From: Ben Skeggs Date: Wed, 5 Mar 2014 04:46:48 +0000 (+1000) Subject: drm/gk104/gr: therm magic needed on some kepler boards X-Git-Tag: v3.15-rc1~51^2~39^2~16 X-Git-Url: https://openfabrics.org/gitweb/?a=commitdiff_plain;h=1f1ac3bf2a804911ad46076b10ce15364237a698;p=~emulex%2Finfiniband.git drm/gk104/gr: therm magic needed on some kepler boards Not needed everywhere, and potentially not safe to do depending on how the rest of PTHERM is configured... Signed-off-by: Ben Skeggs --- diff --git a/drivers/gpu/drm/nouveau/core/engine/graph/nve4.c b/drivers/gpu/drm/nouveau/core/engine/graph/nve4.c index 6e73b18920b..f7c01121717 100644 --- a/drivers/gpu/drm/nouveau/core/engine/graph/nve4.c +++ b/drivers/gpu/drm/nouveau/core/engine/graph/nve4.c @@ -189,6 +189,29 @@ nve4_graph_pack_mmio[] = { * PGRAPH engine/subdev functions ******************************************************************************/ +static int +nve4_graph_fini(struct nouveau_object *object, bool suspend) +{ + struct nvc0_graph_priv *priv = (void *)object; + + /*XXX: this is a nasty hack to power on gr on certain boards + * where it's disabled by therm, somehow. ideally it'd + * be nice to know when we should be doing this, and why, + * but, it's yet to be determined. for now we test for + * the particular mmio error that occurs in the situation, + * and then bash therm in the way nvidia do. + */ + nv_mask(priv, 0x000200, 0x08001000, 0x08001000); + nv_rd32(priv, 0x000200); + if (nv_rd32(priv, 0x400700) == 0xbadf1000) { + nv_mask(priv, 0x000200, 0x08001000, 0x00000000); + nv_rd32(priv, 0x000200); + nv_mask(priv, 0x020004, 0xc0000000, 0x40000000); + } + + return nouveau_graph_fini(&priv->base, suspend); +} + int nve4_graph_init(struct nouveau_object *object) { @@ -327,7 +350,7 @@ nve4_graph_oclass = &(struct nvc0_graph_oclass) { .ctor = nvc0_graph_ctor, .dtor = nvc0_graph_dtor, .init = nve4_graph_init, - .fini = _nouveau_graph_fini, + .fini = nve4_graph_fini, }, .cclass = &nve4_grctx_oclass, .sclass = nve4_graph_sclass,