> -----Original Message----- > From: amd-gfx [mailto:amd-gfx-bounces at lists.freedesktop.org] On Behalf > Of Tom St Denis > Sent: Tuesday, September 12, 2017 10:18 AM > To: amd-gfx mailing list > Subject: write to GRBM_GFX_INDEX without lock > > In the file smu7_powertune.c we have > > int smu7_enable_didt_config(struct pp_hwmgr *hwmgr) > { > int result; > uint32_t num_se = 0; > uint32_t count, value, value2; > struct cgs_system_info sys_info = {0}; > > sys_info.size = sizeof(struct cgs_system_info); > sys_info.info_id = CGS_SYSTEM_INFO_GFX_SE_INFO; > result = cgs_query_system_info(hwmgr->device, &sys_info); > > if (result == 0) > num_se = sys_info.value; > > if (PP_CAP(PHM_PlatformCaps_SQRamping) || > PP_CAP(PHM_PlatformCaps_DBRamping) || > PP_CAP(PHM_PlatformCaps_TDRamping) || > PP_CAP(PHM_PlatformCaps_TCPRamping)) { > > cgs_enter_safe_mode(hwmgr->device, true); > value = 0; > value2 = cgs_read_register(hwmgr->device, > mmGRBM_GFX_INDEX); > for (count = 0; count < num_se; count++) { > value = > SYS_GRBM_GFX_INDEX_DATA__INSTANCE_BROADCAST_WRITES_MASK > | > SYS_GRBM_GFX_INDEX_DATA__SH_BROADCAST_WRITES_MASK > | (count << > SYS_GRBM_GFX_INDEX_DATA__SE_INDEX__SHIFT); > cgs_write_register(hwmgr->device, > mmGRBM_GFX_INDEX, value); > > > The call to "enter safe mode" doesn't hold the lock which means that > this could in theory occur while other things are going on. Granted > this function is likely called during init it should have a lock no? Yes, it should. I think there is a cgs function to handle the locking. Alex > > Cheers, > Tom > _______________________________________________ > amd-gfx mailing list > amd-gfx at lists.freedesktop.org > https://lists.freedesktop.org/mailman/listinfo/amd-gfx
