On Thu, Sep 25, 2014 at 12:07 PM, Daniel Kurtz djkurtz@chromium.org wrote:
On Thu, Sep 25, 2014 at 5:32 PM, Geert Uytterhoeven geert@linux-m68k.org wrote:
On Sun, Aug 24, 2014 at 4:50 PM, Daniel Kurtz djkurtz@chromium.org wrote:
Commit [0] stopped setting fix.smem_start and fix.smem_len when creating the fbdev.
[0] 2f1eab8d8ab59e799f7d51d62410b398607a7bc3 drm/exynos/fbdev: don't set fix.smem/mmio_{start,len}
However, smem_len is used by some userland applications to calculate the size for mmap. In particular, it is used by xf86-video-fbdev:
http://cgit.freedesktop.org/xorg/xserver/tree/hw/xfree86/fbdevhw/fbdevhw.c?i...
So, let's restore setting the smem_len to unbreak things for these users.
Note: we are still leaving smem_start set to 0.
Doesn't this cause a system crash when userspace (e.g. fbtest) writes into the mmap()ed /dev/fb*, as the wrong MMIO region is mapped?
Do you see a crash during testing, or is your question hypothetical?
It was hypothetical.
I don't think there will be one. exynos's fbev defines its own fb_ops.fb_mmap, which uses dma_mmap_attrs() to mmap the fb's gem buffer, which was allocated by dma_alloc_attrs(). This bypasses the code in drivers/video/fbmem.c:fb_mmap() that references fix.smem_start.
But, perhaps I am missing something?
Thanks, having your own mapping function explains the lack of crashes.
Still, some weird software may map /dev/mem instead of /dev/fb0, and use fb_fix.smem_start. But that's also unsafe in the context of LPAE.
Gr{oetje,eeting}s,
Geert
-- Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org
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