Date:   Wed, 14 Feb 2018 15:33:59 +0000
From:   John Garry <john.garry@...wei.com>
To:     Andy Shevchenko <andy.shevchenko@...il.com>
CC:     Mika Westerberg <mika.westerberg@...ux.intel.com>,
        "Rafael J. Wysocki" <rafael@...nel.org>,
        Lorenzo Pieralisi <lorenzo.pieralisi@....com>,
        "Rafael J. Wysocki" <rjw@...ysocki.net>,
        Hanjun Guo <hanjun.guo@...aro.org>,
        "Rob Herring" <robh+dt@...nel.org>,
        Bjorn Helgaas <bhelgaas@...gle.com>,
        "Arnd Bergmann" <arnd@...db.de>,
        Mark Rutland <mark.rutland@....com>,
        Olof Johansson <olof@...om.net>,
        Dann Frazier <dann.frazier@...onical.com>,
        Rob Herring <robh@...nel.org>, Joe Perches <joe@...ches.com>,
        Benjamin Herrenschmidt <benh@...nel.crashing.org>,
        <linux-pci@...r.kernel.org>,
        "Linux Kernel Mailing List" <linux-kernel@...r.kernel.org>,
        ACPI Devel Maling List <linux-acpi@...r.kernel.org>,
        Linuxarm <linuxarm@...wei.com>, Corey Minyard <minyard@....org>,
        devicetree <devicetree@...r.kernel.org>,
        Linux-Arch <linux-arch@...r.kernel.org>,
        Randy Dunlap <rdunlap@...radead.org>
Subject: Re: [PATCH v13 7/9] ACPI: Translate the I/O range of non-MMIO devices
 before scanning

On 14/02/2018 13:53, Andy Shevchenko wrote:
> On Tue, Feb 13, 2018 at 7:45 PM, John Garry <john.garry@...wei.com> wrote:
>> On some platforms (such as arm64-based hip06/hip07), access to legacy
>> ISA/LPC devices through access IO space is required, similar to x86
>> platforms. As the I/O for these devices are not memory mapped like
>> PCI/PCIE MMIO host bridges, they require special low-level device
>> operations through some host to generate IO accesses, i.e. a non-
>> transparent bridge.
>>
>> Through the logical PIO framework, hosts are able to register address
>> ranges in the logical PIO space for IO accesses. For hosts which require
>> a LLDD to generate the IO accesses, through the logical PIO framework
>> the host also registers accessors as a backend to generate the physical
>> bus transactions for IO space accesses (called indirect IO).
>>
>> When describing the indirect IO child device in APCI tables, the IO
>> resource is the host-specific address for the child (generally a
>> bus address).
>> An example is as follows:
>>   Device (LPC0) {
>>     Name (_HID, "HISI0191")  // HiSi LPC
>>     Name (_CRS, ResourceTemplate () {
>>       Memory32Fixed (ReadWrite, 0xa01b0000, 0x1000)
>>     })
>>   }
>>
>>   Device (LPC0.IPMI) {
>>     Name (_HID, "IPI0001")
>>     Name (LORS, ResourceTemplate() {
>>       QWordIO (
>>         ResourceConsumer,
>>         MinNotFixed,     // _MIF
>>         MaxNotFixed,     // _MAF
>>         PosDecode,
>>         EntireRange,
>>         0x0,             // _GRA
>>         0xe4,            // _MIN
>>         0x3fff,          // _MAX
>>         0x0,             // _TRA
>>         0x04,            // _LEN
>>         , ,
>>         BTIO
>>       )
>>     })
>>
>> Since the IO resource for the child is a host-specific address,
>> special translation are required to retrieve the logical PIO address
>> for that child.
>>
>> To overcome the problem of associating this logical PIO address
>> with the child device, a scan handler is added to scan the ACPI
>> namespace for known indirect IO hosts. This scan handler creates an
>> MFD per child with the translated logical PIO address as it's IO
>> resource, as a substitute for the normal platform device which ACPI
>> would create during device enumeration.
>

Hi Andy,

>> +       unsigned long sys_port;
>
>> +       sys_port = logic_pio_trans_hwaddr(&host->fwnode, res->start, len);
>> +       if (sys_port == -1UL)
>
> Wouldn't it be better to compare with ULONG_MAX?

Could do, being the same thing. Maybe people prefer -1UL as it saves 
having to figure out what ULONG_MAX is :)

>
>> +               return -EFAULT;
>
>
>> +/*
>
> Shouldn't be a kernel-doc?

Right, I'll make it /**

>
>> + * acpi_indirect_io_set_res - set the resources for a child device
>> + * (MFD) of an "indirect IO" host.
>
> In that case this would be one line w/o period at the end.
>
>> + * @child: the device node to be updated the I/O resource
>> + * @hostdev: the device node associated with the "indirect IO" host
>> + * @res: double pointer to be set to the address of translated resources
>> + * @num_res: pointer to variable to hold the number of translated resources
>> + *
>> + * Returns 0 when successful, and a negative value for failure.
>> + *
>> + * For a given "indirect IO" host, each child device will have associated
>> + * host-relevative address resource. This function will return the translated
>> + * logical PIO addresses for each child devices resources.
>> + */
>> +static int acpi_indirect_io_set_res(struct device *child,
>> +                                   struct device *hostdev,
>> +                                   const struct resource **res,
>> +                                   int *num_res)
>> +{
>> +       struct acpi_device *adev;
>> +       struct acpi_device *host;
>> +       struct resource_entry *rentry;
>> +       LIST_HEAD(resource_list);
>> +       struct resource *resources;
>> +       int count;
>> +       int i;
>> +       int ret = -EIO;
>> +
>> +       if (!child || !hostdev)
>> +               return -EINVAL;
>> +
>> +       host = to_acpi_device(hostdev);
>> +       adev = to_acpi_device(child);
>

***

>> +       count = acpi_dev_get_resources(adev, &resource_list, NULL, NULL);
>> +       if (count <= 0) {
>> +               dev_err(child, "failed to get resources\n");
>> +               return count ? count : -EIO;
>> +       }
>> +
>> +       resources = kcalloc(count, sizeof(*resources), GFP_KERNEL);
>> +       if (!resources) {
>> +               acpi_dev_free_resource_list(&resource_list);
>> +               return -ENOMEM;
>> +       }
>> +       count = 0;
>> +       list_for_each_entry(rentry, &resource_list, node)
>> +               resources[count++] = *rentry->res;
>> +
>> +       acpi_dev_free_resource_list(&resource_list);
>
> It has similarities with acpi_create_platform_device().
> I guess we can utilize existing code.
>

For sure, this particular segment is effectively same as part of 
acpi_create_platform_device():

struct platform_device *acpi_create_platform_device(struct acpi_device 
*adev,
                     struct property_entry *properties)
{
     struct platform_device *pdev = NULL;
     struct platform_device_info pdevinfo;
     struct resource_entry *rentry;
     struct list_head resource_list;
     struct resource *resources = NULL;
     int count;

     /* If the ACPI node already has a physical device attached, skip it. */
     if (adev->physical_node_count)
         return NULL;

     if (!acpi_match_device_ids(adev, forbidden_id_list))
         return ERR_PTR(-EINVAL);

***>
     INIT_LIST_HEAD(&resource_list);
     count = acpi_dev_get_resources(adev, &resource_list, NULL, NULL);
     if (count < 0) {
         return NULL;
     } else if (count > 0) {
         resources = kzalloc(count * sizeof(struct resource),
                     GFP_KERNEL);
         if (!resources) {
             dev_err(&adev->dev, "No memory for resources\n");
             acpi_dev_free_resource_list(&resource_list);
             return ERR_PTR(-ENOMEM);
         }
         count = 0;
         list_for_each_entry(rentry, &resource_list, node)
             acpi_platform_fill_resource(adev, rentry->res,
                             &resources[count++]);

         acpi_dev_free_resource_list(&resource_list);
     }
<****
     memset(&pdevinfo, 0, sizeof(pdevinfo));
     /*
      * If the ACPI node has a parent and that parent has a physical

So is your idea to refactor this common segment into a helper function?

>> +       /* translate the I/O resources */
>> +       for (i = 0; i < count; i++) {
>> +               if (!(resources[i].flags & IORESOURCE_IO))
>> +                       continue;
>
>> +               ret = acpi_indirect_io_xlat_res(adev, host, &resources[i]);
>> +               if (ret) {
>> +                       kfree(resources);
>> +                       dev_err(child, "translate IO range failed(%d)\n", ret);
>> +                       return ret;
>> +               }
>> +       }
>> +       *res = resources;
>> +       *num_res = count;
>> +
>> +       return ret;
>
> Perhaps,
>
>    ret = ...
>    if (ret)
>     break;
>   }
>
>   if (ret) {
>                        kfree(resources);
>                        dev_err(child, "translate IO range failed(%d)\n", ret);
>                        return ret;
>   }
>
>   *res = resources;
>   *num_res = count;
>   return 0;

seems fine

>
> ?
>
>> +}
>> +
>> +/*
>> + * acpi_indirect_io_setup - scan handler for "indirect IO" host.
>> + * @adev: "indirect IO" host ACPI device pointer
>> + * Returns 0 when successful, and a negative value for failure.
>> + *
>> + * Setup an "indirect IO" host by scanning all child devices, and
>> + * create a per-device MFD with logical PIO translated IO resources.
>> + */
>> +static int acpi_indirect_io_setup(struct acpi_device *adev)
>> +{
>> +       struct platform_device *pdev;
>> +       struct mfd_cell *mfd_cells;
>> +       struct logic_pio_hwaddr *range;
>> +       struct acpi_device *child;
>> +       struct acpi_indirect_io_mfd_cell *acpi_indirect_io_mfd_cells;
>> +       int size, ret, count = 0, cell_num = 0;
>> +
>> +       range = kzalloc(sizeof(*range), GFP_KERNEL);
>> +       if (!range)
>> +               return -ENOMEM;
>> +       range->fwnode = &adev->fwnode;
>> +       range->flags = PIO_INDIRECT;
>> +       range->size = PIO_INDIRECT_SIZE;
>> +
>> +       ret = logic_pio_register_range(range);
>> +       if (ret)
>> +               goto free_range;
>> +
>> +       list_for_each_entry(child, &adev->children, node)
>> +               cell_num++;
>> +
>> +       /* allocate the mfd cell and companion acpi info, one per child */
>> +       size = sizeof(*mfd_cells) + sizeof(*acpi_indirect_io_mfd_cells);
>> +       mfd_cells = kcalloc(cell_num, size, GFP_KERNEL);
>> +       if (!mfd_cells) {
>> +               ret = -ENOMEM;
>> +               goto free_range;
>> +       }
>> +
>> +       acpi_indirect_io_mfd_cells = (struct acpi_indirect_io_mfd_cell *)
>> +                                       &mfd_cells[cell_num];
>> +       /* Only consider the children of the host */
>> +       list_for_each_entry(child, &adev->children, node) {
>> +               struct mfd_cell *mfd_cell = &mfd_cells[count];
>> +               struct acpi_indirect_io_mfd_cell *acpi_indirect_io_mfd_cell =
>> +                                       &acpi_indirect_io_mfd_cells[count];
>> +               const struct mfd_cell_acpi_match *acpi_match =
>> +                                       &acpi_indirect_io_mfd_cell->acpi_match;
>
>> +               char *name = &acpi_indirect_io_mfd_cell[count].name[0];
>> +               char *pnpid = &acpi_indirect_io_mfd_cell[count].pnpid[0];
>
> Plain x is equivalent to &x[0].

Right, but I thought for arrays that we should use address of x rather 
than x itself, no?

>
>> +               struct mfd_cell_acpi_match match = {
>> +                       .pnpid = pnpid,
>> +               };
>> +
>> +               snprintf(name, ACPI_INDIRECT_IO_NAME_LEN, "indirect-io-%s",
>> +                        acpi_device_hid(child));
>> +               snprintf(pnpid, ACPI_INDIRECT_IO_NAME_LEN, "%s",
>> +                        acpi_device_hid(child))
>
>> +               memcpy((void *)acpi_match, (void *)&match, sizeof(*acpi_match));
>
> Casting to void * is pointless. In both cases.

I rechecked this. The casting to void * was there to mask another issue 
which I've now fixed.

>
>> +               mfd_cell->name = name;
>> +               mfd_cell->acpi_match = acpi_match;
>> +
>> +               ret = acpi_indirect_io_set_res(&child->dev, &adev->dev,
>> +                                              &mfd_cell->resources,
>> +                                              &mfd_cell->num_resources);
>> +               if (ret) {
>> +                       dev_err(&child->dev, "set resource failed (%d)\n", ret);
>> +                       goto free_mfd_resources;
>> +               }
>> +               count++;
>> +       }
>> +
>> +       pdev = acpi_create_platform_device(adev, NULL);
>> +       if (IS_ERR_OR_NULL(pdev)) {
>> +               dev_err(&adev->dev, "create platform device for host failed\n");
>
>> +               ret = PTR_ERR(pdev);
>
> So, NULL case will return 0. Is it expected?
>

Should error in that case also, so I'll change.

>> +               goto free_mfd_resources;
>> +       }
>> +       acpi_device_set_enumerated(adev);
>> +
>> +       ret = mfd_add_devices(&pdev->dev, PLATFORM_DEVID_NONE,
>> +                             mfd_cells, cell_num, NULL, 0, NULL);
>> +       if (ret) {
>> +               dev_err(&pdev->dev, "failed to add mfd cells (%d)\n", ret);
>> +               goto free_mfd_resources;
>> +       }
>> +
>> +       return ret;
>> +
>> +free_mfd_resources:
>> +       while (cell_num--)
>> +               kfree(mfd_cells[cell_num].resources);
>> +       kfree(mfd_cells);
>> +free_range:
>> +       kfree(range);
>> +
>> +       return ret;
>> +}
>
> One question, what a scope of use of this function? Is it ->probe() time?
> If it's so, can we use devm_* variants?

It is called from a scan handler, so prior to device probing.

>

Thanks,
John

Powered by blists - more mailing lists