On 18.03.2025 16:31, Roger Pau Monné wrote:
> On Tue, Mar 18, 2025 at 03:28:32PM +0100, Jan Beulich wrote:
>> On 18.03.2025 14:11, Andrew Cooper wrote:
>>> On 18/03/2025 9:19 am, Roger Pau Monne wrote:
>>>> The call to ioremap_wc() in video_init() will always fail, because
>>>> video_init() is called ahead of vm_init_type(), and so the underlying
>>>> __vmap() call will fail to allocate the linear address space.
>>>>
>>>> Fix by reverting to the previous behavior and using ioremap() for the VGA
>>>> text buffer.
>>>>
>>>> Fixes: 81d195c6c0e2 ('x86: introduce ioremap_wc()')
>>>> Signed-off-by: Roger Pau Monné <roger....@citrix.com>
>>>
>>> This is somewhat ugly.
>>>
>>> ioremap() isn't really any better than ioremap_wc(); this only works
>>> because plain ioremap() has a special case for the bottom 1M where it
>>> does nothing and exits.
>>
>> And this is exactly why I screwed up back then. Imo we would be better
>> off moving to using __va() directly here. Otherwise the same mistake
>> could be made again by someone (perhaps even me) noticing the less
>> efficient ioremap(), when ioremap_wc() really would be wanted.
>
> I can switch to using __va(), that's fine. I guess we should then
> remove the special casing for the low 1MB in ioremap()?
I'm not sure we can - DMI and ACPI might use that, and who knows what
else. But of course if you're sure nothing depends on that anymore ...
Jan
