Hi

On Sat, Aug 3, 2013 at 5:53 PM, H. Peter Anvin [off-list ref] wrote:

When compiled for i386 PAE phys_addr_t is 64 bits but pointers are 32 bits.

Yepp, that makes sense. I am not really comfortable fixing this as I
have no idea how PAE actually works, but I digged through git history
and sent a patch to linux-fbdev (I put you on CC). The (long)-cast
should be sufficient, I guess.

Were there any other build warnings you encountered?

Thanks
David

David Herrmann [off-list ref] wrote:

quoted

Hi

On Sat, Aug 3, 2013 at 1:39 AM, H. Peter Anvin [off-list ref] wrote:

quoted

On 08/02/2013 05:05 AM, David Herrmann wrote:

quoted

Hi

I cut down my previous series to no longer include the SimpleDRM

driver. If

quoted

quoted

anyone is interested, you can find it here:
  http://lwn.net/Articles/558104/
I will resend it once these preparation patches are in.

Changes since v2:
 - added common x86 formats (reported by hpa) (patch #5)

This whole series (including simpledrm) is tested by Stephen and me.

I would be

quoted

quoted

glad if maintainers could ack/nack this so I can continue my work.

This series is pretty small and just converts x86 to use

platform-devices

quoted

quoted

instead of global objects to pass framebuffer data to drivers. The

commit

quoted

quoted

messages explain everything in detail.
The idea is to create a "platform-framebuffer" device which drivers

can bind to.

quoted

quoted

If x86 boot code detectes efi or vesa framebuffers, it creates

efi-framebuffer

quoted

quoted

or vesa-framebuffer devices instead.

Additionally, if the modes are compatible, "simple-framebuffer"

devices are

quoted

quoted

created so simplefb can be used on x86. This feature is only enabled

if

quoted

quoted

CONFIG_X86_SYSFB is selected (off by default) so users without

simplefb still

quoted

quoted

get boot logs.

 @Stephen: I wasn't sure whether you tested the efi/vesa framebuffer

changes,

quoted

quoted

too, so I didn't add your tested-by there. And I changed patch #5 so

I dropped

quoted

quoted

it there, too. Thanks for testing!

I am getting a bunch of new warnings with this patchset, typically of
the form:


/home/hpa/kernel/distwork/drivers/video/arkfb.c:1019:23: warning:

cast

quoted

to pointer from integer of different size [-Wint-to-pointer-cast]
  par->state.vgabase = (void __iomem *) vga_res.start;
                       ^
/home/hpa/kernel/distwork/drivers/video/s3fb.c: In function

‘s3_pci_probe’:

quoted

/home/hpa/kernel/distwork/drivers/video/s3fb.c:1186:23: warning: cast

to

quoted

pointer from integer of different size [-Wint-to-pointer-cast]
  par->state.vgabase = (void __iomem *) vga_res.start;
                       ^
I have pushed it out to a topic branch in the tip tree, partly to

give

quoted

Fengguang's build bot a run at it (it is excellent at spotting the
origin of new warnings), but this needs to be fixed; we will not

merge

quoted

this branch in its current form.

Thanks for looking at it. However, these warnings seem unrelated to my
patchset. I didn't change any vga/fb headers nor did I touch
arkfb/s3fb. Furthermore, I cannot reproduce these warnings. "vga_res"
is a "struct resource", "->start" is of type "resource_size_t" which
is typedef'ed to "phys_addr_t". Seems like the build-bot used some
random config with CONFIG_PHYS_ADDR_T_64BIT set but building on 32bit
(or something like that).

Is there any place where I can see the compiler log? Or the used
.config file?

Thanks
David

--
Sent from my mobile phone. Please excuse brevity and lack of formatting.