CVE-2024-50221

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> drm/amd/pm: Vangogh: Fix kernel memory out of bounds write<br /> <br /> KASAN reports that the GPU metrics table allocated in<br /> vangogh_tables_init() is not large enough for the memset done in<br /> smu_cmn_init_soft_gpu_metrics(). Condensed report follows:<br /> <br /> [ 33.861314] BUG: KASAN: slab-out-of-bounds in smu_cmn_init_soft_gpu_metrics+0x73/0x200 [amdgpu]<br /> [ 33.861799] Write of size 168 at addr ffff888129f59500 by task mangoapp/1067<br /> ...<br /> [ 33.861808] CPU: 6 UID: 1000 PID: 1067 Comm: mangoapp Tainted: G W 6.12.0-rc4 #356 1a56f59a8b5182eeaf67eb7cb8b13594dd23b544<br /> [ 33.861816] Tainted: [W]=WARN<br /> [ 33.861818] Hardware name: Valve Galileo/Galileo, BIOS F7G0107 12/01/2023<br /> [ 33.861822] Call Trace:<br /> [ 33.861826] <br /> [ 33.861829] dump_stack_lvl+0x66/0x90<br /> [ 33.861838] print_report+0xce/0x620<br /> [ 33.861853] kasan_report+0xda/0x110<br /> [ 33.862794] kasan_check_range+0xfd/0x1a0<br /> [ 33.862799] __asan_memset+0x23/0x40<br /> [ 33.862803] smu_cmn_init_soft_gpu_metrics+0x73/0x200 [amdgpu 13b1bc364ec578808f676eba412c20eaab792779]<br /> [ 33.863306] vangogh_get_gpu_metrics_v2_4+0x123/0xad0 [amdgpu 13b1bc364ec578808f676eba412c20eaab792779]<br /> [ 33.864257] vangogh_common_get_gpu_metrics+0xb0c/0xbc0 [amdgpu 13b1bc364ec578808f676eba412c20eaab792779]<br /> [ 33.865682] amdgpu_dpm_get_gpu_metrics+0xcc/0x110 [amdgpu 13b1bc364ec578808f676eba412c20eaab792779]<br /> [ 33.866160] amdgpu_get_gpu_metrics+0x154/0x2d0 [amdgpu 13b1bc364ec578808f676eba412c20eaab792779]<br /> [ 33.867135] dev_attr_show+0x43/0xc0<br /> [ 33.867147] sysfs_kf_seq_show+0x1f1/0x3b0<br /> [ 33.867155] seq_read_iter+0x3f8/0x1140<br /> [ 33.867173] vfs_read+0x76c/0xc50<br /> [ 33.867198] ksys_read+0xfb/0x1d0<br /> [ 33.867214] do_syscall_64+0x90/0x160<br /> ...<br /> [ 33.867353] Allocated by task 378 on cpu 7 at 22.794876s:<br /> [ 33.867358] kasan_save_stack+0x33/0x50<br /> [ 33.867364] kasan_save_track+0x17/0x60<br /> [ 33.867367] __kasan_kmalloc+0x87/0x90<br /> [ 33.867371] vangogh_init_smc_tables+0x3f9/0x840 [amdgpu]<br /> [ 33.867835] smu_sw_init+0xa32/0x1850 [amdgpu]<br /> [ 33.868299] amdgpu_device_init+0x467b/0x8d90 [amdgpu]<br /> [ 33.868733] amdgpu_driver_load_kms+0x19/0xf0 [amdgpu]<br /> [ 33.869167] amdgpu_pci_probe+0x2d6/0xcd0 [amdgpu]<br /> [ 33.869608] local_pci_probe+0xda/0x180<br /> [ 33.869614] pci_device_probe+0x43f/0x6b0<br /> <br /> Empirically we can confirm that the former allocates 152 bytes for the<br /> table, while the latter memsets the 168 large block.<br /> <br /> Root cause appears that when GPU metrics tables for v2_4 parts were added<br /> it was not considered to enlarge the table to fit.<br /> <br /> The fix in this patch is rather "brute force" and perhaps later should be<br /> done in a smarter way, by extracting and consolidating the part version to<br /> size logic to a common helper, instead of brute forcing the largest<br /> possible allocation. Nevertheless, for now this works and fixes the out of<br /> bounds write.<br /> <br /> v2:<br /> * Drop impossible v3_0 case. (Mario)<br /> <br /> (cherry picked from commit 0880f58f9609f0200483a49429af0f050d281703)