[PATCH] [RFC] lmb: Optionally limit available memory to 4 GiB on limited systems

[Peter Robinson]

On Sat, 30 May 2026 at 16:06, Marek Vasut <marek.vasut at mailbox.org> wrote:
>
> On 5/30/26 4:45 PM, Marek Vasut wrote:
> > Some architectures can not DMA above 4 GiB boundary,
> > limit available memory to memory below 4 GiB boundary.
> >
> > Signed-off-by: Marek Vasut <marek.vasut+renesas at mailbox.org>
> > ---
> > Cc: Andrew Goodbody <andrew.goodbody at linaro.org>
> > Cc: David Lechner <dlechner at baylibre.com>
> > Cc: Heiko Schocher <hs at nabladev.com>
> > Cc: Heinrich Schuchardt <xypron.glpk at gmx.de>
> > Cc: Holger Brunck <holger.brunck at hitachienergy.com>
> > Cc: Ilias Apalodimas <ilias.apalodimas at linaro.org>
> > Cc: Peter Robinson <pbrobinson at gmail.com>
> > Cc: Quentin Schulz <quentin.schulz at cherry.de>
> > Cc: Simon Glass <sjg at chromium.org>
> > Cc: Sughosh Ganu <sughosh.ganu at arm.com>
> > Cc: Tom Rini <trini at konsulko.com>
> > Cc: u-boot at lists.denx.de
> > ---
> >   lib/Kconfig |  8 ++++++++
> >   lib/lmb.c   | 20 +++++++++++++++++++-
> >   2 files changed, 27 insertions(+), 1 deletion(-)
> >
> > diff --git a/lib/Kconfig b/lib/Kconfig
> > index 77ebc79e1db..29911068a69 100644
> > --- a/lib/Kconfig
> > +++ b/lib/Kconfig
> > @@ -1295,6 +1295,14 @@ config SPL_LMB_ARCH_MEM_MAP
> >         memory map. Enable this config in such scenarios which allow
> >         architectures and boards to define their own memory map.
> >
> > +config LMB_LIMIT_DMA_BELOW_4G
> > +     bool
> > +     depends on LMB
> > +     default y if ARCH_BCM283X
> > +     help
> > +       Some architectures can not DMA above 4 GiB boundary,
> > +       limit available memory to memory below 4 GiB boundary.
> > +
> >   config PHANDLE_CHECK_SEQ
> >       bool "Enable phandle check while getting sequence number"
> >       help
> > diff --git a/lib/lmb.c b/lib/lmb.c
> > index 8f12c6ad8e5..a8a988eedc0 100644
> > --- a/lib/lmb.c
> > +++ b/lib/lmb.c
> > @@ -611,6 +611,7 @@ static __maybe_unused void lmb_reserve_common_spl(void)
> >   static void lmb_add_memory(void)
> >   {
> >       int i;
> > +     phys_addr_t bank_end;
> >       phys_size_t size;
> >       u64 ram_top = gd->ram_top;
> >       struct bd_info *bd = gd->bd;
> > @@ -625,8 +626,25 @@ static void lmb_add_memory(void)
> >       for (i = 0; i < CONFIG_NR_DRAM_BANKS; i++) {
> >               size = bd->bi_dram[i].size;
> >
> > -             if (size)
> > +             if (size) {
> >                       lmb_add(bd->bi_dram[i].start, size);
> > +                     if (IS_ENABLED(CONFIG_LMB_LIMIT_DMA_BELOW_4G))
>
> This conditional should be inverted.
>
> Peter Robinson , can you test this change on RPi ?

That doesn't fix the problem for me on a RPi4 4GB or a RPi5 8Gb.

> > +                             continue;
> > +
> > +                     bank_end = bd->bi_dram[i].start + size;
> > +
> > +                     /*
> > +                      * Reserve memory above ram_top as
> > +                      * no-overwrite so that it cannot be
> > +                      * allocated
> > +                      */
> > +                     if (bd->bi_dram[i].start >= ram_top)
> > +                             lmb_reserve(bd->bi_dram[i].start, size,
> > +                                         LMB_NOOVERWRITE);
> > +                     else if (bank_end > ram_top)
> > +                             lmb_reserve(ram_top, bank_end - ram_top,
> > +                                         LMB_NOOVERWRITE);
> > +             }
> >       }
> >   }
> >
>
> [...]