Date:   Tue, 19 May 2020 16:32:01 +0200
From:   Ard Biesheuvel <ardb@...nel.org>
To:     Russell King - ARM Linux admin <linux@...linux.org.uk>
Cc:     Geert Uytterhoeven <geert@...ux-m68k.org>,
        Lukasz Stelmach <l.stelmach@...sung.com>,
        Dmitry Osipenko <digetx@...il.com>,
        Nicolas Pitre <nico@...xnic.net>,
        Arnd Bergmann <arnd@...db.de>,
        Eric Miao <eric.miao@...dia.com>,
        Uwe Kleine-König 
        <u.kleine-koenig@...gutronix.de>,
        Masahiro Yamada <masahiroy@...nel.org>,
        Marek Szyprowski <m.szyprowski@...sung.com>,
        Chris Brandt <chris.brandt@...esas.com>,
        Linux ARM <linux-arm-kernel@...ts.infradead.org>,
        Linux-Renesas <linux-renesas-soc@...r.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        Bartlomiej Zolnierkiewicz <b.zolnierkie@...sung.com>,
        "open list:OPEN FIRMWARE AND FLATTENED DEVICE TREE BINDINGS" 
        <devicetree@...r.kernel.org>, Rob Herring <robh+dt@...nel.org>,
        Grant Likely <grant.likely@....com>
Subject: Re: [PATCH v6] ARM: boot: Obtain start of physical memory from DTB

On Tue, 19 May 2020 at 16:29, Russell King - ARM Linux admin
<linux@...linux.org.uk> wrote:
>
> On Tue, May 19, 2020 at 03:56:59PM +0200, Ard Biesheuvel wrote:
> > On Tue, 19 May 2020 at 13:21, Geert Uytterhoeven <geert@...ux-m68k.org> wrote:
> > >
> > > Hi Russell,
> > >
> > > CC devicetree
> > >
> > > On Tue, May 19, 2020 at 11:46 AM Russell King - ARM Linux admin
> > > <linux@...linux.org.uk> wrote:
> > > > On Tue, May 19, 2020 at 11:44:17AM +0200, Geert Uytterhoeven wrote:
> > > > > On Tue, May 19, 2020 at 10:54 AM Lukasz Stelmach <l.stelmach@...sung.com> wrote:
> > > > > > It was <2020-04-29 śro 10:21>, when Geert Uytterhoeven wrote:
> > > > > > > Currently, the start address of physical memory is obtained by masking
> > > > > > > the program counter with a fixed mask of 0xf8000000.  This mask value
> > > > > > > was chosen as a balance between the requirements of different platforms.
> > > > > > > However, this does require that the start address of physical memory is
> > > > > > > a multiple of 128 MiB, precluding booting Linux on platforms where this
> > > > > > > requirement is not fulfilled.
> > > > > > >
> > > > > > > Fix this limitation by obtaining the start address from the DTB instead,
> > > > > > > if available (either explicitly passed, or appended to the kernel).
> > > > > > > Fall back to the traditional method when needed.
> > > > > > >
> > > > > > > This allows to boot Linux on r7s9210/rza2mevb using the 64 MiB of SDRAM
> > > > > > > on the RZA2MEVB sub board, which is located at 0x0C000000 (CS3 space),
> > > > > > > i.e. not at a multiple of 128 MiB.
> > > > > > >
> > > > > > > Suggested-by: Nicolas Pitre <nico@...xnic.net>
> > > > > > > Signed-off-by: Geert Uytterhoeven <geert+renesas@...der.be>
> > > > > > > Reviewed-by: Nicolas Pitre <nico@...xnic.net>
> > > > > > > Reviewed-by: Ard Biesheuvel <ardb@...nel.org>
> > > > > > > Tested-by: Marek Szyprowski <m.szyprowski@...sung.com>
> > > > > > > Tested-by: Dmitry Osipenko <digetx@...il.com>
> > > > > > > ---
> > > > > >
> > > > > > [...]
> > > > > >
> > > > > > Apparently reading physical memory layout from DTB breaks crashdump
> > > > > > kernels. A crashdump kernel is loaded into a region of memory, that is
> > > > > > reserved in the original (i.e. to be crashed) kernel. The reserved
> > > > > > region is large enough for the crashdump kernel to run completely inside
> > > > > > it and don't modify anything outside it, just read and dump the remains
> > > > > > of the crashed kernel. Using the information from DTB makes the
> > > > > > decompressor place the kernel outside of the dedicated region.
> > > > > >
> > > > > > The log below shows that a zImage and DTB are loaded at 0x18eb8000 and
> > > > > > 0x193f6000 (physical). The kernel is expected to run at 0x18008000, but
> > > > > > it is decompressed to 0x00008000 (see r4 reported before jumping from
> > > > > > within __enter_kernel). If I were to suggest something, there need to be
> > > > > > one more bit of information passed in the DTB telling the decompressor
> > > > > > to use the old masking technique to determain kernel address. It would
> > > > > > be set in the DTB loaded along with the crashdump kernel.
> > > > >
> > > > > Shouldn't the DTB passed to the crashkernel describe which region of
> > > > > memory is to be used instead?
> > > >
> > > > Definitely not.  The crashkernel needs to know where the RAM in the
> > > > machine is, so that it can create a coredump of the crashed kernel.
> > >
> > > So the DTB should describe both ;-)
> > >
> > > > > Describing "to use the old masking technique" sounds a bit hackish to me.
> > > > > I guess it cannot just restrict the /memory node to the reserved region,
> > > > > as the crashkernel needs to be able to dump the remains of the crashed
> > > > > kernel, which lie outside this region.
> > > >
> > > > Correct.
> > > >
> > > > > However, something under /chosen should work.
> > > >
> > > > Yet another sticky plaster...
> > >
> > > IMHO the old masking technique is the hacky solution covered by
> > > plasters.
> > >
> >
> > I think debating which solution is the hacky one will not get us anywhere.
> >
> > The simple reality is that the existing solution works fine for
> > existing platforms, and so any changes in the logic will have to be
> > opt-in in one way or the other.
> >
> > Since U-boot supports EFI boot these days, one potential option is to
> > rely on that. I have some changes implementing this that go on top of
> > this patch, but they don't actually rely on it - it was just to
> > prevent lexical conflicts.
> >
> > The only remaining options imo are a kernel command line option, or a
> > DT property that tells the decompressor to look at the memory nodes.
> > But using the DT memory nodes on all platforms like this patch does is
> > obviously just too risky.
> >
> > On another note, I do think the usable-memory-region property should
> > be implemented for ARM as well - relying on this rounding to ensure
> > that the decompressor does the right thing is too fragile.
>
> What is "too fragile" is trying to change this and expecting everything
> to continue working as it did before.
>

That is my point.

> How will switching to EFI help?  Won't the crashdump kernel detect EFI
> and try to get the memory map from EFI, whereby it runs into exactly
> the same issue that the DT approach does?
>

No. If you boot from kexec, then the EFI stub is completely
circumvented, and things work as before.

> The current crashkernel situation works precisely because of the 128M
> masking that is being done.
>

Indeed. That is precisely my point.

Powered by blists - more mailing lists