On 28.06.2024 10:48 AM, Varadarajan Narayanan wrote: > On Thu, Jun 27, 2024 at 12:00:35AM +0200, Konrad Dybcio wrote: >> On 19.06.2024 9:36 AM, Varadarajan Narayanan wrote: >> >> [...] >> >> >>> Tested the patches with both gcc and nsscc providers having >>> 'sync_state' set to icc_sync_state. >>> >>> # dmesg | grep synced >>> [ 3.029820] qcom,gcc-ipq9574 1800000.clock-controller: interconnect provider is in synced state >>> [ 3.470106] qcom,nsscc-ipq9574 39b00000.clock-controller: interconnect provider is in synced state >>> >>> I can see that icc_sync_state is getting called and clocks >>> related to paths with zero bandwidth are getting disabled. >>> >>> Will post the NSSCC patches to get the full picture. >> >> Going back to the original question, does removing interconnects = from >> things like PCIe now make them not work / crash the device, which would >> indicate the NoC clocks were indeed gated? > > Yes. With and without 'interconnects =', the following behaviour > is same > * Boot completes > * PCIe devices were probed succesfully and can be > seen in /proc/bus/pci/devices. > * icc_sync_state is called. The system has 4 pcie nodes > in the DT, out of which pcie0 is not enabled. > > The difference is seen in icc_sync_state > > With 'interconnects =' > > * During icc_sync_state, the following 2 clocks > corresponding to the interconnects of 'pcie0' get > disabled. > > [ 2.986356] ---> clk_core_disable_lock: gcc_anoc_pcie0_1lane_m_clk > [ 3.012486] ---> clk_core_disable_lock: gcc_snoc_pcie0_1lane_s_clk > > * System shutdown also completes without issues > > Without the 'interconnects =', > > * During icc_sync_state, the following clocks > corresponding to the interconnects of all the 4 PCIe > nodes get disabled. > > [ 2.887860] ---> clk_core_disable_lock: gcc_anoc_pcie0_1lane_m_clk > [ 2.913988] ---> clk_core_disable_lock: gcc_snoc_pcie0_1lane_s_clk > [ 2.939857] ---> clk_core_disable_lock: gcc_anoc_pcie1_1lane_m_clk > [ 2.965725] ---> clk_core_disable_lock: gcc_snoc_pcie1_1lane_s_clk > [ 2.991594] ---> clk_core_disable_lock: gcc_anoc_pcie2_2lane_m_clk > [ 3.017463] ---> clk_core_disable_lock: gcc_snoc_pcie2_2lane_s_clk > [ 3.043328] ---> clk_core_disable_lock: gcc_anoc_pcie3_2lane_m_clk > [ 3.069201] ---> clk_core_disable_lock: gcc_snoc_pcie3_2lane_s_clk > > * System shutdown hangs (possibly due to un-clocked > access of PCIe register) in pcie_pme_interrupt_enable > > [ 10.773134] dump_stack+0x18/0x24 > [ 10.776779] pcie_pme_remove+0x2c/0x88 > [ 10.780078] pcie_port_remove_service+0x50/0x74 > [ 10.783725] device_remove+0x12c/0x148 > [ 10.788151] __device_release_driver+0x65c/0x8cc > [ 10.791972] device_release_driver+0x2c/0x44 > [ 10.796746] bus_remove_device+0xcc/0x10c > [ 10.800999] device_del+0x14c/0x400 > [ 10.804904] device_unregister+0x18/0x34 > [ 10.808203] remove_iter+0x2c/0x3c > [ 10.812369] device_for_each_child+0x60/0xb4 > [ 10.815583] pcie_portdrv_shutdown+0x34/0x90 > [ 10.820009] pci_device_shutdown+0x34/0x74 > [ 10.824263] device_shutdown+0x150/0x258 > [ 10.828169] kernel_restart_prepare+0x98/0xbc > [ 10.832249] kernel_restart+0x44/0x110 > [ 10.836502] __do_sys_reboot+0x18c/0x304 > > I believe, this is confirms NOC clocks getting disabled by > icc_sync_state. Yes, this looks good now, thanks. Konrad
