On Tue, Mar 05, 2013 at 12:13:57PM +0100, Martin Peres wrote:
> On 04/03/2013 22:41, Konrad Rzeszutek Wilk wrote:
> >Pls CC me in case you would like me also to test them with the
> >mdelay patch.
> 
> Hi Konrad,
> 
> Marcin proposed me another explanation for the issue you are seeing
> and it made me look again at the code.
> 
> I don't have enough nv4x hw to test all the conditions but with the
> attached patches, you may get a saner
> behaviour than a computer that shut-downs whenever you turn it on
> (like a "most useless machine ever").
> The most important patch is the 8th one.
> 
> Please try applying them on top of your 3.9-rc1 kernel and send me
> back your kernel logs + sensors output.

I also added on top of this a debug patch to twidle with the values:

diff --git a/drivers/gpu/drm/nouveau/core/subdev/therm/nv40.c b/drivers/gpu/drm/nouveau/core/subdev/therm/nv40.c
index 92f3fca..a5a8abe 100644
--- a/drivers/gpu/drm/nouveau/core/subdev/therm/nv40.c
+++ b/drivers/gpu/drm/nouveau/core/subdev/therm/nv40.c
@@ -31,22 +31,28 @@ struct nv40_therm_priv {
 
 enum nv40_sensor_style { INVALID_STYLE = -1, OLD_STYLE = 0, NEW_STYLE = 1 };
 
+extern int hack_old_style;
+extern int hack_mdelay;
 static enum nv40_sensor_style
 nv40_is_older_style_sensor(struct nouveau_therm *therm)
 {
 	struct nouveau_device *device = nv_device(therm);
 
+	if (hack_old_style) {
+		if (device->chipset == 0x4c)
+			return OLD_STYLE;
+	}
 	switch (device->chipset) {
 	case 0x43:
 	case 0x44:
 	case 0x4a:
 	case 0x47:
-	case 0x4c:
 		return OLD_STYLE;
 
 	case 0x46:
 	case 0x49:
 	case 0x4b:
+	case 0x4c:
 	case 0x4e:
 	case 0x67:
 	case 0x68:
@@ -66,11 +72,17 @@ nv40_sensor_setup(struct nouveau_therm *therm)
 	if (style == NEW_STYLE) {
 		nv_mask(therm, 0x15b8, 0x80000000, 0);
 		nv_wr32(therm, 0x15b0, 0x80003fff);
-		mdelay(20); /* wait for the temperature to stabilize */
+		if (hack_mdelay)
+			mdelay(hack_mdelay);
+		else
+			mdelay(20); /* wait for the temperature to stabilize */
 		return nv_rd32(therm, 0x15b4) & 0x3fff;
 	} else if (style == OLD_STYLE) {
 		nv_wr32(therm, 0x15b0, 0xff);
-		mdelay(20); /* wait for the temperature to stabilize */
+		if (hack_mdelay)
+			mdelay(hack_mdelay);
+		else
+			mdelay(20); /* wait for the temperature to stabilize */
 		return nv_rd32(therm, 0x15b4) & 0xff;
 	} else
 		return -ENODEV;
diff --git a/drivers/gpu/drm/nouveau/nouveau_drm.c b/drivers/gpu/drm/nouveau/nouveau_drm.c
index d109936..d51bf21 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_drm.c
@@ -69,6 +69,12 @@ MODULE_PARM_DESC(modeset, "enable driver (default: auto, "
 int nouveau_modeset = -1;
 module_param_named(modeset, nouveau_modeset, int, 0400);
 
+int hack_mdelay = 0;
+module_param_named(mdelay, hack_mdelay, int, 0400);
+
+int hack_old_style = 1;
+module_param_named(old_style, hack_old_style, int, 0400);
+
 static struct drm_driver driver;
 
 static int

With that I am still getting the issues (even with an insance delay of 100 seconds).
Here is the serial log with various runs.