Commit e41115d0 authored by Ben Skeggs's avatar Ben Skeggs

drm/nouveau: rework gpu-specific instmem interfaces

Reviewed-by: default avatarFrancisco Jerez <currojerez@riseup.net>
Signed-off-by: default avatarBen Skeggs <bskeggs@redhat.com>
parent dc1e5c0d
...@@ -146,15 +146,16 @@ enum nouveau_flags { ...@@ -146,15 +146,16 @@ enum nouveau_flags {
#define NVOBJ_FLAG_ZERO_ALLOC (1 << 1) #define NVOBJ_FLAG_ZERO_ALLOC (1 << 1)
#define NVOBJ_FLAG_ZERO_FREE (1 << 2) #define NVOBJ_FLAG_ZERO_FREE (1 << 2)
#define NVOBJ_CINST_GLOBAL 0xdeadbeef
struct nouveau_gpuobj { struct nouveau_gpuobj {
struct drm_device *dev; struct drm_device *dev;
struct kref refcount; struct kref refcount;
struct list_head list; struct list_head list;
struct drm_mm_node *im_pramin; void *node;
struct nouveau_bo *im_backing;
u32 *suspend; u32 *suspend;
int im_bound;
uint32_t flags; uint32_t flags;
...@@ -288,11 +289,11 @@ struct nouveau_instmem_engine { ...@@ -288,11 +289,11 @@ struct nouveau_instmem_engine {
int (*suspend)(struct drm_device *dev); int (*suspend)(struct drm_device *dev);
void (*resume)(struct drm_device *dev); void (*resume)(struct drm_device *dev);
int (*populate)(struct drm_device *, struct nouveau_gpuobj *, int (*get)(struct nouveau_gpuobj *, u32 size, u32 align);
u32 *size, u32 align); void (*put)(struct nouveau_gpuobj *);
void (*clear)(struct drm_device *, struct nouveau_gpuobj *); int (*map)(struct nouveau_gpuobj *);
int (*bind)(struct drm_device *, struct nouveau_gpuobj *); void (*unmap)(struct nouveau_gpuobj *);
int (*unbind)(struct drm_device *, struct nouveau_gpuobj *);
void (*flush)(struct drm_device *); void (*flush)(struct drm_device *);
}; };
...@@ -1182,11 +1183,10 @@ extern int nv04_instmem_init(struct drm_device *); ...@@ -1182,11 +1183,10 @@ extern int nv04_instmem_init(struct drm_device *);
extern void nv04_instmem_takedown(struct drm_device *); extern void nv04_instmem_takedown(struct drm_device *);
extern int nv04_instmem_suspend(struct drm_device *); extern int nv04_instmem_suspend(struct drm_device *);
extern void nv04_instmem_resume(struct drm_device *); extern void nv04_instmem_resume(struct drm_device *);
extern int nv04_instmem_populate(struct drm_device *, struct nouveau_gpuobj *, extern int nv04_instmem_get(struct nouveau_gpuobj *, u32 size, u32 align);
u32 *size, u32 align); extern void nv04_instmem_put(struct nouveau_gpuobj *);
extern void nv04_instmem_clear(struct drm_device *, struct nouveau_gpuobj *); extern int nv04_instmem_map(struct nouveau_gpuobj *);
extern int nv04_instmem_bind(struct drm_device *, struct nouveau_gpuobj *); extern void nv04_instmem_unmap(struct nouveau_gpuobj *);
extern int nv04_instmem_unbind(struct drm_device *, struct nouveau_gpuobj *);
extern void nv04_instmem_flush(struct drm_device *); extern void nv04_instmem_flush(struct drm_device *);
/* nv50_instmem.c */ /* nv50_instmem.c */
...@@ -1194,11 +1194,10 @@ extern int nv50_instmem_init(struct drm_device *); ...@@ -1194,11 +1194,10 @@ extern int nv50_instmem_init(struct drm_device *);
extern void nv50_instmem_takedown(struct drm_device *); extern void nv50_instmem_takedown(struct drm_device *);
extern int nv50_instmem_suspend(struct drm_device *); extern int nv50_instmem_suspend(struct drm_device *);
extern void nv50_instmem_resume(struct drm_device *); extern void nv50_instmem_resume(struct drm_device *);
extern int nv50_instmem_populate(struct drm_device *, struct nouveau_gpuobj *, extern int nv50_instmem_get(struct nouveau_gpuobj *, u32 size, u32 align);
u32 *size, u32 align); extern void nv50_instmem_put(struct nouveau_gpuobj *);
extern void nv50_instmem_clear(struct drm_device *, struct nouveau_gpuobj *); extern int nv50_instmem_map(struct nouveau_gpuobj *);
extern int nv50_instmem_bind(struct drm_device *, struct nouveau_gpuobj *); extern void nv50_instmem_unmap(struct nouveau_gpuobj *);
extern int nv50_instmem_unbind(struct drm_device *, struct nouveau_gpuobj *);
extern void nv50_instmem_flush(struct drm_device *); extern void nv50_instmem_flush(struct drm_device *);
extern void nv84_instmem_flush(struct drm_device *); extern void nv84_instmem_flush(struct drm_device *);
extern void nv50_vm_flush(struct drm_device *, int engine); extern void nv50_vm_flush(struct drm_device *, int engine);
...@@ -1208,11 +1207,10 @@ extern int nvc0_instmem_init(struct drm_device *); ...@@ -1208,11 +1207,10 @@ extern int nvc0_instmem_init(struct drm_device *);
extern void nvc0_instmem_takedown(struct drm_device *); extern void nvc0_instmem_takedown(struct drm_device *);
extern int nvc0_instmem_suspend(struct drm_device *); extern int nvc0_instmem_suspend(struct drm_device *);
extern void nvc0_instmem_resume(struct drm_device *); extern void nvc0_instmem_resume(struct drm_device *);
extern int nvc0_instmem_populate(struct drm_device *, struct nouveau_gpuobj *, extern int nvc0_instmem_get(struct nouveau_gpuobj *, u32 size, u32 align);
u32 *size, u32 align); extern void nvc0_instmem_put(struct nouveau_gpuobj *);
extern void nvc0_instmem_clear(struct drm_device *, struct nouveau_gpuobj *); extern int nvc0_instmem_map(struct nouveau_gpuobj *);
extern int nvc0_instmem_bind(struct drm_device *, struct nouveau_gpuobj *); extern void nvc0_instmem_unmap(struct nouveau_gpuobj *);
extern int nvc0_instmem_unbind(struct drm_device *, struct nouveau_gpuobj *);
extern void nvc0_instmem_flush(struct drm_device *); extern void nvc0_instmem_flush(struct drm_device *);
/* nv04_mc.c */ /* nv04_mc.c */
......
...@@ -168,17 +168,14 @@ nouveau_gpuobj_new(struct drm_device *dev, struct nouveau_channel *chan, ...@@ -168,17 +168,14 @@ nouveau_gpuobj_new(struct drm_device *dev, struct nouveau_channel *chan,
struct nouveau_gpuobj **gpuobj_ret) struct nouveau_gpuobj **gpuobj_ret)
{ {
struct drm_nouveau_private *dev_priv = dev->dev_private; struct drm_nouveau_private *dev_priv = dev->dev_private;
struct nouveau_engine *engine = &dev_priv->engine; struct nouveau_instmem_engine *instmem = &dev_priv->engine.instmem;
struct nouveau_gpuobj *gpuobj; struct nouveau_gpuobj *gpuobj;
struct drm_mm_node *ramin = NULL; struct drm_mm_node *ramin = NULL;
int ret; int ret, i;
NV_DEBUG(dev, "ch%d size=%u align=%d flags=0x%08x\n", NV_DEBUG(dev, "ch%d size=%u align=%d flags=0x%08x\n",
chan ? chan->id : -1, size, align, flags); chan ? chan->id : -1, size, align, flags);
if (!dev_priv || !gpuobj_ret || *gpuobj_ret != NULL)
return -EINVAL;
gpuobj = kzalloc(sizeof(*gpuobj), GFP_KERNEL); gpuobj = kzalloc(sizeof(*gpuobj), GFP_KERNEL);
if (!gpuobj) if (!gpuobj)
return -ENOMEM; return -ENOMEM;
...@@ -193,88 +190,45 @@ nouveau_gpuobj_new(struct drm_device *dev, struct nouveau_channel *chan, ...@@ -193,88 +190,45 @@ nouveau_gpuobj_new(struct drm_device *dev, struct nouveau_channel *chan,
spin_unlock(&dev_priv->ramin_lock); spin_unlock(&dev_priv->ramin_lock);
if (chan) { if (chan) {
NV_DEBUG(dev, "channel heap\n");
ramin = drm_mm_search_free(&chan->ramin_heap, size, align, 0); ramin = drm_mm_search_free(&chan->ramin_heap, size, align, 0);
if (ramin) if (ramin)
ramin = drm_mm_get_block(ramin, size, align); ramin = drm_mm_get_block(ramin, size, align);
if (!ramin) { if (!ramin) {
nouveau_gpuobj_ref(NULL, &gpuobj); nouveau_gpuobj_ref(NULL, &gpuobj);
return -ENOMEM; return -ENOMEM;
} }
} else {
NV_DEBUG(dev, "global heap\n");
/* allocate backing pages, sets vinst */
ret = engine->instmem.populate(dev, gpuobj, &size, align);
if (ret) {
nouveau_gpuobj_ref(NULL, &gpuobj);
return ret;
}
/* try and get aperture space */ gpuobj->pinst = chan->ramin->pinst;
do { if (gpuobj->pinst != ~0)
if (drm_mm_pre_get(&dev_priv->ramin_heap)) gpuobj->pinst += ramin->start;
return -ENOMEM;
spin_lock(&dev_priv->ramin_lock);
ramin = drm_mm_search_free(&dev_priv->ramin_heap, size,
align, 0);
if (ramin == NULL) {
spin_unlock(&dev_priv->ramin_lock);
nouveau_gpuobj_ref(NULL, &gpuobj);
return -ENOMEM;
}
ramin = drm_mm_get_block_atomic(ramin, size, align);
spin_unlock(&dev_priv->ramin_lock);
} while (ramin == NULL);
/* on nv50 it's ok to fail, we have a fallback path */ if (dev_priv->card_type < NV_50)
if (!ramin && dev_priv->card_type < NV_50) { gpuobj->cinst = gpuobj->pinst;
nouveau_gpuobj_ref(NULL, &gpuobj); else
return -ENOMEM; gpuobj->cinst = ramin->start;
}
}
/* if we got a chunk of the aperture, map pages into it */ gpuobj->vinst = ramin->start + chan->ramin->vinst;
gpuobj->im_pramin = ramin; gpuobj->node = ramin;
if (!chan && gpuobj->im_pramin && dev_priv->ramin_available) { } else {
ret = engine->instmem.bind(dev, gpuobj); ret = instmem->get(gpuobj, size, align);
if (ret) { if (ret) {
nouveau_gpuobj_ref(NULL, &gpuobj); nouveau_gpuobj_ref(NULL, &gpuobj);
return ret; return ret;
} }
}
/* calculate the various different addresses for the object */
if (chan) {
gpuobj->pinst = chan->ramin->pinst;
if (gpuobj->pinst != ~0)
gpuobj->pinst += gpuobj->im_pramin->start;
if (dev_priv->card_type < NV_50) { ret = -ENOSYS;
gpuobj->cinst = gpuobj->pinst; if (dev_priv->ramin_available)
} else { ret = instmem->map(gpuobj);
gpuobj->cinst = gpuobj->im_pramin->start; if (ret)
gpuobj->vinst = gpuobj->im_pramin->start +
chan->ramin->vinst;
}
} else {
if (gpuobj->im_pramin)
gpuobj->pinst = gpuobj->im_pramin->start;
else
gpuobj->pinst = ~0; gpuobj->pinst = ~0;
gpuobj->cinst = 0xdeadbeef;
gpuobj->cinst = NVOBJ_CINST_GLOBAL;
} }
if (gpuobj->flags & NVOBJ_FLAG_ZERO_ALLOC) { if (gpuobj->flags & NVOBJ_FLAG_ZERO_ALLOC) {
int i;
for (i = 0; i < gpuobj->size; i += 4) for (i = 0; i < gpuobj->size; i += 4)
nv_wo32(gpuobj, i, 0); nv_wo32(gpuobj, i, 0);
engine->instmem.flush(dev); instmem->flush(dev);
} }
...@@ -326,26 +280,34 @@ nouveau_gpuobj_del(struct kref *ref) ...@@ -326,26 +280,34 @@ nouveau_gpuobj_del(struct kref *ref)
container_of(ref, struct nouveau_gpuobj, refcount); container_of(ref, struct nouveau_gpuobj, refcount);
struct drm_device *dev = gpuobj->dev; struct drm_device *dev = gpuobj->dev;
struct drm_nouveau_private *dev_priv = dev->dev_private; struct drm_nouveau_private *dev_priv = dev->dev_private;
struct nouveau_engine *engine = &dev_priv->engine; struct nouveau_instmem_engine *instmem = &dev_priv->engine.instmem;
int i; int i;
NV_DEBUG(dev, "gpuobj %p\n", gpuobj); NV_DEBUG(dev, "gpuobj %p\n", gpuobj);
if (gpuobj->im_pramin && (gpuobj->flags & NVOBJ_FLAG_ZERO_FREE)) { if (gpuobj->node && (gpuobj->flags & NVOBJ_FLAG_ZERO_FREE)) {
for (i = 0; i < gpuobj->size; i += 4) for (i = 0; i < gpuobj->size; i += 4)
nv_wo32(gpuobj, i, 0); nv_wo32(gpuobj, i, 0);
engine->instmem.flush(dev); instmem->flush(dev);
} }
if (gpuobj->dtor) if (gpuobj->dtor)
gpuobj->dtor(dev, gpuobj); gpuobj->dtor(dev, gpuobj);
if (gpuobj->im_backing) if (gpuobj->cinst == NVOBJ_CINST_GLOBAL) {
engine->instmem.clear(dev, gpuobj); if (gpuobj->node) {
instmem->unmap(gpuobj);
instmem->put(gpuobj);
}
} else {
if (gpuobj->node) {
spin_lock(&dev_priv->ramin_lock);
drm_mm_put_block(gpuobj->node);
spin_unlock(&dev_priv->ramin_lock);
}
}
spin_lock(&dev_priv->ramin_lock); spin_lock(&dev_priv->ramin_lock);
if (gpuobj->im_pramin)
drm_mm_put_block(gpuobj->im_pramin);
list_del(&gpuobj->list); list_del(&gpuobj->list);
spin_unlock(&dev_priv->ramin_lock); spin_unlock(&dev_priv->ramin_lock);
...@@ -385,7 +347,7 @@ nouveau_gpuobj_new_fake(struct drm_device *dev, u32 pinst, u64 vinst, ...@@ -385,7 +347,7 @@ nouveau_gpuobj_new_fake(struct drm_device *dev, u32 pinst, u64 vinst,
kref_init(&gpuobj->refcount); kref_init(&gpuobj->refcount);
gpuobj->size = size; gpuobj->size = size;
gpuobj->pinst = pinst; gpuobj->pinst = pinst;
gpuobj->cinst = 0xdeadbeef; gpuobj->cinst = NVOBJ_CINST_GLOBAL;
gpuobj->vinst = vinst; gpuobj->vinst = vinst;
if (gpuobj->flags & NVOBJ_FLAG_ZERO_ALLOC) { if (gpuobj->flags & NVOBJ_FLAG_ZERO_ALLOC) {
...@@ -935,7 +897,7 @@ nouveau_gpuobj_suspend(struct drm_device *dev) ...@@ -935,7 +897,7 @@ nouveau_gpuobj_suspend(struct drm_device *dev)
int i; int i;
list_for_each_entry(gpuobj, &dev_priv->gpuobj_list, list) { list_for_each_entry(gpuobj, &dev_priv->gpuobj_list, list) {
if (gpuobj->cinst != 0xdeadbeef) if (gpuobj->cinst != NVOBJ_CINST_GLOBAL)
continue; continue;
gpuobj->suspend = vmalloc(gpuobj->size); gpuobj->suspend = vmalloc(gpuobj->size);
......
...@@ -53,10 +53,10 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev) ...@@ -53,10 +53,10 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
engine->instmem.takedown = nv04_instmem_takedown; engine->instmem.takedown = nv04_instmem_takedown;
engine->instmem.suspend = nv04_instmem_suspend; engine->instmem.suspend = nv04_instmem_suspend;
engine->instmem.resume = nv04_instmem_resume; engine->instmem.resume = nv04_instmem_resume;
engine->instmem.populate = nv04_instmem_populate; engine->instmem.get = nv04_instmem_get;
engine->instmem.clear = nv04_instmem_clear; engine->instmem.put = nv04_instmem_put;
engine->instmem.bind = nv04_instmem_bind; engine->instmem.map = nv04_instmem_map;
engine->instmem.unbind = nv04_instmem_unbind; engine->instmem.unmap = nv04_instmem_unmap;
engine->instmem.flush = nv04_instmem_flush; engine->instmem.flush = nv04_instmem_flush;
engine->mc.init = nv04_mc_init; engine->mc.init = nv04_mc_init;
engine->mc.takedown = nv04_mc_takedown; engine->mc.takedown = nv04_mc_takedown;
...@@ -106,10 +106,10 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev) ...@@ -106,10 +106,10 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
engine->instmem.takedown = nv04_instmem_takedown; engine->instmem.takedown = nv04_instmem_takedown;
engine->instmem.suspend = nv04_instmem_suspend; engine->instmem.suspend = nv04_instmem_suspend;
engine->instmem.resume = nv04_instmem_resume; engine->instmem.resume = nv04_instmem_resume;
engine->instmem.populate = nv04_instmem_populate; engine->instmem.get = nv04_instmem_get;
engine->instmem.clear = nv04_instmem_clear; engine->instmem.put = nv04_instmem_put;
engine->instmem.bind = nv04_instmem_bind; engine->instmem.map = nv04_instmem_map;
engine->instmem.unbind = nv04_instmem_unbind; engine->instmem.unmap = nv04_instmem_unmap;
engine->instmem.flush = nv04_instmem_flush; engine->instmem.flush = nv04_instmem_flush;
engine->mc.init = nv04_mc_init; engine->mc.init = nv04_mc_init;
engine->mc.takedown = nv04_mc_takedown; engine->mc.takedown = nv04_mc_takedown;
...@@ -163,10 +163,10 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev) ...@@ -163,10 +163,10 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
engine->instmem.takedown = nv04_instmem_takedown; engine->instmem.takedown = nv04_instmem_takedown;
engine->instmem.suspend = nv04_instmem_suspend; engine->instmem.suspend = nv04_instmem_suspend;
engine->instmem.resume = nv04_instmem_resume; engine->instmem.resume = nv04_instmem_resume;
engine->instmem.populate = nv04_instmem_populate; engine->instmem.get = nv04_instmem_get;
engine->instmem.clear = nv04_instmem_clear; engine->instmem.put = nv04_instmem_put;
engine->instmem.bind = nv04_instmem_bind; engine->instmem.map = nv04_instmem_map;
engine->instmem.unbind = nv04_instmem_unbind; engine->instmem.unmap = nv04_instmem_unmap;
engine->instmem.flush = nv04_instmem_flush; engine->instmem.flush = nv04_instmem_flush;
engine->mc.init = nv04_mc_init; engine->mc.init = nv04_mc_init;
engine->mc.takedown = nv04_mc_takedown; engine->mc.takedown = nv04_mc_takedown;
...@@ -220,10 +220,10 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev) ...@@ -220,10 +220,10 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
engine->instmem.takedown = nv04_instmem_takedown; engine->instmem.takedown = nv04_instmem_takedown;
engine->instmem.suspend = nv04_instmem_suspend; engine->instmem.suspend = nv04_instmem_suspend;
engine->instmem.resume = nv04_instmem_resume; engine->instmem.resume = nv04_instmem_resume;
engine->instmem.populate = nv04_instmem_populate; engine->instmem.get = nv04_instmem_get;
engine->instmem.clear = nv04_instmem_clear; engine->instmem.put = nv04_instmem_put;
engine->instmem.bind = nv04_instmem_bind; engine->instmem.map = nv04_instmem_map;
engine->instmem.unbind = nv04_instmem_unbind; engine->instmem.unmap = nv04_instmem_unmap;
engine->instmem.flush = nv04_instmem_flush; engine->instmem.flush = nv04_instmem_flush;
engine->mc.init = nv04_mc_init; engine->mc.init = nv04_mc_init;
engine->mc.takedown = nv04_mc_takedown; engine->mc.takedown = nv04_mc_takedown;
...@@ -280,10 +280,10 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev) ...@@ -280,10 +280,10 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
engine->instmem.takedown = nv04_instmem_takedown; engine->instmem.takedown = nv04_instmem_takedown;
engine->instmem.suspend = nv04_instmem_suspend; engine->instmem.suspend = nv04_instmem_suspend;
engine->instmem.resume = nv04_instmem_resume; engine->instmem.resume = nv04_instmem_resume;
engine->instmem.populate = nv04_instmem_populate; engine->instmem.get = nv04_instmem_get;
engine->instmem.clear = nv04_instmem_clear; engine->instmem.put = nv04_instmem_put;
engine->instmem.bind = nv04_instmem_bind; engine->instmem.map = nv04_instmem_map;
engine->instmem.unbind = nv04_instmem_unbind; engine->instmem.unmap = nv04_instmem_unmap;
engine->instmem.flush = nv04_instmem_flush; engine->instmem.flush = nv04_instmem_flush;
engine->mc.init = nv40_mc_init; engine->mc.init = nv40_mc_init;
engine->mc.takedown = nv40_mc_takedown; engine->mc.takedown = nv40_mc_takedown;
...@@ -343,10 +343,10 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev) ...@@ -343,10 +343,10 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
engine->instmem.takedown = nv50_instmem_takedown; engine->instmem.takedown = nv50_instmem_takedown;
engine->instmem.suspend = nv50_instmem_suspend; engine->instmem.suspend = nv50_instmem_suspend;
engine->instmem.resume = nv50_instmem_resume; engine->instmem.resume = nv50_instmem_resume;
engine->instmem.populate = nv50_instmem_populate; engine->instmem.get = nv50_instmem_get;
engine->instmem.clear = nv50_instmem_clear; engine->instmem.put = nv50_instmem_put;
engine->instmem.bind = nv50_instmem_bind; engine->instmem.map = nv50_instmem_map;
engine->instmem.unbind = nv50_instmem_unbind; engine->instmem.unmap = nv50_instmem_unmap;
if (dev_priv->chipset == 0x50) if (dev_priv->chipset == 0x50)
engine->instmem.flush = nv50_instmem_flush; engine->instmem.flush = nv50_instmem_flush;
else else
...@@ -449,10 +449,10 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev) ...@@ -449,10 +449,10 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
engine->instmem.takedown = nvc0_instmem_takedown; engine->instmem.takedown = nvc0_instmem_takedown;
engine->instmem.suspend = nvc0_instmem_suspend; engine->instmem.suspend = nvc0_instmem_suspend;
engine->instmem.resume = nvc0_instmem_resume; engine->instmem.resume = nvc0_instmem_resume;
engine->instmem.populate = nvc0_instmem_populate; engine->instmem.get = nvc0_instmem_get;
engine->instmem.clear = nvc0_instmem_clear; engine->instmem.put = nvc0_instmem_put;
engine->instmem.bind = nvc0_instmem_bind; engine->instmem.map = nvc0_instmem_map;
engine->instmem.unbind = nvc0_instmem_unbind; engine->instmem.unmap = nvc0_instmem_unmap;
engine->instmem.flush = nvc0_instmem_flush; engine->instmem.flush = nvc0_instmem_flush;
engine->mc.init = nv50_mc_init; engine->mc.init = nv50_mc_init;
engine->mc.takedown = nv50_mc_takedown; engine->mc.takedown = nv50_mc_takedown;
......
...@@ -98,42 +98,66 @@ nv04_instmem_takedown(struct drm_device *dev) ...@@ -98,42 +98,66 @@ nv04_instmem_takedown(struct drm_device *dev)
} }
int int
nv04_instmem_populate(struct drm_device *dev, struct nouveau_gpuobj *gpuobj, nv04_instmem_suspend(struct drm_device *dev)
u32 *size, u32 align)
{ {
return 0; return 0;
} }
void void
nv04_instmem_clear(struct drm_device *dev, struct nouveau_gpuobj *gpuobj) nv04_instmem_resume(struct drm_device *dev)
{
}
int
nv04_instmem_bind(struct drm_device *dev, struct nouveau_gpuobj *gpuobj)
{ {
return 0;
} }
int int
nv04_instmem_unbind(struct drm_device *dev, struct nouveau_gpuobj *gpuobj) nv04_instmem_get(struct nouveau_gpuobj *gpuobj, u32 size, u32 align)
{ {
struct drm_nouveau_private *dev_priv = gpuobj->dev->dev_private;
struct drm_mm_node *ramin = NULL;
do {
if (drm_mm_pre_get(&dev_priv->ramin_heap))
return -ENOMEM;
spin_lock(&dev_priv->ramin_lock);
ramin = drm_mm_search_free(&dev_priv->ramin_heap, size, align, 0);
if (ramin == NULL) {
spin_unlock(&dev_priv->ramin_lock);
return -ENOMEM;
}
ramin = drm_mm_get_block_atomic(ramin, size, align);
spin_unlock(&dev_priv->ramin_lock);
} while (ramin == NULL);
gpuobj->node = ramin;
gpuobj->vinst = ramin->start;
return 0; return 0;
} }
void void
nv04_instmem_flush(struct drm_device *dev) nv04_instmem_put(struct nouveau_gpuobj *gpuobj)
{ {
struct drm_nouveau_private *dev_priv = gpuobj->dev->dev_private;
spin_lock(&dev_priv->ramin_lock);
drm_mm_put_block(gpuobj->node);
gpuobj->node = NULL;
spin_unlock(&dev_priv->ramin_lock);
} }
int int
nv04_instmem_suspend(struct drm_device *dev) nv04_instmem_map(struct nouveau_gpuobj *gpuobj)
{ {
gpuobj->pinst = gpuobj->vinst;
return 0; return 0;
} }
void void
nv04_instmem_resume(struct drm_device *dev) nv04_instmem_unmap(struct nouveau_gpuobj *gpuobj)
{ {
} }
void
nv04_instmem_flush(struct drm_device *dev)
{
}
...@@ -157,10 +157,7 @@ nv50_instmem_init(struct drm_device *dev) ...@@ -157,10 +157,7 @@ nv50_instmem_init(struct drm_device *dev)
nv_wo32(priv->pramin_bar, 0x10, 0x00000000); nv_wo32(priv->pramin_bar, 0x10, 0x00000000);
nv_wo32(priv->pramin_bar, 0x14, 0x00000000); nv_wo32(priv->pramin_bar, 0x14, 0x00000000);
/* map channel into PRAMIN, gpuobj didn't do it for us */ nv50_instmem_map(chan->ramin);
ret = nv50_instmem_bind(dev, chan->ramin);
if (ret)
return ret;
/* poke regs... */ /* poke regs... */
nv_wr32(dev, 0x001704, 0x00000000 | (chan->ramin->vinst >> 12)); nv_wr32(dev, 0x001704, 0x00000000 | (chan->ramin->vinst >> 12));
...@@ -305,72 +302,91 @@ nv50_instmem_resume(struct drm_device *dev) ...@@ -305,72 +302,91 @@ nv50_instmem_resume(struct drm_device *dev)
dev_priv->ramin_available = true; dev_priv->ramin_available = true;
} }
struct nv50_gpuobj_node {
struct nouveau_bo *vram;
struct drm_mm_node *ramin;
u32 align;
};
int int
nv50_instmem_populate(struct drm_device *dev, struct nouveau_gpuobj *gpuobj, nv50_instmem_get(struct nouveau_gpuobj *gpuobj, u32 size, u32 align)
u32 *size, u32 align)
{ {
struct drm_device *dev = gpuobj->dev;
struct nv50_gpuobj_node *node = NULL;
int ret; int ret;
if (gpuobj->im_backing) node = kzalloc(sizeof(*node), GFP_KERNEL);
return -EINVAL; if (!node)
return -ENOMEM;
*size = ALIGN(*size, 4096); node->align = align;
if (*size == 0)
return -EINVAL;
ret = nouveau_bo_new(dev, NULL, *size, align, TTM_PL_FLAG_VRAM, ret = nouveau_bo_new(dev, NULL, size, align, TTM_PL_FLAG_VRAM,
0, 0x0000, true, false, &gpuobj->im_backing); 0, 0x0000, true, false, &node->vram);
if (ret) { if (ret) {
NV_ERROR(dev, "error getting PRAMIN backing pages: %d\n", ret); NV_ERROR(dev, "error getting PRAMIN backing pages: %d\n", ret);
return ret; return ret;
} }
ret = nouveau_bo_pin(gpuobj->im_backing, TTM_PL_FLAG_VRAM); ret = nouveau_bo_pin(node->vram, TTM_PL_FLAG_VRAM);
if (ret) { if (ret) {
NV_ERROR(dev, "error pinning PRAMIN backing VRAM: %d\n", ret); NV_ERROR(dev, "error pinning PRAMIN backing VRAM: %d\n", ret);
nouveau_bo_ref(NULL, &gpuobj->im_backing); nouveau_bo_ref(NULL, &node->vram);
return ret; return ret;
} }
gpuobj->vinst = gpuobj->im_backing->bo.mem.start << PAGE_SHIFT; gpuobj->vinst = node->vram->bo.mem.start << PAGE_SHIFT;
gpuobj->size = node->vram->bo.mem.num_pages << PAGE_SHIFT;
gpuobj->node = node;
return 0; return 0;
} }
void void
nv50_instmem_clear(struct drm_device *dev, struct nouveau_gpuobj *gpuobj) nv50_instmem_put(struct nouveau_gpuobj *gpuobj)
{ {
struct drm_nouveau_private *dev_priv = dev->dev_private; struct nv50_gpuobj_node *node;
if (gpuobj && gpuobj->im_backing) { node = gpuobj->node;
if (gpuobj->im_bound) gpuobj->node = NULL;
dev_priv->engine.instmem.unbind(dev, gpuobj);
nouveau_bo_unpin(gpuobj->im_backing); nouveau_bo_unpin(node->vram);
nouveau_bo_ref(NULL, &gpuobj->im_backing); nouveau_bo_ref(NULL, &node->vram);
gpuobj->im_backing = NULL; kfree(node);
}
} }
int int
nv50_instmem_bind(struct drm_device *dev, struct nouveau_gpuobj *gpuobj) nv50_instmem_map(struct nouveau_gpuobj *gpuobj)
{ {
struct drm_nouveau_private *dev_priv = dev->dev_private; struct drm_nouveau_private *dev_priv = gpuobj->dev->dev_private;
struct nv50_instmem_priv *priv = dev_priv->engine.instmem.priv; struct nv50_instmem_priv *priv = dev_priv->engine.instmem.priv;
struct nouveau_gpuobj *pramin_pt = priv->pramin_pt; struct nv50_gpuobj_node *node = gpuobj->node;
uint32_t pte, pte_end; struct drm_device *dev = gpuobj->dev;
uint64_t vram; struct drm_mm_node *ramin = NULL;
u32 pte, pte_end;
if (!gpuobj->im_backing || !gpuobj->im_pramin || gpuobj->im_bound) u64 vram;
return -EINVAL;
do {
if (drm_mm_pre_get(&dev_priv->ramin_heap))
return -ENOMEM;
spin_lock(&dev_priv->ramin_lock);
ramin = drm_mm_search_free(&dev_priv->ramin_heap, gpuobj->size,
node->align, 0);
if (ramin == NULL) {
spin_unlock(&dev_priv->ramin_lock);
return -ENOMEM;
}
NV_DEBUG(dev, "st=0x%lx sz=0x%lx\n", ramin = drm_mm_get_block_atomic(ramin, gpuobj->size, node->align);
gpuobj->im_pramin->start, gpuobj->im_pramin->size); spin_unlock(&dev_priv->ramin_lock);
} while (ramin == NULL);
pte = (gpuobj->im_pramin->start >> 12) << 1; pte = (ramin->start >> 12) << 1;
pte_end = ((gpuobj->im_pramin->size >> 12) << 1) + pte; pte_end = ((ramin->size >> 12) << 1) + pte;
vram = gpuobj->vinst; vram = gpuobj->vinst;
NV_DEBUG(dev, "pramin=0x%lx, pte=%d, pte_end=%d\n", NV_DEBUG(dev, "pramin=0x%lx, pte=%d, pte_end=%d\n",
gpuobj->im_pramin->start, pte, pte_end); ramin->start, pte, pte_end);
NV_DEBUG(dev, "first vram page: 0x%010llx\n", gpuobj->vinst); NV_DEBUG(dev, "first vram page: 0x%010llx\n", gpuobj->vinst);
vram |= 1; vram |= 1;
...@@ -380,8 +396,8 @@ nv50_instmem_bind(struct drm_device *dev, struct nouveau_gpuobj *gpuobj) ...@@ -380,8 +396,8 @@ nv50_instmem_bind(struct drm_device *dev, struct nouveau_gpuobj *gpuobj)
} }
while (pte < pte_end) { while (pte < pte_end) {
nv_wo32(pramin_pt, (pte * 4) + 0, lower_32_bits(vram)); nv_wo32(priv->pramin_pt, (pte * 4) + 0, lower_32_bits(vram));
nv_wo32(pramin_pt, (pte * 4) + 4, upper_32_bits(vram)); nv_wo32(priv->pramin_pt, (pte * 4) + 4, upper_32_bits(vram));
vram += 0x1000; vram += 0x1000;
pte += 2; pte += 2;
} }
...@@ -389,36 +405,36 @@ nv50_instmem_bind(struct drm_device *dev, struct nouveau_gpuobj *gpuobj) ...@@ -389,36 +405,36 @@ nv50_instmem_bind(struct drm_device *dev, struct nouveau_gpuobj *gpuobj)
nv50_vm_flush(dev, 6); nv50_vm_flush(dev, 6);
gpuobj->im_bound = 1; node->ramin = ramin;
gpuobj->pinst = ramin->start;
return 0; return 0;
} }
int void
nv50_instmem_unbind(struct drm_device *dev, struct nouveau_gpuobj *gpuobj) nv50_instmem_unmap(struct nouveau_gpuobj *gpuobj)
{ {
struct drm_nouveau_private *dev_priv = dev->dev_private; struct drm_nouveau_private *dev_priv = gpuobj->dev->dev_private;
struct nv50_instmem_priv *priv = dev_priv->engine.instmem.priv; struct nv50_instmem_priv *priv = dev_priv->engine.instmem.priv;
uint32_t pte, pte_end; struct nv50_gpuobj_node *node = gpuobj->node;
u32 pte, pte_end;
if (gpuobj->im_bound == 0)
return -EINVAL;
/* can happen during late takedown */ if (!node->ramin || !dev_priv->ramin_available)
if (unlikely(!dev_priv->ramin_available)) return;
return 0;
pte = (gpuobj->im_pramin->start >> 12) << 1; pte = (node->ramin->start >> 12) << 1;
pte_end = ((gpuobj->im_pramin->size >> 12) << 1) + pte; pte_end = ((node->ramin->size >> 12) << 1) + pte;
while (pte < pte_end) { while (pte < pte_end) {
nv_wo32(priv->pramin_pt, (pte * 4) + 0, 0x00000000); nv_wo32(priv->pramin_pt, (pte * 4) + 0, 0x00000000);
nv_wo32(priv->pramin_pt, (pte * 4) + 4, 0x00000000); nv_wo32(priv->pramin_pt, (pte * 4) + 4, 0x00000000);
pte += 2; pte += 2;
} }
dev_priv->engine.instmem.flush(dev); dev_priv->engine.instmem.flush(gpuobj->dev);
gpuobj->im_bound = 0; spin_lock(&dev_priv->ramin_lock);
return 0; drm_mm_put_block(node->ramin);
node->ramin = NULL;
spin_unlock(&dev_priv->ramin_lock);
} }
void void
......
...@@ -26,67 +26,89 @@ ...@@ -26,67 +26,89 @@
#include "nouveau_drv.h" #include "nouveau_drv.h"
struct nvc0_gpuobj_node {
struct nouveau_bo *vram;
struct drm_mm_node *ramin;
u32 align;
};
int int
nvc0_instmem_populate(struct drm_device *dev, struct nouveau_gpuobj *gpuobj, nvc0_instmem_get(struct nouveau_gpuobj *gpuobj, u32 size, u32 align)
u32 *size, u32 align)
{ {
struct drm_device *dev = gpuobj->dev;
struct nvc0_gpuobj_node *node = NULL;
int ret; int ret;
*size = ALIGN(*size, 4096); node = kzalloc(sizeof(*node), GFP_KERNEL);
if (*size == 0) if (!node)
return -EINVAL; return -ENOMEM;
node->align = align;
ret = nouveau_bo_new(dev, NULL, *size, align, TTM_PL_FLAG_VRAM, ret = nouveau_bo_new(dev, NULL, size, align, TTM_PL_FLAG_VRAM,
0, 0x0000, true, false, &gpuobj->im_backing); 0, 0x0000, true, false, &node->vram);
if (ret) { if (ret) {
NV_ERROR(dev, "error getting PRAMIN backing pages: %d\n", ret); NV_ERROR(dev, "error getting PRAMIN backing pages: %d\n", ret);
return ret; return ret;
} }
ret = nouveau_bo_pin(gpuobj->im_backing, TTM_PL_FLAG_VRAM); ret = nouveau_bo_pin(node->vram, TTM_PL_FLAG_VRAM);
if (ret) { if (ret) {
NV_ERROR(dev, "error pinning PRAMIN backing VRAM: %d\n", ret); NV_ERROR(dev, "error pinning PRAMIN backing VRAM: %d\n", ret);
nouveau_bo_ref(NULL, &gpuobj->im_backing); nouveau_bo_ref(NULL, &node->vram);
return ret; return ret;
} }
gpuobj->vinst = gpuobj->im_backing->bo.mem.start << PAGE_SHIFT; gpuobj->vinst = node->vram->bo.mem.start << PAGE_SHIFT;
gpuobj->size = node->vram->bo.mem.num_pages << PAGE_SHIFT;
gpuobj->node = node;
return 0; return 0;
} }
void void
nvc0_instmem_clear(struct drm_device *dev, struct nouveau_gpuobj *gpuobj) nvc0_instmem_put(struct nouveau_gpuobj *gpuobj)
{ {
struct drm_nouveau_private *dev_priv = dev->dev_private; struct nvc0_gpuobj_node *node;
if (gpuobj && gpuobj->im_backing) { node = gpuobj->node;
if (gpuobj->im_bound) gpuobj->node = NULL;
dev_priv->engine.instmem.unbind(dev, gpuobj);
nouveau_bo_unpin(gpuobj->im_backing); nouveau_bo_unpin(node->vram);
nouveau_bo_ref(NULL, &gpuobj->im_backing); nouveau_bo_ref(NULL, &node->vram);
gpuobj->im_backing = NULL; kfree(node);
}
} }
int int
nvc0_instmem_bind(struct drm_device *dev, struct nouveau_gpuobj *gpuobj) nvc0_instmem_map(struct nouveau_gpuobj *gpuobj)
{ {
struct drm_nouveau_private *dev_priv = dev->dev_private; struct drm_nouveau_private *dev_priv = gpuobj->dev->dev_private;
uint32_t pte, pte_end; struct nvc0_gpuobj_node *node = gpuobj->node;
uint64_t vram; struct drm_device *dev = gpuobj->dev;
struct drm_mm_node *ramin = NULL;
if (!gpuobj->im_backing || !gpuobj->im_pramin || gpuobj->im_bound) u32 pte, pte_end;
return -EINVAL; u64 vram;
NV_DEBUG(dev, "st=0x%lx sz=0x%lx\n", do {
gpuobj->im_pramin->start, gpuobj->im_pramin->size); if (drm_mm_pre_get(&dev_priv->ramin_heap))
return -ENOMEM;
pte = gpuobj->im_pramin->start >> 12;
pte_end = (gpuobj->im_pramin->size >> 12) + pte; spin_lock(&dev_priv->ramin_lock);
ramin = drm_mm_search_free(&dev_priv->ramin_heap, gpuobj->size,
node->align, 0);
if (ramin == NULL) {
spin_unlock(&dev_priv->ramin_lock);
return -ENOMEM;
}
ramin = drm_mm_get_block_atomic(ramin, gpuobj->size, node->align);
spin_unlock(&dev_priv->ramin_lock);
} while (ramin == NULL);
pte = (ramin->start >> 12) << 1;
pte_end = ((ramin->size >> 12) << 1) + pte;
vram = gpuobj->vinst; vram = gpuobj->vinst;
NV_DEBUG(dev, "pramin=0x%lx, pte=%d, pte_end=%d\n", NV_DEBUG(dev, "pramin=0x%lx, pte=%d, pte_end=%d\n",
gpuobj->im_pramin->start, pte, pte_end); ramin->start, pte, pte_end);
NV_DEBUG(dev, "first vram page: 0x%010llx\n", gpuobj->vinst); NV_DEBUG(dev, "first vram page: 0x%010llx\n", gpuobj->vinst);
while (pte < pte_end) { while (pte < pte_end) {
...@@ -103,30 +125,35 @@ nvc0_instmem_bind(struct drm_device *dev, struct nouveau_gpuobj *gpuobj) ...@@ -103,30 +125,35 @@ nvc0_instmem_bind(struct drm_device *dev, struct nouveau_gpuobj *gpuobj)
nv_wr32(dev, 0x100cbc, 0x80000005); nv_wr32(dev, 0x100cbc, 0x80000005);
} }
gpuobj->im_bound = 1; node->ramin = ramin;
gpuobj->pinst = ramin->start;
return 0; return 0;
} }
int void
nvc0_instmem_unbind(struct drm_device *dev, struct nouveau_gpuobj *gpuobj) nvc0_instmem_unmap(struct nouveau_gpuobj *gpuobj)
{ {
struct drm_nouveau_private *dev_priv = dev->dev_private; struct drm_nouveau_private *dev_priv = gpuobj->dev->dev_private;
uint32_t pte, pte_end; struct nvc0_gpuobj_node *node = gpuobj->node;
u32 pte, pte_end;
if (gpuobj->im_bound == 0) if (!node->ramin || !dev_priv->ramin_available)
return -EINVAL; return;
pte = (node->ramin->start >> 12) << 1;
pte_end = ((node->ramin->size >> 12) << 1) + pte;
pte = gpuobj->im_pramin->start >> 12;
pte_end = (gpuobj->im_pramin->size >> 12) + pte;
while (pte < pte_end) { while (pte < pte_end) {
nv_wr32(dev, 0x702000 + (pte * 8), 0); nv_wr32(gpuobj->dev, 0x702000 + (pte * 8), 0);
nv_wr32(dev, 0x702004 + (pte * 8), 0); nv_wr32(gpuobj->dev, 0x702004 + (pte * 8), 0);
pte++; pte++;
} }
dev_priv->engine.instmem.flush(dev); dev_priv->engine.instmem.flush(gpuobj->dev);
gpuobj->im_bound = 0; spin_lock(&dev_priv->ramin_lock);
return 0; drm_mm_put_block(node->ramin);
node->ramin = NULL;
spin_unlock(&dev_priv->ramin_lock);
} }
void void
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment