On Tue, Nov 16, 2021 at 6:06 PM Emil Renner Berthing [off-list ref] wrote:

On Tue, 16 Nov 2021 at 17:01, Andy Shevchenko [off-list ref] wrote:

quoted

On Tue, Nov 16, 2021 at 5:06 PM Emil Renner Berthing [off-list ref] wrote:

...

quoted

quoted

+#if BITS_PER_LONG == 64
+#define jh7100_reset_read      readq
+#define jh7100_reset_write     writeq
+#else
+#define jh7100_reset_read      readl
+#define jh7100_reset_write     writel
+#endif

No need, just use always readq() / writeq() and include io-64-nonatomic-lo-hi

Hmm.. how would that work on a 32bit architecture where sizeof(long)
== 4? The whole point of this is to read chunks in sizes matching the
words in the bitmap later on.

Why not? You just put some amendments, see below an example.

...

quoted

quoted

+       void __iomem *reg_assert = data->base + JH7100_RESET_ASSERT0 + offset * sizeof(long);
+       void __iomem *reg_status = data->base + JH7100_RESET_STATUS0 + offset * sizeof(long);
+       unsigned long value;

      void __iomem *reg_assert = data->base + JH7100_RESET_ASSERT0 + off64;
      void __iomem *reg_status = data->base + JH7100_RESET_STATUS0 + off64;
      u64 value;

quoted

quoted

+       value = jh7100_reset_read(reg_assert);
+       if (assert)
+               value |= mask;
+       else
+               value &= ~mask;
+       jh7100_reset_write(value, reg_assert);

value = readq(...);
...update...
writeq(value, ...);

quoted

quoted

+       /* if the associated clock is gated, deasserting might otherwise hang forever */
+       ret = readx_poll_timeout_atomic(jh7100_reset_read, reg_status, value,
+                                       (value & mask) == done, 0, 1000);

readq_poll_timeout_atomic()

Maybe at the end of the day simple use of u64 will be better than bitmap.
Up to you, you have got the idea. Either way you may put my tag.

-- 
With Best Regards,
Andy Shevchenko