Hi,
On 2023/4/20 17:04, Geert Uytterhoeven wrote:
Hi Sui,
On Thu, Apr 20, 2023 at 5:09 AM Sui Jingfeng <suijingfeng@xxxxxxxxxxx> wrote:
The fbdev test of IGT may write after EOF, which lead to out-of-bound
access for drm drivers hire fbdev-generic. For example, run fbdev test
on a x86+ast2400 platform, with 1680x1050 resolution, will cause the
linux kernel hang with the following call trace:
Oops: 0000 [#1] PREEMPT SMP PTI
[IGT] fbdev: starting subtest eof
Workqueue: events drm_fb_helper_damage_work [drm_kms_helper]
[IGT] fbdev: starting subtest nullptr
RIP: 0010:memcpy_erms+0xa/0x20
RSP: 0018:ffffa17d40167d98 EFLAGS: 00010246
RAX: ffffa17d4eb7fa80 RBX: ffffa17d40e0aa80 RCX: 00000000000014c0
RDX: 0000000000001a40 RSI: ffffa17d40e0b000 RDI: ffffa17d4eb80000
RBP: ffffa17d40167e20 R08: 0000000000000000 R09: ffff89522ecff8c0
R10: ffffa17d4e4c5000 R11: 0000000000000000 R12: ffffa17d4eb7fa80
R13: 0000000000001a40 R14: 000000000000041a R15: ffffa17d40167e30
FS: 0000000000000000(0000) GS:ffff895257380000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: ffffa17d40e0b000 CR3: 00000001eaeca006 CR4: 00000000001706e0
Call Trace:
<TASK>
? drm_fbdev_generic_helper_fb_dirty+0x207/0x330 [drm_kms_helper]
drm_fb_helper_damage_work+0x8f/0x170 [drm_kms_helper]
process_one_work+0x21f/0x430
worker_thread+0x4e/0x3c0
? __pfx_worker_thread+0x10/0x10
kthread+0xf4/0x120
? __pfx_kthread+0x10/0x10
ret_from_fork+0x2c/0x50
</TASK>
CR2: ffffa17d40e0b000
---[ end trace 0000000000000000 ]---
The is because damage rectangles computed by
drm_fb_helper_memory_range_to_clip() function does not guaranteed to be
bound in the screen's active display area. Possible reasons are:
1) Buffers are allocated in the granularity of page size, for mmap system
call support. The shadow screen buffer consumed by fbdev emulation may
also choosed be page size aligned.
2) The DIV_ROUND_UP() used in drm_fb_helper_memory_range_to_clip()
will introduce off-by-one error.
For example, on a 16KB page size system, in order to store a 1920x1080
XRGB framebuffer, we need allocate 507 pages. Unfortunately, the size
1920*1080*4 can not be divided exactly by 16KB.
1920 * 1080 * 4 = 8294400 bytes
506 * 16 * 1024 = 8290304 bytes
507 * 16 * 1024 = 8306688 bytes
line_length = 1920*4 = 7680 bytes
507 * 16 * 1024 / 7680 = 1081.6
off / line_length = 507 * 16 * 1024 / 7680 = 1081
DIV_ROUND_UP(507 * 16 * 1024, 7680) will yeild 1082
memcpy_toio() typically issue the copy line by line, when copy the last
line, out-of-bound access will be happen. Because:
1082 * line_length = 1082 * 7680 = 8309760, and 8309760 > 8306688
Note that userspace may stil write to the invisiable area if a larger
buffer than width x stride is exposed. But it is not a big issue as
long as there still have memory resolve the access if not drafting so
far.
- Also limit the y1 (Daniel)
- keep fix patch it to minimal (Daniel)
- screen_size is page size aligned because of it need mmap (Thomas)
- Adding fixes tag (Thomas)
Fixes: aa15c677cc34 ("drm/fb-helper: Fix vertical damage clipping")
Signed-off-by: Sui Jingfeng <suijingfeng@xxxxxxxxxxx>
Reviewed-by: Thomas Zimmermann <tzimmermann@xxxxxxx>
Tested-by: Geert Uytterhoeven <geert+renesas@xxxxxxxxx>
Thanks for the update! This v5 is completely different from the v3
I tested before, so keeping my Tested-by is not really appropriate...
Indeed, I will be carefully next time.
I have retested fbtest with shmob-drm on Armadillo-800-EVA
(800x480@RG16, i.e. 187.5 pages), and fortunately this version still
works fine, so
Tested-by: Geert Uytterhoeven <geert+renesas@xxxxxxxxx>
Thanks a lot.
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx
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
