Re: [PATCH v1 1/2] platform/x86: intel_pmc_ipc: Use MFD framework to create dependent devices

[sathya]

Hi Andy,

Thanks for the review.

On 01/26/2018 08:17 AM, Andy Shevchenko wrote:
> On Fri, Jan 26, 2018 at 12:53 AM,
> <sathyanarayanan.kuppuswamy@xxxxxxxxxxxxxxx> wrote:
> > From: Kuppuswamy Sathyanarayanan <sathyanarayanan.kuppuswamy@xxxxxxxxxxxxxxx>
> >
> > Currently, we have lot of repetitive code in dependent device resource
> > allocation and device creation handling code. This logic can be improved if
> > we use MFD framework for dependent device creation. This patch adds this
> > support.
> Thanks for an update. My comments below.
>
> > Signed-off-by: Kuppuswamy Sathyanarayanan <sathyanarayanan.kuppuswamy@xxxxxxxxxxxxxxx>
> > Signed-off-by: Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx>
> First of all, I barely remember what I did to this patch.
Sorry, I know I took a long break. But its over now. I will be active in 
coming months.
> In any case
> this one is redundant since it will have mine when I push it to our
> repo.
>
> > @@ -508,7 +492,7 @@ static int ipc_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
> >          ret = devm_request_irq(&pdev->dev, pdev->irq, ioc, 0, "intel_pmc_ipc",
> >                                  pmc);
> >          if (ret) {
> > -               dev_err(&pdev->dev, "Failed to request irq\n");
> > +               dev_err(&pdev->dev, "Failed to request IRQ\n");
Will split these renames into a separate patch.
> >                  return ret;
> >          }
> Split this kind of changes in a separate patch.
>
> > +static int ipc_create_punit_device(struct platform_device *pdev)
> >   {
> > +       static struct resource punit_res[PLAT_RESOURCE_MEM_MAX_INDEX];
> > +       static struct mfd_cell punit_cell;
> > +       struct resource *res;
> > +       int ret, mindex, pindex = 0;
> > +
> > +       for (mindex = 0; mindex <= PLAT_RESOURCE_MEM_MAX_INDEX; mindex++) {
> '<=' ??? (Why = is here?)
Good catch. It should be only <. I will fix it in next release.
> > +               res = platform_get_resource(pdev, IORESOURCE_MEM, mindex);
> > +
> > +               switch (mindex) {
> > +               /* Get PUNIT resources */
> > +               case PLAT_RESOURCE_BIOS_DATA_INDEX:
> > +               case PLAT_RESOURCE_BIOS_IFACE_INDEX:
> > +                       /* BIOS resources are required, so return error if not
> > +                        * available
> > +                        */
> It's not the network subsystem, please, do a proper style for
> multi-line comments.
Will fix it in next release.
> > +                       if (!res) {
> Would the following work for you?
>
> if (res)
>   break;
>       dev_err(&pdev->dev, "Failed to get PUNIT MEM resource %d\n", pindex);
>       return -ENXIO;
> case ...:
> ...
> if (res)
>   break;
> default:
>   continue;
>
> memcpy(...);
> ...
If you move memcpy outside the switch statement, then it will be called 
for cases (non punit cases) like PLAT_RESOURCE_TELEM_SSRAM_INDEX or 
PLAT_RESOURCE_ACPI_IO_INDEX which is logically incorrect.
> > +                               dev_err(&pdev->dev,
> > +                                       "Failed to get PUNIT MEM resource %d\n",
> > +                                       pindex);
> > +                               return -ENXIO;
> > +                       }
> > +               case PLAT_RESOURCE_ISP_DATA_INDEX:
> > +               case PLAT_RESOURCE_ISP_IFACE_INDEX:
> > +               case PLAT_RESOURCE_GTD_DATA_INDEX:
> > +               case PLAT_RESOURCE_GTD_IFACE_INDEX:
> > +                       /* if valid resource found, copy the resource to PUNIT
> > +                        * resource
> > +                        */
> > +                       if (res)
> > +                               memcpy(&punit_res[pindex], res, sizeof(*res));
> > +                       punit_res[pindex].flags = IORESOURCE_MEM;
> > +                       pindex++;
> > +                       break;
> >                  };
> > +       }
> > +       ret = devm_mfd_add_devices(&pdev->dev, PLATFORM_DEVID_NONE, &punit_cell,
> > +                                  1, NULL, 0, NULL);
> >
> > +       dev_dbg(&pdev->dev, "Successfully created PUNIT device\n");
> Wrong. If ret is not 0 the message is misleading.
> Just remove it.
>
> Same for the rest cases.
I will remove it.
> > +       return ret;
> >   }
> > +static int ipc_create_wdt_device(struct platform_device *pdev)
> >   {
> > +       static struct resource wdt_ipc_res[2];
> > +       static struct mfd_cell wdt_cell;
> >          struct resource *res;
> > +       int ret;
> >
> > +       /* If we have ACPI based watchdog use that instead, othewise create
> > +        * a MFD cell for iTCO watchdog
> > +        */
> Style.
Got it. I will be fixed in next version.
> > +       if (acpi_has_watchdog())
> > +               return 0;
> > +       ret = devm_mfd_add_devices(&pdev->dev, PLATFORM_DEVID_NONE, &wdt_cell,
> > +                                  1, NULL, 0, NULL);
> > +
> > +       dev_dbg(&pdev->dev, "Successfully created watchdog device\n");
> > +
> > +       return ret;
> What Heikki meant is to fill cells by those helper functions and call
> mfd_add_devices() only once.
Ok. It will be fixed in next version.

I could not find the actual BUG reported by Heikki. So I did not 
understand the reason behind his proposal.
> See lpc_ich.c as an example.
>
> >   }
> > +static int ipc_create_pmc_devices(struct platform_device *pdev)
> >   {
> >          int ret;
> >
> > +       ret = ipc_create_punit_device(pdev);
> > +       if (ret < 0)
> > +               return ret;
> Is it fatal? (Hint: it's quite likely not)
Logically not. But this logic exist in originally driver. I did not want 
to change the behavior without knowing the full details. Let me know 
your opinion.
> > +       ret = ipc_create_wdt_device(pdev);
> > +       if (ret < 0)
> > +               return ret;
> Is it fatal?
Same as above.
> > +       ret = ipc_create_telemetry_device(pdev);
> > +       if (ret < 0)
> > +               return ret;
> Is it fatal?
Same as above.
> > +       return 0;
> >   }
> > +       ret = devm_request_irq(&pdev->dev, ipcdev.irq, ioc, IRQF_NO_SUSPEND,
> > +                              "intel_pmc_ipc", &ipcdev);
> > +       if (ret) {
> > +               dev_err(&pdev->dev, "Failed to request IRQ\n");
> > +               return ret;
> >          }
> >
> >          ret = sysfs_create_group(&pdev->dev.kobj, &intel_ipc_group);
> >          if (ret) {
> >                  dev_err(&pdev->dev, "Failed to create sysfs group %d\n",
> >                          ret);
> > -               goto err_sys;
> > +               devm_free_irq(&pdev->dev, ipcdev.irq, &ipcdev);
> Why do you need this one?
This was added by you in one of the previous submissions.

I think you added it because we have a separate device remove function 
in this driver, and not explicitly freeing IRQ could mess up the 
resource cleanup order.
> > +               return ret;
> >          }
> >
> >          ipcdev.has_gcr_regs = true;
> >
> >          return 0;
> >   }
> And to the main question, what this is doing in PDx86 now? There
> should be a patch to move it under drivers/mfd.
Drivers like drivers/platform/chrome/cros_ec_dev.c already use MFD calls 
outside MFD framework. I am not sure what is the norm.

If you agree with the move, I will submit a patch for it.
> In _any case_ I need an Ack from Lee.