On 20/09/17 03:07 AM, Christian König wrote: > Am 19.09.2017 um 23:38 schrieb Tom St Denis: >> On 19/09/17 02:33 PM, Christian König wrote: >>>> [root at carrizo ~]# xxd -l 1024 -s 0xC0000 >>>> /sys/kernel/debug/dri/0/amdgpu_iova >>> >>> Actually 0xC0000 is a special address, e.g. video BIOS if I'm not >>> completely mistaken. >> >> Not sure why that would be mapped by the driver but I can also read >> the kernel's bss section with this... >> >> $ xxd -l 1048576 -s 0x01e4c000 /sys/kernel/debug/dri/0/amdgpu_iova >> >> ...<snip>... >> 01e6a430: 4c69 6e75 7820 7665 7273 696f 6e20 342e Linux version 4. >> 01e6a440: 3133 2e30 2d72 6335 2b20 2872 6f6f 7440 13.0-rc5+ (root@ >> 01e6a450: 6361 7272 697a 6f29 2028 6763 6320 7665 carrizo) (gcc ve >> 01e6a460: 7273 696f 6e20 362e 332e 3120 3230 3136 rsion 6.3.1 2016 >> 01e6a470: 3132 3231 2028 5265 6420 4861 7420 362e 1221 (Red Hat 6. >> 01e6a480: 332e 312d 3129 2028 4743 4329 2920 2333 3.1-1) (GCC)) #3 >> 01e6a490: 3120 534d 5020 5475 6520 5365 7020 3139 1 SMP Tue Sep 19 >> 01e6a4a0: 2030 373a 3138 3a33 3120 4544 5420 3230  07:18:31 EDT 20 >> <snip> >> >> That's part of the dmesg buffer apparently. >> >> I pointed it at all sorts of address (bios/system ram/etc) it pretty >> much will read anything. > > Yeah, feared that this would be the case. OK in this case your original > concern was completely correct and we can't allow this in general. Which means we want to go to a different v2 of this patch https://lists.freedesktop.org/archives/amd-gfx/2017-September/013329.html Where I check if domain==NULL and return the input or use the iova_to_phys to translate. Then in umr I have to revert the last commit on master and apply the original patch I had sent to the list: https://lists.freedesktop.org/archives/amd-gfx/2017-September/013332.html Tom
