On 10/13/2015 08:20 PM, Arnd Bergmann wrote:

On Tuesday 13 October 2015 17:47:02 zhangfei wrote:

quoted

On 10/12/2015 11:21 PM, Arnd Bergmann wrote:

quoted

On Monday 12 October 2015 23:20:19 John Garry wrote:

quoted

+int hisi_sas_ioremap(struct hisi_hba *hisi_hba)
+{
+       struct platform_device *pdev = hisi_hba->pdev;
+       struct device *dev = &pdev->dev;
+       struct resource *res;
+
+       res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       hisi_hba->regs = devm_ioremap(dev,
+                                     res->start,
+                                     resource_size(res));
+       if (!hisi_hba->regs)
+               return -ENOMEM;
+
+       res = platform_get_resource(pdev, IORESOURCE_MEM, 1);
+       hisi_hba->ctrl_regs = devm_ioremap(dev,
+                                          res->start,
+                                          resource_size(res));
+       if (!hisi_hba->ctrl_regs)
+               return -ENOMEM;
+
+       return 0;
+}

   static const struct of_device_id sas_of_match[] = {

Better use devm_ioremap_resource() here, which registers the resource so they
are checked for conflicts and listed in /proc/iomem.

Yes, hisi_hba->regs can use devm_ioremap_resource.

However ctrl_regs have to use devm_ioremap, since the address are
sharing among different nodes, unfortunately, and devm_ioremap_resource
will fail.

This sounds like it should be fixed in the DT binding then, to ensure
that the ranges don't overlap.

Mapping the same register region multiple times is generally considered
a bad idea because the drivers that map them often don't have global
locks that serialize the access, so it's better to have code in place
that ensures that they are distinct.

What is the purpose of the ctrl_regs region, and why is it shared
across multiple devices?

Are all users of these registers in the same driver?

We are considering using syscon for ctrl_regs.

Thanks Arnd.