Re: [PATCH v6 3/4] of: reserved_mem: Use unflatten_devicetree APIs to scan reserved memory nodes

[Rob Herring]

On Mon, Jul 8, 2024 at 5:12 PM Oreoluwa Babatunde
<quic_obabatun@xxxxxxxxxxx> wrote:
>
>
> On 7/5/2024 8:38 AM, Klara Modin wrote:
> > On 2024-07-05 15:05, Klara Modin wrote:
> >> Hi,
> >>
> >> On 2024-05-29 00:36, Oreoluwa Babatunde wrote:
> >>> The unflatten_devicetree APIs have been setup and are available to be
> >>> used by the time the fdt_init_reserved_mem() function is called.
> >>> Since the unflatten_devicetree APIs are a more efficient way of scanning
> >>> through the DT nodes, switch to using these APIs to facilitate the rest
> >>> of the reserved memory processing.
> >>
> >> With this patch series, I've observed significantly less memory available to userspace on my Raspberry Pi 1 and 3.
> >>
> >> I see this message on the kernel console:
> >> Jul  4 23:13:49 bonnet kernel: OF: reserved mem: 0x1b000000..0x1effffff (65536 KiB) map non-reusable linux
> >>
> >> where it was previously marked as reusable:
> >> Jul  4 22:23:22 bonnet kernel: OF: reserved mem: 0x1b000000..0x1effffff (65536 KiB) map reusable linux,cma
> >>
> >> If I look at bcm283x.dtsi, it definitely has the reusable property.
> >>
> >> I've below pointed out the snippet I think could be suspicous.
> >>
> >>>
> >>> Signed-off-by: Oreoluwa Babatunde <quic_obabatun@xxxxxxxxxxx>
> >>> ---
> >>>   drivers/of/of_reserved_mem.c    | 93 ++++++++++++++++++++-------------
> >>>   include/linux/of_reserved_mem.h |  2 +-
> >>>   kernel/dma/coherent.c           | 10 ++--
> >>>   kernel/dma/contiguous.c         |  8 +--
> >>>   kernel/dma/swiotlb.c            | 10 ++--
> >>>   5 files changed, 72 insertions(+), 51 deletions(-)
> >>>
> >>> diff --git a/drivers/of/of_reserved_mem.c b/drivers/of/of_reserved_mem.c
> >>> index 113d593ea031..05283cd24c3b 100644
> >>> --- a/drivers/of/of_reserved_mem.c
> >>> +++ b/drivers/of/of_reserved_mem.c
> >
> >>> @@ -447,7 +476,7 @@ static int __init __reserved_mem_alloc_size(unsigned long node, const char *unam
> >>>                  uname, (unsigned long)(size / SZ_1M));
> >>>           return -ENOMEM;
> >>>       }
> >>
> >>
> >>> -    fdt_reserved_mem_save_node(node, uname, base, size);
> >>> +    fdt_reserved_mem_save_node(NULL, uname, base, size);
> >>
> >> This could perhaps be suspicious?
> >>
> >> The above message seems to come from of_init_reserved_mem_node when
> >> called from of_reserved_mem_save_node when called from here. This would mean that the node is not actually saved to rmem and thus not marked reusable?
> >>
> >>
> >>>       return 0;
> >>>   }
> >
> >>
> >> Regards,
> >> Klara Modin
> >
> > Attaching kernel logs of old and new behavior, and my config for reference.
> Hi Klara,
>
> Thanks for pointing this out. I have uploaded a fix here:
> https://lore.kernel.org/all/20240708230613.448846-1-quic_obabatun@xxxxxxxxxxx/

Sorry, but I'm not taking reverts of half the series. I've dropped it
all for 6.11.

Rob