Re: [PATCH] of: Drop reserved mem dependency on DMA_DECLARE_COHERENT and DMA_CMA

[Rob Herring]

On Fri, May 28, 2021 at 5:32 PM Randy Dunlap <rdunlap@xxxxxxxxxxxxx> wrote:
>
> On 5/28/21 3:09 PM, Randy Dunlap wrote:
> > On 5/27/21 3:32 PM, Rob Herring wrote:
> >> Reserved memory regions can be used for more than just DMA regions, so
> >> only enabling on DMA_DECLARE_COHERENT (via HAS_DMA) or DMA_CMA is wrong.
> >> This effectively doesn't matter except for the few cases arches select
> >> NO_DMA.
> >>
> >> At least, these users of RESERVEDMEM_OF_DECLARE depend on reserved memory
> >> support:
> >>
> >> arch/riscv/mm/init.c:RESERVEDMEM_OF_DECLARE(elfcorehdr, "linux,elfcorehdr", elfcore_hdr_setup);
> >> drivers/memory/tegra/tegra210-emc-table.c:RESERVEDMEM_OF_DECLARE(tegra210_emc_table, "nvidia,tegra210-emc-table",
> >> drivers/soc/fsl/qbman/bman_ccsr.c:RESERVEDMEM_OF_DECLARE(bman_fbpr, "fsl,bman-fbpr", bman_fbpr);
> >> drivers/soc/fsl/qbman/qman_ccsr.c:RESERVEDMEM_OF_DECLARE(qman_fqd, "fsl,qman-fqd", qman_fqd);
> >> drivers/soc/fsl/qbman/qman_ccsr.c:RESERVEDMEM_OF_DECLARE(qman_pfdr, "fsl,qman-pfdr", qman_pfdr);
> >>
> >> Let's simplify things and enable OF_RESERVED_MEM when OF_EARLY_FLATTREE is
> >> enabled.
> >>
> >> Cc: Christoph Hellwig <hch@xxxxxx>
> >> Signed-off-by: Rob Herring <robh@xxxxxxxxxx>
> >
> > Hi Rob,
> >
> > I'm OK with this patch, but with or without this patch,
> > compiling kernel/dma/coherent.c without HAS_IOMEM (!)
> > has build errors:
> >
> > s390-linux-ld: kernel/dma/coherent.o: in function `dma_init_coherent_memory':
> > coherent.c:(.text+0x5ae): undefined reference to `memremap'
> > s390-linux-ld: coherent.c:(.text+0x7a4): undefined reference to `memunmap'
> > s390-linux-ld: kernel/dma/coherent.o: in function `dma_declare_coherent_memory':
> > coherent.c:(.text+0xd84): undefined reference to `memunmap'
> >
> >
> > along with a slew of other build errors (arch/s390/, CONFIG_PCI is not set,
> > CONFIG_HAS_IOMEM is not set):
> >
> > s390-linux-ld: drivers/irqchip/irq-al-fic.o: in function `al_fic_init_dt':
> > irq-al-fic.c:(.init.text+0x6a0): undefined reference to `iounmap'
> > s390-linux-ld: drivers/char/xillybus/xillybus_of.o: in function `xilly_drv_probe':
> > xillybus_of.c:(.text+0x2c8): undefined reference to `devm_platform_ioremap_resource'
> > s390-linux-ld: drivers/pcmcia/cistpl.o: in function `set_cis_map':
> > cistpl.c:(.text+0x87a): undefined reference to `ioremap'
> > s390-linux-ld: cistpl.c:(.text+0x99a): undefined reference to `iounmap'
> > s390-linux-ld: cistpl.c:(.text+0xa34): undefined reference to `iounmap'
> > s390-linux-ld: cistpl.c:(.text+0xaa6): undefined reference to `ioremap'
> > s390-linux-ld: drivers/pcmcia/cistpl.o: in function `release_cis_mem':
> > cistpl.c:(.text+0x2976): undefined reference to `iounmap'
> > s390-linux-ld: drivers/of/address.o: in function `of_iomap':
> > address.c:(.text+0x1f02): undefined reference to `ioremap'
> > s390-linux-ld: drivers/of/address.o: in function `of_io_request_and_map':
> > address.c:(.text+0x2030): undefined reference to `ioremap'
> >
>
> Oops, sorry. I should have said that this is also with my
> experimental path that I thought you suggested earlier this
> week (although I could have misunderstood you):

Yes, that's what I'd like, but no doubt that change alone will make
things worse.

> --- linux-next-20210526.orig/drivers/of/Kconfig
> +++ linux-next-20210526/drivers/of/Kconfig
> @@ -64,7 +64,7 @@ config OF_DYNAMIC
>
>  config OF_ADDRESS
>         def_bool y
> -       depends on !SPARC && (HAS_IOMEM || UML)
> +       depends on !SPARC
>
>  config OF_IRQ
>         def_bool y
>
> [rebuild]
>
> OK, even without this small patch, the result is about the same (just
> a few different function names):
>
> s390-linux-ld: kernel/dma/coherent.o: in function `dma_init_coherent_memory':
> coherent.c:(.text+0x5ae): undefined reference to `memremap'
> s390-linux-ld: coherent.c:(.text+0x7a4): undefined reference to `memunmap'
> s390-linux-ld: kernel/dma/coherent.o: in function `dma_declare_coherent_memory':
> coherent.c:(.text+0xd84): undefined reference to `memunmap'
>
> s390-linux-ld: drivers/irqchip/irq-al-fic.o: in function `al_fic_init_dt':
> irq-al-fic.c:(.init.text+0x64): undefined reference to `of_iomap'
> s390-linux-ld: irq-al-fic.c:(.init.text+0x6a0): undefined reference to `iounmap'
> s390-linux-ld: drivers/char/xillybus/xillybus_of.o: in function `xilly_drv_probe':
> xillybus_of.c:(.text+0x2c8): undefined reference to `devm_platform_ioremap_resource'
> s390-linux-ld: drivers/pcmcia/cistpl.o: in function `set_cis_map':
> cistpl.c:(.text+0x87a): undefined reference to `ioremap'
> s390-linux-ld: cistpl.c:(.text+0x99a): undefined reference to `iounmap'
> s390-linux-ld: cistpl.c:(.text+0xa34): undefined reference to `iounmap'
> s390-linux-ld: cistpl.c:(.text+0xaa6): undefined reference to `ioremap'
> s390-linux-ld: drivers/pcmcia/cistpl.o: in function `release_cis_mem':
> cistpl.c:(.text+0x2976): undefined reference to `iounmap'

Empty stubs for ioremap, iounmap, memremap, and memunmap would fix all
these, right? Though maybe adding stubs is more complicated than I'm
thinking given all the architecture specifics.

Rob