Re: [PATCH v11 3/3] PCI: layerscape: Add power management support for ls1028a

[Manivannan Sadhasivam]

On Mon, Aug 21, 2023 at 10:21:05AM +0200, Lorenzo Pieralisi wrote:
> On Thu, Aug 17, 2023 at 06:42:50PM -0400, Frank Li wrote:
> > On Wed, Aug 16, 2023 at 05:30:10PM +0200, Lorenzo Pieralisi wrote:
> > > On Wed, Aug 09, 2023 at 11:35:40AM -0400, Frank Li wrote:
> > > > From: Hou Zhiqiang <Zhiqiang.Hou@xxxxxxx>
> > > > 
> > > > Add PME_Turn_off/PME_TO_Ack handshake sequence for ls1028a platform. Call
> > > > common dwc dw_pcie_suspend(resume)_noirq() function when system enter/exit
> > > > suspend state.
> > > > 
> > > > Acked-by: Manivannan Sadhasivam <mani@xxxxxxxxxx>
> > > > Signed-off-by: Hou Zhiqiang <Zhiqiang.Hou@xxxxxxx>
> > > > Signed-off-by: Frank Li <Frank.Li@xxxxxxx>
> > > > ---
> > > >  drivers/pci/controller/dwc/pci-layerscape.c | 130 ++++++++++++++++++--
> > > >  1 file changed, 121 insertions(+), 9 deletions(-)
> > > > 
> > > > diff --git a/drivers/pci/controller/dwc/pci-layerscape.c b/drivers/pci/controller/dwc/pci-layerscape.c
> > > > index ed5fb492fe084..b49f654335fd7 100644
> > > > --- a/drivers/pci/controller/dwc/pci-layerscape.c
> > > > +++ b/drivers/pci/controller/dwc/pci-layerscape.c
> > > > @@ -8,9 +8,11 @@
> > > >   * Author: Minghuan Lian <Minghuan.Lian@xxxxxxxxxxxxx>
> > > >   */
> > > >  
> > > > +#include <linux/delay.h>
> > > >  #include <linux/kernel.h>
> > > >  #include <linux/interrupt.h>
> > > >  #include <linux/init.h>
> > > > +#include <linux/iopoll.h>
> > > >  #include <linux/of_pci.h>
> > > >  #include <linux/of_platform.h>
> > > >  #include <linux/of_address.h>
> > > > @@ -20,6 +22,7 @@
> > > >  #include <linux/mfd/syscon.h>
> > > >  #include <linux/regmap.h>
> > > >  
> > > > +#include "../../pci.h"
> > > >  #include "pcie-designware.h"
> > > >  
> > > >  /* PEX Internal Configuration Registers */
> > > > @@ -27,12 +30,26 @@
> > > >  #define PCIE_ABSERR		0x8d0 /* Bridge Slave Error Response Register */
> > > >  #define PCIE_ABSERR_SETTING	0x9401 /* Forward error of non-posted request */
> > > >  
> > > > +/* PF Message Command Register */
> > > > +#define LS_PCIE_PF_MCR		0x2c
> > > > +#define PF_MCR_PTOMR		BIT(0)
> > > > +#define PF_MCR_EXL2S		BIT(1)
> > > > +
> > > >  #define PCIE_IATU_NUM		6
> > > >  
> > > > +struct ls_pcie_drvdata {
> > > > +	const u32 pf_off;
> > > > +	bool pm_support;
> > > > +};
> > > > +
> > > >  struct ls_pcie {
> > > >  	struct dw_pcie *pci;
> > > > +	const struct ls_pcie_drvdata *drvdata;
> > > > +	void __iomem *pf_base;
> > > > +	bool big_endian;
> > > >  };
> > > >  
> > > > +#define ls_pcie_pf_readl_addr(addr)	ls_pcie_pf_readl(pcie, addr)
> > > >  #define to_ls_pcie(x)	dev_get_drvdata((x)->dev)
> > > >  
> > > >  static bool ls_pcie_is_bridge(struct ls_pcie *pcie)
> > > > @@ -73,6 +90,60 @@ static void ls_pcie_fix_error_response(struct ls_pcie *pcie)
> > > >  	iowrite32(PCIE_ABSERR_SETTING, pci->dbi_base + PCIE_ABSERR);
> > > >  }
> > > >  
> > > > +static u32 ls_pcie_pf_readl(struct ls_pcie *pcie, u32 off)
> > > > +{
> > > > +	if (pcie->big_endian)
> > > > +		return ioread32be(pcie->pf_base + off);
> > > > +
> > > > +	return ioread32(pcie->pf_base + off);
> > > > +}
> > > > +
> > > > +static void ls_pcie_pf_writel(struct ls_pcie *pcie, u32 off, u32 val)
> > > > +{
> > > > +	if (pcie->big_endian)
> > > > +		iowrite32be(val, pcie->pf_base + off);
> > > > +	else
> > > > +		iowrite32(val, pcie->pf_base + off);
> > > > +}
> > > > +
> > > > +static void ls_pcie_send_turnoff_msg(struct dw_pcie_rp *pp)
> > > > +{
> > > > +	struct dw_pcie *pci = to_dw_pcie_from_pp(pp);
> > > > +	struct ls_pcie *pcie = to_ls_pcie(pci);
> > > > +	u32 val;
> > > > +	int ret;
> > > > +
> > > > +	val = ls_pcie_pf_readl(pcie, LS_PCIE_PF_MCR);
> > > > +	val |= PF_MCR_PTOMR;
> > > > +	ls_pcie_pf_writel(pcie, LS_PCIE_PF_MCR, val);
> > > > +
> > > > +	ret = readx_poll_timeout(ls_pcie_pf_readl_addr, LS_PCIE_PF_MCR,
> > > > +				 val, !(val & PF_MCR_PTOMR),
> > > > +				 PCIE_PME_TO_L2_TIMEOUT_US/10,
> > > > +				 PCIE_PME_TO_L2_TIMEOUT_US);
> > > > +	if (ret)
> > > > +		dev_err(pcie->pci->dev, "PME_Turn_off timeout\n");
> > > > +}
> > > > +
> > > > +static void ls_pcie_exit_from_l2(struct dw_pcie_rp *pp)
> > > > +{
> > > > +	struct dw_pcie *pci = to_dw_pcie_from_pp(pp);
> > > > +	struct ls_pcie *pcie = to_ls_pcie(pci);
> > > > +	u32 val;
> > > > +	int ret;
> > > > +
> > > > +	val = ls_pcie_pf_readl(pcie, LS_PCIE_PF_MCR);
> > > > +	val |= PF_MCR_EXL2S;
> > > > +	ls_pcie_pf_writel(pcie, LS_PCIE_PF_MCR, val);
> > > 
> > > What is this write transaction generating in HW ?
> > 
> > I don't think send anything to to pci bus because it was called before
> > host init.
> > 
> > The spec of ls1028 is not clear enough.
> > 
> > `EXL2S: exit l2 state command. when set to 1, an L2 exit command is
> > generated. The bit is self clearing. Once the bit is set. SW needs to wait
> > for the bit to selfclear before sending a new command'
> > 
> > > 
> > > Why is it needed ? Shouldn't L2 exit happen automatically
> > > in HW ?
> > 
> > I tried remove this, PCI can't resume. I think this is specific for ls1028
> > chip to clear internal logic.
> 
> Well, if you don't even know what this does how can you write a sane
> device driver ?
> 
> Can you ask designers a more detailed description please ?
> 

I often encounter hw quirks like this one and the hw designers will just say
that "set bit X to make Y happpen". IMO a comment saying that the driver need to
set PF_MCR_EXL2S bit in LS_PCIE_PF_MCR register for the link to exit L2 state is
good enough.

> > > > +
> > > > +	ret = readx_poll_timeout(ls_pcie_pf_readl_addr, LS_PCIE_PF_MCR,
> > > > +				 val, !(val & PF_MCR_EXL2S),
> > > > +				 PCIE_PME_TO_L2_TIMEOUT_US/10,
> > > > +			PCIE_PME_TO_L2_TIMEOUT_US);
> > > 
> > > And why is the timeout value the same used for the PME_turn_off message ?
> > 
> > I think No spec define it, just reused it. use PCIE_PME_TO_L2_TIMEOUT_US
> > may cause confuse. What's do you prefered? Just use number,such as 10ms.
> 
> This delay value is misleading, it is not good to reuse a value for
> a delay that is most certainly controller specific.
> 
> From this discussion I would say that having pme_turn_off() and
> exit_from_l2() hooks is generalizing something we don't know yet
> it is needed for all DWC based controllers.
> 
> It is probably worth keeping the layerscape specific changes in
> the layerscape driver and from there call the "generic" DWC
> suspend/resume functions:
> 
> dw_pcie_suspend_noirq()
> dw_pcie_resume_noirq()
> 
> rather than adding hooks that we barely know what they are needed for.
> 
> Mani, what do you think ?
> 

PME_Turn_off procedure may vary between controllers and is really required
from core DWC perspective. So I'd prefer to keep the pme_turn_off() callback
and leave exit_from_l2() since later seems to be only required for layerscape.

- Mani

-- 
மணிவண்ணன் சதாசிவம்