On 07/13/2016 06:04 PM, Deak, Imre wrote:
Hi Yakui,
thanks for taking a look at these, see my comment below.
On ke, 2016-07-13 at 10:22 +0800, Zhao Yakui wrote:
On 07/01/2016 09:40 PM, Deak, Imre wrote:
The purpose for each MOCS entry isn't well defined atm. Defining these
is important to remove any uncertainty about the use of these entries
for example in terms of performance and GPU/CPU coherency.
Suggested by Ville.
CC: Rong R Yang<rong.r.yang@xxxxxxxxx>
CC: Yakui Zhao<yakui.zhao@xxxxxxxxx>
CC: Ville Syrjälä<ville.syrjala@xxxxxxxxxxxxxxx>
CC: Chris Wilson<chris@xxxxxxxxxxxxxxxxxx>
Signed-off-by: Imre Deak<imre.deak@xxxxxxxxx>
This looks readable and meaningful after giving proper names to MOCS
entry index.
But not sure whether the comment of I915_MOCS_CACHE has one typo?
---
drivers/gpu/drm/i915/intel_mocs.c | 13 +++++++------
include/uapi/drm/i915_drm.h | 24 ++++++++++++++++++++++++
2 files changed, 31 insertions(+), 6 deletions(-)
diff --git a/drivers/gpu/drm/i915/intel_mocs.c b/drivers/gpu/drm/i915/intel_mocs.c
index 927825f..86adc11 100644
--- a/drivers/gpu/drm/i915/intel_mocs.c
+++ b/drivers/gpu/drm/i915/intel_mocs.c
@@ -97,7 +97,8 @@ struct drm_i915_mocs_table {
* end.
*/
static const struct drm_i915_mocs_entry skylake_mocs_table[] = {
- { /* 0x00000009 */
+ [I915_MOCS_UNCACHED] = {
+ /* 0x00000009 */
.control_value = LE_CACHEABILITY(LE_UC) |
LE_TGT_CACHE(LE_TC_LLC_ELLC) |
LE_LRUM(0) | LE_AOM(0) | LE_RSC(0) | LE_SCC(0) |
@@ -106,7 +107,7 @@ static const struct drm_i915_mocs_entry skylake_mocs_table[] = {
/* 0x0010 */
.l3cc_value = L3_ESC(0) | L3_SCC(0) | L3_CACHEABILITY(L3_UC),
},
- {
+ [I915_MOCS_AUTO] = {
/* 0x00000038 */
.control_value = LE_CACHEABILITY(LE_PAGETABLE) |
LE_TGT_CACHE(LE_TC_LLC_ELLC) |
@@ -115,7 +116,7 @@ static const struct drm_i915_mocs_entry skylake_mocs_table[] = {
/* 0x0030 */
.l3cc_value = L3_ESC(0) | L3_SCC(0) | L3_CACHEABILITY(L3_WB),
},
- {
+ [I915_MOCS_CACHED] = {
/* 0x0000003b */
.control_value = LE_CACHEABILITY(LE_WB) |
LE_TGT_CACHE(LE_TC_LLC_ELLC) |
@@ -128,7 +129,7 @@ static const struct drm_i915_mocs_entry skylake_mocs_table[] = {
/* NOTE: the LE_TGT_CACHE is not used on Broxton */
static const struct drm_i915_mocs_entry broxton_mocs_table[] = {
- {
+ [I915_MOCS_UNCACHED] = {
/* 0x00000009 */
.control_value = LE_CACHEABILITY(LE_UC) |
LE_TGT_CACHE(LE_TC_LLC_ELLC) |
@@ -138,7 +139,7 @@ static const struct drm_i915_mocs_entry broxton_mocs_table[] = {
/* 0x0010 */
.l3cc_value = L3_ESC(0) | L3_SCC(0) | L3_CACHEABILITY(L3_UC),
},
- {
+ [I915_MOCS_AUTO] = {
/* 0x00000038 */
.control_value = LE_CACHEABILITY(LE_PAGETABLE) |
LE_TGT_CACHE(LE_TC_LLC_ELLC) |
@@ -148,7 +149,7 @@ static const struct drm_i915_mocs_entry broxton_mocs_table[] = {
/* 0x0030 */
.l3cc_value = L3_ESC(0) | L3_SCC(0) | L3_CACHEABILITY(L3_WB),
},
- {
+ [I915_MOCS_CACHED] = {
/* 0x00000039 */
.control_value = LE_CACHEABILITY(LE_UC) |
LE_TGT_CACHE(LE_TC_LLC_ELLC) |
diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h
index c17d63d..a5d116f 100644
--- a/include/uapi/drm/i915_drm.h
+++ b/include/uapi/drm/i915_drm.h
@@ -62,6 +62,30 @@ extern "C" {
#define I915_ERROR_UEVENT "ERROR"
#define I915_RESET_UEVENT "RESET"
+/*
+ * MOCS indexes used for GPU surfaces, defining the cacheability of the
+ * surface data and the coherency for this data wrt. CPU vs. GPU accesses.
+ */
+enum i915_mocs_table_index {
+ /*
+ * Not cached anywhere, coherency between CPU and GPU accesses is
+ * guaranteed.
+ */
+ I915_MOCS_UNCACHED,
+ /*
+ * Cacheability and coherency controlled by the kernel automatically
+ * based on the DRM_I915_GEM_SET_CACHING IOCTL setting and the current
+ * usage of the surface (used for display scanout or not).
+ */
+ I915_MOCS_AUTO,
+ /*
+ * Cached in all GPU caches available on the platform.
+ * Coherency between CPU and GPU accesses to the surface is not
+ * guaranteed without extra synchronization.
+ */
IMO the coherency is guaranteed without extra synchronization for the
MOCS_CACHED.
No. On BXT it will make the data cached in GPU caches but will not keep
the data coherent between GPU and CPU without extra synchronization.
For that we would need to enable snooping, but that has considerable
overhead, so we turn that off in patch 2/3. On SKL using this entry
happens to give you a coherent mapping, but that's just because the HW
doesn't allow us to turn off snooping on that platform (supposedly
because there snooping doesn't have a considerable overhead thanks to
LLC).
thanks for the detailed explanation.
Now it is clear to me.
Thanks
Yakui
--Imre
+ I915_MOCS_CACHED,
+};
+
/* Each region is a minimum of 16k, and there are at most 255 of them.
*/
#define I915_NR_TEX_REGIONS 255 /* table size 2k - maximum due to use
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
