mirror of
https://github.com/torvalds/linux.git
synced 2024-11-24 05:02:12 +00:00
Merge remote branch 'nouveau/drm-nouveau-fixes' of /ssd/git/drm-nouveau-next into drm-fixes
* 'nouveau/drm-nouveau-fixes' of /ssd/git/drm-nouveau-next: drm/nouveau: fix allocation of notifier object drm/nouveau: fix notifier memory corruption bug drm/nouveau: fix pinning of notifier block drm/nouveau: populate ttm_alloced with false, when it's not drm/nouveau: fix nv30 pcie boards drm/nouveau: split ramin_lock into two locks, one hardirq safe
This commit is contained in:
commit
be761d5ebd
@ -83,7 +83,7 @@ nouveau_dma_init(struct nouveau_channel *chan)
|
||||
return ret;
|
||||
|
||||
/* NV_MEMORY_TO_MEMORY_FORMAT requires a notifier object */
|
||||
ret = nouveau_notifier_alloc(chan, NvNotify0, 32, 0xfd0, 0x1000,
|
||||
ret = nouveau_notifier_alloc(chan, NvNotify0, 32, 0xfe0, 0x1000,
|
||||
&chan->m2mf_ntfy);
|
||||
if (ret)
|
||||
return ret;
|
||||
|
@ -682,6 +682,9 @@ struct drm_nouveau_private {
|
||||
/* For PFIFO and PGRAPH. */
|
||||
spinlock_t context_switch_lock;
|
||||
|
||||
/* VM/PRAMIN flush, legacy PRAMIN aperture */
|
||||
spinlock_t vm_lock;
|
||||
|
||||
/* RAMIN configuration, RAMFC, RAMHT and RAMRO offsets */
|
||||
struct nouveau_ramht *ramht;
|
||||
struct nouveau_gpuobj *ramfc;
|
||||
|
@ -181,13 +181,13 @@ nouveau_fbcon_sync(struct fb_info *info)
|
||||
OUT_RING (chan, 0);
|
||||
}
|
||||
|
||||
nouveau_bo_wr32(chan->notifier_bo, chan->m2mf_ntfy + 3, 0xffffffff);
|
||||
nouveau_bo_wr32(chan->notifier_bo, chan->m2mf_ntfy/4 + 3, 0xffffffff);
|
||||
FIRE_RING(chan);
|
||||
mutex_unlock(&chan->mutex);
|
||||
|
||||
ret = -EBUSY;
|
||||
for (i = 0; i < 100000; i++) {
|
||||
if (!nouveau_bo_rd32(chan->notifier_bo, chan->m2mf_ntfy + 3)) {
|
||||
if (!nouveau_bo_rd32(chan->notifier_bo, chan->m2mf_ntfy/4 + 3)) {
|
||||
ret = 0;
|
||||
break;
|
||||
}
|
||||
|
@ -398,7 +398,7 @@ nouveau_mem_vram_init(struct drm_device *dev)
|
||||
dma_bits = 40;
|
||||
} else
|
||||
if (drm_pci_device_is_pcie(dev) &&
|
||||
dev_priv->chipset != 0x40 &&
|
||||
dev_priv->chipset > 0x40 &&
|
||||
dev_priv->chipset != 0x45) {
|
||||
if (pci_dma_supported(dev->pdev, DMA_BIT_MASK(39)))
|
||||
dma_bits = 39;
|
||||
|
@ -35,19 +35,22 @@ nouveau_notifier_init_channel(struct nouveau_channel *chan)
|
||||
{
|
||||
struct drm_device *dev = chan->dev;
|
||||
struct nouveau_bo *ntfy = NULL;
|
||||
uint32_t flags;
|
||||
uint32_t flags, ttmpl;
|
||||
int ret;
|
||||
|
||||
if (nouveau_vram_notify)
|
||||
if (nouveau_vram_notify) {
|
||||
flags = NOUVEAU_GEM_DOMAIN_VRAM;
|
||||
else
|
||||
ttmpl = TTM_PL_FLAG_VRAM;
|
||||
} else {
|
||||
flags = NOUVEAU_GEM_DOMAIN_GART;
|
||||
ttmpl = TTM_PL_FLAG_TT;
|
||||
}
|
||||
|
||||
ret = nouveau_gem_new(dev, NULL, PAGE_SIZE, 0, flags, 0, 0, &ntfy);
|
||||
if (ret)
|
||||
return ret;
|
||||
|
||||
ret = nouveau_bo_pin(ntfy, flags);
|
||||
ret = nouveau_bo_pin(ntfy, ttmpl);
|
||||
if (ret)
|
||||
goto out_err;
|
||||
|
||||
|
@ -1039,19 +1039,20 @@ nv_ro32(struct nouveau_gpuobj *gpuobj, u32 offset)
|
||||
{
|
||||
struct drm_nouveau_private *dev_priv = gpuobj->dev->dev_private;
|
||||
struct drm_device *dev = gpuobj->dev;
|
||||
unsigned long flags;
|
||||
|
||||
if (gpuobj->pinst == ~0 || !dev_priv->ramin_available) {
|
||||
u64 ptr = gpuobj->vinst + offset;
|
||||
u32 base = ptr >> 16;
|
||||
u32 val;
|
||||
|
||||
spin_lock(&dev_priv->ramin_lock);
|
||||
spin_lock_irqsave(&dev_priv->vm_lock, flags);
|
||||
if (dev_priv->ramin_base != base) {
|
||||
dev_priv->ramin_base = base;
|
||||
nv_wr32(dev, 0x001700, dev_priv->ramin_base);
|
||||
}
|
||||
val = nv_rd32(dev, 0x700000 + (ptr & 0xffff));
|
||||
spin_unlock(&dev_priv->ramin_lock);
|
||||
spin_unlock_irqrestore(&dev_priv->vm_lock, flags);
|
||||
return val;
|
||||
}
|
||||
|
||||
@ -1063,18 +1064,19 @@ nv_wo32(struct nouveau_gpuobj *gpuobj, u32 offset, u32 val)
|
||||
{
|
||||
struct drm_nouveau_private *dev_priv = gpuobj->dev->dev_private;
|
||||
struct drm_device *dev = gpuobj->dev;
|
||||
unsigned long flags;
|
||||
|
||||
if (gpuobj->pinst == ~0 || !dev_priv->ramin_available) {
|
||||
u64 ptr = gpuobj->vinst + offset;
|
||||
u32 base = ptr >> 16;
|
||||
|
||||
spin_lock(&dev_priv->ramin_lock);
|
||||
spin_lock_irqsave(&dev_priv->vm_lock, flags);
|
||||
if (dev_priv->ramin_base != base) {
|
||||
dev_priv->ramin_base = base;
|
||||
nv_wr32(dev, 0x001700, dev_priv->ramin_base);
|
||||
}
|
||||
nv_wr32(dev, 0x700000 + (ptr & 0xffff), val);
|
||||
spin_unlock(&dev_priv->ramin_lock);
|
||||
spin_unlock_irqrestore(&dev_priv->vm_lock, flags);
|
||||
return;
|
||||
}
|
||||
|
||||
|
@ -55,6 +55,7 @@ nouveau_sgdma_populate(struct ttm_backend *be, unsigned long num_pages,
|
||||
be->func->clear(be);
|
||||
return -EFAULT;
|
||||
}
|
||||
nvbe->ttm_alloced[nvbe->nr_pages] = false;
|
||||
}
|
||||
|
||||
nvbe->nr_pages++;
|
||||
@ -427,7 +428,7 @@ nouveau_sgdma_init(struct drm_device *dev)
|
||||
u32 aper_size, align;
|
||||
int ret;
|
||||
|
||||
if (dev_priv->card_type >= NV_50 || drm_pci_device_is_pcie(dev))
|
||||
if (dev_priv->card_type >= NV_40 && drm_pci_device_is_pcie(dev))
|
||||
aper_size = 512 * 1024 * 1024;
|
||||
else
|
||||
aper_size = 64 * 1024 * 1024;
|
||||
@ -457,7 +458,7 @@ nouveau_sgdma_init(struct drm_device *dev)
|
||||
dev_priv->gart_info.func = &nv50_sgdma_backend;
|
||||
} else
|
||||
if (drm_pci_device_is_pcie(dev) &&
|
||||
dev_priv->chipset != 0x40 && dev_priv->chipset != 0x45) {
|
||||
dev_priv->chipset > 0x40 && dev_priv->chipset != 0x45) {
|
||||
if (nv44_graph_class(dev)) {
|
||||
dev_priv->gart_info.func = &nv44_sgdma_backend;
|
||||
align = 512 * 1024;
|
||||
|
@ -608,6 +608,7 @@ nouveau_card_init(struct drm_device *dev)
|
||||
spin_lock_init(&dev_priv->channels.lock);
|
||||
spin_lock_init(&dev_priv->tile.lock);
|
||||
spin_lock_init(&dev_priv->context_switch_lock);
|
||||
spin_lock_init(&dev_priv->vm_lock);
|
||||
|
||||
/* Make the CRTCs and I2C buses accessible */
|
||||
ret = engine->display.early_init(dev);
|
||||
|
@ -404,23 +404,25 @@ void
|
||||
nv50_instmem_flush(struct drm_device *dev)
|
||||
{
|
||||
struct drm_nouveau_private *dev_priv = dev->dev_private;
|
||||
unsigned long flags;
|
||||
|
||||
spin_lock(&dev_priv->ramin_lock);
|
||||
spin_lock_irqsave(&dev_priv->vm_lock, flags);
|
||||
nv_wr32(dev, 0x00330c, 0x00000001);
|
||||
if (!nv_wait(dev, 0x00330c, 0x00000002, 0x00000000))
|
||||
NV_ERROR(dev, "PRAMIN flush timeout\n");
|
||||
spin_unlock(&dev_priv->ramin_lock);
|
||||
spin_unlock_irqrestore(&dev_priv->vm_lock, flags);
|
||||
}
|
||||
|
||||
void
|
||||
nv84_instmem_flush(struct drm_device *dev)
|
||||
{
|
||||
struct drm_nouveau_private *dev_priv = dev->dev_private;
|
||||
unsigned long flags;
|
||||
|
||||
spin_lock(&dev_priv->ramin_lock);
|
||||
spin_lock_irqsave(&dev_priv->vm_lock, flags);
|
||||
nv_wr32(dev, 0x070000, 0x00000001);
|
||||
if (!nv_wait(dev, 0x070000, 0x00000002, 0x00000000))
|
||||
NV_ERROR(dev, "PRAMIN flush timeout\n");
|
||||
spin_unlock(&dev_priv->ramin_lock);
|
||||
spin_unlock_irqrestore(&dev_priv->vm_lock, flags);
|
||||
}
|
||||
|
||||
|
@ -174,10 +174,11 @@ void
|
||||
nv50_vm_flush_engine(struct drm_device *dev, int engine)
|
||||
{
|
||||
struct drm_nouveau_private *dev_priv = dev->dev_private;
|
||||
unsigned long flags;
|
||||
|
||||
spin_lock(&dev_priv->ramin_lock);
|
||||
spin_lock_irqsave(&dev_priv->vm_lock, flags);
|
||||
nv_wr32(dev, 0x100c80, (engine << 16) | 1);
|
||||
if (!nv_wait(dev, 0x100c80, 0x00000001, 0x00000000))
|
||||
NV_ERROR(dev, "vm flush timeout: engine %d\n", engine);
|
||||
spin_unlock(&dev_priv->ramin_lock);
|
||||
spin_unlock_irqrestore(&dev_priv->vm_lock, flags);
|
||||
}
|
||||
|
@ -104,11 +104,12 @@ nvc0_vm_flush(struct nouveau_vm *vm)
|
||||
struct nouveau_instmem_engine *pinstmem = &dev_priv->engine.instmem;
|
||||
struct drm_device *dev = vm->dev;
|
||||
struct nouveau_vm_pgd *vpgd;
|
||||
unsigned long flags;
|
||||
u32 engine = (dev_priv->chan_vm == vm) ? 1 : 5;
|
||||
|
||||
pinstmem->flush(vm->dev);
|
||||
|
||||
spin_lock(&dev_priv->ramin_lock);
|
||||
spin_lock_irqsave(&dev_priv->vm_lock, flags);
|
||||
list_for_each_entry(vpgd, &vm->pgd_list, head) {
|
||||
/* looks like maybe a "free flush slots" counter, the
|
||||
* faster you write to 0x100cbc to more it decreases
|
||||
@ -125,5 +126,5 @@ nvc0_vm_flush(struct nouveau_vm *vm)
|
||||
nv_rd32(dev, 0x100c80), engine);
|
||||
}
|
||||
}
|
||||
spin_unlock(&dev_priv->ramin_lock);
|
||||
spin_unlock_irqrestore(&dev_priv->vm_lock, flags);
|
||||
}
|
||||
|
Loading…
Reference in New Issue
Block a user