2011/9/13 Chen Jie <[email protected]>: > Hi, > > 2011/9/7 Michel Dänzer <[email protected]> >> >> What if you only comment out one of the invocations? Do you get >> corruption in one case but not the other one? > I got corruption in either case. > > We got a patch from the BSP team of the CPU vendor, which seems solved > the problem. > > The patch replaces both RADEON_GPU_PAGE_SIZE in function > r600_copy_blit() with PAGE_SIZE. (We use 16K PAGE_SIZE, and > RADEON_GPU_PAGE_SIZE is 4K) > > Any idea? If it had copy problems, why bo moving test couldn't detect > it? Also, after applied the patch, the bo moving test was still > passed, curious.
The patch is correct. r600_copy_blit() takes CPU pages as a parameter and the r600_blit_*_copy() functions take bytes as a parameter. evergreen_copy_blit() needs a similar fix. > > BTW, if the patch was correct, then it seems more suitable to be fixed > in radeon_move_blit(): > * The num_pages field of ttm_mem_reg need to be changed if moved > between VRAM and GTT && PAGE != RADEON_GPU_PAGE_SIZE. There may be other places where we are using the wrong page size for GPU vs CPU pages. > * What is a drm_mm_node? Represent a block resides at VRAM/GTT/System Mem? > drm_mm is basically a set of helper functions for managing address ranges. Alex _______________________________________________ xorg-driver-ati mailing list [email protected] http://lists.x.org/mailman/listinfo/xorg-driver-ati
