Re: [PATCH RFC v1 3/3] PCI: qcom: ep: Add wake up host op to dw_pcie_ep_ops

[Manivannan Sadhasivam]

On Wed, Jun 28, 2023 at 08:01:45AM +0530, Krishna Chaitanya Chundru wrote:
> 
> On 6/27/2023 7:23 PM, Manivannan Sadhasivam wrote:
> > On Mon, Jun 26, 2023 at 07:18:49PM +0530, Krishna Chaitanya Chundru wrote:
> > > On 6/23/2023 11:48 AM, Manivannan Sadhasivam wrote:
> > > > On Wed, Jun 14, 2023 at 08:30:49PM +0530, Krishna chaitanya chundru wrote:
> > > > > Add wakeup host op to dw_pcie_ep_ops to wake up host from D3cold
> > > > > or D3hot.
> > > > > 
> > > > Commit message should describe how the wakeup is implemented in the driver.
> > > I will correct this in next series.
> > > > > Signed-off-by: Krishna chaitanya chundru <quic_krichai@xxxxxxxxxxx>
> > > > > ---
> > > > >    drivers/pci/controller/dwc/pcie-qcom-ep.c | 34 +++++++++++++++++++++++++++++++
> > > > >    1 file changed, 34 insertions(+)
> > > > > 
> > > > > diff --git a/drivers/pci/controller/dwc/pcie-qcom-ep.c b/drivers/pci/controller/dwc/pcie-qcom-ep.c
> > > > > index 5d146ec..916a138 100644
> > > > > --- a/drivers/pci/controller/dwc/pcie-qcom-ep.c
> > > > > +++ b/drivers/pci/controller/dwc/pcie-qcom-ep.c
> > > > > @@ -91,6 +91,7 @@
> > > > >    /* PARF_PM_CTRL register fields */
> > > > >    #define PARF_PM_CTRL_REQ_EXIT_L1		BIT(1)
> > > > >    #define PARF_PM_CTRL_READY_ENTR_L23		BIT(2)
> > > > > +#define PARF_PM_CTRL_XMT_PME			BIT(4)
> > > > >    #define PARF_PM_CTRL_REQ_NOT_ENTR_L1		BIT(5)
> > > > >    /* PARF_MHI_CLOCK_RESET_CTRL fields */
> > > > > @@ -794,10 +795,43 @@ static void qcom_pcie_ep_init(struct dw_pcie_ep *ep)
> > > > >    		dw_pcie_ep_reset_bar(pci, bar);
> > > > >    }
> > > > > +static int qcom_pcie_ep_wakeup_host(struct dw_pcie_ep *ep, u8 func_no)
> > > > > +{
> > > > > +	struct dw_pcie *pci = to_dw_pcie_from_ep(ep);
> > > > > +	struct qcom_pcie_ep *pcie_ep = to_pcie_ep(pci);
> > > > > +	struct device *dev = pci->dev;
> > > > > +	u32 perst, dstate, val;
> > > > > +
> > > > > +	perst = gpiod_get_value(pcie_ep->reset);
> > > > > +	/* Toggle wake GPIO when device is in D3 cold */
> > > > > +	if (perst) {
> > > > > +		dev_info(dev, "Device is in D3 cold toggling wake\n");
> > > > dev_dbg(). "Waking up the host by toggling WAKE#"
> > > > 
> > > > > +		gpiod_set_value_cansleep(pcie_ep->wake, 1);
> > > > Waking a device from D3cold requires power-on sequence by the host and in the
> > > > presence of Vaux, the EPF should be prepared for that. In that case, the mode of
> > > > wakeup should be decided by the EPF driver. So the wakeup API should have an
> > > > argument to decide whether the wakeup is through PME or sideband WAKE#.
> > > > 
> > > > Also note that as per PCIe Spec 3.0, the devices can support PME generation from
> > > > D3cold provided that the Vaux is supplied to the device. I do not know if that
> > > > is supported by Qcom devices but API should honor the spec. So the wakeup
> > > > control should come from EPF driver as I suggested above.
> > > I aggre with you, but how will EPF know the PCI device state is in D3cold or
> > > D3hot.
> > > 
> > We should add a notifier in the controller driver which signals EPF when it
> > receives the DState events.. Take a look at pci_epc_linkdown().
> Ok I will add this kind of Dstate change event
> > 
> > > And how the EPF knows whether Vaux is supported or not in D3cold?
> > > 
> > > If there is any existing mechanism can you please point me that.
> > > 
> > > FYI Qcom does not support vaux power in D3 cold.
> > > 
> > If the EPF can trigger wakeup event during D3Cold then it means it is powered
> > by Vaux, isn't it?
> > 
> > - Mani
> 
> EPF needs to know that if the endpoint is getting vaux in D3cold or not
> without this info
> 
> EPF doesn't know wheather to send toggle wake or send Inband PME right.
> 
> I mean EPF should have some  prior information wheather vaux is supplied or
> not to decide
> 
> wheather to toggle wake or send in band PME.
> 

Controller driver can use the #PERST level to detect D3Cold. Then it can pass
that info to EPF over notifiers. EPF may decide whether to toggle #WAKE or
send PME based on its configuration. For MHI EPF, it can toggle #WAKE as PME
during D3Cold is not supported.

- Mani

> -KC
> 
> > 
> > > > > +		usleep_range(WAKE_DELAY_US, WAKE_DELAY_US + 500);
> > > > > +		gpiod_set_value_cansleep(pcie_ep->wake, 0);
> > > > > +		return 0;
> > > > > +	}
> > > > > +
> > > > > +	dstate = dw_pcie_readl_dbi(pci, DBI_CON_STATUS) &
> > > > > +				   DBI_CON_STATUS_POWER_STATE_MASK;
> > > > > +	if (dstate == 3) {
> > > > > +		dev_info(dev, "Device is in D3 hot sending inband PME\n");
> > > > dev_dbg(). As I said, the device can sent PME from D3cold also. So the log could
> > > > be "Waking up the host using PME".
> > > > 
> > > > > +		val = readl_relaxed(pcie_ep->parf + PARF_PM_CTRL);
> > > > > +		val |= PARF_PM_CTRL_XMT_PME;
> > > > > +		writel_relaxed(val, pcie_ep->parf + PARF_PM_CTRL);
> > > > > +	} else {
> > > > > +		dev_err(dev, "Device is not in D3 state wakeup is not supported\n");
> > > > > +		return -EPERM;
> > > > -ENOTSUPP
> > > > 
> > > > - Mani
> > > > 
> > > > > +	}
> > > > > +
> > > > > +	return 0;
> > > > > +}
> > > > > +
> > > > >    static const struct dw_pcie_ep_ops pci_ep_ops = {
> > > > >    	.ep_init = qcom_pcie_ep_init,
> > > > >    	.raise_irq = qcom_pcie_ep_raise_irq,
> > > > >    	.get_features = qcom_pcie_epc_get_features,
> > > > > +	.wakeup_host = qcom_pcie_ep_wakeup_host,
> > > > >    };
> > > > >    static int qcom_pcie_ep_probe(struct platform_device *pdev)
> > > > > -- 
> > > > > 2.7.4
> > > > > 

-- 
மணிவண்ணன் சதாசிவம்