提交 235b87af 编写于 作者: D Dave Airlie

Merge remote branch 'nouveau/drm-nouveau-next' of ../drm-nouveau-next into drm-core-next

* 'nouveau/drm-nouveau-next' of ../drm-nouveau-next:
  drm/nouveau: fix __nouveau_fence_wait performance
  drm/nv40: attempt to reserve just enough vram for all 32 channels
  drm/nv50: check for vm traps on every gr irq
  drm/nv50: decode vm faults some more
  drm/nouveau: add nouveau_enum_find() util function
  drm/nouveau: properly handle pushbuffer check failures
  drm/nvc0: remove vm hack forcing large/small pages to not share a PDE
......@@ -1076,7 +1076,7 @@ extern void nv40_fb_set_tile_region(struct drm_device *dev, int i);
/* nv50_fb.c */
extern int nv50_fb_init(struct drm_device *);
extern void nv50_fb_takedown(struct drm_device *);
extern void nv50_fb_vm_trap(struct drm_device *, int display, const char *);
extern void nv50_fb_vm_trap(struct drm_device *, int display);
/* nvc0_fb.c */
extern int nvc0_fb_init(struct drm_device *);
......
......@@ -27,6 +27,9 @@
#include "drmP.h"
#include "drm.h"
#include <linux/ktime.h>
#include <linux/hrtimer.h>
#include "nouveau_drv.h"
#include "nouveau_ramht.h"
#include "nouveau_dma.h"
......@@ -229,7 +232,8 @@ int
__nouveau_fence_wait(void *sync_obj, void *sync_arg, bool lazy, bool intr)
{
unsigned long timeout = jiffies + (3 * DRM_HZ);
unsigned long sleep_time = jiffies + 1;
unsigned long sleep_time = NSEC_PER_MSEC / 1000;
ktime_t t;
int ret = 0;
while (1) {
......@@ -243,8 +247,13 @@ __nouveau_fence_wait(void *sync_obj, void *sync_arg, bool lazy, bool intr)
__set_current_state(intr ? TASK_INTERRUPTIBLE
: TASK_UNINTERRUPTIBLE);
if (lazy && time_after_eq(jiffies, sleep_time))
schedule_timeout(1);
if (lazy) {
t = ktime_set(0, sleep_time);
schedule_hrtimeout(&t, HRTIMER_MODE_REL);
sleep_time *= 2;
if (sleep_time > NSEC_PER_MSEC)
sleep_time = NSEC_PER_MSEC;
}
if (intr && signal_pending(current)) {
ret = -ERESTARTSYS;
......
......@@ -600,7 +600,7 @@ nouveau_gem_ioctl_pushbuf(struct drm_device *dev, void *data,
if (push[i].bo_index >= req->nr_buffers) {
NV_ERROR(dev, "push %d buffer not in list\n", i);
ret = -EINVAL;
goto out;
goto out_prevalid;
}
bo[push[i].bo_index].read_domains |= (1 << 31);
......@@ -612,7 +612,7 @@ nouveau_gem_ioctl_pushbuf(struct drm_device *dev, void *data,
if (ret) {
if (ret != -ERESTARTSYS)
NV_ERROR(dev, "validate: %d\n", ret);
goto out;
goto out_prevalid;
}
/* Apply any relocations that are required */
......@@ -705,6 +705,8 @@ nouveau_gem_ioctl_pushbuf(struct drm_device *dev, void *data,
out:
validate_fini(&op, fence);
nouveau_fence_unref(&fence);
out_prevalid:
kfree(bo);
kfree(push);
......
......@@ -424,14 +424,32 @@ nouveau_mem_vram_init(struct drm_device *dev)
}
/* reserve space at end of VRAM for PRAMIN */
if (dev_priv->chipset == 0x40 || dev_priv->chipset == 0x47 ||
dev_priv->chipset == 0x49 || dev_priv->chipset == 0x4b)
dev_priv->ramin_rsvd_vram = (2 * 1024 * 1024);
else
if (dev_priv->card_type >= NV_40)
dev_priv->ramin_rsvd_vram = (1 * 1024 * 1024);
else
dev_priv->ramin_rsvd_vram = (512 * 1024);
if (dev_priv->card_type >= NV_50) {
dev_priv->ramin_rsvd_vram = 1 * 1024 * 1024;
} else
if (dev_priv->card_type >= NV_40) {
u32 vs = hweight8((nv_rd32(dev, 0x001540) & 0x0000ff00) >> 8);
u32 rsvd;
/* estimate grctx size, the magics come from nv40_grctx.c */
if (dev_priv->chipset == 0x40) rsvd = 0x6aa0 * vs;
else if (dev_priv->chipset < 0x43) rsvd = 0x4f00 * vs;
else if (nv44_graph_class(dev)) rsvd = 0x4980 * vs;
else rsvd = 0x4a40 * vs;
rsvd += 16 * 1024;
rsvd *= dev_priv->engine.fifo.channels;
/* pciegart table */
if (drm_pci_device_is_pcie(dev))
rsvd += 512 * 1024;
/* object storage */
rsvd += 512 * 1024;
dev_priv->ramin_rsvd_vram = round_up(rsvd, 4096);
} else {
dev_priv->ramin_rsvd_vram = 512 * 1024;
}
ret = dev_priv->engine.vram.init(dev);
if (ret)
......
......@@ -427,8 +427,7 @@ nouveau_sgdma_init(struct drm_device *dev)
u32 aper_size, align;
int ret;
if (dev_priv->card_type >= NV_50 ||
dev_priv->ramin_rsvd_vram >= 2 * 1024 * 1024)
if (dev_priv->card_type >= NV_50 || drm_pci_device_is_pcie(dev))
aper_size = 512 * 1024 * 1024;
else
aper_size = 64 * 1024 * 1024;
......
......@@ -47,18 +47,27 @@ nouveau_bitfield_print(const struct nouveau_bitfield *bf, u32 value)
printk(" (unknown bits 0x%08x)", value);
}
void
nouveau_enum_print(const struct nouveau_enum *en, u32 value)
const struct nouveau_enum *
nouveau_enum_find(const struct nouveau_enum *en, u32 value)
{
while (en->name) {
if (value == en->value) {
printk("%s", en->name);
return;
}
if (en->value == value)
return en;
en++;
}
return NULL;
}
void
nouveau_enum_print(const struct nouveau_enum *en, u32 value)
{
en = nouveau_enum_find(en, value);
if (en) {
printk("%s", en->name);
return;
}
printk("(unknown enum 0x%08x)", value);
}
......
......@@ -36,10 +36,14 @@ struct nouveau_bitfield {
struct nouveau_enum {
u32 value;
const char *name;
void *data;
};
void nouveau_bitfield_print(const struct nouveau_bitfield *, u32 value);
void nouveau_enum_print(const struct nouveau_enum *, u32 value);
const struct nouveau_enum *
nouveau_enum_find(const struct nouveau_enum *, u32 value);
int nouveau_ratelimit(void);
#endif
......@@ -314,18 +314,7 @@ nouveau_vm_new(struct drm_device *dev, u64 offset, u64 length, u64 mm_offset,
vm->spg_shift = 12;
vm->lpg_shift = 17;
pgt_bits = 27;
/* Should be 4096 everywhere, this is a hack that's
* currently necessary to avoid an elusive bug that
* causes corruption when mixing small/large pages
*/
if (length < (1ULL << 40))
block = 4096;
else {
block = (1 << pgt_bits);
if (length < block)
block = length;
}
block = 4096;
} else {
kfree(vm);
return -ENOSYS;
......
......@@ -516,7 +516,7 @@ nv04_fifo_isr(struct drm_device *dev)
if (dev_priv->card_type == NV_50) {
if (status & 0x00000010) {
nv50_fb_vm_trap(dev, 1, "PFIFO_BAR_FAULT");
nv50_fb_vm_trap(dev, nouveau_ratelimit());
status &= ~0x00000010;
nv_wr32(dev, 0x002100, 0x00000010);
}
......
......@@ -114,12 +114,109 @@ nv50_fb_takedown(struct drm_device *dev)
nv50_fb_destroy(dev);
}
static struct nouveau_enum vm_dispatch_subclients[] = {
{ 0x00000000, "GRCTX", NULL },
{ 0x00000001, "NOTIFY", NULL },
{ 0x00000002, "QUERY", NULL },
{ 0x00000003, "COND", NULL },
{ 0x00000004, "M2M_IN", NULL },
{ 0x00000005, "M2M_OUT", NULL },
{ 0x00000006, "M2M_NOTIFY", NULL },
{}
};
static struct nouveau_enum vm_ccache_subclients[] = {
{ 0x00000000, "CB", NULL },
{ 0x00000001, "TIC", NULL },
{ 0x00000002, "TSC", NULL },
{}
};
static struct nouveau_enum vm_prop_subclients[] = {
{ 0x00000000, "RT0", NULL },
{ 0x00000001, "RT1", NULL },
{ 0x00000002, "RT2", NULL },
{ 0x00000003, "RT3", NULL },
{ 0x00000004, "RT4", NULL },
{ 0x00000005, "RT5", NULL },
{ 0x00000006, "RT6", NULL },
{ 0x00000007, "RT7", NULL },
{ 0x00000008, "ZETA", NULL },
{ 0x00000009, "LOCAL", NULL },
{ 0x0000000a, "GLOBAL", NULL },
{ 0x0000000b, "STACK", NULL },
{ 0x0000000c, "DST2D", NULL },
{}
};
static struct nouveau_enum vm_pfifo_subclients[] = {
{ 0x00000000, "PUSHBUF", NULL },
{ 0x00000001, "SEMAPHORE", NULL },
{}
};
static struct nouveau_enum vm_bar_subclients[] = {
{ 0x00000000, "FB", NULL },
{ 0x00000001, "IN", NULL },
{}
};
static struct nouveau_enum vm_client[] = {
{ 0x00000000, "STRMOUT", NULL },
{ 0x00000003, "DISPATCH", vm_dispatch_subclients },
{ 0x00000004, "PFIFO_WRITE", NULL },
{ 0x00000005, "CCACHE", vm_ccache_subclients },
{ 0x00000006, "PPPP", NULL },
{ 0x00000007, "CLIPID", NULL },
{ 0x00000008, "PFIFO_READ", NULL },
{ 0x00000009, "VFETCH", NULL },
{ 0x0000000a, "TEXTURE", NULL },
{ 0x0000000b, "PROP", vm_prop_subclients },
{ 0x0000000c, "PVP", NULL },
{ 0x0000000d, "PBSP", NULL },
{ 0x0000000e, "PCRYPT", NULL },
{ 0x0000000f, "PCOUNTER", NULL },
{ 0x00000011, "PDAEMON", NULL },
{}
};
static struct nouveau_enum vm_engine[] = {
{ 0x00000000, "PGRAPH", NULL },
{ 0x00000001, "PVP", NULL },
{ 0x00000004, "PEEPHOLE", NULL },
{ 0x00000005, "PFIFO", vm_pfifo_subclients },
{ 0x00000006, "BAR", vm_bar_subclients },
{ 0x00000008, "PPPP", NULL },
{ 0x00000009, "PBSP", NULL },
{ 0x0000000a, "PCRYPT", NULL },
{ 0x0000000b, "PCOUNTER", NULL },
{ 0x0000000c, "SEMAPHORE_BG", NULL },
{ 0x0000000d, "PCOPY", NULL },
{ 0x0000000e, "PDAEMON", NULL },
{}
};
static struct nouveau_enum vm_fault[] = {
{ 0x00000000, "PT_NOT_PRESENT", NULL },
{ 0x00000001, "PT_TOO_SHORT", NULL },
{ 0x00000002, "PAGE_NOT_PRESENT", NULL },
{ 0x00000003, "PAGE_SYSTEM_ONLY", NULL },
{ 0x00000004, "PAGE_READ_ONLY", NULL },
{ 0x00000006, "NULL_DMAOBJ", NULL },
{ 0x00000007, "WRONG_MEMTYPE", NULL },
{ 0x0000000b, "VRAM_LIMIT", NULL },
{ 0x0000000f, "DMAOBJ_LIMIT", NULL },
{}
};
void
nv50_fb_vm_trap(struct drm_device *dev, int display, const char *name)
nv50_fb_vm_trap(struct drm_device *dev, int display)
{
struct drm_nouveau_private *dev_priv = dev->dev_private;
const struct nouveau_enum *en, *cl;
unsigned long flags;
u32 trap[6], idx, chinst;
u8 st0, st1, st2, st3;
int i, ch;
idx = nv_rd32(dev, 0x100c90);
......@@ -136,8 +233,8 @@ nv50_fb_vm_trap(struct drm_device *dev, int display, const char *name)
if (!display)
return;
/* lookup channel id */
chinst = (trap[2] << 16) | trap[1];
spin_lock_irqsave(&dev_priv->channels.lock, flags);
for (ch = 0; ch < dev_priv->engine.fifo.channels; ch++) {
struct nouveau_channel *chan = dev_priv->channels.ptr[ch];
......@@ -150,9 +247,48 @@ nv50_fb_vm_trap(struct drm_device *dev, int display, const char *name)
}
spin_unlock_irqrestore(&dev_priv->channels.lock, flags);
NV_INFO(dev, "%s - VM: Trapped %s at %02x%04x%04x status %08x "
"channel %d (0x%08x)\n",
name, (trap[5] & 0x100 ? "read" : "write"),
trap[5] & 0xff, trap[4] & 0xffff, trap[3] & 0xffff,
trap[0], ch, chinst);
/* decode status bits into something more useful */
if (dev_priv->chipset < 0xa3 ||
dev_priv->chipset == 0xaa || dev_priv->chipset == 0xac) {
st0 = (trap[0] & 0x0000000f) >> 0;
st1 = (trap[0] & 0x000000f0) >> 4;
st2 = (trap[0] & 0x00000f00) >> 8;
st3 = (trap[0] & 0x0000f000) >> 12;
} else {
st0 = (trap[0] & 0x000000ff) >> 0;
st1 = (trap[0] & 0x0000ff00) >> 8;
st2 = (trap[0] & 0x00ff0000) >> 16;
st3 = (trap[0] & 0xff000000) >> 24;
}
NV_INFO(dev, "VM: trapped %s at 0x%02x%04x%04x on ch %d [0x%08x] ",
(trap[5] & 0x00000100) ? "read" : "write",
trap[5] & 0xff, trap[4] & 0xffff, trap[3] & 0xffff, ch, chinst);
en = nouveau_enum_find(vm_engine, st0);
if (en)
printk("%s/", en->name);
else
printk("%02x/", st0);
cl = nouveau_enum_find(vm_client, st2);
if (cl)
printk("%s/", cl->name);
else
printk("%02x/", st2);
if (cl && cl->data) cl = nouveau_enum_find(cl->data, st3);
else if (en && en->data) cl = nouveau_enum_find(en->data, st3);
else cl = NULL;
if (cl)
printk("%s", cl->name);
else
printk("%02x", st3);
printk(" reason: ");
en = nouveau_enum_find(vm_fault, st1);
if (en)
printk("%s\n", en->name);
else
printk("0x%08x\n", st1);
}
......@@ -550,11 +550,11 @@ nv86_graph_tlb_flush(struct drm_device *dev)
static struct nouveau_enum nv50_mp_exec_error_names[] =
{
{ 3, "STACK_UNDERFLOW" },
{ 4, "QUADON_ACTIVE" },
{ 8, "TIMEOUT" },
{ 0x10, "INVALID_OPCODE" },
{ 0x40, "BREAKPOINT" },
{ 3, "STACK_UNDERFLOW", NULL },
{ 4, "QUADON_ACTIVE", NULL },
{ 8, "TIMEOUT", NULL },
{ 0x10, "INVALID_OPCODE", NULL },
{ 0x40, "BREAKPOINT", NULL },
{}
};
......@@ -582,47 +582,47 @@ static struct nouveau_bitfield nv50_graph_trap_ccache[] = {
/* There must be a *lot* of these. Will take some time to gather them up. */
struct nouveau_enum nv50_data_error_names[] = {
{ 0x00000003, "INVALID_QUERY_OR_TEXTURE" },
{ 0x00000004, "INVALID_VALUE" },
{ 0x00000005, "INVALID_ENUM" },
{ 0x00000008, "INVALID_OBJECT" },
{ 0x00000009, "READ_ONLY_OBJECT" },
{ 0x0000000a, "SUPERVISOR_OBJECT" },
{ 0x0000000b, "INVALID_ADDRESS_ALIGNMENT" },
{ 0x0000000c, "INVALID_BITFIELD" },
{ 0x0000000d, "BEGIN_END_ACTIVE" },
{ 0x0000000e, "SEMANTIC_COLOR_BACK_OVER_LIMIT" },
{ 0x0000000f, "VIEWPORT_ID_NEEDS_GP" },
{ 0x00000010, "RT_DOUBLE_BIND" },
{ 0x00000011, "RT_TYPES_MISMATCH" },
{ 0x00000012, "RT_LINEAR_WITH_ZETA" },
{ 0x00000015, "FP_TOO_FEW_REGS" },
{ 0x00000016, "ZETA_FORMAT_CSAA_MISMATCH" },
{ 0x00000017, "RT_LINEAR_WITH_MSAA" },
{ 0x00000018, "FP_INTERPOLANT_START_OVER_LIMIT" },
{ 0x00000019, "SEMANTIC_LAYER_OVER_LIMIT" },
{ 0x0000001a, "RT_INVALID_ALIGNMENT" },
{ 0x0000001b, "SAMPLER_OVER_LIMIT" },
{ 0x0000001c, "TEXTURE_OVER_LIMIT" },
{ 0x0000001e, "GP_TOO_MANY_OUTPUTS" },
{ 0x0000001f, "RT_BPP128_WITH_MS8" },
{ 0x00000021, "Z_OUT_OF_BOUNDS" },
{ 0x00000023, "XY_OUT_OF_BOUNDS" },
{ 0x00000027, "CP_MORE_PARAMS_THAN_SHARED" },
{ 0x00000028, "CP_NO_REG_SPACE_STRIPED" },
{ 0x00000029, "CP_NO_REG_SPACE_PACKED" },
{ 0x0000002a, "CP_NOT_ENOUGH_WARPS" },
{ 0x0000002b, "CP_BLOCK_SIZE_MISMATCH" },
{ 0x0000002c, "CP_NOT_ENOUGH_LOCAL_WARPS" },
{ 0x0000002d, "CP_NOT_ENOUGH_STACK_WARPS" },
{ 0x0000002e, "CP_NO_BLOCKDIM_LATCH" },
{ 0x00000031, "ENG2D_FORMAT_MISMATCH" },
{ 0x0000003f, "PRIMITIVE_ID_NEEDS_GP" },
{ 0x00000044, "SEMANTIC_VIEWPORT_OVER_LIMIT" },
{ 0x00000045, "SEMANTIC_COLOR_FRONT_OVER_LIMIT" },
{ 0x00000046, "LAYER_ID_NEEDS_GP" },
{ 0x00000047, "SEMANTIC_CLIP_OVER_LIMIT" },
{ 0x00000048, "SEMANTIC_PTSZ_OVER_LIMIT" },
{ 0x00000003, "INVALID_QUERY_OR_TEXTURE", NULL },
{ 0x00000004, "INVALID_VALUE", NULL },
{ 0x00000005, "INVALID_ENUM", NULL },
{ 0x00000008, "INVALID_OBJECT", NULL },
{ 0x00000009, "READ_ONLY_OBJECT", NULL },
{ 0x0000000a, "SUPERVISOR_OBJECT", NULL },
{ 0x0000000b, "INVALID_ADDRESS_ALIGNMENT", NULL },
{ 0x0000000c, "INVALID_BITFIELD", NULL },
{ 0x0000000d, "BEGIN_END_ACTIVE", NULL },
{ 0x0000000e, "SEMANTIC_COLOR_BACK_OVER_LIMIT", NULL },
{ 0x0000000f, "VIEWPORT_ID_NEEDS_GP", NULL },
{ 0x00000010, "RT_DOUBLE_BIND", NULL },
{ 0x00000011, "RT_TYPES_MISMATCH", NULL },
{ 0x00000012, "RT_LINEAR_WITH_ZETA", NULL },
{ 0x00000015, "FP_TOO_FEW_REGS", NULL },
{ 0x00000016, "ZETA_FORMAT_CSAA_MISMATCH", NULL },
{ 0x00000017, "RT_LINEAR_WITH_MSAA", NULL },
{ 0x00000018, "FP_INTERPOLANT_START_OVER_LIMIT", NULL },
{ 0x00000019, "SEMANTIC_LAYER_OVER_LIMIT", NULL },
{ 0x0000001a, "RT_INVALID_ALIGNMENT", NULL },
{ 0x0000001b, "SAMPLER_OVER_LIMIT", NULL },
{ 0x0000001c, "TEXTURE_OVER_LIMIT", NULL },
{ 0x0000001e, "GP_TOO_MANY_OUTPUTS", NULL },
{ 0x0000001f, "RT_BPP128_WITH_MS8", NULL },
{ 0x00000021, "Z_OUT_OF_BOUNDS", NULL },
{ 0x00000023, "XY_OUT_OF_BOUNDS", NULL },
{ 0x00000027, "CP_MORE_PARAMS_THAN_SHARED", NULL },
{ 0x00000028, "CP_NO_REG_SPACE_STRIPED", NULL },
{ 0x00000029, "CP_NO_REG_SPACE_PACKED", NULL },
{ 0x0000002a, "CP_NOT_ENOUGH_WARPS", NULL },
{ 0x0000002b, "CP_BLOCK_SIZE_MISMATCH", NULL },
{ 0x0000002c, "CP_NOT_ENOUGH_LOCAL_WARPS", NULL },
{ 0x0000002d, "CP_NOT_ENOUGH_STACK_WARPS", NULL },
{ 0x0000002e, "CP_NO_BLOCKDIM_LATCH", NULL },
{ 0x00000031, "ENG2D_FORMAT_MISMATCH", NULL },
{ 0x0000003f, "PRIMITIVE_ID_NEEDS_GP", NULL },
{ 0x00000044, "SEMANTIC_VIEWPORT_OVER_LIMIT", NULL },
{ 0x00000045, "SEMANTIC_COLOR_FRONT_OVER_LIMIT", NULL },
{ 0x00000046, "LAYER_ID_NEEDS_GP", NULL },
{ 0x00000047, "SEMANTIC_CLIP_OVER_LIMIT", NULL },
{ 0x00000048, "SEMANTIC_PTSZ_OVER_LIMIT", NULL },
{}
};
......@@ -702,7 +702,6 @@ nv50_pgraph_tp_trap(struct drm_device *dev, int type, uint32_t ustatus_old,
tps++;
switch (type) {
case 6: /* texture error... unknown for now */
nv50_fb_vm_trap(dev, display, name);
if (display) {
NV_ERROR(dev, "magic set %d:\n", i);
for (r = ustatus_addr + 4; r <= ustatus_addr + 0x10; r += 4)
......@@ -725,7 +724,6 @@ nv50_pgraph_tp_trap(struct drm_device *dev, int type, uint32_t ustatus_old,
uint32_t e1c = nv_rd32(dev, ustatus_addr + 0x14);
uint32_t e20 = nv_rd32(dev, ustatus_addr + 0x18);
uint32_t e24 = nv_rd32(dev, ustatus_addr + 0x1c);
nv50_fb_vm_trap(dev, display, name);
/* 2d engine destination */
if (ustatus & 0x00000010) {
if (display) {
......@@ -1068,6 +1066,7 @@ nv50_graph_isr(struct drm_device *dev)
NV_INFO(dev, "PGRAPH - ch %d (0x%010llx) subc %d "
"class 0x%04x mthd 0x%04x data 0x%08x\n",
chid, inst, subc, class, mthd, data);
nv50_fb_vm_trap(dev, 1);
}
}
......
......@@ -136,5 +136,5 @@ nv84_crypt_isr(struct drm_device *dev)
nv_wr32(dev, 0x102130, stat);
nv_wr32(dev, 0x10200c, 0x10);
nv50_fb_vm_trap(dev, show, "PCRYPT");
nv50_fb_vm_trap(dev, show);
}
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册