[PATCH] regmap: Ensure range selector registers are updated after cache sync

[Mark Brown <broonie@kernel.org>]

When we sync the register cache we do so with the cache bypassed in order
to avoid overhead from writing the synced values back into the cache. If
the regmap has ranges and the selector register for those ranges is in a
register which is cached this has the unfortunate side effect of meaning
that the physical and cached copies of the selector register can be out of
sync after a cache sync. The cache will have whatever the selector was when
the sync started and the hardware will have the selector for the register
that was synced last.

Fix this by rewriting all cached selector registers after every sync,
ensuring that the hardware and cache have the same content. This will
result in extra writes that wouldn't otherwise be needed but is simple
so hopefully robust. We don't read from the hardware since not all
devices have physical read support.

Given that nobody noticed this until now it is likely that we are rarely if
ever hitting this case.

Reported-by: Hector Martin <marcan@marcan.st>
Signed-off-by: Mark Brown <broonie@kernel.org>
Cc: stable@vger.kernel.org
---
 drivers/base/regmap/regcache.c | 30 ++++++++++++++++++++++++++++++
 1 file changed, 30 insertions(+)

diff --git a/drivers/base/regmap/regcache.c b/drivers/base/regmap/regcache.c
index c5d151e9c481..92592f944a3d 100644
--- a/drivers/base/regmap/regcache.c
+++ b/drivers/base/regmap/regcache.c
@@ -334,6 +334,11 @@ static int regcache_default_sync(struct regmap *map, unsigned int min,
 	return 0;
 }
 
+static int rbtree_all(const void *key, const struct rb_node *node)
+{
+	return 0;
+}
+
 /**
  * regcache_sync - Sync the register cache with the hardware.
  *
@@ -351,6 +356,7 @@ int regcache_sync(struct regmap *map)
 	unsigned int i;
 	const char *name;
 	bool bypass;
+	struct rb_node *node;
 
 	if (WARN_ON(map->cache_type == REGCACHE_NONE))
 		return -EINVAL;
@@ -392,6 +398,30 @@ int regcache_sync(struct regmap *map)
 	/* Restore the bypass state */
 	map->cache_bypass = bypass;
 	map->no_sync_defaults = false;
+
+	/*
+	 * If we did any paging with cache bypassed and a cached
+	 * paging register then the register and cache state might
+	 * have gone out of sync, force writes of all the paging
+	 * registers.
+	 */
+	rb_for_each(node, 0, &map->range_tree, rbtree_all) {
+		struct regmap_range_node *this =
+			rb_entry(node, struct regmap_range_node, node);
+
+		/* If there's nothing in the cache there's nothing to sync */
+		ret = regcache_read(map, this->selector_reg, &i);
+		if (ret != 0)
+			continue;
+
+		ret = _regmap_write(map, this->selector_reg, i);
+		if (ret != 0) {
+			dev_err(map->dev, "Failed to write %x = %x: %d\n",
+				this->selector_reg, i, ret);
+			break;
+		}
+	}
+
 	map->unlock(map->lock_arg);
 
 	regmap_async_complete(map);

---
base-commit: 611da07b89fdd53f140d7b33013f255bf0ed8f34
change-id: 20231026-regmap-fix-selector-sync-ad1514fd15df

Best regards,
-- 
Mark Brown <broonie@kernel.org>