| Message ID | 20170405190506.2227-1-robert@sixbynine.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 5 April 2017 at 20:05, Robert Bragg <robert@sixbynine.org> wrote: > An oa_exponent_to_ns() utility and per-gen timebase constants where were > recently removed when updating the tail pointer race condition WA, and > this restores those so we can update the _PROP_OA_EXPONENT validation > done in read_properties_unlocked() to not assume we have a 12.5MHz > timebase as we did for Haswell. > > Accordingly the oa_sample_rate_hard_limit value that's referenced by > proc_dointvec_minmax defining the absolute limit for the OA sampling > frequency is now initialized to (timestamp_frequency / 2) instead of the > 6.25MHz constant for Haswell. > > v2: > Specify frequency of 19.2MHz for BXT (Ville) > Initialize oa_sample_rate_hard_limit per-gen too (Lionel) > > Signed-off-by: Robert Bragg <robert@sixbynine.org> > Cc: Lionel Landwerlin <lionel.g.landwerlin@linux.intel.com> > Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> > --- > drivers/gpu/drm/i915/i915_drv.h | 1 + > drivers/gpu/drm/i915/i915_perf.c | 31 ++++++++++++++++++++++--------- > 2 files changed, 23 insertions(+), 9 deletions(-) > > diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h > index 3a22b6fd0ee6..48b07d706f06 100644 > --- a/drivers/gpu/drm/i915/i915_drv.h > +++ b/drivers/gpu/drm/i915/i915_drv.h > @@ -2463,6 +2463,7 @@ struct drm_i915_private { > > bool periodic; > int period_exponent; > + int timestamp_frequency; > > int metrics_set; > > diff --git a/drivers/gpu/drm/i915/i915_perf.c b/drivers/gpu/drm/i915/i915_perf.c > index 98eb6415b63a..980b4a1fd7cc 100644 > --- a/drivers/gpu/drm/i915/i915_perf.c > +++ b/drivers/gpu/drm/i915/i915_perf.c > @@ -288,10 +288,12 @@ static u32 i915_perf_stream_paranoid = true; > > /* For sysctl proc_dointvec_minmax of i915_oa_max_sample_rate > * > - * 160ns is the smallest sampling period we can theoretically program the OA > - * unit with on Haswell, corresponding to 6.25MHz. > + * The highest sampling frequency we can theoretically program the OA unit > + * with is always half the timestamp frequency: E.g. 6.25Mhz for Haswell. > + * > + * Initialized just before we register the sysctl parameter. > */ > -static int oa_sample_rate_hard_limit = 6250000; > +static int oa_sample_rate_hard_limit; > > /* Theoretically we can program the OA unit to sample every 160ns but don't > * allow that by default unless root... > @@ -2549,6 +2551,12 @@ i915_perf_open_ioctl_locked(struct drm_i915_private *dev_priv, > return ret; > } > > +static u64 oa_exponent_to_ns(struct drm_i915_private *dev_priv, int exponent) > +{ > + return div_u64(1000000000ULL * (2ULL << exponent), > + dev_priv->perf.oa.timestamp_frequency); > +} > + > /** > * read_properties_unlocked - validate + copy userspace stream open properties > * @dev_priv: i915 device instance > @@ -2647,14 +2655,9 @@ static int read_properties_unlocked(struct drm_i915_private *dev_priv, > /* Theoretically we can program the OA unit to sample > * every 160ns but don't allow that by default unless hmm, that's not actually true if we consider BXT, right? Reviewed-by: Matthew Auld <matthew.auld@intel.com>
On Wed, Apr 12, 2017 at 1:34 PM, Matthew Auld < matthew.william.auld@gmail.com> wrote: > On 5 April 2017 at 20:05, Robert Bragg <robert@sixbynine.org> wrote: > > An oa_exponent_to_ns() utility and per-gen timebase constants where > were > > > recently removed when updating the tail pointer race condition WA, and > > this restores those so we can update the _PROP_OA_EXPONENT validation > > done in read_properties_unlocked() to not assume we have a 12.5MHz > > timebase as we did for Haswell. > > > > Accordingly the oa_sample_rate_hard_limit value that's referenced by > > proc_dointvec_minmax defining the absolute limit for the OA sampling > > frequency is now initialized to (timestamp_frequency / 2) instead of the > > 6.25MHz constant for Haswell. > > > > v2: > > Specify frequency of 19.2MHz for BXT (Ville) > > Initialize oa_sample_rate_hard_limit per-gen too (Lionel) > > > > Signed-off-by: Robert Bragg <robert@sixbynine.org> > > Cc: Lionel Landwerlin <lionel.g.landwerlin@linux.intel.com> > > Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> > > --- > > drivers/gpu/drm/i915/i915_drv.h | 1 + > > drivers/gpu/drm/i915/i915_perf.c | 31 ++++++++++++++++++++++--------- > > 2 files changed, 23 insertions(+), 9 deletions(-) > > > > diff --git a/drivers/gpu/drm/i915/i915_drv.h > b/drivers/gpu/drm/i915/i915_drv.h > > index 3a22b6fd0ee6..48b07d706f06 100644 > > --- a/drivers/gpu/drm/i915/i915_drv.h > > +++ b/drivers/gpu/drm/i915/i915_drv.h > > @@ -2463,6 +2463,7 @@ struct drm_i915_private { > > > > bool periodic; > > int period_exponent; > > + int timestamp_frequency; > > > > int metrics_set; > > > > diff --git a/drivers/gpu/drm/i915/i915_perf.c > b/drivers/gpu/drm/i915/i915_perf.c > > index 98eb6415b63a..980b4a1fd7cc 100644 > > --- a/drivers/gpu/drm/i915/i915_perf.c > > +++ b/drivers/gpu/drm/i915/i915_perf.c > > @@ -288,10 +288,12 @@ static u32 i915_perf_stream_paranoid = true; > > > > /* For sysctl proc_dointvec_minmax of i915_oa_max_sample_rate > > * > > - * 160ns is the smallest sampling period we can theoretically program > the OA > > - * unit with on Haswell, corresponding to 6.25MHz. > > + * The highest sampling frequency we can theoretically program the OA > unit > > + * with is always half the timestamp frequency: E.g. 6.25Mhz for > Haswell. > > + * > > + * Initialized just before we register the sysctl parameter. > > */ > > -static int oa_sample_rate_hard_limit = 6250000; > > +static int oa_sample_rate_hard_limit; > > > > /* Theoretically we can program the OA unit to sample every 160ns but > don't > > * allow that by default unless root... > > @@ -2549,6 +2551,12 @@ i915_perf_open_ioctl_locked(struct > drm_i915_private *dev_priv, > > return ret; > > } > > > > +static u64 oa_exponent_to_ns(struct drm_i915_private *dev_priv, int > exponent) > > +{ > > + return div_u64(1000000000ULL * (2ULL << exponent), > > + dev_priv->perf.oa.timestamp_frequency); > > +} > > + > > /** > > * read_properties_unlocked - validate + copy userspace stream open > properties > > * @dev_priv: i915 device instance > > @@ -2647,14 +2655,9 @@ static int read_properties_unlocked(struct > drm_i915_private *dev_priv, > > /* Theoretically we can program the OA unit to > sample > > * every 160ns but don't allow that by default > unless > hmm, that's not actually true if we consider BXT, right? > right, I've updated this comment now. > > Reviewed-by: Matthew Auld <matthew.auld@intel.com> > thanks
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h index 3a22b6fd0ee6..48b07d706f06 100644 --- a/drivers/gpu/drm/i915/i915_drv.h +++ b/drivers/gpu/drm/i915/i915_drv.h @@ -2463,6 +2463,7 @@ struct drm_i915_private { bool periodic; int period_exponent; + int timestamp_frequency; int metrics_set; diff --git a/drivers/gpu/drm/i915/i915_perf.c b/drivers/gpu/drm/i915/i915_perf.c index 98eb6415b63a..980b4a1fd7cc 100644 --- a/drivers/gpu/drm/i915/i915_perf.c +++ b/drivers/gpu/drm/i915/i915_perf.c @@ -288,10 +288,12 @@ static u32 i915_perf_stream_paranoid = true; /* For sysctl proc_dointvec_minmax of i915_oa_max_sample_rate * - * 160ns is the smallest sampling period we can theoretically program the OA - * unit with on Haswell, corresponding to 6.25MHz. + * The highest sampling frequency we can theoretically program the OA unit + * with is always half the timestamp frequency: E.g. 6.25Mhz for Haswell. + * + * Initialized just before we register the sysctl parameter. */ -static int oa_sample_rate_hard_limit = 6250000; +static int oa_sample_rate_hard_limit; /* Theoretically we can program the OA unit to sample every 160ns but don't * allow that by default unless root... @@ -2549,6 +2551,12 @@ i915_perf_open_ioctl_locked(struct drm_i915_private *dev_priv, return ret; } +static u64 oa_exponent_to_ns(struct drm_i915_private *dev_priv, int exponent) +{ + return div_u64(1000000000ULL * (2ULL << exponent), + dev_priv->perf.oa.timestamp_frequency); +} + /** * read_properties_unlocked - validate + copy userspace stream open properties * @dev_priv: i915 device instance @@ -2647,14 +2655,9 @@ static int read_properties_unlocked(struct drm_i915_private *dev_priv, /* Theoretically we can program the OA unit to sample * every 160ns but don't allow that by default unless * root. - * - * On Haswell the period is derived from the exponent - * as: - * - * period = 80ns * 2^(exponent + 1) */ BUILD_BUG_ON(sizeof(oa_period) != 8); - oa_period = 80ull * (2ull << value); + oa_period = oa_exponent_to_ns(dev_priv, value); /* This check is primarily to ensure that oa_period <= * UINT32_MAX (before passing to do_div which only @@ -2910,6 +2913,8 @@ void i915_perf_init(struct drm_i915_private *dev_priv) dev_priv->perf.oa.ops.oa_hw_tail_read = gen7_oa_hw_tail_read; + dev_priv->perf.oa.timestamp_frequency = 12500000; + dev_priv->perf.oa.oa_formats = hsw_oa_formats; dev_priv->perf.oa.n_builtin_sets = @@ -2923,6 +2928,8 @@ void i915_perf_init(struct drm_i915_private *dev_priv) */ if (IS_GEN8(dev_priv)) { + dev_priv->perf.oa.timestamp_frequency = 12500000; + dev_priv->perf.oa.ctx_oactxctrl_offset = 0x120; dev_priv->perf.oa.ctx_flexeu0_offset = 0x2ce; dev_priv->perf.oa.gen8_valid_ctx_bit = (1<<25); @@ -2939,6 +2946,8 @@ void i915_perf_init(struct drm_i915_private *dev_priv) i915_oa_select_metric_set_chv; } } else if (IS_GEN9(dev_priv)) { + dev_priv->perf.oa.timestamp_frequency = 12000000; + dev_priv->perf.oa.ctx_oactxctrl_offset = 0x128; dev_priv->perf.oa.ctx_flexeu0_offset = 0x3de; dev_priv->perf.oa.gen8_valid_ctx_bit = (1<<16); @@ -2959,6 +2968,8 @@ void i915_perf_init(struct drm_i915_private *dev_priv) dev_priv->perf.oa.ops.select_metric_set = i915_oa_select_metric_set_sklgt4; } else if (IS_BROXTON(dev_priv)) { + dev_priv->perf.oa.timestamp_frequency = 19200000; + dev_priv->perf.oa.n_builtin_sets = i915_oa_n_builtin_metric_sets_bxt; dev_priv->perf.oa.ops.select_metric_set = @@ -2993,6 +3004,8 @@ void i915_perf_init(struct drm_i915_private *dev_priv) spin_lock_init(&dev_priv->perf.hook_lock); spin_lock_init(&dev_priv->perf.oa.oa_buffer.ptr_lock); + oa_sample_rate_hard_limit = + dev_priv->perf.oa.timestamp_frequency / 2; dev_priv->perf.sysctl_header = register_sysctl_table(dev_root); dev_priv->perf.initialized = true;
An oa_exponent_to_ns() utility and per-gen timebase constants where recently removed when updating the tail pointer race condition WA, and this restores those so we can update the _PROP_OA_EXPONENT validation done in read_properties_unlocked() to not assume we have a 12.5MHz timebase as we did for Haswell. Accordingly the oa_sample_rate_hard_limit value that's referenced by proc_dointvec_minmax defining the absolute limit for the OA sampling frequency is now initialized to (timestamp_frequency / 2) instead of the 6.25MHz constant for Haswell. v2: Specify frequency of 19.2MHz for BXT (Ville) Initialize oa_sample_rate_hard_limit per-gen too (Lionel) Signed-off-by: Robert Bragg <robert@sixbynine.org> Cc: Lionel Landwerlin <lionel.g.landwerlin@linux.intel.com> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> --- drivers/gpu/drm/i915/i915_drv.h | 1 + drivers/gpu/drm/i915/i915_perf.c | 31 ++++++++++++++++++++++--------- 2 files changed, 23 insertions(+), 9 deletions(-)