Re: [PATCH] riscv: dts: thead: set dma-noncoherent to soc bus

[Drew Fustini]

On Tue, Sep 05, 2023 at 12:28:41AM +0800, Xi Ruoyao wrote:
> On Mon, 2023-09-04 at 22:45 +0800, Jisheng Zhang wrote:
> > On Sun, Sep 03, 2023 at 04:32:21PM -0700, Drew Fustini wrote:
> > > On Sun, Aug 20, 2023 at 07:53:53PM +0800, Jisheng Zhang wrote:
> > > > riscv select ARCH_DMA_DEFAULT_COHERENT by default, and th1520 isn't
> > > > dma coherent, so set dma-noncoherent to reflect this fact.
> > > > 
> > > > Signed-off-by: Jisheng Zhang <jszhang@xxxxxxxxxx>
> > > > ---
> > > >  arch/riscv/boot/dts/thead/th1520.dtsi | 1 +
> > > >  1 file changed, 1 insertion(+)
> > > > 
> > > > diff --git a/arch/riscv/boot/dts/thead/th1520.dtsi b/arch/riscv/boot/dts/thead/th1520.dtsi
> > > > index 56a73134b49e..58108f0eb3fd 100644
> > > > --- a/arch/riscv/boot/dts/thead/th1520.dtsi
> > > > +++ b/arch/riscv/boot/dts/thead/th1520.dtsi
> > > > @@ -139,6 +139,7 @@ soc {
> > > >                 interrupt-parent = <&plic>;
> > > >                 #address-cells = <2>;
> > > >                 #size-cells = <2>;
> > > > +               dma-noncoherent;
> > > >                 ranges;
> > > >  
> > > >                 plic: interrupt-controller@ffd8000000 {
> > > > -- 
> > > > 2.40.1
> > > > 
> > > 
> > > Tested-by: Drew Fustini <dfustini@xxxxxxxxxxxx>
> > > 
> > > I tried this on the BeagleV Ahead. They system booted as expected and I
> > > did not notice any problems in the boot log.
> > > 
> > > Are there other patches such as the dwmac series that I should test this
> > > with?
> > 
> > Hmm, this patch is necessary to test dwmac, emmc ;)
> 
> Drew: does this fix the "broken DMA" issue you've mentioned in the EMMC
> support patch?

After I sent my reply yesterday, I then tried applying this patch on top
of my eMMC series. I removed SDHCI_QUIRK_BROKEN_DMA and I discovered
that SDMA did not immediately break and the mmcblk0 partitions appeared
in /dev [1]. Note - I am using CONFIG_INITRAMFS_SOURCE set to my
buidlroot rootfs.cpio so that the system will boot okay regardless of
the emmc.

I was able to mount the emmc boot partition /dev/mmcblk0p2. However,
after unmounting it, I noticed that I could not mount it again:

  # mount /dev/mmcblk0p2 /mnt/p2 
  [   41.684899] EXT4-fs (mmcblk0p2): mounted filesystem 6ea98ccf-2178-48a9-a1ad-0eb454f57268 r/w with ordered data mode
  # ls -la /mnt/p2
  total 22696
  drwxr-xr-x    5 root     root          4096 Sep  4  2023 .
  drwxr-xr-x    4 root     root             0 Jan  1 00:00 ..
  -rw-rw-r--    1 1000     1000      26455552 Sep  4  2023 Image
  drwxr-xr-x    2 root     root          4096 Jul 29  2023 extlinux
  <snip>
  # umount /mnt/p2
  [   52.101003] EXT4-fs (mmcblk0p2): unmounting filesystem 6ea98ccf-2178-48a9-a1ad-0eb454f57268.
  # mount /dev/mmcblk0p2 /mnt/p2
  [   58.826078] JBD2: no valid journal superblock found
  [   58.831041] EXT4-fs (mmcblk0p2): Could not load journal inode
  mount: mounting /dev/mmcblk0p2 on /mnt/p2 failed: Invalid argument

I was completely unable to mount the root fs partition /dev/mmcblk0p3:

  # mount /dev/mmcblk0p2 /mnt/p2
  [   58.826078] JBD2: no valid journal superblock found
  [   58.831041] EXT4-fs (mmcblk0p2): Could not load journal inode
  mount: mounting /dev/mmcblk0p2 on /mnt/p2 failed: Invalid argument
  # umount /mnt/p2
  umount: can't unmount /mnt/p2: Invalid argument

I suspect there is some corruption of the buffers happening. I see there
are also these patch series from Jisheng so I will try applying them and
see if there is any difference:

  [PATCH v2 0/2] riscv: errata: improve T-Head CMO
  [PATCH] riscv: mm: update T-Head memory type definitions

As for ADMA, when I removed SDHCI_QUIRK_BROKEN_ADMA, something fails
silently [2] during boot and the mmcblk0 partitions do not appear under
/dev. 

-Drew

[1] https://gist.github.com/pdp7/6d27c57f8280fe1ae35e585d66e73f61
[2] https://gist.github.com/pdp7/23e2bcc2aba09b5c88eb579b61ea45ce