Am Dienstag, den 20.12.2016, 11:55 +0530 schrieb Kishon Vijay Abraham I: > Hi Bjorn, > > On Friday 09 December 2016 02:13 AM, Bjorn Helgaas wrote: > > On Thu, Dec 08, 2016 at 02:19:00PM +0530, Kishon Vijay Abraham I wrote: > >> Hi Bjorn, > >> > >> On Thursday 08 December 2016 03:32 AM, Bjorn Helgaas wrote: > >>> Hi Kishon, et al, > >>> > >>> Does dra7xx suspend/resume work? I'm not sure dra7xx_pcie_resume() > >>> and dra7xx_pcie_resume_noirq() restore everything necessary. For > >>> example, the probe path has this: > >>> > >>> dra7xx_pcie_probe > >>> dra7xx_add_pcie_port > >>> dw_pcie_host_init > >>> dra7xx_pcie_host_init # .host_init > >>> dw_pcie_setup_rc > >>> dw_pcie_prog_outbound_atu > >>> > >>> so I think it programs the ATU in dw_pcie_setup_rc(). But the resume > >>> path doesn't call dw_pcie_setup_rc(), so I don't see where the ATU > >>> setup would be restored. > >> > >> DRA7xx only supported shallow power state and not deep power state. In shallow > >> power state, the register settings are not lost and hence we didn't have to > >> restore anything. > > > > I'm not really a PM guy, so I'm not familiar with shallow power state. > > Could I have discovered from the code somehow that dra7xx suspend to > > shallow power state preserves register state? It's nice if a reader > > can verify correctness without knowing the system-specific details. > > Some drivers like drivers/mmc/host/omap_hsmmc.c saves the register contents > (omap_hsmmc_context_save) and then in runtime_resume checks if the register > contents are still same (omap_hsmmc_context_restore). Do you think we have to > do something like that for dra7xx? > > omap_hsmmc.c is used by a lot of platforms some of which loses the context > during suspend. Since dra7xx doesn't loose context, not sure if we have to do > something like omap_hsmmc. I think we need some DW-PCIe helper functions that save and restore the register state and can be called from drivers on platforms where the core losses its state over PM. I think this should be fully in the control of the individual glue drivers, as it is really platform dependent. On MX6 the older generations only support shallow PM states, where we don't need the save/restore dance, but newer generations could go into deeper idle where the core looses state. Regards, Lucas -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
