[PATCH v2 2/2] ARM: move device tree mapping out of linear region

[Ard Biesheuvel]

On Wed, 28 Oct 2020 at 13:05, Joel Stanley <joel at jms.id.au> wrote:
>
> On Wed, 28 Oct 2020 at 09:19, Marek Szyprowski <m.szyprowski at samsung.com> wrote:
> >
> > Hi,
> >
> > On 07.10.2020 10:39, Ard Biesheuvel wrote:
> > > On ARM, setting up the linear region is tricky, given the constraints
> > > around placement and alignment of the memblocks, and how the kernel
> > > itself as well as the DT are placed in physical memory.
> > >
> > > Let's simplify matters a bit, by moving the device tree mapping to the
> > > top of the address space, right between the end of the vmalloc region
> > > and the start of the the fixmap region, and create a read-only mapping
> > > for it that is independent of the size of the linear region, and how it
> > > is organized.
> > >
> > > Since this region was formerly used as a guard region, which will now be
> > > populated fully on LPAE builds by this read-only mapping (which will
> > > still be able to function as a guard region for stray writes), bump the
> > > start of the [underutilized] fixmap region by 512 KB as well, to ensure
> > > that there is always a proper guard region here. Doing so still leaves
> > > ample room for the fixmap space, even with NR_CPUS set to its maximum
> > > value of 32.
> > >
> > > Tested-by: Linus Walleij <linus.walleij at linaro.org>
> > > Reviewed-by: Linus Walleij <linus.walleij at linaro.org>
> > > Signed-off-by: Ard Biesheuvel <ardb at kernel.org>
> >
> > This patch landed in linux-next 20201028 as commit 7a1be318f579 ("ARM:
> > 9012/1: move device tree mapping out of linear region"). Sadly it broke
> > booting  almost all Samsung Exynos-based boards. The only one which
> > booted, used an appended device tree. I can provide more information if
> > needed, just let me know what to check. "Starting kernel ..." is the
> > last message I see here. No output from earlycon.
>
> A bisection lead me to this patch after the next-20201028 failed to
> boot on the aspeed systems in testing (aspeed_g5_defconfig).
>
> You can reproduce this with today's next and qemu 5.1:
>
> qemu-system-arm -M romulus-bmc -nographic \
>  -kernel arch/arm/boot/zImage \
>  -dtb arch/arm/boot/dts/aspeed-bmc-opp-romulus.dtb \
>  -initrd any-old-file
>
> It requires the initrd option to reproduce, but the initrd doesn't
> need to be valid as we don't make it that far.
>
> There is no output but attaching gdb shows the kernel is stuck in
> setup_machine_tags. (If we enable CONFIG_ATAGS it is instead stuck in
> calibrate_delay).
>
> (gdb) bt
> #0  setup_machine_tags (machine_nr=<optimized out>,
> __atags_vaddr=<optimized out>) at ../arch/arm/kernel/atags.h:12
> #1  setup_arch (cmdline_p=0x80c01fc4) at ../arch/arm/kernel/setup.c:1100
> #2  0x80b00d2c in start_kernel () at ../init/main.c:862
> #3  0x00000000 in ?? ()
>
> Reverting 7a1be318f579 on top of next allowed the system to boot again.
>

Does this help?

diff --git a/arch/arm/include/asm/memory.h b/arch/arm/include/asm/memory.h
index bb79e52aeb90..4f355bda872a 100644
--- a/arch/arm/include/asm/memory.h
+++ b/arch/arm/include/asm/memory.h
@@ -68,8 +68,8 @@
 #define XIP_VIRT_ADDR(physaddr)  (MODULES_VADDR + ((physaddr) & 0x000fffff))

 #define FDT_FIXED_BASE         UL(0xff800000)
-#define FDT_FIXED_SIZE         (2 * PMD_SIZE)
-#define FDT_VIRT_ADDR(physaddr)        ((void *)(FDT_FIXED_BASE |
(physaddr) % PMD_SIZE))
+#define FDT_FIXED_SIZE         (2 * SECTION_SIZE)
+#define FDT_VIRT_ADDR(physaddr)        ((void *)(FDT_FIXED_BASE |
(physaddr) % SECTION_SIZE))

 #if !defined(CONFIG_SMP) && !defined(CONFIG_ARM_LPAE)
 /*