user/sven/linux.git/drivers/gpu/drm/amd, branch v6.7.2

drm/amdgpu: fall back to INPUT power for AVG power via INFO IOCTL

2024-01-25T23:45:28Z

[ Upstream commit d02069850fc102b07ae923535d5e212f2c8a34e9 ] For backwards compatibility with userspace. Fixes: 47f1724db4fe ("drm/amd: Introduce `AMDGPU_PP_SENSOR_GPU_INPUT_POWER`") Link: https://gitlab.freedesktop.org/drm/amd/-/issues/2897 Reviewed-by: Yang Wang Signed-off-by: Alex Deucher Signed-off-by: Sasha Levin

drm/amdkfd: fixes for HMM mem allocation

2024-01-25T23:45:28Z

[ Upstream commit 02eed83abc1395a1207591aafad9bcfc5cb1abcb ] Fix err return value and reset pgmap->type after checking it. Fixes: c83dee9b6394 ("drm/amdkfd: add SPM support for SVM") Reviewed-by: Felix Kuehling Signed-off-by: Dafna Hirschfeld Signed-off-by: Felix Kuehling Signed-off-by: Alex Deucher Signed-off-by: Sasha Levin

drm/amdgpu: revert "Adjust removal control flow for smu v13_0_2"

2024-01-25T23:45:16Z

commit fb1c93c2e9604a884467a773790016199f78ca08 upstream. Calling amdgpu_device_ip_resume_phase1() during shutdown leaves the HW in an active state and is an unbalanced use of the IP callbacks. Using the IP callbacks like this can lead to memory leaks, double free and imbalanced reference counters. Leaving the HW in an active state can lead to DMA accesses to memory now freed by the driver. Both is a complete no-go for driver unload so completely revert the workaround for now. This reverts commit f5c7e7797060255dbc8160734ccc5ad6183c5e04. Signed-off-by: Christian König Acked-by: Alex Deucher Signed-off-by: Alex Deucher Cc: stable@vger.kernel.org Signed-off-by: Greg Kroah-Hartman

drm/amd: Enable PCIe PME from D3

2024-01-25T23:45:14Z

commit bd1f6a31e7762ebc99b97f3eda5e5ea3708fa792 upstream. When dGPU is put into BOCO it may be in D3cold but still able send PME on display hotplug event. For this to work it must be enabled as wake source from D3. When runpm is enabled use pci_wake_from_d3() to mark wakeup as enabled by default. Cc: stable@vger.kernel.org # 6.1+ Signed-off-by: Mario Limonciello Acked-by: Alex Deucher Signed-off-by: Alex Deucher Signed-off-by: Greg Kroah-Hartman

Revert "drm/amdkfd: Relocate TBA/TMA to opposite side of VM hole"

2024-01-25T23:45:09Z

commit 0f35b0a7b8fa402adbffa2565047cdcc4c480153 upstream. That commit causes NULL pointer dereferences in dmesgs when running applications using ROCm, including clinfo, blender, and PyTorch, since v6.6.1. Revert it to fix blender again. This reverts commit 96c211f1f9ef82183493f4ceed4e347b52849149. Closes: https://github.com/ROCm/ROCm/issues/2596 Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/2991 Reviewed-by: Jay Cornwall Signed-off-by: Kaibo Ma Signed-off-by: Alex Deucher Signed-off-by: Greg Kroah-Hartman

drm/amd/display: avoid stringop-overflow warnings for dp_decide_lane_settings()

2024-01-25T23:45:08Z

[ Upstream commit c966dc0e9d96dc44423c404a2628236f1200c24e ] gcc prints a warning about a possible array overflow for a couple of callers of dp_decide_lane_settings() after commit 1b56c90018f0 ("Makefile: Enable -Wstringop-overflow globally"): drivers/gpu/drm/amd/amdgpu/../display/dc/link/protocols/link_dp_training_fixed_vs_pe_retimer.c: In function 'dp_perform_fixed_vs_pe_training_sequence_legacy': drivers/gpu/drm/amd/amdgpu/../display/dc/link/protocols/link_dp_training_fixed_vs_pe_retimer.c:426:25: error: 'dp_decide_lane_settings' accessing 4 bytes in a region of size 1 [-Werror=stringop-overflow=] 426 | dp_decide_lane_settings(lt_settings, dpcd_lane_adjust, | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 427 | lt_settings->hw_lane_settings, lt_settings->dpcd_lane_settings); | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ drivers/gpu/drm/amd/amdgpu/../display/dc/link/protocols/link_dp_training_fixed_vs_pe_retimer.c:426:25: note: referencing argument 4 of type 'union dpcd_training_lane[4]' I'm not entirely sure what caused this, but changing the prototype to expect a pointer instead of an array avoids the warnings. Fixes: 7727e7b60f82 ("drm/amd/display: Improve robustness of FIXED_VS link training at DP1 rates") Acked-by: Randy Dunlap Tested-by: Randy Dunlap # build-tested Signed-off-by: Arnd Bergmann Signed-off-by: Alex Deucher Signed-off-by: Sasha Levin

drm/amd/pm/smu7: fix a memleak in smu7_hwmgr_backend_init

2024-01-25T23:45:08Z

[ Upstream commit 2f3be3ca779b11c332441b10e00443a2510f4d7b ] The hwmgr->backend, (i.e. data) allocated by kzalloc is not freed in the error-handling paths of smu7_get_evv_voltages and smu7_update_edc_leakage_table. However, it did be freed in the error-handling of phm_initializa_dynamic_state_adjustment_rule_settings, by smu7_hwmgr_backend_fini. So the lack of free in smu7_get_evv_voltages and smu7_update_edc_leakage_table is considered a memleak in this patch. Fixes: 599a7e9fe1b6 ("drm/amd/powerplay: implement smu7 hwmgr to manager asics with smu ip version 7.") Fixes: 8f0804c6b7d0 ("drm/amd/pm: add edc leakage controller setting") Signed-off-by: Zhipeng Lu Signed-off-by: Alex Deucher Signed-off-by: Sasha Levin

drm/amdkfd: Confirm list is non-empty before utilizing list_first_entry in kfd_topology.c

2024-01-25T23:45:08Z

[ Upstream commit 499839eca34ad62d43025ec0b46b80e77065f6d8 ] Before using list_first_entry, make sure to check that list is not empty, if list is empty return -ENODATA. Fixes the below: drivers/gpu/drm/amd/amdgpu/../amdkfd/kfd_topology.c:1347 kfd_create_indirect_link_prop() warn: can 'gpu_link' even be NULL? drivers/gpu/drm/amd/amdgpu/../amdkfd/kfd_topology.c:1428 kfd_add_peer_prop() warn: can 'iolink1' even be NULL? drivers/gpu/drm/amd/amdgpu/../amdkfd/kfd_topology.c:1433 kfd_add_peer_prop() warn: can 'iolink2' even be NULL? Fixes: 0f28cca87e9a ("drm/amdkfd: Extend KFD device topology to surface peer-to-peer links") Cc: Felix Kuehling Cc: Christian König Cc: Alex Deucher Signed-off-by: Srinivasan Shanmugam Suggested-by: Felix Kuehling Suggested-by: Lijo Lazar Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher Signed-off-by: Sasha Levin

drm/amdkfd: Fix type of 'dbg_flags' in 'struct kfd_process'

2024-01-25T23:45:08Z

[ Upstream commit 217e85f97031791fb48a2d374c7bdcf439365b21 ] dbg_flags looks to be defined with incorrect data type; to process multiple debug flag options, and hence defined dbg_flags as u32. Fixes the below: drivers/gpu/drm/amd/amdgpu/../amdkfd/kfd_packet_manager_v9.c:117 pm_map_process_aldebaran() warn: maybe use && instead of & Fixes: 0de4ec9a0353 ("drm/amdgpu: prepare map process for multi-process debug devices") Suggested-by: Lijo Lazar Cc: Felix Kuehling Cc: Christian König Cc: Alex Deucher Signed-off-by: Srinivasan Shanmugam Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher Signed-off-by: Sasha Levin

drm/amd/pm: fix a double-free in amdgpu_parse_extended_power_table

2024-01-25T23:45:05Z

[ Upstream commit a6582701178a47c4d0cb2188c965c59c0c0647c8 ] The amdgpu_free_extended_power_table is called in every error-handling paths of amdgpu_parse_extended_power_table. However, after the following call chain of returning: amdgpu_parse_extended_power_table |-> kv_dpm_init / si_dpm_init (the only two caller of amdgpu_parse_extended_power_table) |-> kv_dpm_sw_init / si_dpm_sw_init (the only caller of kv_dpm_init / si_dpm_init, accordingly) |-> kv_dpm_fini / si_dpm_fini (goto dpm_failed in xx_dpm_sw_init) |-> amdgpu_free_extended_power_table As above, the amdgpu_free_extended_power_table is called twice in this returning chain and thus a double-free is triggered. Similarily, the last kfree in amdgpu_parse_extended_power_table also cause a double free with amdgpu_free_extended_power_table in kv_dpm_fini. Fixes: 84176663e70d ("drm/amd/pm: create a new holder for those APIs used only by legacy ASICs(si/kv)") Signed-off-by: Zhipeng Lu Signed-off-by: Alex Deucher Signed-off-by: Sasha Levin