[Intel-gfx] [PATCH 4/4] drm/i915/perf: add a parameter to control the size of OA buffer
Lionel Landwerlin
lionel.g.landwerlin at intel.com
Wed Oct 10 16:55:33 UTC 2018
The way our hardware is designed doesn't seem to let us use the
MI_RECORD_PERF_COUNT command without setting up a circular buffer.
In the case where the user didn't request OA reports to be available
through the i915 perf stream, we can set the OA buffer to the minimum
size to avoid consuming memory which won't be used by the driver.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin at intel.com>
---
drivers/gpu/drm/i915/i915_drv.h | 2 +
drivers/gpu/drm/i915/i915_perf.c | 111 +++++++++++++++++++++++--------
drivers/gpu/drm/i915/i915_reg.h | 2 +
include/uapi/drm/i915_drm.h | 8 +++
4 files changed, 95 insertions(+), 28 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 65eaac2d7e3c..a0069add9d39 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -2053,6 +2053,8 @@ struct drm_i915_private {
u32 last_ctx_id;
int format;
int format_size;
+ u32 size;
+ int size_exponent;
/**
* Locks reads and writes to all head/tail state
diff --git a/drivers/gpu/drm/i915/i915_perf.c b/drivers/gpu/drm/i915/i915_perf.c
index a648ded97969..1eaef711160a 100644
--- a/drivers/gpu/drm/i915/i915_perf.c
+++ b/drivers/gpu/drm/i915/i915_perf.c
@@ -212,13 +212,7 @@
#include "i915_oa_icl.h"
#include "intel_lrc_reg.h"
-/* HW requires this to be a power of two, between 128k and 16M, though driver
- * is currently generally designed assuming the largest 16M size is used such
- * that the overflow cases are unlikely in normal operation.
- */
-#define OA_BUFFER_SIZE SZ_16M
-
-#define OA_TAKEN(tail, head) ((tail - head) & (OA_BUFFER_SIZE - 1))
+#define OA_TAKEN(tail, head) ((tail - head) & (dev_priv->perf.oa.oa_buffer.size - 1))
/**
* DOC: OA Tail Pointer Race
@@ -361,6 +355,8 @@ struct perf_open_properties {
int oa_format;
bool oa_periodic;
int oa_period_exponent;
+ u32 oa_buffer_size;
+ u32 oa_buffer_size_exponent;
};
static void free_oa_config(struct drm_i915_private *dev_priv,
@@ -523,7 +519,7 @@ static bool oa_buffer_check_unlocked(struct drm_i915_private *dev_priv)
* could put the tail out of bounds...
*/
if (hw_tail >= gtt_offset &&
- hw_tail < (gtt_offset + OA_BUFFER_SIZE)) {
+ hw_tail < (gtt_offset + dev_priv->perf.oa.oa_buffer.size)) {
dev_priv->perf.oa.oa_buffer.tails[!aged_idx].offset =
aging_tail = hw_tail;
dev_priv->perf.oa.oa_buffer.aging_timestamp = now;
@@ -652,7 +648,7 @@ static int gen8_append_oa_reports(struct i915_perf_stream *stream,
int report_size = dev_priv->perf.oa.oa_buffer.format_size;
u8 *oa_buf_base = dev_priv->perf.oa.oa_buffer.vaddr;
u32 gtt_offset = i915_ggtt_offset(dev_priv->perf.oa.oa_buffer.vma);
- u32 mask = (OA_BUFFER_SIZE - 1);
+ u32 mask = (dev_priv->perf.oa.oa_buffer.size - 1);
size_t start_offset = *offset;
unsigned long flags;
unsigned int aged_tail_idx;
@@ -692,8 +688,8 @@ static int gen8_append_oa_reports(struct i915_perf_stream *stream,
* only be incremented by multiples of the report size (notably also
* all a power of two).
*/
- if (WARN_ONCE(head > OA_BUFFER_SIZE || head % report_size ||
- tail > OA_BUFFER_SIZE || tail % report_size,
+ if (WARN_ONCE(head > dev_priv->perf.oa.oa_buffer.size || head % report_size ||
+ tail > dev_priv->perf.oa.oa_buffer.size || tail % report_size,
"Inconsistent OA buffer pointers: head = %u, tail = %u\n",
head, tail))
return -EIO;
@@ -716,7 +712,7 @@ static int gen8_append_oa_reports(struct i915_perf_stream *stream,
* here would imply a driver bug that would result
* in an overrun.
*/
- if (WARN_ON((OA_BUFFER_SIZE - head) < report_size)) {
+ if (WARN_ON((dev_priv->perf.oa.oa_buffer.size - head) < report_size)) {
DRM_ERROR("Spurious OA head ptr: non-integral report offset\n");
break;
}
@@ -941,7 +937,7 @@ static int gen7_append_oa_reports(struct i915_perf_stream *stream,
int report_size = dev_priv->perf.oa.oa_buffer.format_size;
u8 *oa_buf_base = dev_priv->perf.oa.oa_buffer.vaddr;
u32 gtt_offset = i915_ggtt_offset(dev_priv->perf.oa.oa_buffer.vma);
- u32 mask = (OA_BUFFER_SIZE - 1);
+ u32 mask = (dev_priv->perf.oa.oa_buffer.size - 1);
size_t start_offset = *offset;
unsigned long flags;
unsigned int aged_tail_idx;
@@ -978,8 +974,8 @@ static int gen7_append_oa_reports(struct i915_perf_stream *stream,
* only be incremented by multiples of the report size (notably also
* all a power of two).
*/
- if (WARN_ONCE(head > OA_BUFFER_SIZE || head % report_size ||
- tail > OA_BUFFER_SIZE || tail % report_size,
+ if (WARN_ONCE(head > dev_priv->perf.oa.oa_buffer.size || head % report_size ||
+ tail > dev_priv->perf.oa.oa_buffer.size || tail % report_size,
"Inconsistent OA buffer pointers: head = %u, tail = %u\n",
head, tail))
return -EIO;
@@ -999,7 +995,7 @@ static int gen7_append_oa_reports(struct i915_perf_stream *stream,
* here would imply a driver bug that would result
* in an overrun.
*/
- if (WARN_ON((OA_BUFFER_SIZE - head) < report_size)) {
+ if (WARN_ON((dev_priv->perf.oa.oa_buffer.size - head) < report_size)) {
DRM_ERROR("Spurious OA head ptr: non-integral report offset\n");
break;
}
@@ -1396,7 +1392,9 @@ static void gen7_init_oa_buffer(struct drm_i915_private *dev_priv)
I915_WRITE(GEN7_OABUFFER, gtt_offset);
- I915_WRITE(GEN7_OASTATUS1, gtt_offset | OABUFFER_SIZE_16M); /* tail */
+ I915_WRITE(GEN7_OASTATUS1, gtt_offset |
+ (dev_priv->perf.oa.oa_buffer.size_exponent <<
+ GEN7_OASTATUS1_BUFFER_SIZE_SHIFT)); /* tail */
/* Mark that we need updated tail pointers to read from... */
dev_priv->perf.oa.oa_buffer.tails[0].offset = INVALID_TAIL_PTR;
@@ -1421,7 +1419,8 @@ static void gen7_init_oa_buffer(struct drm_i915_private *dev_priv)
* the assumption that new reports are being written to zeroed
* memory...
*/
- memset(dev_priv->perf.oa.oa_buffer.vaddr, 0, OA_BUFFER_SIZE);
+ memset(dev_priv->perf.oa.oa_buffer.vaddr, 0,
+ dev_priv->perf.oa.oa_buffer.vma->size);
/* Maybe make ->pollin per-stream state if we support multiple
* concurrent streams in the future.
@@ -1451,7 +1450,9 @@ static void gen8_init_oa_buffer(struct drm_i915_private *dev_priv)
* bit."
*/
I915_WRITE(GEN8_OABUFFER, gtt_offset |
- OABUFFER_SIZE_16M | GEN8_OABUFFER_MEM_SELECT_GGTT);
+ (dev_priv->perf.oa.oa_buffer.size_exponent <<
+ GEN8_OABUFFER_BUFFER_SIZE_SHIFT) |
+ GEN8_OABUFFER_MEM_SELECT_GGTT);
I915_WRITE(GEN8_OATAILPTR, gtt_offset & GEN8_OATAILPTR_MASK);
/* Mark that we need updated tail pointers to read from... */
@@ -1479,7 +1480,8 @@ static void gen8_init_oa_buffer(struct drm_i915_private *dev_priv)
* the assumption that new reports are being written to zeroed
* memory...
*/
- memset(dev_priv->perf.oa.oa_buffer.vaddr, 0, OA_BUFFER_SIZE);
+ memset(dev_priv->perf.oa.oa_buffer.vaddr, 0,
+ dev_priv->perf.oa.oa_buffer.vma->size);
/*
* Maybe make ->pollin per-stream state if we support multiple
@@ -1488,7 +1490,8 @@ static void gen8_init_oa_buffer(struct drm_i915_private *dev_priv)
dev_priv->perf.oa.pollin = false;
}
-static int alloc_oa_buffer(struct drm_i915_private *dev_priv)
+static int alloc_oa_buffer(struct drm_i915_private *dev_priv, u32 size,
+ int size_exponent)
{
struct drm_i915_gem_object *bo;
struct i915_vma *vma;
@@ -1503,10 +1506,9 @@ static int alloc_oa_buffer(struct drm_i915_private *dev_priv)
goto unlock;
}
- BUILD_BUG_ON_NOT_POWER_OF_2(OA_BUFFER_SIZE);
- BUILD_BUG_ON(OA_BUFFER_SIZE < SZ_128K || OA_BUFFER_SIZE > SZ_16M);
+ BUG_ON(size < SZ_128K || size > SZ_16M);
- bo = i915_gem_object_create(dev_priv, OA_BUFFER_SIZE);
+ bo = i915_gem_object_create(dev_priv, size);
if (IS_ERR(bo)) {
DRM_ERROR("Failed to allocate OA buffer\n");
ret = PTR_ERR(bo);
@@ -1518,12 +1520,14 @@ static int alloc_oa_buffer(struct drm_i915_private *dev_priv)
goto err_unref;
/* PreHSW required 512K alignment, HSW requires 16M */
- vma = i915_gem_object_ggtt_pin(bo, NULL, 0, SZ_16M, 0);
+ vma = i915_gem_object_ggtt_pin(bo, NULL, 0, size, 0);
if (IS_ERR(vma)) {
ret = PTR_ERR(vma);
goto err_unref;
}
dev_priv->perf.oa.oa_buffer.vma = vma;
+ dev_priv->perf.oa.oa_buffer.size = size;
+ dev_priv->perf.oa.oa_buffer.size_exponent = size_exponent;
dev_priv->perf.oa.oa_buffer.vaddr =
i915_gem_object_pin_map(bo, I915_MAP_WB);
@@ -1532,9 +1536,10 @@ static int alloc_oa_buffer(struct drm_i915_private *dev_priv)
goto err_unpin;
}
- DRM_DEBUG_DRIVER("OA Buffer initialized, gtt offset = 0x%x, vaddr = %p\n",
+ DRM_DEBUG_DRIVER("OA Buffer initialized, gtt offset = 0x%x, vaddr = %p, size = %u\n",
i915_ggtt_offset(dev_priv->perf.oa.oa_buffer.vma),
- dev_priv->perf.oa.oa_buffer.vaddr);
+ dev_priv->perf.oa.oa_buffer.vaddr,
+ dev_priv->perf.oa.oa_buffer.size);
goto unlock;
@@ -2094,7 +2099,8 @@ static int i915_oa_stream_init(struct i915_perf_stream *stream,
intel_runtime_pm_get(dev_priv);
intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
- ret = alloc_oa_buffer(dev_priv);
+ ret = alloc_oa_buffer(dev_priv, props->oa_buffer_size,
+ props->oa_buffer_size_exponent);
if (ret)
goto err_oa_buf_alloc;
@@ -2653,6 +2659,39 @@ static u64 oa_exponent_to_ns(struct drm_i915_private *dev_priv, int exponent)
1000ULL * INTEL_INFO(dev_priv)->cs_timestamp_frequency_khz);
}
+static int
+select_oa_buffer_size(struct drm_i915_private *i915,
+ u64 requested_size,
+ u32 *selected_size,
+ int *selected_exponent)
+{
+ const u32 max_size = SZ_16M;
+
+ /*
+ * When no size is specified, use the largest size supported
+ * by all generations.
+ */
+ if (!requested_size) {
+ *selected_size = SZ_16M;
+ *selected_exponent = 7;
+ return 0;
+ }
+
+ /* Start with the smallest OA buffer size. */
+ *selected_size = 128 * 1024;
+ *selected_exponent = 0;
+ while (requested_size > *selected_size &&
+ *selected_size < max_size) {
+ *selected_size *= 2;
+ *selected_exponent += 1;
+ }
+
+ if (requested_size > *selected_size)
+ return -EINVAL; /* TODO: ENOMEM? ENODEV? */
+
+ return 0;
+}
+
/**
* read_properties_unlocked - validate + copy userspace stream open properties
* @dev_priv: i915 device instance
@@ -2780,6 +2819,13 @@ static int read_properties_unlocked(struct drm_i915_private *dev_priv,
props->oa_periodic = true;
props->oa_period_exponent = value;
break;
+ case DRM_I915_PERF_PROP_OA_BUFFER_SIZE:
+ ret = select_oa_buffer_size(dev_priv, value,
+ &props->oa_buffer_size,
+ &props->oa_buffer_size_exponent);
+ if (ret)
+ return ret;
+ break;
case DRM_I915_PERF_PROP_MAX:
MISSING_CASE(id);
return -EINVAL;
@@ -2788,6 +2834,15 @@ static int read_properties_unlocked(struct drm_i915_private *dev_priv,
uprop += 2;
}
+ /* If no buffer size was requested, select the default one. */
+ if (!props->oa_buffer_size) {
+ if (select_oa_buffer_size(dev_priv, 0, &props->oa_buffer_size,
+ &props->oa_buffer_size_exponent)) {
+ DRM_ERROR("Unable to select default OA buffer size\n");
+ return -EINVAL;
+ }
+ }
+
return 0;
}
diff --git a/drivers/gpu/drm/i915/i915_reg.h b/drivers/gpu/drm/i915/i915_reg.h
index 3bbf21ea5c57..e9fc5e4f6d87 100644
--- a/drivers/gpu/drm/i915/i915_reg.h
+++ b/drivers/gpu/drm/i915/i915_reg.h
@@ -595,12 +595,14 @@ static inline bool i915_mmio_reg_valid(i915_reg_t reg)
#define GEN8_OABUFFER_UDW _MMIO(0x23b4)
#define GEN8_OABUFFER _MMIO(0x2b14)
#define GEN8_OABUFFER_MEM_SELECT_GGTT (1 << 0) /* 0: PPGTT, 1: GGTT */
+#define GEN8_OABUFFER_BUFFER_SIZE_SHIFT 3
#define GEN7_OASTATUS1 _MMIO(0x2364)
#define GEN7_OASTATUS1_TAIL_MASK 0xffffffc0
#define GEN7_OASTATUS1_COUNTER_OVERFLOW (1 << 2)
#define GEN7_OASTATUS1_OABUFFER_OVERFLOW (1 << 1)
#define GEN7_OASTATUS1_REPORT_LOST (1 << 0)
+#define GEN7_OASTATUS1_BUFFER_SIZE_SHIFT 3
#define GEN7_OASTATUS2 _MMIO(0x2368)
#define GEN7_OASTATUS2_HEAD_MASK 0xffffffc0
diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h
index f22c4ee82871..b51a5777f627 100644
--- a/include/uapi/drm/i915_drm.h
+++ b/include/uapi/drm/i915_drm.h
@@ -1558,6 +1558,14 @@ enum drm_i915_perf_property_id {
*/
DRM_I915_PERF_PROP_OA_EXPONENT,
+ /**
+ * Specify a global OA buffer size to be allocated in bytes.
+ * The driver will allocate a HW supported size that is at
+ * least as large as specified by this property. Larger sizes
+ * than what the HW supports will fail.
+ */
+ DRM_I915_PERF_PROP_OA_BUFFER_SIZE,
+
DRM_I915_PERF_PROP_MAX /* non-ABI */
};
--
2.19.1
More information about the Intel-gfx
mailing list