drm/nouveau: exec: report max pushs through getparam

Report the maximum number of IBs that can be pushed with a single
DRM_IOCTL_NOUVEAU_EXEC through DRM_IOCTL_NOUVEAU_GETPARAM.

While the maximum number of IBs per ring might vary between chipsets,
the kernel will make sure that userspace can only push a fraction of the
maximum number of IBs per ring per job, such that we avoid a situation
where there's only a single job occupying the ring, which could
potentially lead to the ring run dry.

Using DRM_IOCTL_NOUVEAU_GETPARAM to report the maximum number of IBs
that can be pushed with a single DRM_IOCTL_NOUVEAU_EXEC implies that
all channels of a given device have the same ring size.

Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Acked-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Signed-off-by: Danilo Krummrich <dakr@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231002135008.10651-3-dakr@redhat.com

drivers/gpu/drm/nouveau/nouveau_abi16.c

@@ -31,6 +31,7 @@
 
 #include "nouveau_drv.h"
 #include "nouveau_dma.h"
+#include "nouveau_exec.h"
 #include "nouveau_gem.h"
 #include "nouveau_chan.h"
 #include "nouveau_abi16.h"
@@ -183,6 +184,20 @@ nouveau_abi16_fini(struct nouveau_abi16 *abi16)
 	cli->abi16 = NULL;
 }
 
+static inline int
+getparam_dma_ib_max(struct nvif_device *device)
+{
+	const struct nvif_mclass dmas[] = {
+		{ NV03_CHANNEL_DMA, 0 },
+		{ NV10_CHANNEL_DMA, 0 },
+		{ NV17_CHANNEL_DMA, 0 },
+		{ NV40_CHANNEL_DMA, 0 },
+		{}
+	};
+
+	return nvif_mclass(&device->object, dmas) < 0 ? NV50_DMA_IB_MAX : 0;
+}
+
 int
 nouveau_abi16_ioctl_getparam(ABI16_IOCTL_ARGS)
 {
@@ -247,6 +262,12 @@ nouveau_abi16_ioctl_getparam(ABI16_IOCTL_ARGS)
 	case NOUVEAU_GETPARAM_GRAPH_UNITS:
 		getparam->value = nvkm_gr_units(gr);
 		break;
+	case NOUVEAU_GETPARAM_EXEC_PUSH_MAX: {
+		int ib_max = getparam_dma_ib_max(device);
+
+		getparam->value = nouveau_exec_push_max_from_ib_max(ib_max);
+		break;
+	}
 	default:
 		NV_PRINTK(dbg, cli, "unknown parameter %lld\n", getparam->param);
 		return -EINVAL;

drivers/gpu/drm/nouveau/nouveau_chan.c

@@ -454,7 +454,7 @@ nouveau_channel_init(struct nouveau_channel *chan, u32 vram, u32 gart)
 		chan->user_get = 0x44;
 		chan->user_get_hi = 0x60;
 		chan->dma.ib_base =  0x10000 / 4;
-		chan->dma.ib_max  = (0x02000 / 8) - 1;
+		chan->dma.ib_max  = NV50_DMA_IB_MAX;
 		chan->dma.ib_put  = 0;
 		chan->dma.ib_free = chan->dma.ib_max - chan->dma.ib_put;
 		chan->dma.max = chan->dma.ib_base;

drivers/gpu/drm/nouveau/nouveau_dma.h

@@ -49,6 +49,9 @@ void nv50_dma_push(struct nouveau_channel *, u64 addr, u32 length,
 /* Maximum push buffer size. */
 #define NV50_DMA_PUSH_MAX_LENGTH 0x7fffff
 
+/* Maximum IBs per ring. */
+#define NV50_DMA_IB_MAX ((0x02000 / 8) - 1)
+
 /* Object handles - for stuff that's doesn't use handle == oclass. */
 enum {
 	NvDmaFB		= 0x80000002,

drivers/gpu/drm/nouveau/nouveau_exec.c

@@ -379,7 +379,7 @@ nouveau_exec_ioctl_exec(struct drm_device *dev,
 	struct nouveau_channel *chan = NULL;
 	struct nouveau_exec_job_args args = {};
 	struct drm_nouveau_exec *req = data;
-	int ret = 0;
+	int push_max, ret = 0;
 
 	if (unlikely(!abi16))
 		return -ENOMEM;
@@ -404,9 +404,10 @@ nouveau_exec_ioctl_exec(struct drm_device *dev,
 	if (!chan->dma.ib_max)
 		return nouveau_abi16_put(abi16, -ENOSYS);
 
-	if (unlikely(req->push_count > NOUVEAU_GEM_MAX_PUSH)) {
+	push_max = nouveau_exec_push_max_from_ib_max(chan->dma.ib_max);
+	if (unlikely(req->push_count > push_max)) {
 		NV_PRINTK(err, cli, "pushbuf push count exceeds limit: %d max %d\n",
-			 req->push_count, NOUVEAU_GEM_MAX_PUSH);
+			  req->push_count, push_max);
 		return nouveau_abi16_put(abi16, -EINVAL);
 	}
 

drivers/gpu/drm/nouveau/nouveau_exec.h

@@ -51,4 +51,14 @@ int nouveau_exec_job_init(struct nouveau_exec_job **job,
 int nouveau_exec_ioctl_exec(struct drm_device *dev, void *data,
 			    struct drm_file *file_priv);
 
+static inline unsigned int
+nouveau_exec_push_max_from_ib_max(int ib_max)
+{
+	/* Limit the number of IBs per job to half the size of the ring in order
+	 * to avoid the ring running dry between submissions and preserve one
+	 * more slot for the job's HW fence.
+	 */
+	return ib_max > 1 ? ib_max / 2 - 1 : 0;
+}
+
 #endif

include/uapi/drm/nouveau_drm.h

@@ -44,6 +44,16 @@ extern "C" {
 #define NOUVEAU_GETPARAM_PTIMER_TIME     14
 #define NOUVEAU_GETPARAM_HAS_BO_USAGE    15
 #define NOUVEAU_GETPARAM_HAS_PAGEFLIP    16
+
+/**
+ * @NOUVEAU_GETPARAM_EXEC_PUSH_MAX
+ *
+ * Query the maximum amount of IBs that can be pushed through a single
+ * &drm_nouveau_exec structure and hence a single &DRM_IOCTL_NOUVEAU_EXEC
+ * ioctl().
+ */
+#define NOUVEAU_GETPARAM_EXEC_PUSH_MAX   17
+
 struct drm_nouveau_getparam {
 	__u64 param;
 	__u64 value;