Waitboost is a legacy feature implemented in the Host Turbo algorithm. This patch set implements it for the SLPC path. A "boost" happens when user calls gem_wait ioctl on a submission that has not landed on HW yet. GT frequency gets temporarily bumped to RP0 to allow the previous request to finish quickly. We achieve this on SLPC by setting the min frequency, SLPC will set that as the requested frequency.
The boost will occur through a worker thread that will be scheduled when the required conditions are met.
Like before, boost frequency is configurable through sysfs, so we can adjust it to any specific value as long as it is between [min, RP0].
v2: Add a worker thread to perform freq boost.
Cc: Ashutosh Dixit ashutosh.dixit@intel.com Signed-off-by: Vinay Belgaumkar vinay.belgaumkar@intel.com
Vinay Belgaumkar (3): drm/i915/guc/slpc: Define and initialize boost frequency drm/i915/guc/slpc: Add waitboost functionality for SLPC drm/i915/guc/slpc: Update boost sysfs hooks for SLPC
drivers/gpu/drm/i915/gt/intel_rps.c | 73 +++++++++ drivers/gpu/drm/i915/gt/intel_rps.h | 3 + drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.c | 149 +++++++++++++++--- drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.h | 3 + .../gpu/drm/i915/gt/uc/intel_guc_slpc_types.h | 13 ++ drivers/gpu/drm/i915/i915_request.c | 2 +- drivers/gpu/drm/i915/i915_sysfs.c | 19 +-- 7 files changed, 222 insertions(+), 40 deletions(-)
Define helpers and struct members required to record boost info. Boost frequency is initialized to RP0 at SLPC init. Also define num_waiters which can track the pending boost requests.
Boost will be done by scheduling a worker thread. This will allow us to make H2G calls inside an interrupt context. Initialize the worker function during SLPC init as well. Had to move intel_guc_slpc_init a few lines below to accomodate this.
v2: Add a workqueue to handle waitboost
Cc: Ashutosh Dixit ashutosh.dixit@intel.com Signed-off-by: Vinay Belgaumkar vinay.belgaumkar@intel.com --- drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.c | 101 ++++++++++++++---- drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.h | 1 + .../gpu/drm/i915/gt/uc/intel_guc_slpc_types.h | 13 +++ 3 files changed, 92 insertions(+), 23 deletions(-)
diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.c b/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.c index 65a3e7fdb2b2..cc51987b2535 100644 --- a/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.c +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.c @@ -79,29 +79,6 @@ static void slpc_mem_set_disabled(struct slpc_shared_data *data, slpc_mem_set_param(data, enable_id, 0); }
-int intel_guc_slpc_init(struct intel_guc_slpc *slpc) -{ - struct intel_guc *guc = slpc_to_guc(slpc); - struct drm_i915_private *i915 = slpc_to_i915(slpc); - u32 size = PAGE_ALIGN(sizeof(struct slpc_shared_data)); - int err; - - GEM_BUG_ON(slpc->vma); - - err = intel_guc_allocate_and_map_vma(guc, size, &slpc->vma, (void **)&slpc->vaddr); - if (unlikely(err)) { - drm_err(&i915->drm, - "Failed to allocate SLPC struct (err=%pe)\n", - ERR_PTR(err)); - return err; - } - - slpc->max_freq_softlimit = 0; - slpc->min_freq_softlimit = 0; - - return err; -} - static u32 slpc_get_state(struct intel_guc_slpc *slpc) { struct slpc_shared_data *data; @@ -203,6 +180,81 @@ static int slpc_unset_param(struct intel_guc_slpc *slpc, return guc_action_slpc_unset_param(guc, id); }
+static int slpc_force_min_freq(struct intel_guc_slpc *slpc, u32 freq) +{ + struct drm_i915_private *i915 = slpc_to_i915(slpc); + intel_wakeref_t wakeref; + int ret = 0; + + lockdep_assert_held(&slpc->lock); + + /** + * This function is a little different as compared to + * intel_guc_slpc_set_min_freq(). Softlimit will not be updated + * here since this is used to temporarily change min freq, + * for example, during a waitboost. Caller is responsible for + * checking bounds. + */ + + with_intel_runtime_pm(&i915->runtime_pm, wakeref) { + ret = slpc_set_param(slpc, + SLPC_PARAM_GLOBAL_MIN_GT_UNSLICE_FREQ_MHZ, + freq); + if (ret) + drm_err(&i915->drm, "Unable to force min freq to %u: %d", + freq, ret); + } + + return ret; +} + +static void slpc_boost_work(struct work_struct *work) +{ + struct intel_guc_slpc *slpc = container_of(work, typeof(*slpc), boost_work); + + /* Raise min freq to boost. It's possible that + * this is greater than current max. But it will + * certainly be limited by RP0. An error setting + * the min param is not fatal. + */ + mutex_lock(&slpc->lock); + if (atomic_read(&slpc->num_waiters)) { + slpc_force_min_freq(slpc, slpc->boost_freq); + slpc->num_boosts++; + } + mutex_unlock(&slpc->lock); +} + +int intel_guc_slpc_init(struct intel_guc_slpc *slpc) +{ + struct intel_guc *guc = slpc_to_guc(slpc); + struct drm_i915_private *i915 = slpc_to_i915(slpc); + u32 size = PAGE_ALIGN(sizeof(struct slpc_shared_data)); + int err; + + GEM_BUG_ON(slpc->vma); + + err = intel_guc_allocate_and_map_vma(guc, size, &slpc->vma, (void **)&slpc->vaddr); + if (unlikely(err)) { + drm_err(&i915->drm, + "Failed to allocate SLPC struct (err=%pe)\n", + ERR_PTR(err)); + return err; + } + + slpc->max_freq_softlimit = 0; + slpc->min_freq_softlimit = 0; + + slpc->boost_freq = 0; + atomic_set(&slpc->num_waiters, 0); + slpc->num_boosts = 0; + + mutex_init(&slpc->lock); + INIT_WORK(&slpc->boost_work, slpc_boost_work); + + return err; +} + static const char *slpc_global_state_to_string(enum slpc_global_state state) { switch (state) { @@ -522,6 +574,9 @@ static void slpc_get_rp_values(struct intel_guc_slpc *slpc) GT_FREQUENCY_MULTIPLIER; slpc->min_freq = REG_FIELD_GET(RPN_CAP_MASK, rp_state_cap) * GT_FREQUENCY_MULTIPLIER; + + if (!slpc->boost_freq) + slpc->boost_freq = slpc->rp0_freq; }
/* diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.h b/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.h index e45054d5b9b4..b62528647770 100644 --- a/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.h +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.h @@ -38,5 +38,6 @@ int intel_guc_slpc_get_max_freq(struct intel_guc_slpc *slpc, u32 *val); int intel_guc_slpc_get_min_freq(struct intel_guc_slpc *slpc, u32 *val); int intel_guc_slpc_print_info(struct intel_guc_slpc *slpc, struct drm_printer *p); void intel_guc_pm_intrmsk_enable(struct intel_gt *gt); +void intel_guc_slpc_boost(struct intel_guc_slpc *slpc);
#endif diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc_types.h b/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc_types.h index 41d13527666f..bf5b9a563c09 100644 --- a/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc_types.h +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc_types.h @@ -6,6 +6,9 @@ #ifndef _INTEL_GUC_SLPC_TYPES_H_ #define _INTEL_GUC_SLPC_TYPES_H_
+#include <linux/atomic.h> +#include <linux/workqueue.h> +#include <linux/mutex.h> #include <linux/types.h>
#define SLPC_RESET_TIMEOUT_MS 5 @@ -20,10 +23,20 @@ struct intel_guc_slpc { u32 min_freq; u32 rp0_freq; u32 rp1_freq; + u32 boost_freq;
/* frequency softlimits */ u32 min_freq_softlimit; u32 max_freq_softlimit; + + /* Protects set/reset of boost freq + * and value of num_waiters + */ + struct mutex lock; + + struct work_struct boost_work; + atomic_t num_waiters; + u32 num_boosts; };
#endif
On Sun, 31 Oct 2021 21:39:35 -0700, Belgaumkar, Vinay wrote:
Define helpers and struct members required to record boost info. Boost frequency is initialized to RP0 at SLPC init. Also define num_waiters which can track the pending boost requests.
Boost will be done by scheduling a worker thread. This will allow us to make H2G calls inside an interrupt context. Initialize the
"to not make H2G calls from interrupt context" is probably better.
+static int slpc_force_min_freq(struct intel_guc_slpc *slpc, u32 freq) +{
- struct drm_i915_private *i915 = slpc_to_i915(slpc);
- intel_wakeref_t wakeref;
- int ret = 0;
- lockdep_assert_held(&slpc->lock);
- /**
nit: this I believe should just be
/*
/** I believe shows up in kerneldoc so shouldn't be used unless we want something in kerneldoc.
* This function is a little different as compared to* intel_guc_slpc_set_min_freq(). Softlimit will not be updated* here since this is used to temporarily change min freq,* for example, during a waitboost. Caller is responsible for* checking bounds.*/- with_intel_runtime_pm(&i915->runtime_pm, wakeref) {
ret = slpc_set_param(slpc,SLPC_PARAM_GLOBAL_MIN_GT_UNSLICE_FREQ_MHZ,freq);if (ret)drm_err(&i915->drm, "Unable to force min freq to %u: %d",
Probably drm_err_ratelimited since it's called at run time not only at init? Not sure if drm_err_once suffizes, probably not.
freq, ret);- }
- return ret;
+}
+static void slpc_boost_work(struct work_struct *work) +{
- struct intel_guc_slpc *slpc = container_of(work, typeof(*slpc), boost_work);
- /* Raise min freq to boost. It's possible that
* this is greater than current max. But it will* certainly be limited by RP0. An error setting* the min param is not fatal.*/
nit: do we follow the following format for multi-line comments, Documentation/process/coding-style.rst mentions this:
/* * Line 1 * Line 2 */
On 11/1/2021 1:26 PM, Dixit, Ashutosh wrote:
On Sun, 31 Oct 2021 21:39:35 -0700, Belgaumkar, Vinay wrote:
Define helpers and struct members required to record boost info. Boost frequency is initialized to RP0 at SLPC init. Also define num_waiters which can track the pending boost requests.
Boost will be done by scheduling a worker thread. This will allow us to make H2G calls inside an interrupt context. Initialize the
"to not make H2G calls from interrupt context" is probably better.
+static int slpc_force_min_freq(struct intel_guc_slpc *slpc, u32 freq) +{
- struct drm_i915_private *i915 = slpc_to_i915(slpc);
- intel_wakeref_t wakeref;
- int ret = 0;
- lockdep_assert_held(&slpc->lock);
- /**
nit: this I believe should just be
/*
ok.
/** I believe shows up in kerneldoc so shouldn't be used unless we want something in kerneldoc.
* This function is a little different as compared to* intel_guc_slpc_set_min_freq(). Softlimit will not be updated* here since this is used to temporarily change min freq,* for example, during a waitboost. Caller is responsible for* checking bounds.*/- with_intel_runtime_pm(&i915->runtime_pm, wakeref) {
ret = slpc_set_param(slpc,SLPC_PARAM_GLOBAL_MIN_GT_UNSLICE_FREQ_MHZ,freq);if (ret)drm_err(&i915->drm, "Unable to force min freq to %u: %d",Probably drm_err_ratelimited since it's called at run time not only at init? Not sure if drm_err_once suffizes, probably not.
Keeping it drm_err as discussed offline.
freq, ret);- }
- return ret;
+}
+static void slpc_boost_work(struct work_struct *work) +{
- struct intel_guc_slpc *slpc = container_of(work, typeof(*slpc), boost_work);
- /* Raise min freq to boost. It's possible that
* this is greater than current max. But it will* certainly be limited by RP0. An error setting* the min param is not fatal.*/nit: do we follow the following format for multi-line comments, Documentation/process/coding-style.rst mentions this:
/*
- Line 1
- Line 2
*/
Ok.
Thanks, Vinay.
Add helper in RPS code for handling SLPC and non-SLPC paths. When boost is requested in the SLPC path, we can ask GuC to ramp up the frequency req by setting the minimum frequency to boost freq. Reset freq back to the min softlimit when there are no more waiters.
v2: Schedule a worker thread which can boost freq from within an interrupt context as well.
Cc: Ashutosh Dixit ashutosh.dixit@intel.com Signed-off-by: Vinay Belgaumkar vinay.belgaumkar@intel.com --- drivers/gpu/drm/i915/gt/intel_rps.c | 26 +++++++++++++++++++++ drivers/gpu/drm/i915/gt/intel_rps.h | 1 + drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.c | 19 +++++++++++++++ drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.h | 1 + drivers/gpu/drm/i915/i915_request.c | 2 +- 5 files changed, 48 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/i915/gt/intel_rps.c b/drivers/gpu/drm/i915/gt/intel_rps.c index 5e275f8dda8c..b2d5b1747086 100644 --- a/drivers/gpu/drm/i915/gt/intel_rps.c +++ b/drivers/gpu/drm/i915/gt/intel_rps.c @@ -936,8 +936,23 @@ void intel_rps_park(struct intel_rps *rps) GT_TRACE(rps_to_gt(rps), "park:%x\n", rps->cur_freq); }
+void intel_rps_dec_waiters(struct intel_rps *rps) +{ + struct intel_guc_slpc *slpc; + + if (rps_uses_slpc(rps)) { + slpc = rps_to_slpc(rps); + + intel_guc_slpc_dec_waiters(slpc); + } else { + atomic_dec(&rps->num_waiters); + } +} + void intel_rps_boost(struct i915_request *rq) { + struct intel_guc_slpc *slpc; + if (i915_request_signaled(rq) || i915_request_has_waitboost(rq)) return;
@@ -945,6 +960,17 @@ void intel_rps_boost(struct i915_request *rq) if (!test_and_set_bit(I915_FENCE_FLAG_BOOST, &rq->fence.flags)) { struct intel_rps *rps = &READ_ONCE(rq->engine)->gt->rps;
+ if (rps_uses_slpc(rps)) { + slpc = rps_to_slpc(rps); + + /* Return if old value is non zero */ + if (atomic_fetch_inc(&slpc->num_waiters)) + return; + + if (intel_rps_get_requested_frequency(rps) < slpc->boost_freq) + schedule_work(&slpc->boost_work); + } + if (atomic_fetch_inc(&rps->num_waiters)) return;
diff --git a/drivers/gpu/drm/i915/gt/intel_rps.h b/drivers/gpu/drm/i915/gt/intel_rps.h index 11960d64ca82..407e878d5006 100644 --- a/drivers/gpu/drm/i915/gt/intel_rps.h +++ b/drivers/gpu/drm/i915/gt/intel_rps.h @@ -23,6 +23,7 @@ void intel_rps_disable(struct intel_rps *rps); void intel_rps_park(struct intel_rps *rps); void intel_rps_unpark(struct intel_rps *rps); void intel_rps_boost(struct i915_request *rq); +void intel_rps_dec_waiters(struct intel_rps *rps);
int intel_rps_set(struct intel_rps *rps, u8 val); void intel_rps_mark_interactive(struct intel_rps *rps, bool interactive); diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.c b/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.c index cc51987b2535..65da454b6693 100644 --- a/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.c +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.c @@ -445,7 +445,11 @@ int intel_guc_slpc_set_min_freq(struct intel_guc_slpc *slpc, u32 val) val > slpc->max_freq_softlimit) return -EINVAL;
+ /* Need a lock now since waitboost can be modifying min as well */ + mutex_lock(&slpc->lock); + with_intel_runtime_pm(&i915->runtime_pm, wakeref) { + ret = slpc_set_param(slpc, SLPC_PARAM_GLOBAL_MIN_GT_UNSLICE_FREQ_MHZ, val); @@ -458,6 +462,8 @@ int intel_guc_slpc_set_min_freq(struct intel_guc_slpc *slpc, u32 val) if (!ret) slpc->min_freq_softlimit = val;
+ mutex_unlock(&slpc->lock); + return ret; }
@@ -643,6 +649,19 @@ int intel_guc_slpc_enable(struct intel_guc_slpc *slpc) return 0; }
+void intel_guc_slpc_dec_waiters(struct intel_guc_slpc *slpc) +{ + /* Return min back to the softlimit. + * This is called during request retire, + * so we don't need to fail that if the + * set_param fails. + */ + mutex_lock(&slpc->lock); + if (atomic_dec_and_test(&slpc->num_waiters)) + slpc_force_min_freq(slpc, slpc->min_freq_softlimit); + mutex_unlock(&slpc->lock); +} + int intel_guc_slpc_print_info(struct intel_guc_slpc *slpc, struct drm_printer *p) { struct drm_i915_private *i915 = slpc_to_i915(slpc); diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.h b/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.h index b62528647770..d74d6d749bdc 100644 --- a/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.h +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.h @@ -39,5 +39,6 @@ int intel_guc_slpc_get_min_freq(struct intel_guc_slpc *slpc, u32 *val); int intel_guc_slpc_print_info(struct intel_guc_slpc *slpc, struct drm_printer *p); void intel_guc_pm_intrmsk_enable(struct intel_gt *gt); void intel_guc_slpc_boost(struct intel_guc_slpc *slpc); +void intel_guc_slpc_dec_waiters(struct intel_guc_slpc *slpc);
#endif diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c index 2c3cd6e635b5..08f38e86231d 100644 --- a/drivers/gpu/drm/i915/i915_request.c +++ b/drivers/gpu/drm/i915/i915_request.c @@ -339,7 +339,7 @@ bool i915_request_retire(struct i915_request *rq) }
if (test_and_set_bit(I915_FENCE_FLAG_BOOST, &rq->fence.flags)) - atomic_dec(&rq->engine->gt->rps.num_waiters); + intel_rps_dec_waiters(&rq->engine->gt->rps);
/* * We only loosely track inflight requests across preemption,
On Sun, 31 Oct 2021 21:39:36 -0700, Belgaumkar, Vinay wrote:
@@ -945,6 +960,17 @@ void intel_rps_boost(struct i915_request *rq) if (!test_and_set_bit(I915_FENCE_FLAG_BOOST, &rq->fence.flags)) { struct intel_rps *rps = &READ_ONCE(rq->engine)->gt->rps;
if (rps_uses_slpc(rps)) {slpc = rps_to_slpc(rps);/* Return if old value is non zero */if (atomic_fetch_inc(&slpc->num_waiters))return;if (intel_rps_get_requested_frequency(rps) < slpc->boost_freq)
I think this check is not needed because:
a. The waitboost code only changes min_freq. i915 code should not depend on how GuC changes requested_freq in response to change in min_freq.
b. What is more worrisome is that when we "de-boost" we set min_freq to min_freq_softlimit. If GuC e.g. has a delay in bringing requested_freq down and intel_rps_boost() gets called meanwhile we will miss the one opportunity we have to boost the freq (when num_waiters goes from 0 to 1. Asking GuC to boost when actual_freq is already boost_freq is harmless in comparison). So to avoid this risk of missing the chance to boost I think we should delete this check and replace the code above with something like:
if (rps_uses_slpc(rps)) { struct intel_guc_slpc *slpc = rps_to_slpc(rps);
if (slpc->boost_freq <= slpc->min_freq_softlimit) return;
if (!atomic_fetch_inc(&slpc->num_waiters)) schedule_work(&slpc->boost_work);
return; }
Note that this check:
if (slpc->boost_freq <= slpc->min_freq_softlimit) return;
(which is basically a degenerate case in which we don't have to do anything), can be probably be implemented when boost_freq is set in sysfs, or may already be encompassed in "val < slpc->min_freq" in intel_guc_slpc_set_boost_freq() in which case this check can also be skipped from this function.
+void intel_guc_slpc_dec_waiters(struct intel_guc_slpc *slpc) +{
- /* Return min back to the softlimit.
* This is called during request retire,* so we don't need to fail that if the* set_param fails.*/
nit: maybe follow kernel multi-line comment format.
On 11/1/2021 1:28 PM, Dixit, Ashutosh wrote:
On Sun, 31 Oct 2021 21:39:36 -0700, Belgaumkar, Vinay wrote:
@@ -945,6 +960,17 @@ void intel_rps_boost(struct i915_request *rq) if (!test_and_set_bit(I915_FENCE_FLAG_BOOST, &rq->fence.flags)) { struct intel_rps *rps = &READ_ONCE(rq->engine)->gt->rps;
if (rps_uses_slpc(rps)) {slpc = rps_to_slpc(rps);/* Return if old value is non zero */if (atomic_fetch_inc(&slpc->num_waiters))return;if (intel_rps_get_requested_frequency(rps) < slpc->boost_freq)I think this check is not needed because:
a. The waitboost code only changes min_freq. i915 code should not depend on how GuC changes requested_freq in response to change in min_freq.
b. What is more worrisome is that when we "de-boost" we set min_freq to min_freq_softlimit. If GuC e.g. has a delay in bringing requested_freq down and intel_rps_boost() gets called meanwhile we will miss the one opportunity we have to boost the freq (when num_waiters goes from 0 to 1. Asking GuC to boost when actual_freq is already boost_freq is harmless in comparison). So to avoid this risk of missing the chance to boost I think we should delete this check and replace the code above with something like:
if (rps_uses_slpc(rps)) { struct intel_guc_slpc *slpc = rps_to_slpc(rps); if (slpc->boost_freq <= slpc->min_freq_softlimit) return; if (!atomic_fetch_inc(&slpc->num_waiters)) schedule_work(&slpc->boost_work); return; }Note that this check:
if (slpc->boost_freq <= slpc->min_freq_softlimit) return;(which is basically a degenerate case in which we don't have to do anything), can be probably be implemented when boost_freq is set in sysfs, or may already be encompassed in "val < slpc->min_freq" in intel_guc_slpc_set_boost_freq() in which case this check can also be skipped from this function.
We already have that check in set_boost_freq function. So, just adding the atomic_fetch_inc check.
+void intel_guc_slpc_dec_waiters(struct intel_guc_slpc *slpc) +{
- /* Return min back to the softlimit.
* This is called during request retire,* so we don't need to fail that if the* set_param fails.*/nit: maybe follow kernel multi-line comment format.
Ok.
Thanks, Vinay.
Add a helper to sort through the SLPC/RPS paths of get/set methods. Boost frequency will be modified as long as it is within the constraints of RP0 and if it is different from the existing one. We will set min freq to boost only if there is at least one active waiter.
v2: Add num_boosts to guc_slpc_info and changes for worker function
Cc: Ashutosh Dixit ashutosh.dixit@intel.com Signed-off-by: Vinay Belgaumkar vinay.belgaumkar@intel.com --- drivers/gpu/drm/i915/gt/intel_rps.c | 47 +++++++++++++++++++++ drivers/gpu/drm/i915/gt/intel_rps.h | 2 + drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.c | 29 +++++++++++++ drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.h | 1 + drivers/gpu/drm/i915/i915_sysfs.c | 19 ++------- 5 files changed, 82 insertions(+), 16 deletions(-)
diff --git a/drivers/gpu/drm/i915/gt/intel_rps.c b/drivers/gpu/drm/i915/gt/intel_rps.c index b2d5b1747086..21f60fba864f 100644 --- a/drivers/gpu/drm/i915/gt/intel_rps.c +++ b/drivers/gpu/drm/i915/gt/intel_rps.c @@ -936,6 +936,53 @@ void intel_rps_park(struct intel_rps *rps) GT_TRACE(rps_to_gt(rps), "park:%x\n", rps->cur_freq); }
+u32 intel_rps_get_boost_frequency(struct intel_rps *rps) +{ + struct intel_guc_slpc *slpc; + + if (rps_uses_slpc(rps)) { + slpc = rps_to_slpc(rps); + + return slpc->boost_freq; + } else { + return intel_gpu_freq(rps, rps->boost_freq); + } +} + +static int set_boost_freq(struct intel_rps *rps, u32 val) +{ + bool boost = false; + + /* Validate against (static) hardware limits */ + val = intel_freq_opcode(rps, val); + if (val < rps->min_freq || val > rps->max_freq) + return -EINVAL; + + mutex_lock(&rps->lock); + if (val != rps->boost_freq) { + rps->boost_freq = val; + boost = atomic_read(&rps->num_waiters); + } + mutex_unlock(&rps->lock); + if (boost) + schedule_work(&rps->work); + + return 0; +} + +int intel_rps_set_boost_frequency(struct intel_rps *rps, u32 freq) +{ + struct intel_guc_slpc *slpc; + + if (rps_uses_slpc(rps)) { + slpc = rps_to_slpc(rps); + + return intel_guc_slpc_set_boost_freq(slpc, freq); + } else { + return set_boost_freq(rps, freq); + } +} + void intel_rps_dec_waiters(struct intel_rps *rps) { struct intel_guc_slpc *slpc; diff --git a/drivers/gpu/drm/i915/gt/intel_rps.h b/drivers/gpu/drm/i915/gt/intel_rps.h index 407e878d5006..aee12f37d38a 100644 --- a/drivers/gpu/drm/i915/gt/intel_rps.h +++ b/drivers/gpu/drm/i915/gt/intel_rps.h @@ -24,6 +24,8 @@ void intel_rps_park(struct intel_rps *rps); void intel_rps_unpark(struct intel_rps *rps); void intel_rps_boost(struct i915_request *rq); void intel_rps_dec_waiters(struct intel_rps *rps); +u32 intel_rps_get_boost_frequency(struct intel_rps *rps); +int intel_rps_set_boost_frequency(struct intel_rps *rps, u32 freq);
int intel_rps_set(struct intel_rps *rps, u8 val); void intel_rps_mark_interactive(struct intel_rps *rps, bool interactive); diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.c b/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.c index 65da454b6693..285133ae47b0 100644 --- a/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.c +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.c @@ -649,6 +649,33 @@ int intel_guc_slpc_enable(struct intel_guc_slpc *slpc) return 0; }
+int intel_guc_slpc_set_boost_freq(struct intel_guc_slpc *slpc, u32 val) +{ + int ret = 0; + + if (val < slpc->min_freq || val > slpc->rp0_freq) + return -EINVAL; + + mutex_lock(&slpc->lock); + + if (slpc->boost_freq != val) { + /* Apply only if there are active waiters */ + if (atomic_read(&slpc->num_waiters)) { + ret = slpc_force_min_freq(slpc, val); + if (ret) { + ret = -EIO; + goto done; + } + } + + slpc->boost_freq = val; + } + +done: + mutex_unlock(&slpc->lock); + return ret; +} + void intel_guc_slpc_dec_waiters(struct intel_guc_slpc *slpc) { /* Return min back to the softlimit. @@ -685,6 +712,8 @@ int intel_guc_slpc_print_info(struct intel_guc_slpc *slpc, struct drm_printer *p slpc_decode_max_freq(slpc)); drm_printf(p, "\tMin freq: %u MHz\n", slpc_decode_min_freq(slpc)); + drm_printf(p, "\twaitboosts: %u\n", + slpc->num_boosts); } }
diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.h b/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.h index d74d6d749bdc..0caa8fee3c04 100644 --- a/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.h +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_slpc.h @@ -34,6 +34,7 @@ int intel_guc_slpc_enable(struct intel_guc_slpc *slpc); void intel_guc_slpc_fini(struct intel_guc_slpc *slpc); int intel_guc_slpc_set_max_freq(struct intel_guc_slpc *slpc, u32 val); int intel_guc_slpc_set_min_freq(struct intel_guc_slpc *slpc, u32 val); +int intel_guc_slpc_set_boost_freq(struct intel_guc_slpc *slpc, u32 val); int intel_guc_slpc_get_max_freq(struct intel_guc_slpc *slpc, u32 *val); int intel_guc_slpc_get_min_freq(struct intel_guc_slpc *slpc, u32 *val); int intel_guc_slpc_print_info(struct intel_guc_slpc *slpc, struct drm_printer *p); diff --git a/drivers/gpu/drm/i915/i915_sysfs.c b/drivers/gpu/drm/i915/i915_sysfs.c index 1804f4142740..59d441cedc75 100644 --- a/drivers/gpu/drm/i915/i915_sysfs.c +++ b/drivers/gpu/drm/i915/i915_sysfs.c @@ -279,7 +279,7 @@ static ssize_t gt_boost_freq_mhz_show(struct device *kdev, struct device_attribu struct drm_i915_private *i915 = kdev_minor_to_i915(kdev); struct intel_rps *rps = &i915->gt.rps;
- return sysfs_emit(buf, "%d\n", intel_gpu_freq(rps, rps->boost_freq)); + return sysfs_emit(buf, "%d\n", intel_rps_get_boost_frequency(rps)); }
static ssize_t gt_boost_freq_mhz_store(struct device *kdev, @@ -288,7 +288,6 @@ static ssize_t gt_boost_freq_mhz_store(struct device *kdev, { struct drm_i915_private *dev_priv = kdev_minor_to_i915(kdev); struct intel_rps *rps = &dev_priv->gt.rps; - bool boost = false; ssize_t ret; u32 val;
@@ -296,21 +295,9 @@ static ssize_t gt_boost_freq_mhz_store(struct device *kdev, if (ret) return ret;
- /* Validate against (static) hardware limits */ - val = intel_freq_opcode(rps, val); - if (val < rps->min_freq || val > rps->max_freq) - return -EINVAL; - - mutex_lock(&rps->lock); - if (val != rps->boost_freq) { - rps->boost_freq = val; - boost = atomic_read(&rps->num_waiters); - } - mutex_unlock(&rps->lock); - if (boost) - schedule_work(&rps->work); + ret = intel_rps_set_boost_frequency(rps, val);
- return count; + return ret ?: count; }
static ssize_t vlv_rpe_freq_mhz_show(struct device *kdev,
On Sun, 31 Oct 2021 21:39:37 -0700, Belgaumkar, Vinay wrote:
+static int set_boost_freq(struct intel_rps *rps, u32 val)
Since this is legacy rps code path maybe change function name to rps_set_boost_freq?
On Mon, 01 Nov 2021 13:28:14 -0700, Dixit, Ashutosh wrote:
On Sun, 31 Oct 2021 21:39:37 -0700, Belgaumkar, Vinay wrote:
+static int set_boost_freq(struct intel_rps *rps, u32 val)
Since this is legacy rps code path maybe change function name to rps_set_boost_freq?
Not being able to find v3 of this patch so giving a R-b on v2 but the R-b applies to v3:
Reviewed-by: Ashutosh Dixit ashutosh.dixit@intel.com
On Sun, 31 Oct 2021 21:39:34 -0700, Belgaumkar, Vinay wrote:
Waitboost is a legacy feature implemented in the Host Turbo algorithm. This patch set implements it for the SLPC path. A "boost" happens when user calls gem_wait ioctl on a submission that has not landed on HW yet.
Afaiu user doesn't have to call gem_wait, the boost will happen whenever a request waits to be submitted to GuC because of an unmet depedency. This has to be done from i915 because GuC has not yet seen the request.
Rest of the cover letter is fine.
On 11/1/2021 1:24 PM, Dixit, Ashutosh wrote:
On Sun, 31 Oct 2021 21:39:34 -0700, Belgaumkar, Vinay wrote:
Waitboost is a legacy feature implemented in the Host Turbo algorithm. This patch set implements it for the SLPC path. A "boost" happens when user calls gem_wait ioctl on a submission that has not landed on HW yet.
Afaiu user doesn't have to call gem_wait, the boost will happen whenever a request waits to be submitted to GuC because of an unmet depedency. This has to be done from i915 because GuC has not yet seen the request.
Rest of the cover letter is fine.
Ok, thanks, Vinay.
dri-devel@lists.freedesktop.org
-
Belgaumkar, Vinay -
Dixit, Ashutosh -
Vinay Belgaumkar