xiaomi-sm8450/android_kernel_xiaomi_sm8450
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

drm/i915/perf: add a parameter to control the size of OA buffer

Browse Source 
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.

v2: Simplify oa buffer size exponent selection (Chris)
    Reuse vma size field (Lionel)

v3: Restrict size opening parameter to values supported by HW (Chris)

v4: Drop out of date comment (Matt)
    Add debug message when buffer size is rejected (Matt)

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181023100707.31738-5-lionel.g.landwerlin@intel.com
This commit is contained in:
Lionel Landwerlin
2018-10-23 11:07:07 +01:00
parent 5728de2f4f
commit cd956bfcd0
4 changed files with 76 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
drivers/gpu/drm/i915/i915_drv.h
View File

@@ -2006,6 +2006,7 @@ struct drm_i915_private {
u32 last_ctx_id; u32 last_ctx_id;
int format; int format;
int format_size; int format_size;
int size_exponent;
/** /**
* Locks reads and writes to all head/tail state * Locks reads and writes to all head/tail state

99
drivers/gpu/drm/i915/i915_perf.c
View File

@@ -212,13 +212,7 @@
#include "i915_oa_icl.h" #include "i915_oa_icl.h"
#include "intel_lrc_reg.h" #include "intel_lrc_reg.h"
/* HW requires this to be a power of two, between 128k and 16M, though driver #define OA_TAKEN(tail, head) (((tail) - (head)) & (dev_priv->perf.oa.oa_buffer.vma->size - 1))
* 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))
/** /**
* DOC: OA Tail Pointer Race * DOC: OA Tail Pointer Race
@@ -361,6 +355,7 @@ struct perf_open_properties {
int oa_format; int oa_format;
bool oa_periodic; bool oa_periodic;
int oa_period_exponent; int oa_period_exponent;
u32 oa_buffer_size_exponent;
}; };
static void free_oa_config(struct drm_i915_private *dev_priv, static void free_oa_config(struct drm_i915_private *dev_priv,
@@ -523,7 +518,7 @@ static bool oa_buffer_check_unlocked(struct drm_i915_private *dev_priv)
* could put the tail out of bounds... * could put the tail out of bounds...
*/ */
if (hw_tail >= gtt_offset && if (hw_tail >= gtt_offset &&
hw_tail < (gtt_offset + OA_BUFFER_SIZE)) { hw_tail < (gtt_offset + dev_priv->perf.oa.oa_buffer.vma->size)) {
dev_priv->perf.oa.oa_buffer.tails[!aged_idx].offset = dev_priv->perf.oa.oa_buffer.tails[!aged_idx].offset =
aging_tail = hw_tail; aging_tail = hw_tail;
dev_priv->perf.oa.oa_buffer.aging_timestamp = now; dev_priv->perf.oa.oa_buffer.aging_timestamp = now;
@@ -652,7 +647,7 @@ static int gen8_append_oa_reports(struct i915_perf_stream *stream,
int report_size = dev_priv->perf.oa.oa_buffer.format_size; int report_size = dev_priv->perf.oa.oa_buffer.format_size;
u8 *oa_buf_base = dev_priv->perf.oa.oa_buffer.vaddr; 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 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.vma->size - 1);
size_t start_offset = *offset; size_t start_offset = *offset;
unsigned long flags; unsigned long flags;
unsigned int aged_tail_idx; unsigned int aged_tail_idx;
@@ -692,8 +687,8 @@ static int gen8_append_oa_reports(struct i915_perf_stream *stream,
* only be incremented by multiples of the report size (notably also * only be incremented by multiples of the report size (notably also
* all a power of two). * all a power of two).
*/ */
if (WARN_ONCE(head > OA_BUFFER_SIZE || head % report_size || if (WARN_ONCE(head > dev_priv->perf.oa.oa_buffer.vma->size || head % report_size ||
tail > OA_BUFFER_SIZE || tail % report_size, tail > dev_priv->perf.oa.oa_buffer.vma->size || tail % report_size,
"Inconsistent OA buffer pointers: head = %u, tail = %u\n", "Inconsistent OA buffer pointers: head = %u, tail = %u\n",
head, tail)) head, tail))
return -EIO; return -EIO;
@@ -716,7 +711,7 @@ static int gen8_append_oa_reports(struct i915_perf_stream *stream,
* here would imply a driver bug that would result * here would imply a driver bug that would result
* in an overrun. * in an overrun.
*/ */
if (WARN_ON((OA_BUFFER_SIZE - head) < report_size)) { if (WARN_ON((dev_priv->perf.oa.oa_buffer.vma->size - head) < report_size)) {
DRM_ERROR("Spurious OA head ptr: non-integral report offset\n"); DRM_ERROR("Spurious OA head ptr: non-integral report offset\n");
break; break;
} }
@@ -875,11 +870,6 @@ static int gen8_oa_read(struct i915_perf_stream *stream,
* automatically triggered reports in this condition and so we * automatically triggered reports in this condition and so we
* have to assume that old reports are now being trampled * have to assume that old reports are now being trampled
* over. * over.
*
* Considering how we don't currently give userspace control
* over the OA buffer size and always configure a large 16MB
* buffer, then a buffer overflow does anyway likely indicate
* that something has gone quite badly wrong.
*/ */
if (oastatus & GEN8_OASTATUS_OABUFFER_OVERFLOW) { if (oastatus & GEN8_OASTATUS_OABUFFER_OVERFLOW) {
ret = append_oa_status(stream, buf, count, offset, ret = append_oa_status(stream, buf, count, offset,
@@ -941,7 +931,7 @@ static int gen7_append_oa_reports(struct i915_perf_stream *stream,
int report_size = dev_priv->perf.oa.oa_buffer.format_size; int report_size = dev_priv->perf.oa.oa_buffer.format_size;
u8 *oa_buf_base = dev_priv->perf.oa.oa_buffer.vaddr; 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 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.vma->size - 1);
size_t start_offset = *offset; size_t start_offset = *offset;
unsigned long flags; unsigned long flags;
unsigned int aged_tail_idx; unsigned int aged_tail_idx;
@@ -978,8 +968,8 @@ static int gen7_append_oa_reports(struct i915_perf_stream *stream,
* only be incremented by multiples of the report size (notably also * only be incremented by multiples of the report size (notably also
* all a power of two). * all a power of two).
*/ */
if (WARN_ONCE(head > OA_BUFFER_SIZE || head % report_size || if (WARN_ONCE(head > dev_priv->perf.oa.oa_buffer.vma->size || head % report_size ||
tail > OA_BUFFER_SIZE || tail % report_size, tail > dev_priv->perf.oa.oa_buffer.vma->size || tail % report_size,
"Inconsistent OA buffer pointers: head = %u, tail = %u\n", "Inconsistent OA buffer pointers: head = %u, tail = %u\n",
head, tail)) head, tail))
return -EIO; return -EIO;
@@ -999,7 +989,7 @@ static int gen7_append_oa_reports(struct i915_perf_stream *stream,
* here would imply a driver bug that would result * here would imply a driver bug that would result
* in an overrun. * in an overrun.
*/ */
if (WARN_ON((OA_BUFFER_SIZE - head) < report_size)) { if (WARN_ON((dev_priv->perf.oa.oa_buffer.vma->size - head) < report_size)) {
DRM_ERROR("Spurious OA head ptr: non-integral report offset\n"); DRM_ERROR("Spurious OA head ptr: non-integral report offset\n");
break; break;
} }
@@ -1394,7 +1384,9 @@ static void gen7_init_oa_buffer(struct drm_i915_private *dev_priv)
I915_WRITE(GEN7_OABUFFER, gtt_offset); 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 - 17) <<
GEN7_OASTATUS1_BUFFER_SIZE_SHIFT)); /* tail */
/* Mark that we need updated tail pointers to read from... */ /* Mark that we need updated tail pointers to read from... */
dev_priv->perf.oa.oa_buffer.tails[0].offset = INVALID_TAIL_PTR; dev_priv->perf.oa.oa_buffer.tails[0].offset = INVALID_TAIL_PTR;
@@ -1419,7 +1411,8 @@ static void gen7_init_oa_buffer(struct drm_i915_private *dev_priv)
* the assumption that new reports are being written to zeroed * the assumption that new reports are being written to zeroed
* memory... * 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 /* Maybe make ->pollin per-stream state if we support multiple
* concurrent streams in the future. * concurrent streams in the future.
@@ -1449,7 +1442,9 @@ static void gen8_init_oa_buffer(struct drm_i915_private *dev_priv)
* bit." * bit."
*/ */
I915_WRITE(GEN8_OABUFFER, gtt_offset | I915_WRITE(GEN8_OABUFFER, gtt_offset |
OABUFFER_SIZE_16M | GEN8_OABUFFER_MEM_SELECT_GGTT); ((dev_priv->perf.oa.oa_buffer.size_exponent - 17) <<
GEN8_OABUFFER_BUFFER_SIZE_SHIFT) |
GEN8_OABUFFER_MEM_SELECT_GGTT);
I915_WRITE(GEN8_OATAILPTR, gtt_offset & GEN8_OATAILPTR_MASK); I915_WRITE(GEN8_OATAILPTR, gtt_offset & GEN8_OATAILPTR_MASK);
/* Mark that we need updated tail pointers to read from... */ /* Mark that we need updated tail pointers to read from... */
@@ -1477,7 +1472,8 @@ static void gen8_init_oa_buffer(struct drm_i915_private *dev_priv)
* the assumption that new reports are being written to zeroed * the assumption that new reports are being written to zeroed
* memory... * 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 * Maybe make ->pollin per-stream state if we support multiple
@@ -1486,23 +1482,24 @@ static void gen8_init_oa_buffer(struct drm_i915_private *dev_priv)
dev_priv->perf.oa.pollin = false; 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, int size_exponent)
{ {
struct drm_i915_gem_object *bo; struct drm_i915_gem_object *bo;
struct i915_vma *vma; struct i915_vma *vma;
size_t size = 1U << size_exponent;
int ret; int ret;
if (WARN_ON(dev_priv->perf.oa.oa_buffer.vma)) if (WARN_ON(dev_priv->perf.oa.oa_buffer.vma))
return -ENODEV; return -ENODEV;
if (WARN_ON(size < SZ_128K || size > SZ_16M))
return -EINVAL;
ret = i915_mutex_lock_interruptible(&dev_priv->drm); ret = i915_mutex_lock_interruptible(&dev_priv->drm);
if (ret) if (ret)
return ret; return ret;
BUILD_BUG_ON_NOT_POWER_OF_2(OA_BUFFER_SIZE); bo = i915_gem_object_create(dev_priv, size);
BUILD_BUG_ON(OA_BUFFER_SIZE < SZ_128K || OA_BUFFER_SIZE > SZ_16M);
bo = i915_gem_object_create(dev_priv, OA_BUFFER_SIZE);
if (IS_ERR(bo)) { if (IS_ERR(bo)) {
DRM_ERROR("Failed to allocate OA buffer\n"); DRM_ERROR("Failed to allocate OA buffer\n");
ret = PTR_ERR(bo); ret = PTR_ERR(bo);
@@ -1520,6 +1517,7 @@ static int alloc_oa_buffer(struct drm_i915_private *dev_priv)
goto err_unref; goto err_unref;
} }
dev_priv->perf.oa.oa_buffer.vma = vma; dev_priv->perf.oa.oa_buffer.vma = vma;
dev_priv->perf.oa.oa_buffer.size_exponent = size_exponent;
dev_priv->perf.oa.oa_buffer.vaddr = dev_priv->perf.oa.oa_buffer.vaddr =
i915_gem_object_pin_map(bo, I915_MAP_WB); i915_gem_object_pin_map(bo, I915_MAP_WB);
@@ -1528,9 +1526,10 @@ static int alloc_oa_buffer(struct drm_i915_private *dev_priv)
goto err_unpin; 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 = %llu\n",
i915_ggtt_offset(dev_priv->perf.oa.oa_buffer.vma), 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.vma->size);
goto unlock; goto unlock;
@@ -2090,7 +2089,7 @@ static int i915_oa_stream_init(struct i915_perf_stream *stream,
intel_runtime_pm_get(dev_priv); intel_runtime_pm_get(dev_priv);
intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
ret = alloc_oa_buffer(dev_priv); ret = alloc_oa_buffer(dev_priv, props->oa_buffer_size_exponent);
if (ret) if (ret)
goto err_oa_buf_alloc; goto err_oa_buf_alloc;
@@ -2649,6 +2648,26 @@ static u64 oa_exponent_to_ns(struct drm_i915_private *dev_priv, int exponent)
1000ULL * INTEL_INFO(dev_priv)->cs_timestamp_frequency_khz); 1000ULL * INTEL_INFO(dev_priv)->cs_timestamp_frequency_khz);
} }
static int
select_oa_buffer_exponent(struct drm_i915_private *i915,
u64 requested_size)
{
int order;
/*
* When no size is specified, use the largest size supported by all
* generations.
*/
if (!requested_size)
return order_base_2(SZ_16M);
order = order_base_2(clamp_t(u64, requested_size, SZ_128K, SZ_16M));
if (requested_size != (1UL << order))
return -EINVAL;
return order;
}
/** /**
* read_properties_unlocked - validate + copy userspace stream open properties * read_properties_unlocked - validate + copy userspace stream open properties
* @dev_priv: i915 device instance * @dev_priv: i915 device instance
@@ -2776,6 +2795,14 @@ static int read_properties_unlocked(struct drm_i915_private *dev_priv,
props->oa_periodic = true; props->oa_periodic = true;
props->oa_period_exponent = value; props->oa_period_exponent = value;
break; break;
case DRM_I915_PERF_PROP_OA_BUFFER_SIZE:
ret = select_oa_buffer_exponent(dev_priv, value);
if (ret < 0) {
DRM_DEBUG("OA buffer size invalid %llu\n", value);
return ret;
}
props->oa_buffer_size_exponent = ret;
break;
case DRM_I915_PERF_PROP_MAX: case DRM_I915_PERF_PROP_MAX:
MISSING_CASE(id); MISSING_CASE(id);
return -EINVAL; return -EINVAL;
@@ -2784,6 +2811,12 @@ static int read_properties_unlocked(struct drm_i915_private *dev_priv,
uprop += 2; uprop += 2;
} }
/* If no buffer size was requested, select the default one. */
if (!props->oa_buffer_size_exponent) {
props->oa_buffer_size_exponent =
select_oa_buffer_exponent(dev_priv, 0);
}
return 0; return 0;
} }

2
drivers/gpu/drm/i915/i915_reg.h
View File

@@ -569,12 +569,14 @@ static inline bool i915_mmio_reg_valid(i915_reg_t reg)
#define GEN8_OABUFFER_UDW _MMIO(0x23b4) #define GEN8_OABUFFER_UDW _MMIO(0x23b4)
#define GEN8_OABUFFER _MMIO(0x2b14) #define GEN8_OABUFFER _MMIO(0x2b14)
#define GEN8_OABUFFER_MEM_SELECT_GGTT (1 << 0) /* 0: PPGTT, 1: GGTT */ #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 _MMIO(0x2364)
#define GEN7_OASTATUS1_TAIL_MASK 0xffffffc0 #define GEN7_OASTATUS1_TAIL_MASK 0xffffffc0
#define GEN7_OASTATUS1_COUNTER_OVERFLOW (1 << 2) #define GEN7_OASTATUS1_COUNTER_OVERFLOW (1 << 2)
#define GEN7_OASTATUS1_OABUFFER_OVERFLOW (1 << 1) #define GEN7_OASTATUS1_OABUFFER_OVERFLOW (1 << 1)
#define GEN7_OASTATUS1_REPORT_LOST (1 << 0) #define GEN7_OASTATUS1_REPORT_LOST (1 << 0)
#define GEN7_OASTATUS1_BUFFER_SIZE_SHIFT 3
#define GEN7_OASTATUS2 _MMIO(0x2368) #define GEN7_OASTATUS2 _MMIO(0x2368)
#define GEN7_OASTATUS2_HEAD_MASK 0xffffffc0 #define GEN7_OASTATUS2_HEAD_MASK 0xffffffc0

7
include/uapi/drm/i915_drm.h
View File

@@ -1540,6 +1540,13 @@ enum drm_i915_perf_property_id {
*/ */
DRM_I915_PERF_PROP_OA_EXPONENT, DRM_I915_PERF_PROP_OA_EXPONENT,
/**
* Specify a global OA buffer size to be allocated in bytes. The size
* specified must be supported by HW (currently supported sizes are
* powers of 2 ranging from 128Kb to 16Mb).
*/
DRM_I915_PERF_PROP_OA_BUFFER_SIZE,
DRM_I915_PERF_PROP_MAX /* non-ABI */ DRM_I915_PERF_PROP_MAX /* non-ABI */
}; };