Re: [PATCH v3 1/2] PCI: dwc: Implement general suspend/resume functionality for L2/L3 transitions

[Manivannan Sadhasivam <mani@kernel.org>]

On Fri, Jul 21, 2023 at 10:09:18AM +0800, Shawn Lin wrote:
> 
> On 2023/7/21 0:07, Manivannan Sadhasivam wrote:
> > On Thu, Jul 20, 2023 at 10:37:36AM -0400, Frank Li wrote:
> > > On Thu, Jul 20, 2023 at 07:55:09PM +0530, Manivannan Sadhasivam wrote:
> > > > On Tue, Jul 18, 2023 at 03:34:26PM +0530, Manivannan Sadhasivam wrote:
> > > > > On Mon, Jul 17, 2023 at 02:36:19PM -0400, Frank Li wrote:
> > > > > > On Mon, Jul 17, 2023 at 10:15:26PM +0530, Manivannan Sadhasivam wrote:
> > > > > > > On Wed, Apr 19, 2023 at 12:41:17PM -0400, Frank Li wrote:
> > > > > > > > Introduced helper function dw_pcie_get_ltssm to retrieve SMLH_LTSS_STATE.
> > > > > > > > Added API pme_turn_off and exit_from_l2 for managing L2/L3 state transitions.
> > > > > > > > 
> > > > > > > > Typical L2 entry workflow:
> > > > > > > > 
> > > > > > > > 1. Transmit PME turn off signal to PCI devices.
> > > > > > > > 2. Await link entering L2_IDLE state.
> > > > > > > 
> > > > > > > AFAIK, typical workflow is to wait for PME_To_Ack.
> > > > > > 
> > > > > > 1 Already wait for PME_to_ACK,  2, just wait for link actual enter L2.
> > > > > > I think PCI RC needs some time to set link enter L2 after get ACK from
> > > > > > PME.
> > > > > > 
> > > > 
> > > > One more comment. If you transition the device to L2/L3, then it can loose power
> > > > if Vaux was not provided. In that case, can all the devices work after resume?
> > > > Most notably NVMe?
> > > 
> > > I have not hardware to do such test, NVMe driver will reinit everything after
> > > resume if no L1.1\L1.2 support. If there are L1.1\L1.2, NVME expect it leave
> > > at L1.2 at suspend to get better resume latency.
> > > 
> > 
> > To be precise, NVMe driver will shutdown the device if there is no ASPM support
> > and keep it in low power mode otherwise (there are other cases as well but we do
> > not need to worry).
> > 
> > But here you are not checking for ASPM state in the suspend path, and just
> > forcing the link to be in L2/L3 (thereby D3Cold) even though NVMe driver may
> > expect it to be in low power state like ASPM/APST.
> > 
> > So you should only put the link to L2/L3 if there is no ASPM support. Otherwise,
> > you'll ending up with bug reports when users connect NVMe to it.
> > 
> 
> 
> At this topic, it's very interesting to look at
> 
> drivers/pci/controller/dwc/pcie-tegra194.c
> 
> 
> static int tegra_pcie_dw_suspend_noirq(struct device *dev)
> {
>         struct tegra_pcie_dw *pcie = dev_get_drvdata(dev);
> 
>         if (!pcie->link_state)
>                 return 0;
> 
>         tegra_pcie_downstream_dev_to_D0(pcie);
>         tegra_pcie_dw_pme_turnoff(pcie);
>         tegra_pcie_unconfig_controller(pcie);
> 
>         return 0;
> }
> 
> It brings back all the downstream components to D0, as I assumed it was L0
> indeed, before sending PME aiming to enter L2.
> 

The behavior is Tegra specific as mentioned in the comment in
tegra_pcie_downstream_dev_to_D0():

        /*
         * link doesn't go into L2 state with some of the endpoints with Tegra
         * if they are not in D0 state. So, need to make sure that immediate
         * downstream devices are in D0 state before sending PME_TurnOff to put
         * link into L2 state.
         * This is as per PCI Express Base r4.0 v1.0 September 27-2017,
         * 5.2 Link State Power Management (Page #428).
         */

But I couldn't find the behavior documented in the spec as per the comment. Not
sure if I'm reading it wrong!

Also, I can confirm from previous interations with the linux-nvme list that
Tegra also faces the suspend issue with NVMe devices.

- Mani

- Mani

> > - Mani
> > 
> > > This API help remove duplicate codes and it can be improved gradually.
> > > 
> > > 
> > > > 
> > > > - Mani
> > > > 
> > > > 
> > > > -- 
> > > > மணிவண்ணன் சதாசிவம்
> > 

-- 
மணிவண்ணன் சதாசிவம்