Re: Broken CONFIG_SPL_SATA_SUPPORT=y

[Tom Rini <trini@konsulko.com>]

[-- Attachment #1: Type: text/plain, Size: 8427 bytes --]

On Mon, Aug 02, 2021 at 11:56:58PM +0200, Pali Rohár wrote:
> On Monday 02 August 2021 17:43:22 Tom Rini wrote:
> > On Mon, Aug 02, 2021 at 03:30:20PM -0600, Simon Glass wrote:
> > > Hi Pali,
> > > 
> > > On Mon, 2 Aug 2021 at 15:26, Pali Rohár <pali@kernel.org> wrote:>
> > > > On Monday 02 August 2021 15:23:22 Simon Glass wrote:
> > > > > () Hi Pali,
> > > > >
> > > > > On Mon, 2 Aug 2021 at 15:18, Pali Rohár <pali@kernel.org> wrote:
> > > > > >
> > > > > > On Monday 02 August 2021 15:14:30 Simon Glass wrote:
> > > > > > > Hi Pali,
> > > > > > >
> > > > > > > On Mon, 2 Aug 2021 at 15:09, Pali Rohár <pali@kernel.org> wrote:
> > > > > > > >
> > > > > > > > On Monday 02 August 2021 16:55:34 Tom Rini wrote:
> > > > > > > > > On Sun, Aug 01, 2021 at 02:25:16PM +0200, Pali Rohár wrote:
> > > > > > > > >
> > > > > > > > > > Hello!
> > > > > > > > > >
> > > > > > > > > > Option CONFIG_SPL_SATA_SUPPORT=y is currently broken in u-boot master
> > > > > > > > > > branch. If I try to enable it for A38x platform I'm getting following
> > > > > > > > > > compiler error:
> > > > > > > > > >
> > > > > > > > > >     LD      spl/u-boot-spl
> > > > > > > > > >   arm-linux-gnueabihf-ld.bfd: drivers/ata/ahci.o: in function `ahci_probe_scsi_pci':
> > > > > > > > > >   drivers/ata/ahci.c:1205: undefined reference to `dm_pci_map_bar'
> > > > > > > > > >   arm-linux-gnueabihf-ld.bfd: drivers/ata/ahci.c:1215: undefined reference to `dm_pci_read_config16'
> > > > > > > > > >   arm-linux-gnueabihf-ld.bfd: drivers/ata/ahci.c:1216: undefined reference to `dm_pci_read_config16'
> > > > > > > > > >   arm-linux-gnueabihf-ld.bfd: drivers/ata/ahci.c:1220: undefined reference to `dm_pci_map_bar'
> > > > > > > > > >   make[1]: *** [scripts/Makefile.spl:512: spl/u-boot-spl] Error 1
> > > > > > > > > >   make: *** [Makefile:1977: spl/u-boot-spl] Error 2
> > > > > > > > > >
> > > > > > > > > > You can reproduce it by running following commands:
> > > > > > > > > >
> > > > > > > > > >   $ make turris_omnia_defconfig
> > > > > > > > > >   $ echo CONFIG_SPL_SATA_SUPPORT=y >> .config
> > > > > > > > > >   $ make CROSS_COMPILE=arm-linux-gnueabihf-
> > > > > > > > > >
> > > > > > > > > > I workaround it by following patch:
> > > > > > > > > >
> > > > > > > > > > diff --git a/drivers/ata/ahci.c b/drivers/ata/ahci.c
> > > > > > > > > > index d4047c04f5d0..6bad72e4cfa4 100644
> > > > > > > > > > --- a/drivers/ata/ahci.c
> > > > > > > > > > +++ b/drivers/ata/ahci.c
> > > > > > > > > > @@ -1196,7 +1196,7 @@ int ahci_probe_scsi(struct udevice *ahci_dev, ulong base)
> > > > > > > > > >     return 0;
> > > > > > > > > >  }
> > > > > > > > > >
> > > > > > > > > > -#ifdef CONFIG_DM_PCI
> > > > > > > > > > +#if CONFIG_IS_ENABLED(DM_PCI)
> > > > > > > > > >  int ahci_probe_scsi_pci(struct udevice *ahci_dev)
> > > > > > > > > >  {
> > > > > > > > > >     ulong base;
> > > > > > > > > >
> > > > > > > > > > It fixed this particular problem. So it looks like that CONFIG_DM_PCI is
> > > > > > > > > > defined also when building SPL even when it is not enabled for SPL.
> > > > > > > > > > Whole PCI is disabled in SPL.
> > > > > > > > > >
> > > > > > > > > > But then I got another compile error:
> > > > > > > > > >
> > > > > > > > > >     LD      spl/u-boot-spl
> > > > > > > > > >   arm-linux-gnueabihf-ld.bfd: drivers/ata/ahci-pci.o: in function `ahci_pci_probe':
> > > > > > > > > >   drivers/ata/ahci-pci.c:21: undefined reference to `ahci_probe_scsi_pci'
> > > > > > > > > >   make[1]: *** [scripts/Makefile.spl:512: spl/u-boot-spl] Error 1
> > > > > > > > > >   make: *** [Makefile:1977: spl/u-boot-spl] Error 2
> > > > > > > > > >
> > > > > > > > > > Seems that u-boot is trying to compile and link ahci-pci.o into SPL
> > > > > > > > > > binary even when it is not enabled nor used. PCI is completed disabled
> > > > > > > > > > in SPL for this case.
> > > > > > > > > >
> > > > > > > > > > I workaround it by putting whole ahci-pci.c file into one big #idef:
> > > > > > > > > >
> > > > > > > > > > diff --git a/drivers/ata/ahci-pci.c b/drivers/ata/ahci-pci.c
> > > > > > > > > > index b1d231e0f9e1..34afebd2f87f 100644
> > > > > > > > > > --- a/drivers/ata/ahci-pci.c
> > > > > > > > > > +++ b/drivers/ata/ahci-pci.c
> > > > > > > > > > @@ -9,6 +9,8 @@
> > > > > > > > > >  #include <dm.h>
> > > > > > > > > >  #include <pci.h>
> > > > > > > > > >
> > > > > > > > > > +#if CONFIG_IS_ENABLED(DM_PCI)
> > > > > > > > > > +
> > > > > > > > > >  static int ahci_pci_bind(struct udevice *dev)
> > > > > > > > > >  {
> > > > > > > > > >     struct udevice *scsi_dev;
> > > > > > > > > > @@ -42,3 +44,5 @@ static struct pci_device_id ahci_pci_supported[] = {
> > > > > > > > > >  };
> > > > > > > > > >
> > > > > > > > > >  U_BOOT_PCI_DEVICE(ahci_pci, ahci_pci_supported);
> > > > > > > > > > +
> > > > > > > > > > +#endif
> > > > > > > > > >
> > > > > > > > > > And then finally U-Boot produced final target image u-boot-spl.kwb:
> > > > > > > > > >
> > > > > > > > > >   LD      spl/u-boot-spl
> > > > > > > > > >   OBJCOPY spl/u-boot-spl-nodtb.bin
> > > > > > > > > >   SYM     spl/u-boot-spl.sym
> > > > > > > > > >   CAT     spl/u-boot-spl-dtb.bin
> > > > > > > > > >   COPY    spl/u-boot-spl.bin
> > > > > > > > > >   MKIMAGE u-boot-spl.kwb
> > > > > > > > > >
> > > > > > > > > > So this looks like a bug in Kconfig or Makefile dependences that build
> > > > > > > > > > system is trying to compile and link also files which should not be
> > > > > > > > > > linked at all.
> > > > > > > > >
> > > > > > > > > Yes, it's a missing dependency in Kconfig.  You need to enable SPL_DM
> > > > > > > > > and then quite possibly SPL_OF_CONTROL (I forget if that was a hard
> > > > > > > > > dependency on the main conversion or not) on the platform.
> > > > > > > >
> > > > > > > > CONFIG_SPL_DM=y is already enabled
> > > > > > >
> > > > > > > CONFIG_SPL_PCI ?
> > > > > > >
> > > > > > > It seems that pci-uclass.c is not being built for SPL.
> > > > > >
> > > > > > CONFIG_SPL_PCI is not enabled.
> > > > > >
> > > > > > But why should be CONFIG_SPL_PCI needed for SATA? As above my patches
> > > > > > show no PCI is required for SATA...
> > > > >
> > > > > I suggest investigating what is calling those functions you mention...
> > > > >
> > > > > If you look at ahci_probe_scsi_pci() it is bracketed by:
> > > > >
> > > > > #ifdef CONFIG_DM_PCI
> > > > >
> > > > > Perhaps it should be:
> > > > >
> > > > > #if CONFIG_IS_ENABLED(DM_PCI)
> > > > >
> > > > > so it takes account of SPL?
> > > >
> > > > And this is exactly what I did in first patch attempt.
> > > 
> > > Then how about changing the Makefile rule to:
> > > 
> > > obj-$(CONFIG_$(SPL_TPL_)AHCI_PCI) += ahci-pci.o
> > 
> > But there is no SPL_AHCI_PCI nor TPL_AHCI_PCI.  I really want to know in
> > addition to how SATA is hooked up on this board what the expected truth
> > table of U-Boot and SPL CONFIG options is expected to be.
> 
> What do you mean by hooked? SATA support on A38x is provided by standard
> AHCI interface implemented by ahci_mvebu.c driver (CONFIG_AHCI_MVEBU).
> No PCIe is involved.

OK, so why are you building drivers/ata/ahci-pci.c at all?  Should you
be enabling SCSI_AHCI_PLAT like the armada 37xx and 8k platforms are?

> But I think this is not board neither platform specific, but general SPL
> U-Boot issue. As I mentioned in the first email, U-Boot build system is
> trying to compile and link files into SPL which should not and which are
> not enabled in config at all. More precisely, build system is trying to
> link into SPL binary functions from AHCI_PCI even when PCI is not
> enabled in SPL. PCI is enabled only in proper U-Boot.
> 
> So this looks like a broken dependency either in Makefile or Kconfig,
> but I really do not know where nor why it happens.

It's a combination of a lack combination testing really.

> AHCI_PCI depends on PCI and AHCI. AHCI_MVEBU depends only on AHCI. So
> when PCI is not enabled in SPL then also AHCI_PCI must not be enabled in
> SPL. But AHCI still can be enabled in SPL even when AHCI_PCI is
> disabled. Therefore also AHCI_MVEBU can be enabled in SPL when PCI is
> disabled in SPL.

Why would you be enabling AHCI_PCI and not enabling PCI?  Or, what
functionality would you be using in that case?

-- 
Tom

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 659 bytes --]