Extract MAS0(Entry) */
+
+	mfspr   r7,SPRN_MAS1                    /* Insure IPROT set */
+	oris    r7,r7,MAS1_IPROT@h
+	mtspr   SPRN_MAS1,r7
+	tlbwe
+
+/* 2. Invalidate all entries except the entry we're executing in */
+	mfspr   r9,SPRN_TLB1CFG
+	andi.   r9,r9,0xfff
+	li      r6,0                            /* Set Entry counter to 0 */
+1:      lis     r7,0x1000                       /* Set MAS0(TLBSEL) = 1 */
+	rlwimi  r7,r6,16,4,15                   /* Setup MAS0 = TLBSEL | ESEL(r6) */
+	mtspr   SPRN_MAS0,r7
+	tlbre
+	mfspr   r7,SPRN_MAS1
+	rlwinm  r7,r7,0,2,31                    /* Clear MAS1 Valid and IPROT */
+	cmpw    r3,r6
+	beq     skpinv                        /* Dont update the current execution TLB */
+	mtspr   SPRN_MAS1,r7
+	tlbwe
+	isync

114 more lines

milton

Sebastian

On Sep 29, 2008, at 3:04 PM, Sebastian Siewior wrote:

* Milton Miller | 2008-09-23 20:24:02 [-0500]:

quoted

If you have any questions about kdump or what needs to happen,
please feel free to contact me

I copied most of the 64bit code to parse the device tree without the pci
nodes & moved it to 32. The userland *could* work, I'm not sure. My
outout is:

|load: entry = 0x80053c flags = 0
|nr_segments = 2
|segment[0].buf   = 0x1002b8f0
|segment[0].bufsz = 80
|segment[0].mem   = (nil)
|segment[0].memsz = 1000
|segment[1].buf   = 0x4803f008
|segment[1].bufsz = 3a3138
|segment[1].mem   = 0x800000
|segment[1].memsz = 3b0000

I would expect a third segment (kernel/zImage, dtb, and purgatory), but its not clear that you are getting that far yet.

Now. The entry address in image->start is valid and is the entrypoint of
the "custom" cuImage. Custom means that it does not depend any register
values passed from u-boot (the original one needs a pointer to bd_t).
The only requirement is a valid 1:1 memory mapping.

ok sounds good.  does this have the dtb in it too?

I learned, that I can not disable the MMU on Book-E so I have to create
a new temporary mapping in my relocate_new_kernel routine. _start is
doing the same thing what I am trying to accomplish: create a new
mapping and don't kill the current one and switch over. This is done by
disabling all mappings but the current, creating a new mapping with
EFN/RPN = 0 and swapping the TS bit in MAS1. This is my current patch
which is not really working:

I have never actually written or debugged any book-E code, and this deals directly with that.  However, a quick read of ePAPR chapter 5 suggests that rather than just "the other TS", you want to actively decide that the control will transfer to TS0, and establishing the mappings there.  Again, I'm not faimilar with the book-E code, but the kernel itself will use the upper 1/4 of the effective address space by default, so the low offset will likely be available.

ok, but should you not be using PID0 explictly to say global only?

50 more quoted lines

+	/* Just modify the entry ID and EPN for the temp mapping */
+	lis     r7,0x1000       /* Set MAS0(TLBSEL) = 1 */
+	rlwimi  r7,r5,16,4,15   /* Setup MAS0 = TLBSEL | ESEL(r5) */
+	mtspr   SPRN_MAS0,r7
+	xori    r6,r4,1         /* Setup TMP mapping in the other Address space */
+	slwi    r6,r6,12
+	oris    r6,r6,(MAS1_VALID|MAS1_IPROT)@h
+	ori     r6,r6,(MAS1_TSIZE(BOOKE_PAGESZ_1GB))@l
+	mtspr   SPRN_MAS1,r6
+	mfspr   r6,SPRN_MAS2
+	li      r7,0            /* temp EPN = 0 */
+	rlwimi  r7,r6,0,20,31
+
+	mtspr   SPRN_MAS2,r7
+	mtspr   SPRN_MAS3,r8

Here I get:

MAS0: 0x10010000
MAS1: 0xc0001a00
MAS2: 0x00000000
MAS3: 0x00000015

+	tlbwe

I haven't made it to here.

or you do make it, but you turned off all of your debugging access to notice.

hmm... so you are trying to keep one entry but establish mappings in the other ...

+old_copy_code:
+
+	mr r4, r27
+	mr r5, r28
+	mr r6, r29
+
+	li	r0, 0

-1:
 	/* from this point address translation is turned off */
 	/* and interrupts are disabled */


In order to dump the MASx values, I did not invalidate the TLBs entries
(s/beq skpinv/b skpinv) and branched to address 0 before the tlbwe to
receive a register dump. This little trick did not work after the tlbwe
instruction. I probably overwrote the wrong / active TLB entry. Since
the TI bit is fliped, this new TLB entry should not be used anyway.
Without invalidating the TLBs, I may have picked the wrong ESEL index
and overwrote either my current mapping or that one that used by the
expection handler/serial port and the bug is somewhere else.

or you just killed enough of the mappings that the kernel can not take the fault and output the message that yoyu failed.

Milton, do have an idea how I could debug this further? My cuImage does
not expect any register values or anything so I thing it hangs here
somewhere and never enter the cuImage code.

obviously, a jtag or similar hardware debugger would be best.  Second best is probably a simulator (I hear qemu has some support for 440, dont' know if its good ebough for this part or not), and third is some other kind of direct memory access (eg if you have a pci card, a firewire plug-in card can dma to ram once its setup).

Without that, think about where the context is now, and what you want to end up with.  THe kernel normally runs with effective address 0xc0000000 mapped to real 0x00000000 and linear offsets theretwo.  But looking at ePAPR, the boot transfer, and also purgatory and cuboot, expect to run in effective=real.   While the rfi can do the branch at the end and switch which mode the cpu is running in, I think it needs to end up in space 0 for the next code.

Another thing to think about is the device tree the debug io for the boot wrapper.  If you have a virt-addr property or even just a linux,stdout setup so that your console works, you will need to make sure that your shutdown space includes the mappnig of the io region (eg serial port).

Pmce you establish a method to poke characters then it gets easier to debug.

As a final note, it looks like you are currently replacing the code in relocate_new_kernel with book-e code.  Obviously this will need refinement to select or move to heat_xx to merge.

Again, I don't have any direct experience, but mauybe this gives you some ideas.

milton

Milton Miller wrote:

quoted

|load: entry = 0x80053c flags = 0
|nr_segments = 2
|segment[0].buf   = 0x1002b8f0
|segment[0].bufsz = 80
|segment[0].mem   = (nil)
|segment[0].memsz = 1000
|segment[1].buf   = 0x4803f008
|segment[1].bufsz = 3a3138
|segment[1].mem   = 0x800000
|segment[1].memsz = 3b0000

I would expect a third segment (kernel/zImage, dtb, and purgatory), but its not clear that you are getting that far yet.

segment 0 looks like a small segment which should create "boot loader environment". That one does nothing.
Segment 1 is my cuImage. What is purgatory?

quoted

Now. The entry address in image->start is valid and is the entrypoint of
the "custom" cuImage. Custom means that it does not depend any register
values passed from u-boot (the original one needs a pointer to bd_t).
The only requirement is a valid 1:1 memory mapping.

ok sounds good.  does this have the dtb in it too?

Yes it does.

quoted

The branch above is taken, so I've found my current mapping

ok, but should you not be using PID0 explictly to say global only?

The kernel mapping should only be global and therefore that might be a good idea.

obviously, a jtag or similar hardware debugger would be best.  Second I have here CodeWarrior usb tap but after more than one hour playing with that thing I started to hack assembly char put. It helper more :) kexec seems to work now :) I get "nobody cared irq X" from time to time so I thing I have to fix here something.....

As a final note, it looks like you are currently replacing the code in relocate_new_kernel with book-e code.  Obviously this will need refinement to select or move to heat_xx to merge.

Yep, this is next what is going to happen. I would prefer to have them runtime switchable instead of build depend.

Again, I don't have any direct experience, but mauybe this gives you some ideas.

Your hints helped. Thx for that.

milton

Sebastian

On Sep 30, 2008, at 12:21 PM, Sebastian Siewior wrote:

Milton Miller wrote:

quoted

quoted

|load: entry = 0x80053c flags = 0
|nr_segments = 2
|segment[0].buf   = 0x1002b8f0
|segment[0].bufsz = 80
|segment[0].mem   = (nil)
|segment[0].memsz = 1000
|segment[1].buf   = 0x4803f008
|segment[1].bufsz = 3a3138
|segment[1].mem   = 0x800000
|segment[1].memsz = 3b0000

I would expect a third segment (kernel/zImage, dtb, and purgatory), but its not clear that you are getting that far yet.

segment 0 looks like a small segment which should create "boot loader environment". That one does nothing.
Segment 1 is my cuImage. What is purgatory?

Purgatory is the code that runs between the old kernel exiting and the new image loading.   Its supposed to be where any registers, dynamic memory structures, etc get set before calling the image supplied to kexec user space.   Its built as part of the kexec-tools suite as a completley relocatable elf and selected and edited based on type of image being loaded.  For powerpc64 it is where we take the "boot" / master cpu's physical id from r3, put it in the dtb header, and load the address of r3 with the dtb before going into the kernel (for vmlinux, and could do for zImage but don't have support upstream).  If you were booting a cuImage (as opossed to the code you are aparently running, which is what grant called simple image, effectively), then you would set any registers uboot leaves behind in this code.

The standard code supplied by kexec-tools also calculates a checkum (sha1) of each loaded segment (except itself) and checks that vs the sum calculated by kexec-tools userspace (printing a message that on powerpc has no way to be displayed then going into an infinite spinloop.  Oh well, I digress.) and also where, for kdump, any memory backup copy is performed when a specific memory segment is needed to boot (eg initial page for ppc64 and classic32 that require interrupt (exception) vectors to be in page 0-2).

The powerpc64 code reads the existing device tree from /proc/device-tree and modifies a few things (initrd start, end, bootargs = command line, and (for kdump) which memory is available and usable to the kernel (vs reserved because it was used for the old kernel, whose image we want to dump, and which could be under dma).

quoted

quoted

Now. The entry address in image->start is valid and is the entrypoint of
the "custom" cuImage. Custom means that it does not depend any register
values passed from u-boot (the original one needs a pointer to bd_t).
The only requirement is a valid 1:1 memory mapping.

ok sounds good.  does this have the dtb in it too?

Yes it does.

ok.   sounds like a simple image then ... ok to start with, but eventually we want to dtb passed via the tool so we can set command line etc.

I actually developed the powerpc 64 code this way to, and let someone else make the standard tool work.  But the standard tool is useful.

quoted

quoted

The branch above is taken, so I've found my current mapping

ok, but should you not be using PID0 explictly to say global only?

The kernel mapping should only be global and therefore that might be a good idea.

quoted

obviously, a jtag or similar hardware debugger would be best.  Second

I have here CodeWarrior usb tap but after more than one hour playing with that thing I started to hack assembly char put. It helper more :) kexec seems to work now :) I get "nobody cared irq X" from time to time so I thing I have to fix here something.....

kexec is a bit harder than kdump in that you have to make sure all devices have shutdown handlers.   Easier for those that are modules that can be loaded and unloaded (make sure they have a shutdown method that is comparable to unload, or even unload in a script to test).   kdump is harder in that while the dma is left running in the old kernel, the new kernel has to fit in the cracks left over, and has to initialize devices that were not shutdown.

quoted

As a final note, it looks like you are currently replacing the code in relocate_new_kernel with book-e code.  Obviously this will need refinement to select or move to heat_xx to merge.

Yep, this is next what is going to happen. I would prefer to have them runtime switchable instead of build depend.

well, I am thinking that we will end up with one exit condition for all book-e, one for classic 32, and one for powerpc64.   I don't understand what you think should be runtime switchable, unless you were thinking about code that should be in purgatory (supplied by userspace as far as the kernel is concerned).

Remember the exit point of the kernel is a single entry point (we cheat and make it 2 on powerpc64, one for master and a second for slaves, although for book-e we could follow epapr instead), and specified pages of memory with user specified content.  The state is supposed to be an emulation of "mmu off", not "I just ran uboot and am its client loader".

quoted

Again, I don't have any direct experience, but mauybe this gives you some ideas.

Your hints helped. Thx for that.

sure.   Maybe the new hits about purgatory will keep you on track too.

milton