thr3ads.net - Nouveau - [Nouveau] [PATCH v3 2/2] fb/nvaa: Enable non-isometric poller on NVAA/NVAC [Dec 2014]

If this information is useful, please help other people find it:
Share via:

Pierre Moreau

2014-Dec-10 07:53 UTC

[Nouveau] [PATCH RESEND 1/2] Allow noaccel to be a pci address

Signed-off-by: Pierre Moreau <pierre.morrow at free.fr>
---
 drm/nouveau_drm.c | 16 +++++++++++-----
 1 file changed, 11 insertions(+), 5 deletions(-)

diff --git a/drm/nouveau_drm.c b/drm/nouveau_drm.c
index afb93bb..ffa1e4f 100644
--- a/drm/nouveau_drm.c
+++ b/drm/nouveau_drm.c
@@ -61,9 +61,10 @@ MODULE_PARM_DESC(debug, "debug string to pass to driver
core");
 static char *nouveau_debug;
 module_param_named(debug, nouveau_debug, charp, 0400);
 
-MODULE_PARM_DESC(noaccel, "disable kernel/abi16 acceleration");
-static int nouveau_noaccel = 0;
-module_param_named(noaccel, nouveau_noaccel, int, 0400);
+MODULE_PARM_DESC(noaccel, "disable kernel/abi16 acceleration for all
cards,"
+		          "or only for the card given its pci bus name");
+static char *nouveau_noaccel;
+module_param_named(noaccel, nouveau_noaccel, charp, 0400);
 
 MODULE_PARM_DESC(modeset, "enable driver (default: auto, "
 		          "0 = disabled, 1 = enabled, 2 = headless)");
@@ -149,8 +150,13 @@ nouveau_accel_init(struct nouveau_drm *drm)
 	u32 sclass[16];
 	int ret, i;
 
-	if (nouveau_noaccel)
+	if (nouveau_noaccel != NULL &&
+	    (!strcmp(nouveau_noaccel, "1") ||
+	     !strcmp(nouveau_noaccel, nvkm_device(device)->name)))
+	{
+		NV_WARN(drm, "Acceleration disabled for card on bus %s\n",
nvkm_device(device)->name);
 		return;
+	}
 
 	/* initialise synchronisation routines */
 	/*XXX: this is crap, but the fence/channel stuff is a little
@@ -1021,7 +1027,7 @@ static void nouveau_display_options(void)
 	DRM_DEBUG_DRIVER("... nofbaccel    : %d\n", nouveau_nofbaccel);
 	DRM_DEBUG_DRIVER("... config       : %s\n", nouveau_config);
 	DRM_DEBUG_DRIVER("... debug        : %s\n", nouveau_debug);
-	DRM_DEBUG_DRIVER("... noaccel      : %d\n", nouveau_noaccel);
+	DRM_DEBUG_DRIVER("... noaccel      : %s\n", nouveau_noaccel);
 	DRM_DEBUG_DRIVER("... modeset      : %d\n", nouveau_modeset);
 	DRM_DEBUG_DRIVER("... runpm        : %d\n", nouveau_runtime_pm);
 	DRM_DEBUG_DRIVER("... vram_pushbuf : %d\n", nouveau_vram_pushbuf);
-- 
2.1.3

Pierre Moreau

2014-Dec-10 07:53 UTC

head link

[Nouveau] [PATCH v3 2/2] fb/nvaa: Enable non-isometric poller on NVAA/NVAC

(This is a v3 of patch "drm/nouveau/fb/nv50: Add PFB writes")

This fix a GPU lockup on 9400M (NVAC) when using acceleration, see
https://bugs.freedesktop.org/show_bug.cgi?id=27501

v2:
- Move code to subdev/fb/nv50.c as suggested by Roy Spliet;
- Remove arbitrary writes to 100c18/100c24
- Replace write to 100c1c of arbitrary value by the address of a scratch page
  as proposed by Ilia Mirkin;
- Remove enabling of bits 16 and 0 as they don't yield in any changes.

v3:
- Move code to subdev/fb/nvaa.c as suggested by Ilia Mirkin.
The following changes were made thanks to information provided by Robert Morell
from NVidia:
- Allocate a dma page for use by the pollers;
- Re-enable pollers at bits 16 and 0;
- Set pollers address to a proper value.

Signed-off-by: Pierre Moreau <pierre.morrow at free.fr>
---
 drm/core/subdev/fb/nvaa.h |  1 +
 nvkm/subdev/fb/nv50.h     |  4 +++
 nvkm/subdev/fb/nvaa.c     | 74 ++++++++++++++++++++++++++++++++++++++++++++---
 nvkm/subdev/fb/nvaa.h     | 19 ++++++++++++
 4 files changed, 94 insertions(+), 4 deletions(-)
 create mode 120000 drm/core/subdev/fb/nvaa.h
 create mode 100644 nvkm/subdev/fb/nvaa.h

diff --git a/drm/core/subdev/fb/nvaa.h b/drm/core/subdev/fb/nvaa.h
new file mode 120000
index 0000000..b450e8c
--- /dev/null
+++ b/drm/core/subdev/fb/nvaa.h
@@ -0,0 +1 @@
+../../../../nvkm/subdev/fb/nvaa.h
\ No newline at end of file
diff --git a/nvkm/subdev/fb/nv50.h b/nvkm/subdev/fb/nv50.h
index c5e5a88..0b20975 100644
--- a/nvkm/subdev/fb/nv50.h
+++ b/nvkm/subdev/fb/nv50.h
@@ -9,6 +9,10 @@ struct nv50_fb_priv {
 	dma_addr_t r100c08;
 };
 
+#define nv50_fb_create(p,e,c,d,o)                                             
\
+	nv50_fb_ctor((p), (e), (c), (d), sizeof(**o),                          \
+			(struct nouveau_object **)o)
+
 int  nv50_fb_ctor(struct nouveau_object *, struct nouveau_object *,
 		  struct nouveau_oclass *, void *, u32,
 		  struct nouveau_object **);
diff --git a/nvkm/subdev/fb/nvaa.c b/nvkm/subdev/fb/nvaa.c
index cba8e68..b70ab2f 100644
--- a/nvkm/subdev/fb/nvaa.c
+++ b/nvkm/subdev/fb/nvaa.c
@@ -22,15 +22,81 @@
  * Authors: Ben Skeggs
  */
 
-#include "nv50.h"
+#include "nvaa.h"
+
+int
+nvaa_fb_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nouveau_device *device = nv_device(parent);
+	struct nvaa_fb_priv *priv;
+	int ret;
+
+	ret = nv50_fb_create(parent, engine, oclass, data, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv = (struct nvaa_fb_priv *)(*pobject);
+
+	priv->r100c18_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
+	if (priv->r100c18_page) {
+		priv->r100c18 = dma_map_page(nv_device_base(device),
+					     priv->r100c18_page, 0, PAGE_SIZE,
+					     DMA_BIDIRECTIONAL);
+		if (dma_mapping_error(nv_device_base(device), priv->r100c18))
+			return -EFAULT;
+	} else {
+		nv_warn(priv, "failed 0x100c18 page alloc\n");
+	}
+	return 0;
+}
+
+void
+nvaa_fb_dtor(struct nouveau_object *object)
+{
+	struct nouveau_device *device = nv_device(object);
+	struct nvaa_fb_priv *priv = (void *)object;
+
+	if (priv->r100c18_page) {
+		dma_unmap_page(nv_device_base(device), priv->r100c18, PAGE_SIZE,
+			       DMA_BIDIRECTIONAL);
+		__free_page(priv->r100c18_page);
+	}
+
+	nv50_fb_dtor(object);
+}
+
+int
+nvaa_fb_init(struct nouveau_object *object)
+{
+	struct nvaa_fb_priv *priv = (void *)object;
+	int ret;
+
+	ret = nv50_fb_init(object);
+	if (ret)
+		return ret;
+
+	/* Enable NISO poller for various clients and set their associated
+	 * read address, only for MCP77/78 and MCP79/7A. (fd#25701)
+	 */
+	nv_wr32(priv, 0x100c18, priv->r100c18 >> 8);
+	nv_mask(priv, 0x100c14, 0x00000000, 0x00000001);
+	nv_wr32(priv, 0x100c1c, (priv->r100c18 >> 8) + 1);
+	nv_mask(priv, 0x100c14, 0x00000000, 0x00000002);
+	nv_wr32(priv, 0x100c24, (priv->r100c18 >> 8) + 2);
+	nv_mask(priv, 0x100c14, 0x00000000, 0x00010000);
+	return 0;
+}
 
 struct nouveau_oclass *
 nvaa_fb_oclass = &(struct nv50_fb_impl) {
 	.base.base.handle = NV_SUBDEV(FB, 0xaa),
 	.base.base.ofuncs = &(struct nouveau_ofuncs) {
-		.ctor = nv50_fb_ctor,
-		.dtor = nv50_fb_dtor,
-		.init = nv50_fb_init,
+		.ctor = nvaa_fb_ctor,
+		.dtor = nvaa_fb_dtor,
+		.init = nvaa_fb_init,
 		.fini = _nouveau_fb_fini,
 	},
 	.base.memtype = nv50_fb_memtype_valid,
diff --git a/nvkm/subdev/fb/nvaa.h b/nvkm/subdev/fb/nvaa.h
new file mode 100644
index 0000000..84e1eca
--- /dev/null
+++ b/nvkm/subdev/fb/nvaa.h
@@ -0,0 +1,19 @@
+#ifndef __NVKM_FB_NVAA_H__
+#define __NVKM_FB_NVAA_H__
+
+#include "nv50.h"
+
+struct nvaa_fb_priv {
+	struct nv50_fb_priv base;
+	struct page *r100c18_page;
+	dma_addr_t r100c18;
+};
+
+int  nvaa_fb_ctor(struct nouveau_object *, struct nouveau_object *,
+		  struct nouveau_oclass *, void *, u32,
+		  struct nouveau_object **);
+void nvaa_fb_dtor(struct nouveau_object *);
+int  nvaa_fb_init(struct nouveau_object *);
+
+
+#endif
-- 
2.1.3

Ben Skeggs

2014-Dec-11 03:42 UTC

head link

[Nouveau] [PATCH v3 2/2] fb/nvaa: Enable non-isometric poller on NVAA/NVAC

On Wed, Dec 10, 2014 at 5:53 PM, Pierre Moreau <pierre.morrow at free.fr>
wrote:> (This is a v3 of patch "drm/nouveau/fb/nv50: Add PFB writes")
>
> This fix a GPU lockup on 9400M (NVAC) when using acceleration, see
> https://bugs.freedesktop.org/show_bug.cgi?id=27501
>
> v2:
> - Move code to subdev/fb/nv50.c as suggested by Roy Spliet;
> - Remove arbitrary writes to 100c18/100c24
> - Replace write to 100c1c of arbitrary value by the address of a scratch
page
>   as proposed by Ilia Mirkin;
> - Remove enabling of bits 16 and 0 as they don't yield in any changes.
>
> v3:
> - Move code to subdev/fb/nvaa.c as suggested by Ilia Mirkin.
> The following changes were made thanks to information provided by Robert
Morell
> from NVidia:
> - Allocate a dma page for use by the pollers;
> - Re-enable pollers at bits 16 and 0;
> - Set pollers address to a proper value.Hey Pierre,

This patch is incorrect.  As Robert pointed out in an older thread,
the registers don't take the physical address of a page, but a
somewhat odd "negative offset from the end of carveout".

See this example (referring to Robert's last email):

220.926392    read32 #3 +0x00100e10 -> 0x00070000
220.926406    read32 #3 +0x00100e14 -> 0x00010000

carveout_base = 0x70000000
carveout_size = 0x10000000

-- snip --

223.300495    read32 #3 +0x00100c14 -> 0x00000000
223.300521    read32 #3 +0x00100c18 -> 0x00000000
223.300547   write32 #3 +0x00100c18 <- 0x000027ff

base = (0x70000000 + 0x10000000) - ((0x27ff + 1) << 5)
base = 0x7ffb0000 << in carveout

223.300573    read32 #3 +0x00100c14 -> 0x00000000
223.300599   write32 #3 +0x00100c14 <- 0x00000001
223.300625    read32 #3 +0x00100c1c -> 0x00000002
223.300651   write32 #3 +0x00100c1c <- 0x000027fe

base = (0x70000000 + 0x10000000) - ((0x27fe + 1) << 5)
base = 0x7ffb0020 << in carveout

223.300677    read32 #3 +0x00100c14 -> 0x00000001
223.300702   write32 #3 +0x00100c14 <- 0x00000003
223.300728    read32 #3 +0x00100c24 -> 0x00000004
223.300754   write32 #3 +0x00100c24 <- 0x000027fd

base = (0x70000000 + 0x10000000) - ((0x27ff + 1) << 5)
base = 0x7ffb0040 << in carveout

223.300780    read32 #3 +0x00100c14 -> 0x00000003
223.300806   write32 #3 +0x00100c14 <- 0x00010003

It's a bit tricky at that point in the nouveau init process to
allocate device memory, but, we already reserve 1MiB of memory at the
end of "VRAM" (carveout / stolen memory) so we can make use of that.

I've attached 3 patches.  The first two of them, I'll probably merge
as-is.

The third patch is an example of how this should probably be done, in
case you want to continue tackling this some more :)

Cheers,
Ben.
>
> Signed-off-by: Pierre Moreau <pierre.morrow at free.fr>
> ---
>  drm/core/subdev/fb/nvaa.h |  1 +
>  nvkm/subdev/fb/nv50.h     |  4 +++
>  nvkm/subdev/fb/nvaa.c     | 74
++++++++++++++++++++++++++++++++++++++++++++---
>  nvkm/subdev/fb/nvaa.h     | 19 ++++++++++++
>  4 files changed, 94 insertions(+), 4 deletions(-)
>  create mode 120000 drm/core/subdev/fb/nvaa.h
>  create mode 100644 nvkm/subdev/fb/nvaa.h
>
> diff --git a/drm/core/subdev/fb/nvaa.h b/drm/core/subdev/fb/nvaa.h
> new file mode 120000
> index 0000000..b450e8c
> --- /dev/null
> +++ b/drm/core/subdev/fb/nvaa.h
> @@ -0,0 +1 @@
> +../../../../nvkm/subdev/fb/nvaa.h
> \ No newline at end of file
> diff --git a/nvkm/subdev/fb/nv50.h b/nvkm/subdev/fb/nv50.h
> index c5e5a88..0b20975 100644
> --- a/nvkm/subdev/fb/nv50.h
> +++ b/nvkm/subdev/fb/nv50.h
> @@ -9,6 +9,10 @@ struct nv50_fb_priv {
>         dma_addr_t r100c08;
>  };
>
> +#define nv50_fb_create(p,e,c,d,o)                                         
\
> +       nv50_fb_ctor((p), (e), (c), (d), sizeof(**o),                      
\
> +                       (struct nouveau_object **)o)
> +
>  int  nv50_fb_ctor(struct nouveau_object *, struct nouveau_object *,
>                   struct nouveau_oclass *, void *, u32,
>                   struct nouveau_object **);
> diff --git a/nvkm/subdev/fb/nvaa.c b/nvkm/subdev/fb/nvaa.c
> index cba8e68..b70ab2f 100644
> --- a/nvkm/subdev/fb/nvaa.c
> +++ b/nvkm/subdev/fb/nvaa.c
> @@ -22,15 +22,81 @@
>   * Authors: Ben Skeggs
>   */
>
> -#include "nv50.h"
> +#include "nvaa.h"
> +
> +int
> +nvaa_fb_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
> +            struct nouveau_oclass *oclass, void *data, u32 size,
> +            struct nouveau_object **pobject)
> +{
> +       struct nouveau_device *device = nv_device(parent);
> +       struct nvaa_fb_priv *priv;
> +       int ret;
> +
> +       ret = nv50_fb_create(parent, engine, oclass, data, &priv);
> +       *pobject = nv_object(priv);
> +       if (ret)
> +               return ret;
> +
> +       priv = (struct nvaa_fb_priv *)(*pobject);
> +
> +       priv->r100c18_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
> +       if (priv->r100c18_page) {
> +               priv->r100c18 = dma_map_page(nv_device_base(device),
> +                                            priv->r100c18_page, 0,
PAGE_SIZE,
> +                                            DMA_BIDIRECTIONAL);
> +               if (dma_mapping_error(nv_device_base(device),
priv->r100c18))
> +                       return -EFAULT;
> +       } else {
> +               nv_warn(priv, "failed 0x100c18 page alloc\n");
> +       }
> +       return 0;
> +}
> +
> +void
> +nvaa_fb_dtor(struct nouveau_object *object)
> +{
> +       struct nouveau_device *device = nv_device(object);
> +       struct nvaa_fb_priv *priv = (void *)object;
> +
> +       if (priv->r100c18_page) {
> +               dma_unmap_page(nv_device_base(device), priv->r100c18,
PAGE_SIZE,
> +                              DMA_BIDIRECTIONAL);
> +               __free_page(priv->r100c18_page);
> +       }
> +
> +       nv50_fb_dtor(object);
> +}
> +
> +int
> +nvaa_fb_init(struct nouveau_object *object)
> +{
> +       struct nvaa_fb_priv *priv = (void *)object;
> +       int ret;
> +
> +       ret = nv50_fb_init(object);
> +       if (ret)
> +               return ret;
> +
> +       /* Enable NISO poller for various clients and set their associated
> +        * read address, only for MCP77/78 and MCP79/7A. (fd#25701)
> +        */
> +       nv_wr32(priv, 0x100c18, priv->r100c18 >> 8);
> +       nv_mask(priv, 0x100c14, 0x00000000, 0x00000001);
> +       nv_wr32(priv, 0x100c1c, (priv->r100c18 >> 8) + 1);
> +       nv_mask(priv, 0x100c14, 0x00000000, 0x00000002);
> +       nv_wr32(priv, 0x100c24, (priv->r100c18 >> 8) + 2);
> +       nv_mask(priv, 0x100c14, 0x00000000, 0x00010000);
> +       return 0;
> +}
>
>  struct nouveau_oclass *
>  nvaa_fb_oclass = &(struct nv50_fb_impl) {
>         .base.base.handle = NV_SUBDEV(FB, 0xaa),
>         .base.base.ofuncs = &(struct nouveau_ofuncs) {
> -               .ctor = nv50_fb_ctor,
> -               .dtor = nv50_fb_dtor,
> -               .init = nv50_fb_init,
> +               .ctor = nvaa_fb_ctor,
> +               .dtor = nvaa_fb_dtor,
> +               .init = nvaa_fb_init,
>                 .fini = _nouveau_fb_fini,
>         },
>         .base.memtype = nv50_fb_memtype_valid,
> diff --git a/nvkm/subdev/fb/nvaa.h b/nvkm/subdev/fb/nvaa.h
> new file mode 100644
> index 0000000..84e1eca
> --- /dev/null
> +++ b/nvkm/subdev/fb/nvaa.h
> @@ -0,0 +1,19 @@
> +#ifndef __NVKM_FB_NVAA_H__
> +#define __NVKM_FB_NVAA_H__
> +
> +#include "nv50.h"
> +
> +struct nvaa_fb_priv {
> +       struct nv50_fb_priv base;
> +       struct page *r100c18_page;
> +       dma_addr_t r100c18;
> +};
> +
> +int  nvaa_fb_ctor(struct nouveau_object *, struct nouveau_object *,
> +                 struct nouveau_oclass *, void *, u32,
> +                 struct nouveau_object **);
> +void nvaa_fb_dtor(struct nouveau_object *);
> +int  nvaa_fb_init(struct nouveau_object *);
> +
> +
> +#endif
> --
> 2.1.3
>
> _______________________________________________
> Nouveau mailing list
> Nouveau at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/nouveau-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-fb-ram-mcp77-subclass-nouveau_ram.patch
Type: text/x-patch
Size: 2106 bytes
Desc: not available
URL:
<http://lists.freedesktop.org/archives/nouveau/attachments/20141211/06b4b60d/attachment-0003.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0002-fb-ram-mcp77-use-carveout-reg-to-determine-size.patch
Type: text/x-patch
Size: 1261 bytes
Desc: not available
URL:
<http://lists.freedesktop.org/archives/nouveau/attachments/20141211/06b4b60d/attachment-0004.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0003-EXAMPLEfb-ram-mcp77-niso-poller-setup.patch
Type: text/x-patch
Size: 2463 bytes
Desc: not available
URL:
<http://lists.freedesktop.org/archives/nouveau/attachments/20141211/06b4b60d/attachment-0005.bin>

Possibly Parallel Threads

Search for more reasonably related threads

Nouveau - Dec 2014 - [PATCH v3 2/2] fb/nvaa: Enable non-isometric poller on NVAA/NVAC

[Nouveau] [PATCH RESEND 1/2] Allow noaccel to be a pci address

[Nouveau] [PATCH v3 2/2] fb/nvaa: Enable non-isometric poller on NVAA/NVAC

[Nouveau] [PATCH v3 2/2] fb/nvaa: Enable non-isometric poller on NVAA/NVAC

Possibly Parallel Threads