[Intel-gfx] [PATCH v10 12/15] drm/i915/perf: Add OA unit support for Gen 8+
Chris Wilson
chris at chris-wilson.co.uk
Wed May 3 10:31:06 UTC 2017
On Tue, May 02, 2017 at 02:44:59PM -0700, Lionel Landwerlin wrote:
> From: Robert Bragg <robert at sixbynine.org>
>
> Enables access to OA unit metrics for BDW, CHV, SKL and BXT which all
> share (more-or-less) the same OA unit design.
>
> Of particular note in comparison to Haswell: some OA unit HW config
> state has become per-context state and as a consequence it is somewhat
> more complicated to manage synchronous state changes from the cpu while
> there's no guarantee of what context (if any) is currently actively
> running on the gpu.
>
> The periodic sampling frequency which can be particularly useful for
> system-wide analysis (as opposed to command stream synchronised
> MI_REPORT_PERF_COUNT commands) is perhaps the most surprising state to
> have become per-context save and restored (while the OABUFFER
> destination is still a shared, system-wide resource).
>
> This support for gen8+ takes care to consider a number of timing
> challenges involved in synchronously updating per-context state
> primarily by programming all config state from the cpu and updating all
> current and saved contexts synchronously while the OA unit is still
> disabled.
>
> The driver intentionally avoids depending on command streamer
> programming to update OA state considering the lack of synchronization
> between the automatic loading of OACTXCONTROL state (that includes the
> periodic sampling state and enable state) on context restore and the
> parsing of any general purpose BB the driver can control. I.e. this
> implementation is careful to avoid the possibility of a context restore
> temporarily enabling any out-of-date periodic sampling state. In
> addition to the risk of transiently-out-of-date state being loaded
> automatically; there are also internal HW latencies involved in the
> loading of MUX configurations which would be difficult to account for
> from the command streamer (and we only want to enable the unit when once
> the MUX configuration is complete).
>
> Since the Gen8+ OA unit design no longer supports clock gating the unit
> off for a single given context (which effectively stopped any progress
> of counters while any other context was running) and instead supports
> tagging OA reports with a context ID for filtering on the CPU, it means
> we can no longer hide the system-wide progress of counters from a
> non-privileged application only interested in metrics for its own
> context. Although we could theoretically try and subtract the progress
> of other contexts before forwarding reports via read() we aren't in a
> position to filter reports captured via MI_REPORT_PERF_COUNT commands.
> As a result, for Gen8+, we always require the
> dev.i915.perf_stream_paranoid to be unset for any access to OA metrics
> if not root.
>
> v5: Drain submitted requests when enabling metric set to ensure no
> lite-restore erases the context image we just updated (Lionel)
>
> v6: In addition to drain, switch to kernel context & update all
> context in place (Chris)
>
> v7: Add missing mutex_unlock() if switching to kernel context fails
> (Matthew)
>
> v8: Simplify OA period/flex-eu-counters programming by using the
> batchbuffer instead of modifying ctx-image (Lionel)
>
> v9: Back to updating the context image (due to erroneous testing,
> batchbuffer programming the OA unit doesn't actually work)
> (Lionel)
> Pin context before updating context image (Chris)
> Drop MMIO programming now that we switch to a kernel context with
> right values in initial context image (Chris)
>
> v10: Just pin_map the contexts we want to modify or let the
> configuration happen on first use (Chris)
>
> Signed-off-by: Robert Bragg <robert at sixbynine.org>
> Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin at intel.com>
> Reviewed-by: Matthew Auld <matthew.auld at intel.com> \o/
> ---
> drivers/gpu/drm/i915/i915_drv.h | 45 +-
> drivers/gpu/drm/i915/i915_perf.c | 889 +++++++++++++++++++++++++++++++++++----
> drivers/gpu/drm/i915/i915_reg.h | 22 +
> drivers/gpu/drm/i915/intel_lrc.c | 2 +
> include/uapi/drm/i915_drm.h | 19 +-
> 5 files changed, 884 insertions(+), 93 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> index f15751534c29..eaf4ce5e5790 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -2063,9 +2063,17 @@ struct i915_oa_ops {
> void (*init_oa_buffer)(struct drm_i915_private *dev_priv);
>
> /**
> - * @enable_metric_set: Applies any MUX configuration to set up the
> - * Boolean and Custom (B/C) counters that are part of the counter
> - * reports being sampled. May apply system constraints such as
> + * @select_metric_set: The auto generated code that checks whether a
> + * requested OA config is applicable to the system and if so sets up
> + * the mux, oa and flex eu register config pointers according to the
> + * current dev_priv->perf.oa.metrics_set.
> + */
> + int (*select_metric_set)(struct drm_i915_private *dev_priv);
> +
> + /**
> + * @enable_metric_set: Selects and applies any MUX configuration to set
> + * up the Boolean and Custom (B/C) counters that are part of the
> + * counter reports being sampled. May apply system constraints such as
> * disabling EU clock gating as required.
> */
> int (*enable_metric_set)(struct drm_i915_private *dev_priv);
> @@ -2096,20 +2104,13 @@ struct i915_oa_ops {
> size_t *offset);
>
> /**
> - * @oa_buffer_check: Check for OA buffer data + update tail
> - *
> - * This is either called via fops or the poll check hrtimer (atomic
> - * ctx) without any locks taken.
> + * @oa_hw_tail_read: read the OA tail pointer register
> *
> - * It's safe to read OA config state here unlocked, assuming that this
> - * is only called while the stream is enabled, while the global OA
> - * configuration can't be modified.
> - *
> - * Efficiency is more important than avoiding some false positives
> - * here, which will be handled gracefully - likely resulting in an
> - * %EAGAIN error for userspace.
> + * In particular this enables us to share all the fiddly code for
> + * handling the OA unit tail pointer race that affects multiple
> + * generations.
> */
> - bool (*oa_buffer_check)(struct drm_i915_private *dev_priv);
> + u32 (*oa_hw_tail_read)(struct drm_i915_private *dev_priv);
> };
>
> struct intel_cdclk_state {
> @@ -2470,6 +2471,7 @@ struct drm_i915_private {
> struct {
> struct i915_vma *vma;
> u8 *vaddr;
> + u32 last_ctx_id;
> int format;
> int format_size;
>
> @@ -2539,6 +2541,14 @@ struct drm_i915_private {
> } oa_buffer;
>
> u32 gen7_latched_oastatus1;
> + u32 ctx_oactxctrl_offset;
> + u32 ctx_flexeu0_offset;
> +
> + /* The RPT_ID/reason field for Gen8+ includes a bit
> + * to determine if the CTX ID in the report is valid
> + * but the specific bit differs between Gen 8 and 9
> + */
> + u32 gen8_valid_ctx_bit;
>
> struct i915_oa_ops ops;
> const struct i915_oa_format *oa_formats;
> @@ -2849,6 +2859,8 @@ intel_info(const struct drm_i915_private *dev_priv)
> #define IS_KBL_ULX(dev_priv) (INTEL_DEVID(dev_priv) == 0x590E || \
> INTEL_DEVID(dev_priv) == 0x5915 || \
> INTEL_DEVID(dev_priv) == 0x591E)
> +#define IS_SKL_GT2(dev_priv) (IS_SKYLAKE(dev_priv) && \
> + (INTEL_DEVID(dev_priv) & 0x00F0) == 0x0010)
> #define IS_SKL_GT3(dev_priv) (IS_SKYLAKE(dev_priv) && \
> (INTEL_DEVID(dev_priv) & 0x00F0) == 0x0020)
> #define IS_SKL_GT4(dev_priv) (IS_SKYLAKE(dev_priv) && \
> @@ -3612,6 +3624,9 @@ i915_gem_context_lookup_timeline(struct i915_gem_context *ctx,
>
> int i915_perf_open_ioctl(struct drm_device *dev, void *data,
> struct drm_file *file);
> +void i915_oa_init_reg_state(struct intel_engine_cs *engine,
> + struct i915_gem_context *ctx,
> + uint32_t *reg_state);
>
> /* i915_gem_evict.c */
> int __must_check i915_gem_evict_something(struct i915_address_space *vm,
> diff --git a/drivers/gpu/drm/i915/i915_perf.c b/drivers/gpu/drm/i915/i915_perf.c
> index 3277a52ce98e..7a82681d122e 100644
> --- a/drivers/gpu/drm/i915/i915_perf.c
> +++ b/drivers/gpu/drm/i915/i915_perf.c
> @@ -196,6 +196,12 @@
>
> #include "i915_drv.h"
> #include "i915_oa_hsw.h"
> +#include "i915_oa_bdw.h"
> +#include "i915_oa_chv.h"
> +#include "i915_oa_sklgt2.h"
> +#include "i915_oa_sklgt3.h"
> +#include "i915_oa_sklgt4.h"
> +#include "i915_oa_bxt.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
> @@ -215,7 +221,7 @@
> *
> * Although this can be observed explicitly while copying reports to userspace
> * by checking for a zeroed report-id field in tail reports, we want to account
> - * for this earlier, as part of the _oa_buffer_check to avoid lots of redundant
> + * for this earlier, as part of the oa_buffer_check to avoid lots of redundant
> * read() attempts.
> *
> * In effect we define a tail pointer for reading that lags the real tail
> @@ -237,7 +243,7 @@
> * indicates that an updated tail pointer is needed.
> *
> * Most of the implementation details for this workaround are in
> - * gen7_oa_buffer_check_unlocked() and gen7_appand_oa_reports()
> + * oa_buffer_check_unlocked() and _append_oa_reports()
> *
> * Note for posterity: previously the driver used to define an effective tail
> * pointer that lagged the real pointer by a 'tail margin' measured in bytes
> @@ -272,6 +278,13 @@ static u32 i915_perf_stream_paranoid = true;
>
> #define INVALID_CTX_ID 0xffffffff
>
> +/* On Gen8+ automatically triggered OA reports include a 'reason' field... */
> +#define OAREPORT_REASON_MASK 0x3f
> +#define OAREPORT_REASON_SHIFT 19
> +#define OAREPORT_REASON_TIMER (1<<0)
> +#define OAREPORT_REASON_CTX_SWITCH (1<<3)
> +#define OAREPORT_REASON_CLK_RATIO (1<<5)
> +
>
> /* For sysctl proc_dointvec_minmax of i915_oa_max_sample_rate
> *
> @@ -303,6 +316,13 @@ static struct i915_oa_format hsw_oa_formats[I915_OA_FORMAT_MAX] = {
> [I915_OA_FORMAT_C4_B8] = { 7, 64 },
> };
>
> +static struct i915_oa_format gen8_plus_oa_formats[I915_OA_FORMAT_MAX] = {
> + [I915_OA_FORMAT_A12] = { 0, 64 },
> + [I915_OA_FORMAT_A12_B8_C8] = { 2, 128 },
> + [I915_OA_FORMAT_A32u40_A4u32_B8_C8] = { 5, 256 },
> + [I915_OA_FORMAT_C4_B8] = { 7, 64 },
> +};
> +
> #define SAMPLE_OA_REPORT (1<<0)
>
> /**
> @@ -332,8 +352,20 @@ struct perf_open_properties {
> int oa_period_exponent;
> };
>
> +static u32 gen8_oa_hw_tail_read(struct drm_i915_private *dev_priv)
> +{
> + return I915_READ(GEN8_OATAILPTR) & GEN8_OATAILPTR_MASK;
> +}
> +
> +static u32 gen7_oa_hw_tail_read(struct drm_i915_private *dev_priv)
> +{
> + u32 oastatus1 = I915_READ(GEN7_OASTATUS1);
> +
> + return oastatus1 & GEN7_OASTATUS1_TAIL_MASK;
> +}
> +
> /**
> - * gen7_oa_buffer_check_unlocked - check for data and update tail ptr state
> + * oa_buffer_check_unlocked - check for data and update tail ptr state
> * @dev_priv: i915 device instance
> *
> * This is either called via fops (for blocking reads in user ctx) or the poll
> @@ -356,12 +388,11 @@ struct perf_open_properties {
> *
> * Returns: %true if the OA buffer contains data, else %false
> */
> -static bool gen7_oa_buffer_check_unlocked(struct drm_i915_private *dev_priv)
> +static bool oa_buffer_check_unlocked(struct drm_i915_private *dev_priv)
> {
> int report_size = dev_priv->perf.oa.oa_buffer.format_size;
> unsigned long flags;
> unsigned int aged_idx;
> - u32 oastatus1;
> u32 head, hw_tail, aged_tail, aging_tail;
> u64 now;
>
> @@ -381,8 +412,7 @@ static bool gen7_oa_buffer_check_unlocked(struct drm_i915_private *dev_priv)
> aged_tail = dev_priv->perf.oa.oa_buffer.tails[aged_idx].offset;
> aging_tail = dev_priv->perf.oa.oa_buffer.tails[!aged_idx].offset;
>
> - oastatus1 = I915_READ(GEN7_OASTATUS1);
> - hw_tail = oastatus1 & GEN7_OASTATUS1_TAIL_MASK;
> + hw_tail = dev_priv->perf.oa.ops.oa_hw_tail_read(dev_priv);
>
> /* The tail pointer increases in 64 byte increments,
> * not in report_size steps...
> @@ -404,6 +434,7 @@ static bool gen7_oa_buffer_check_unlocked(struct drm_i915_private *dev_priv)
> if (aging_tail != INVALID_TAIL_PTR &&
> ((now - dev_priv->perf.oa.oa_buffer.aging_timestamp) >
> OA_TAIL_MARGIN_NSEC)) {
> +
> aged_idx ^= 1;
> dev_priv->perf.oa.oa_buffer.aged_tail_idx = aged_idx;
>
> @@ -553,6 +584,287 @@ static int append_oa_sample(struct i915_perf_stream *stream,
> *
> * Returns: 0 on success, negative error code on failure.
> */
> +static int gen8_append_oa_reports(struct i915_perf_stream *stream,
> + char __user *buf,
> + size_t count,
> + size_t *offset)
> +{
> + struct drm_i915_private *dev_priv = stream->dev_priv;
> + 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);
> + size_t start_offset = *offset;
> + unsigned long flags;
> + unsigned int aged_tail_idx;
> + u32 head, tail;
> + u32 taken;
> + int ret = 0;
> +
> + if (WARN_ON(!stream->enabled))
> + return -EIO;
> +
> + spin_lock_irqsave(&dev_priv->perf.oa.oa_buffer.ptr_lock, flags);
> +
> + head = dev_priv->perf.oa.oa_buffer.head;
> + aged_tail_idx = dev_priv->perf.oa.oa_buffer.aged_tail_idx;
> + tail = dev_priv->perf.oa.oa_buffer.tails[aged_tail_idx].offset;
> +
> + spin_unlock_irqrestore(&dev_priv->perf.oa.oa_buffer.ptr_lock, flags);
> +
> + /* An invalid tail pointer here means we're still waiting for the poll
> + * hrtimer callback to give us a pointer
> + */
> + if (tail == INVALID_TAIL_PTR)
> + return -EAGAIN;
> +
> + /* NB: oa_buffer.head/tail include the gtt_offset which we don't want
> + * while indexing relative to oa_buf_base.
> + */
> + head -= gtt_offset;
> + tail -= gtt_offset;
> +
> + /* An out of bounds or misaligned head or tail pointer implies a driver
> + * bug since we validate + align the tail pointers we read from the
> + * hardware and we are in full control of the head pointer which should
> + * 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,
> + "Inconsistent OA buffer pointers: head = %u, tail = %u\n",
> + head, tail))
> + return -EIO;
> +
> +
> + for (/* none */;
> + (taken = OA_TAKEN(tail, head));
> + head = (head + report_size) & mask) {
> + u8 *report = oa_buf_base + head;
> + u32 *report32 = (void *)report;
> + u32 ctx_id;
> + u32 reason;
> +
> + /* All the report sizes factor neatly into the buffer
> + * size so we never expect to see a report split
> + * between the beginning and end of the buffer.
> + *
> + * Given the initial alignment check a misalignment
> + * here would imply a driver bug that would result
> + * in an overrun.
> + */
> + if (WARN_ON((OA_BUFFER_SIZE - head) < report_size)) {
> + DRM_ERROR("Spurious OA head ptr: non-integral report offset\n");
> + break;
> + }
> +
> + /* The reason field includes flags identifying what
> + * triggered this specific report (mostly timer
> + * triggered or e.g. due to a context switch).
> + *
> + * This field is never expected to be zero so we can
> + * check that the report isn't invalid before copying
> + * it to userspace...
> + */
> + reason = ((report32[0] >> OAREPORT_REASON_SHIFT) &
> + OAREPORT_REASON_MASK);
> + if (reason == 0) {
> + if (__ratelimit(&dev_priv->perf.oa.spurious_report_rs))
> + DRM_NOTE("Skipping spurious, invalid OA report\n");
> + continue;
> + }
> +
> + /* XXX: Just keep the lower 21 bits for now since I'm not
> + * entirely sure if the HW touches any of the higher bits in
> + * this field
> + */
> + ctx_id = report32[2] & 0x1fffff;
> +
> + /* Squash whatever is in the CTX_ID field if it's marked as
> + * invalid to be sure we avoid false-positive, single-context
> + * filtering below...
> + *
> + * Note: that we don't clear the valid_ctx_bit so userspace can
> + * understand that the ID has been squashed by the kernel.
> + */
> + if (!(report32[0] & dev_priv->perf.oa.gen8_valid_ctx_bit))
> + ctx_id = report32[2] = INVALID_CTX_ID;
> +
> + /* NB: For Gen 8 the OA unit no longer supports clock gating
> + * off for a specific context and the kernel can't securely
> + * stop the counters from updating as system-wide / global
> + * values.
> + *
> + * Automatic reports now include a context ID so reports can be
> + * filtered on the cpu but it's not worth trying to
> + * automatically subtract/hide counter progress for other
> + * contexts while filtering since we can't stop userspace
> + * issuing MI_REPORT_PERF_COUNT commands which would still
> + * provide a side-band view of the real values.
> + *
> + * To allow userspace (such as Mesa/GL_INTEL_performance_query)
> + * to normalize counters for a single filtered context then it
> + * needs be forwarded bookend context-switch reports so that it
> + * can track switches in between MI_REPORT_PERF_COUNT commands
> + * and can itself subtract/ignore the progress of counters
> + * associated with other contexts. Note that the hardware
> + * automatically triggers reports when switching to a new
> + * context which are tagged with the ID of the newly active
> + * context. To avoid the complexity (and likely fragility) of
> + * reading ahead while parsing reports to try and minimize
> + * forwarding redundant context switch reports (i.e. between
> + * other, unrelated contexts) we simply elect to forward them
> + * all.
> + *
> + * We don't rely solely on the reason field to identify context
> + * switches since it's not-uncommon for periodic samples to
> + * identify a switch before any 'context switch' report.
> + */
> + if (!dev_priv->perf.oa.exclusive_stream->ctx ||
> + dev_priv->perf.oa.specific_ctx_id == ctx_id ||
> + (dev_priv->perf.oa.oa_buffer.last_ctx_id ==
> + dev_priv->perf.oa.specific_ctx_id) ||
> + reason & OAREPORT_REASON_CTX_SWITCH) {
> +
> + /* While filtering for a single context we avoid
> + * leaking the IDs of other contexts.
> + */
> + if (dev_priv->perf.oa.exclusive_stream->ctx &&
> + dev_priv->perf.oa.specific_ctx_id != ctx_id) {
> + report32[2] = INVALID_CTX_ID;
> + }
> +
> + ret = append_oa_sample(stream, buf, count, offset,
> + report);
> + if (ret)
> + break;
> +
> + dev_priv->perf.oa.oa_buffer.last_ctx_id = ctx_id;
> + }
> +
> + /* The above reason field sanity check is based on
> + * the assumption that the OA buffer is initially
> + * zeroed and we reset the field after copying so the
> + * check is still meaningful once old reports start
> + * being overwritten.
> + */
> + report32[0] = 0;
> + }
> +
> + if (start_offset != *offset) {
> + spin_lock_irqsave(&dev_priv->perf.oa.oa_buffer.ptr_lock, flags);
> +
> + /* We removed the gtt_offset for the copy loop above, indexing
> + * relative to oa_buf_base so put back here...
> + */
> + head += gtt_offset;
> +
> + I915_WRITE(GEN8_OAHEADPTR, head & GEN8_OAHEADPTR_MASK);
> + dev_priv->perf.oa.oa_buffer.head = head;
> +
> + spin_unlock_irqrestore(&dev_priv->perf.oa.oa_buffer.ptr_lock, flags);
> + }
> +
> + return ret;
> +}
> +
> +/**
> + * gen8_oa_read - copy status records then buffered OA reports
> + * @stream: An i915-perf stream opened for OA metrics
> + * @buf: destination buffer given by userspace
> + * @count: the number of bytes userspace wants to read
> + * @offset: (inout): the current position for writing into @buf
> + *
> + * Checks OA unit status registers and if necessary appends corresponding
> + * status records for userspace (such as for a buffer full condition) and then
> + * initiate appending any buffered OA reports.
> + *
> + * Updates @offset according to the number of bytes successfully copied into
> + * the userspace buffer.
> + *
> + * NB: some data may be successfully copied to the userspace buffer
> + * even if an error is returned, and this is reflected in the
> + * updated @offset.
> + *
> + * Returns: zero on success or a negative error code
> + */
> +static int gen8_oa_read(struct i915_perf_stream *stream,
> + char __user *buf,
> + size_t count,
> + size_t *offset)
> +{
> + struct drm_i915_private *dev_priv = stream->dev_priv;
> + u32 oastatus;
> + int ret;
> +
> + if (WARN_ON(!dev_priv->perf.oa.oa_buffer.vaddr))
> + return -EIO;
> +
> + oastatus = I915_READ(GEN8_OASTATUS);
> +
> + /* We treat OABUFFER_OVERFLOW as a significant error:
> + *
> + * Although theoretically we could handle this more gracefully
> + * sometimes, some Gens don't correctly suppress certain
> + * automatically triggered reports in this condition and so we
> + * have to assume that old reports are now being trampled
> + * 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) {
> + ret = append_oa_status(stream, buf, count, offset,
> + DRM_I915_PERF_RECORD_OA_BUFFER_LOST);
> + if (ret)
> + return ret;
> +
> + DRM_DEBUG("OA buffer overflow (exponent = %d): force restart\n",
> + dev_priv->perf.oa.period_exponent);
> +
> + dev_priv->perf.oa.ops.oa_disable(dev_priv);
> + dev_priv->perf.oa.ops.oa_enable(dev_priv);
> +
> + /* Note: .oa_enable() is expected to re-init the oabuffer
> + * and reset GEN8_OASTATUS for us
> + */
> + oastatus = I915_READ(GEN8_OASTATUS);
> + }
> +
> + if (oastatus & GEN8_OASTATUS_REPORT_LOST) {
> + ret = append_oa_status(stream, buf, count, offset,
> + DRM_I915_PERF_RECORD_OA_REPORT_LOST);
> + if (ret)
> + return ret;
> + I915_WRITE(GEN8_OASTATUS,
> + oastatus & ~GEN8_OASTATUS_REPORT_LOST);
> + }
> +
> + return gen8_append_oa_reports(stream, buf, count, offset);
> +}
> +
> +/**
> + * Copies all buffered OA reports into userspace read() buffer.
> + * @stream: An i915-perf stream opened for OA metrics
> + * @buf: destination buffer given by userspace
> + * @count: the number of bytes userspace wants to read
> + * @offset: (inout): the current position for writing into @buf
> + *
> + * Notably any error condition resulting in a short read (-%ENOSPC or
> + * -%EFAULT) will be returned even though one or more records may
> + * have been successfully copied. In this case it's up to the caller
> + * to decide if the error should be squashed before returning to
> + * userspace.
> + *
> + * Note: reports are consumed from the head, and appended to the
> + * tail, so the tail chases the head?... If you think that's mad
> + * and back-to-front you're not alone, but this follows the
> + * Gen PRM naming convention.
> + *
> + * Returns: 0 on success, negative error code on failure.
> + */
> static int gen7_append_oa_reports(struct i915_perf_stream *stream,
> char __user *buf,
> size_t count,
> @@ -733,7 +1045,8 @@ static int gen7_oa_read(struct i915_perf_stream *stream,
> if (ret)
> return ret;
>
> - DRM_DEBUG("OA buffer overflow: force restart\n");
> + DRM_DEBUG("OA buffer overflow (exponent = %d): force restart\n",
> + dev_priv->perf.oa.period_exponent);
>
> dev_priv->perf.oa.ops.oa_disable(dev_priv);
> dev_priv->perf.oa.ops.oa_enable(dev_priv);
> @@ -776,7 +1089,7 @@ static int i915_oa_wait_unlocked(struct i915_perf_stream *stream)
> return -EIO;
>
> return wait_event_interruptible(dev_priv->perf.oa.poll_wq,
> - dev_priv->perf.oa.ops.oa_buffer_check(dev_priv));
> + oa_buffer_check_unlocked(dev_priv));
> }
>
> /**
> @@ -833,33 +1146,37 @@ static int i915_oa_read(struct i915_perf_stream *stream,
> static int oa_get_render_ctx_id(struct i915_perf_stream *stream)
> {
> struct drm_i915_private *dev_priv = stream->dev_priv;
> - struct intel_engine_cs *engine = dev_priv->engine[RCS];
> - int ret;
>
> - ret = i915_mutex_lock_interruptible(&dev_priv->drm);
> - if (ret)
> - return ret;
> + if (i915.enable_execlists)
> + dev_priv->perf.oa.specific_ctx_id = stream->ctx->hw_id;
> + else {
> + struct intel_engine_cs *engine = dev_priv->engine[RCS];
> + int ret;
>
> - /* As the ID is the gtt offset of the context's vma we pin
> - * the vma to ensure the ID remains fixed.
> - *
> - * NB: implied RCS engine...
> - */
> - ret = engine->context_pin(engine, stream->ctx);
> - if (ret)
> - goto unlock;
> + ret = i915_mutex_lock_interruptible(&dev_priv->drm);
> + if (ret)
> + return ret;
>
> - /* Explicitly track the ID (instead of calling i915_ggtt_offset()
> - * on the fly) considering the difference with gen8+ and
> - * execlists
> - */
> - dev_priv->perf.oa.specific_ctx_id =
> - i915_ggtt_offset(stream->ctx->engine[engine->id].state);
> + /* As the ID is the gtt offset of the context's vma we pin
> + * the vma to ensure the ID remains fixed.
> + */
> + ret = engine->context_pin(engine, stream->ctx);
> + if (ret) {
> + mutex_unlock(&dev_priv->drm.struct_mutex);
> + return ret;
> + }
>
> -unlock:
> - mutex_unlock(&dev_priv->drm.struct_mutex);
> + /* Explicitly track the ID (instead of calling
> + * i915_ggtt_offset() on the fly) considering the difference
> + * with gen8+ and execlists
> + */
> + dev_priv->perf.oa.specific_ctx_id =
> + i915_ggtt_offset(stream->ctx->engine[engine->id].state);
>
> - return ret;
> + mutex_unlock(&dev_priv->drm.struct_mutex);
> + }
> +
> + return 0;
> }
>
> /**
> @@ -874,12 +1191,16 @@ static void oa_put_render_ctx_id(struct i915_perf_stream *stream)
> struct drm_i915_private *dev_priv = stream->dev_priv;
> struct intel_engine_cs *engine = dev_priv->engine[RCS];
>
> - mutex_lock(&dev_priv->drm.struct_mutex);
> + if (i915.enable_execlists) {
> + dev_priv->perf.oa.specific_ctx_id = INVALID_CTX_ID;
> + } else {
> + mutex_lock(&dev_priv->drm.struct_mutex);
>
> - dev_priv->perf.oa.specific_ctx_id = INVALID_CTX_ID;
> - engine->context_unpin(engine, stream->ctx);
> + dev_priv->perf.oa.specific_ctx_id = INVALID_CTX_ID;
> + engine->context_unpin(engine, stream->ctx);
>
> - mutex_unlock(&dev_priv->drm.struct_mutex);
> + mutex_unlock(&dev_priv->drm.struct_mutex);
> + }
> }
>
> static void
> @@ -969,6 +1290,61 @@ static void gen7_init_oa_buffer(struct drm_i915_private *dev_priv)
> dev_priv->perf.oa.pollin = false;
> }
>
> +static void gen8_init_oa_buffer(struct drm_i915_private *dev_priv)
> +{
> + u32 gtt_offset = i915_ggtt_offset(dev_priv->perf.oa.oa_buffer.vma);
> + unsigned long flags;
> +
> + spin_lock_irqsave(&dev_priv->perf.oa.oa_buffer.ptr_lock, flags);
> +
> + I915_WRITE(GEN8_OASTATUS, 0);
> + I915_WRITE(GEN8_OAHEADPTR, gtt_offset);
> + dev_priv->perf.oa.oa_buffer.head = gtt_offset;
> +
> + I915_WRITE(GEN8_OABUFFER_UDW, 0);
> +
> + /* PRM says:
> + *
> + * "This MMIO must be set before the OATAILPTR
> + * register and after the OAHEADPTR register. This is
> + * to enable proper functionality of the overflow
> + * bit."
> + */
> + I915_WRITE(GEN8_OABUFFER, gtt_offset |
> + OABUFFER_SIZE_16M | OA_MEM_SELECT_GGTT);
> + I915_WRITE(GEN8_OATAILPTR, gtt_offset & GEN8_OATAILPTR_MASK);
> +
> + /* 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[1].offset = INVALID_TAIL_PTR;
> +
> + /* Reset state used to recognise context switches, affecting which
> + * reports we will forward to userspace while filtering for a single
> + * context.
> + */
> + dev_priv->perf.oa.oa_buffer.last_ctx_id = INVALID_CTX_ID;
> +
> + spin_unlock_irqrestore(&dev_priv->perf.oa.oa_buffer.ptr_lock, flags);
> +
> + /* NB: although the OA buffer will initially be allocated
> + * zeroed via shmfs (and so this memset is redundant when
> + * first allocating), we may re-init the OA buffer, either
> + * when re-enabling a stream or in error/reset paths.
> + *
> + * The reason we clear the buffer for each re-init is for the
> + * sanity check in gen8_append_oa_reports() that looks at the
> + * reason field to make sure it's non-zero which relies on
> + * the assumption that new reports are being written to zeroed
> + * memory...
> + */
> + memset(dev_priv->perf.oa.oa_buffer.vaddr, 0, OA_BUFFER_SIZE);
> +
> + /* Maybe make ->pollin per-stream state if we support multiple
> + * concurrent streams in the future.
> + */
> + dev_priv->perf.oa.pollin = false;
> +}
> +
> static int alloc_oa_buffer(struct drm_i915_private *dev_priv)
> {
> struct drm_i915_gem_object *bo;
> @@ -1113,6 +1489,212 @@ static void hsw_disable_metric_set(struct drm_i915_private *dev_priv)
> ~GT_NOA_ENABLE));
> }
>
> +/* NB: It must always remain pointer safe to run this even if the OA unit
> + * has been disabled.
> + *
> + * It's fine to put out-of-date values into these per-context registers
> + * in the case that the OA unit has been disabled.
> + */
> +static void gen8_update_reg_state_unlocked(struct i915_gem_context *ctx,
> + u32 *reg_state)
> +{
> + struct drm_i915_private *dev_priv = ctx->i915;
> + const struct i915_oa_reg *flex_regs = dev_priv->perf.oa.flex_regs;
> + int n_flex_regs = dev_priv->perf.oa.flex_regs_len;
> + u32 ctx_oactxctrl = dev_priv->perf.oa.ctx_oactxctrl_offset;
> + u32 ctx_flexeu0 = dev_priv->perf.oa.ctx_flexeu0_offset;
> + /* The MMIO offsets for Flex EU registers aren't contiguous */
> + u32 flex_mmio[] = {
> + i915_mmio_reg_offset(EU_PERF_CNTL0),
> + i915_mmio_reg_offset(EU_PERF_CNTL1),
> + i915_mmio_reg_offset(EU_PERF_CNTL2),
> + i915_mmio_reg_offset(EU_PERF_CNTL3),
> + i915_mmio_reg_offset(EU_PERF_CNTL4),
> + i915_mmio_reg_offset(EU_PERF_CNTL5),
> + i915_mmio_reg_offset(EU_PERF_CNTL6),
> + };
> + int i;
> +
> + reg_state[ctx_oactxctrl] = i915_mmio_reg_offset(GEN8_OACTXCONTROL);
> + reg_state[ctx_oactxctrl+1] = (dev_priv->perf.oa.period_exponent <<
> + GEN8_OA_TIMER_PERIOD_SHIFT) |
> + (dev_priv->perf.oa.periodic ?
> + GEN8_OA_TIMER_ENABLE : 0) |
> + GEN8_OA_COUNTER_RESUME;
> +
> + for (i = 0; i < ARRAY_SIZE(flex_mmio); i++) {
> + u32 state_offset = ctx_flexeu0 + i * 2;
> + u32 mmio = flex_mmio[i];
> +
> + /* This arbitrary default will select the 'EU FPU0 Pipeline
> + * Active' event. In the future it's anticipated that there
> + * will be an explicit 'No Event' we can select, but not yet...
> + */
> + u32 value = 0;
> + int j;
> +
> + for (j = 0; j < n_flex_regs; j++) {
> + if (i915_mmio_reg_offset(flex_regs[j].addr) == mmio) {
> + value = flex_regs[j].value;
> + break;
> + }
> + }
> +
> + reg_state[state_offset] = mmio;
> + reg_state[state_offset+1] = value;
> + }
> +}
> +
> +/* Manages updating the per-context aspects of the OA stream
> + * configuration across all contexts.
> + *
> + * The awkward consideration here is that OACTXCONTROL controls the
> + * exponent for periodic sampling which is primarily used for system
> + * wide profiling where we'd like a consistent sampling period even in
> + * the face of context switches.
> + *
> + * Our approach of updating the register state context (as opposed to
> + * say using a workaround batch buffer) ensures that the hardware
> + * won't automatically reload an out-of-date timer exponent even
> + * transiently before a WA BB could be parsed.
> + *
> + * This function needs to:
> + * - Ensure the currently running context's per-context OA state is
> + * updated
> + * - Ensure that all existing contexts will have the correct per-context
> + * OA state if they are scheduled for use.
> + * - Ensure any new contexts will be initialized with the correct
> + * per-context OA state.
> + *
> + * Note: it's only the RCS/Render context that has any OA state.
> + */
> +static int gen8_configure_all_contexts(struct drm_i915_private *dev_priv)
> +{
> + struct i915_gem_context *ctx;
> + int ret;
> +
> + ret = i915_mutex_lock_interruptible(&dev_priv->drm);
> + if (ret)
> + return ret;
> +
> + /* Switch away from any user context. */
> + ret = i915_gem_switch_to_kernel_context(dev_priv);
> + if (ret)
> + goto out;
> +
> + /* The OA register config is setup through the context image. This image
> + * might be written to by the GPU on context switch (in particular on
> + * lite-restore). This means we can't safely update a context's image,
> + * if this context is scheduled/submitted to run on the GPU.
> + *
> + * We could emit the OA register config through the batch buffer but
> + * this might leave small interval of time where the OA unit is
> + * configured at an invalid sampling period.
> + *
> + * So far the best way to work around this issue seems to be draining
> + * the GPU from any submitted work.
> + */
> + ret = i915_gem_wait_for_idle(dev_priv,
> + I915_WAIT_INTERRUPTIBLE |
> + I915_WAIT_LOCKED);
> + if (ret)
> + goto out;
> +
> + /* Update all contexts now that we've stalled the submission. */
> + list_for_each_entry(ctx, &dev_priv->context_list, link) {
> + struct intel_context *ce = &ctx->engine[RCS];
> + u32 *regs;
> +
> + /* OA settings will be set upon first use */
> + if (!ce->state)
> + continue;
> +
> + regs = i915_gem_object_pin_map(ce->state->obj, I915_MAP_WB);
> + if (IS_ERR(regs)) {
> + ret = PTR_ERR(regs);
> + goto out;
> + }
> +
> + ce->state->obj->mm.dirty = true;
> + regs += LRC_STATE_PN * PAGE_SIZE / sizeof(*regs);
> + /* Probably best since we don't really want to expose
> + * the LRC_STATE_PN trick.
> + */
Sorry, was writing comments to myself :) This is not about the code,
just about the interface to export mapping the context from intel_lrc.c
The only worry I have here is a failure whilst pinning the contexts,
leaving some updated, some not, and upon error we will assume the change
did not occur.
Other than that, this looks correct, and fwiw I do much prefer the idea
of the OA adjustment being synchronous wrt to request submission (that
all previous requests complete using the current OA settings and all new
submissions will use the next OA config). The downside is that it may
stall the driver for an indefinite period of time - but for a global
config that seems acceptable in order to have it well defined.
Per-context OA configs presumably are faster to apply?
For the context image adjustment,
Reviewed-by: Chris Wilson <chris at chris-wilson.co.uk>
and Matthew can take responsibility for having checked the rest ;)
-Chris
--
Chris Wilson, Intel Open Source Technology Centre
More information about the Intel-gfx
mailing list