[PATCH] Revert "arm: bootm: Disable LMB reservation for command line and board info on arm64"

[Tom Rini]

On Mon, Aug 02, 2021 at 05:27:59PM -0400, Tom Rini wrote:
> On Thu, Jul 29, 2021 at 09:22:02AM +0200, Jan Kiszka wrote:
> 
> > From: Jan Kiszka <jan.kiszka at siemens.com>
> > 
> > This reverts commit 2359fa7a87848626bcbd3399e92c657595880cd7.
> > 
> > While the goal is valid and there is surely unused memory in that area,
> > we also have a lot of crucial things still located at the top-of-memory
> > while running lmb_alloc_base. Such things are the page table (tlb_addr),
> > relocated U-Boot and the active stack. Possibly more. So this patch was
> > premature, we will need relocations of those things first if we want to
> > use the range.
> > 
> > Fixes booting on the IOT2050, but likely also on other boards. It got 
> > stuck on relocating the FDT - over the relocated U-Boot code.
> > 
> > Signed-off-by: Jan Kiszka <jan.kiszka at siemens.com>
> > ---
> > 
> > Practically,
> > 
> > void arch_lmb_reserve(struct lmb *lmb)
> > {
> > 	lmb_reserve(lmb, gd->relocaddr, gd->ram_top - gd->relocaddr);
> > }
> > 
> > worked as well for me - but it left the stack in danger.
> 
> I want to cycle back up to this practically part.  Marek, would changing
> the arch_lmb_reserve (or possibly even making this a more global thing /
> option) still let the area you want exposed on rcar3 (I assume) be
> exposed ?  Or would it be covered again?  Part of the problem,
> practically speaking, is that we need to ensure that as part of
> (attempting and likely succeeding in) booting the OS we don't overwrite
> ourself and hang.
> 
> An alternative that I'm not 100% sure I like would be adjusting
> env_get_bootm_size() so that the case of bootm_size and bootm_low are
> unset, so we figure out a value based on gd->ram_top we also take in to
> account where U-Boot itself is atm and exclude that.  It's possible that
> if we had something like that at the start of the DT world, we wouldn't
> have the code to disable fdt relocation, which really feels like it's
> largely been "things crash when we relocate stuff, disable relocation"
> and not so much "save a little boot time, we optimized things VERY
> CAREFULLY".

Digging more.  The comment about "Allocate space for command line and
board info" was introduced in 2d1916e48bd8 ("ARM: add flat device tree
support").  I happen to have the platform this was tested on, a
Beagleboard, around still and was able to boot that commit up and poke a
bit.  It's effectively covering all of the running U-Boot with an LMB.
You can move on to commit b485556be51d ("ARM: enable device tree for
beagle").  That defines CONFIG_SYS_BOOTMAPSZ which is about start of
DRAM + a bit, to reserve space there that would be questionable on
ARM64.  But all of that code area has been reworked since then.

So in sum, I think reverting 2359fa7a87848626bcbd3399e92c657595880cd7 is
right.  U-Boot needs to be protected from being overwritten as it's
still alive at that point and relocating things around.  This has always
been true.  This has always been related to device tree booting.

What rcar is trying to do needs to be better explained first so that we
can figure out the right place to make some sort of update.

-- 
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 659 bytes
Desc: not available
URL: <https://lists.denx.de/pipermail/u-boot/attachments/20210803/8ffdecb6/attachment.sig>