Hi Michael,

On Mon, May 14, 2018 at 1:10 PM, Michael Schmitz <[email protected]> wrote:

On kernels with 020/030 support ...

> get_io_area leaves an IO_SIZE gap between mappings which is added to
> the vm_struct representing the mapping. __ioremap() uses the actual
> requested size (after alignment), while __iounmap() is passed the
> size from the vm_struct.
>
> On 020/030, early termination descriptors are used to set up mappings of
> extent 'size', which are validated on unmapping. The unmapped gap of size
> IO_SIZE defeats the sanity check of the pmd tables, causing __iounmap to
> loop forever on 030.
>
> On 040/040, unmapping of page table entries does not check for a valid

040/060

> mapping, so the umapping loop always completes there.
>
> Adjust size to be unmapped by the gap that had been added in the vm_struct
> prior.
>
> This fixes the hang in atari_platform_init() reported a long time ago, and
> a similar one reported by Finn recently (addressed by removing ioremap()
> use from the SWIM driver.
>
> Tested on my Falcon in 030 mode - untested but should work the same on
> 040/060 (the extra page tables cleared there would never have been set up
> anyway).
>
> Comment on whether the gap size should be considered in looking for a
> suitable address to place the next mapping in get_io_area() would be welcome.

At first sight (and looking in full-history-linux git history), I see no
reason for the gap. I'd assume having a block with address and size aligned
to 256 KiB (which the caller already takes care of: IO_SIZE is 256 KiB if
020/030 support is enabled) should be sufficient to use early termination
tables.

Obviously removing the gap would fix the issue as well, but that's more
risky...

> Signed-off-by: Michael Schmitz <[email protected]>
> ---
>  arch/m68k/mm/kmap.c |    3 ++-
>  1 files changed, 2 insertions(+), 1 deletions(-)
>
> diff --git a/arch/m68k/mm/kmap.c b/arch/m68k/mm/kmap.c
> index c2a3832..3b420f6 100644
> --- a/arch/m68k/mm/kmap.c
> +++ b/arch/m68k/mm/kmap.c
> @@ -89,7 +89,8 @@ static inline void free_io_area(void *addr)
>         for (p = &iolist ; (tmp = *p) ; p = &tmp->next) {
>                 if (tmp->addr == addr) {
>                         *p = tmp->next;
> -                       __iounmap(tmp->addr, tmp->size);
> +                       /* remove gap added in get_io_area() */
> +                       __iounmap(tmp->addr, tmp->size - IO_SIZE);
>                         kfree(tmp);
>                         return;
>                 }

Looks good to me, so I will apply and queue for v4.18 with the patch
description slightly modified, but will wait one more day, in case someone
has a comment.

Thanks a lot!

Gr{oetje,eeting}s,

                        Geert

-- 
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds
--
To unsubscribe from this list: send the line "unsubscribe linux-m68k" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Reply via email to