Re: [PATCH] watchdog: mtk_wdt: Move mt6589-wdt fallback compatible to end of list

From: Guenter Roeck
Date: Tue Jul 26 2022 - 09:25:48 EST


On 7/26/22 01:37, AngeloGioacchino Del Regno wrote:
Il 26/07/22 08:20, Chen-Yu Tsai ha scritto:
The watchdog facility in the MediaTek SoCs are all the same, but the
hardware block also contains a reset controller, which differs in the
number of resets used between different SoCs. This difference is
supported with of_device_get_match_data() supplying the number of reset
lines for the newer compatible strings. The original mt6589-wdt only
covers the watchdog facility.

of_device_is_compatible(), and by extension of_device_get_match_data(),
works by going through the given list of compatible strings sequentially,
and checks if any of the device node's compatible strings match.

To avoid early matching on the "mediatek,mt6589-wdt" fallback compatible,
which only provides watchdog functionality and no reset controller, move
the fallback entry to the end of the list, so that other, more specific
compatible strings have a chance at getting matched.

Fixes: c254e103082b ("watchdog: mtk_wdt: mt8183: Add reset controller")
Fixes: adc318a34066 ("watchdog: mt8192: add wdt support")
Fixes: 8c6b5ea6ac68 ("watchdog: mediatek: mt8195: add wdt support")
Fixes: 4dbabc4d9e8c ("watchdog: mediatek: mt8186: add wdt support")
Fixes: 711a5b25bac9 ("watchdog: mtk_wdt: mt7986: Add toprgu reset controller support")
Signed-off-by: Chen-Yu Tsai <wenst@xxxxxxxxxxxx>

Uhm, I don't think that this is an issue?

Ordering precedence is given to the list that you specify in devicetree, that's why
"the second one" is a fallback, meaning: ("impossible" example below)

compatible = "mediatek,mt8195-wdt", "mediatek,mt8183-wdt", "mediatek,mt6589-wdt";

This gets walked as per the order in which you wrote the compatibles, so:
* Check match for mt8195-wdt, does not exist?
 * Check match for mt8183-wdt, exists!
 * Put everything into dev->of_node (having mediatek,mt8183-wdt only!)

__of_device_is_compatible() gets dev->of_node and compares that to all of the
possible matches.

struct device_node for this device hence does *not* contain any of the other
compatibles that we specified in devicetree, so it does *not* contain any of
"mediatek,mt8195-wdt", or "mediatek,mt6589-wdt", because we have previously
successfully matches 8183.

I don't think that I've misinterpreted this flow, but if I have, let's pull
in devicetree people and check with them?


I don't see the problem either. The fallback needs to be listed last in the
compatible property. If it isn't, having it last in struct of_device_id
won't help either.

Cheers,
Angelo

---

This change complements the removal of the fallback compatible from the
bindings and DTSI files [1].


If the fallback isn't listed in the dtsi files, it won't match at all,
no matter where it is located in struct of_device_id.

Guenter

[1] https://lore.kernel.org/linux-mediatek/20220721014845.19044-1-allen-kh.cheng@xxxxxxxxxxxx/

  drivers/watchdog/mtk_wdt.c | 3 ++-
  1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/watchdog/mtk_wdt.c b/drivers/watchdog/mtk_wdt.c
index e97787536792..3d5a239b93ba 100644
--- a/drivers/watchdog/mtk_wdt.c
+++ b/drivers/watchdog/mtk_wdt.c
@@ -425,12 +425,13 @@ static int mtk_wdt_resume(struct device *dev)
  static const struct of_device_id mtk_wdt_dt_ids[] = {
      { .compatible = "mediatek,mt2712-wdt", .data = &mt2712_data },
-    { .compatible = "mediatek,mt6589-wdt" },
      { .compatible = "mediatek,mt7986-wdt", .data = &mt7986_data },
      { .compatible = "mediatek,mt8183-wdt", .data = &mt8183_data },
      { .compatible = "mediatek,mt8186-wdt", .data = &mt8186_data },
      { .compatible = "mediatek,mt8192-wdt", .data = &mt8192_data },
      { .compatible = "mediatek,mt8195-wdt", .data = &mt8195_data },
+    /* Fallback compatible string must be last entry */
+    { .compatible = "mediatek,mt6589-wdt" },
      { /* sentinel */ }
  };
  MODULE_DEVICE_TABLE(of, mtk_wdt_dt_ids);