Re: [PATCH v3] hisi_lpc: Use acpi_dev_for_each_child()

[John Garry <john.garry@huawei.com>]

On 01/07/2022 13:05, Andy Shevchenko wrote:
> On Fri, Jul 1, 2022 at 1:54 PM John Garry <john.garry@huawei.com> wrote:
>> On 01/07/2022 12:07, Andy Shevchenko wrote:
>>> On Fri, Jul 1, 2022 at 1:06 PM Andy Shevchenko
>>> <andy.shevchenko@gmail.com> wrote:
>>>> On Fri, Jul 1, 2022 at 1:04 PM John Garry <john.garry@huawei.com> wrote:
>>>>> On 30/06/2022 19:13, Rafael J. Wysocki wrote:
> 
> ...
> 
>>>>> However Yang Yingliang spotted a pre-existing bug in the ACPI probe and
>>>>> sent a fix today (coincidence?):
>>>>>
>>>>> https://lore.kernel.org/lkml/20220701094352.2104998-1-yangyingliang@huawei.com/T/#u
>>>>>
>>>>> And they conflict. This code has been this way for years, so I just
>>>>> suggest Yang Yingliang resends the fix on top off Rafael's change.
>>>>
>>>> Wondering if Yang can actually switch that to use
>>>> platform_device_register_full().
>>
>> Maybe that would work and simplify things. Let me check it.
>>
>> BTW, when we originally upstreamed this driver there was some ACPI
>> platform device registration code which you/we thought could be factored
>> out later. I can't remember it. I was looking through lore but couldn't
>> find it. I don't remember it being so important, though.
> 
> My suggestion is definitely not for the fix itself, but as a follow up.

FWIW, it works out quite neatly:

diff --git a/drivers/bus/hisi_lpc.c b/drivers/bus/hisi_lpc.c
index e0fee1f863e6..70198d5644c7 100644
--- a/drivers/bus/hisi_lpc.c
+++ b/drivers/bus/hisi_lpc.c
@@ -472,9 +472,7 @@ static int hisi_lpc_acpi_clear_enumerated(struct 
acpi_device *adev, void *not_us

  struct hisi_lpc_acpi_cell {
  	const char *hid;
-	const char *name;
-	void *pdata;
-	size_t pdata_size;
+	struct platform_device_info pdevinfo;
  };

  static void hisi_lpc_acpi_remove(struct device *hostdev)
@@ -505,28 +503,36 @@ static int hisi_lpc_acpi_add_child(struct 
acpi_device *child, void *data)
  		/* ipmi */
  		{
  			.hid = "IPI0001",
-			.name = "hisi-lpc-ipmi",
+			.pdevinfo = {
+				.name = "hisi-lpc-ipmi",
+				.num_res = num_res,
+				.res = res,
+			},
  		},
  		/* 8250-compatible uart */
  		{
  			.hid = "HISI1031",
-			.name = "serial8250",
-			.pdata = (struct plat_serial8250_port []) {
-				{
-					.iobase = res->start,
-					.uartclk = 1843200,
-					.iotype = UPIO_PORT,
-					.flags = UPF_BOOT_AUTOCONF,
+			.pdevinfo = {
+				.name = "serial8250",
+				.data = (struct plat_serial8250_port []) {
+					{
+						.iobase = res->start,
+						.uartclk = 1843200,
+						.iotype = UPIO_PORT,
+						.flags = UPF_BOOT_AUTOCONF,
+					},
+					{}
  				},
-				{}
+				.size_data = 2 *
+					sizeof(struct plat_serial8250_port),
+				.num_res = num_res,
+				.res = res,
  			},
-			.pdata_size = 2 *
-				sizeof(struct plat_serial8250_port),
  		},
  		{}
  	};

-	for (; cell && cell->name; cell++) {
+	for (; cell && cell->pdevinfo.name; cell++) {
  		if (!strcmp(cell->hid, hid)) {
  			found = true;
  			break;
@@ -540,25 +546,13 @@ static int hisi_lpc_acpi_add_child(struct 
acpi_device *child, void *data)
  		return 0;
  	}

-	pdev = platform_device_alloc(cell->name, PLATFORM_DEVID_AUTO);
+	pdev = platform_device_register_full(&cell->pdevinfo);
  	if (!pdev)
  		return -ENOMEM;

  	pdev->dev.parent = hostdev;
  	ACPI_COMPANION_SET(&pdev->dev, child);

-	ret = platform_device_add_resources(pdev, res, num_res);
-	if (ret)
-		return ret;
-
-	ret = platform_device_add_data(pdev, cell->pdata, cell->pdata_size);
-	if (ret)
-		return ret;
-
-	ret = platform_device_add(pdev);
-	if (ret)
-		return ret;
-
  	acpi_device_set_enumerated(child);

  	return 0;

> 
>>> And for the record, I think the Fixes even for very rare bug hits
>>> should go first.
>>
>> ok, I have to admit that I was going to feel awkward asking Rafael to
>> deal with this fix by having a v4 on top of it.
> 
> I don't think it's a problem as long as we have an immutable branch /
> tag with that patch. Another approach could be that Rafael can take it
> as a precursor for his series and route via ACPI tree, but let's hear
> what he thinks about this himself.
> 

ok, fine.

Thanks,
John