[Intel-gfx] [PATCH v4] drm/i915/mtl: Media GT and Render GT share common GGTT

Iddamsetty, Aravind aravind.iddamsetty at intel.com
Tue Nov 29 11:25:00 UTC 2022



On 29-11-2022 15:41, Tvrtko Ursulin wrote:
> 
> On 22/11/2022 07:01, Aravind Iddamsetty wrote:
>> On XE_LPM+ platforms the media engines are carved out into a separate
>> GT but have a common GGTMMADR address range which essentially makes
>> the GGTT address space to be shared between media and render GT. As a
>> result any updates in GGTT shall invalidate TLB of GTs sharing it and
>> similarly any operation on GGTT requiring an action on a GT will have to
>> involve all GTs sharing it. setup_private_pat was being done on a per
>> GGTT based as that doesn't touch any GGTT structures moved it to per GT
>> based.
>>
>> BSPEC: 63834
>>
>> v2:
>> 1. Add details to commit msg
>> 2. includes fix for failure to add item to ggtt->gt_list, as suggested
>> by Lucas
>> 3. as ggtt_flush() is used only for ggtt drop i915_is_ggtt check within
>> it.
>> 4. setup_private_pat moved out of intel_gt_tiles_init
>>
>> v3:
>> 1. Move out for_each_gt from i915_driver.c (Jani Nikula)
>>
>> v4: drop using RCU primitives on ggtt->gt_list as it is not an RCU list
>> (Matt Roper)
>>
>> Cc: Matt Roper <matthew.d.roper at intel.com>
>> Signed-off-by: Aravind Iddamsetty <aravind.iddamsetty at intel.com>
>> ---
>>   drivers/gpu/drm/i915/gt/intel_ggtt.c      | 54 +++++++++++++++++------
>>   drivers/gpu/drm/i915/gt/intel_gt.c        | 13 +++++-
>>   drivers/gpu/drm/i915/gt/intel_gt_types.h  |  3 ++
>>   drivers/gpu/drm/i915/gt/intel_gtt.h       |  4 ++
>>   drivers/gpu/drm/i915/i915_driver.c        | 12 ++---
>>   drivers/gpu/drm/i915/i915_gem.c           |  2 +
>>   drivers/gpu/drm/i915/i915_gem_evict.c     | 51 +++++++++++++++------
>>   drivers/gpu/drm/i915/i915_vma.c           |  5 ++-
>>   drivers/gpu/drm/i915/selftests/i915_gem.c |  2 +
>>   9 files changed, 111 insertions(+), 35 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/i915/gt/intel_ggtt.c
>> b/drivers/gpu/drm/i915/gt/intel_ggtt.c
>> index 8145851ad23d..7644738b9cdb 100644
>> --- a/drivers/gpu/drm/i915/gt/intel_ggtt.c
>> +++ b/drivers/gpu/drm/i915/gt/intel_ggtt.c
>> @@ -8,6 +8,7 @@
>>   #include <linux/types.h>
>>   #include <linux/stop_machine.h>
>>   +#include <drm/drm_managed.h>
>>   #include <drm/i915_drm.h>
>>   #include <drm/intel-gtt.h>
>>   @@ -196,10 +197,13 @@ void i915_ggtt_suspend_vm(struct
>> i915_address_space *vm)
>>     void i915_ggtt_suspend(struct i915_ggtt *ggtt)
>>   {
>> +    struct intel_gt *gt;
>> +
>>       i915_ggtt_suspend_vm(&ggtt->vm);
>>       ggtt->invalidate(ggtt);
>>   -    intel_gt_check_and_clear_faults(ggtt->vm.gt);
>> +    list_for_each_entry(gt, &ggtt->gt_list, ggtt_link)
>> +        intel_gt_check_and_clear_faults(gt);
>>   }
>>     void gen6_ggtt_invalidate(struct i915_ggtt *ggtt)
>> @@ -225,16 +229,21 @@ static void gen8_ggtt_invalidate(struct
>> i915_ggtt *ggtt)
>>     static void guc_ggtt_invalidate(struct i915_ggtt *ggtt)
>>   {
>> -    struct intel_uncore *uncore = ggtt->vm.gt->uncore;
>>       struct drm_i915_private *i915 = ggtt->vm.i915;
>>         gen8_ggtt_invalidate(ggtt);
>>   -    if (GRAPHICS_VER(i915) >= 12)
>> -        intel_uncore_write_fw(uncore, GEN12_GUC_TLB_INV_CR,
>> -                      GEN12_GUC_TLB_INV_CR_INVALIDATE);
>> -    else
>> -        intel_uncore_write_fw(uncore, GEN8_GTCR, GEN8_GTCR_INVALIDATE);
>> +    if (GRAPHICS_VER(i915) >= 12) {
>> +        struct intel_gt *gt;
>> +
>> +        list_for_each_entry(gt, &ggtt->gt_list, ggtt_link)
>> +            intel_uncore_write_fw(gt->uncore,
>> +                          GEN12_GUC_TLB_INV_CR,
>> +                          GEN12_GUC_TLB_INV_CR_INVALIDATE);
>> +    } else {
>> +        intel_uncore_write_fw(ggtt->vm.gt->uncore,
>> +                      GEN8_GTCR, GEN8_GTCR_INVALIDATE);
>> +    }
>>   }
>>     u64 gen8_ggtt_pte_encode(dma_addr_t addr,
>> @@ -986,8 +995,6 @@ static int gen8_gmch_probe(struct i915_ggtt *ggtt)
>>         ggtt->vm.pte_encode = gen8_ggtt_pte_encode;
>>   -    setup_private_pat(ggtt->vm.gt);
>> -
>>       return ggtt_probe_common(ggtt, size);
>>   }
>>   @@ -1196,7 +1203,14 @@ static int ggtt_probe_hw(struct i915_ggtt
>> *ggtt, struct intel_gt *gt)
>>    */
>>   int i915_ggtt_probe_hw(struct drm_i915_private *i915)
>>   {
>> -    int ret;
>> +    struct intel_gt *gt;
>> +    int ret, i;
>> +
>> +    for_each_gt(gt, i915, i) {
>> +        ret = intel_gt_assign_ggtt(gt);
>> +        if (ret)
>> +            return ret;
>> +    }
>>         ret = ggtt_probe_hw(to_gt(i915)->ggtt, to_gt(i915));
>>       if (ret)
>> @@ -1208,6 +1222,19 @@ int i915_ggtt_probe_hw(struct drm_i915_private
>> *i915)
>>       return 0;
>>   }
>>   +struct i915_ggtt *i915_ggtt_create(struct drm_i915_private *i915)
>> +{
>> +    struct i915_ggtt *ggtt;
>> +
>> +    ggtt = drmm_kzalloc(&i915->drm, sizeof(*ggtt), GFP_KERNEL);
>> +    if (!ggtt)
>> +        return ERR_PTR(-ENOMEM);
>> +
>> +    INIT_LIST_HEAD(&ggtt->gt_list);
>> +
>> +    return ggtt;
>> +}
>> +
>>   int i915_ggtt_enable_hw(struct drm_i915_private *i915)
>>   {
>>       if (GRAPHICS_VER(i915) < 6)
>> @@ -1296,9 +1323,11 @@ bool i915_ggtt_resume_vm(struct
>> i915_address_space *vm)
>>     void i915_ggtt_resume(struct i915_ggtt *ggtt)
>>   {
>> +    struct intel_gt *gt;
>>       bool flush;
>>   -    intel_gt_check_and_clear_faults(ggtt->vm.gt);
>> +    list_for_each_entry(gt, &ggtt->gt_list, ggtt_link)
>> +        intel_gt_check_and_clear_faults(gt);
>>         flush = i915_ggtt_resume_vm(&ggtt->vm);
>>   @@ -1307,9 +1336,6 @@ void i915_ggtt_resume(struct i915_ggtt *ggtt)
>>       if (flush)
>>           wbinvd_on_all_cpus();
>>   -    if (GRAPHICS_VER(ggtt->vm.i915) >= 8)
>> -        setup_private_pat(ggtt->vm.gt);
> 
> Moving this really should have been a separate patch.
> 
>> -
>>       intel_ggtt_restore_fences(ggtt);
>>   }
>>   diff --git a/drivers/gpu/drm/i915/gt/intel_gt.c
>> b/drivers/gpu/drm/i915/gt/intel_gt.c
>> index b5ad9caa5537..b03788d7674e 100644
>> --- a/drivers/gpu/drm/i915/gt/intel_gt.c
>> +++ b/drivers/gpu/drm/i915/gt/intel_gt.c
>> @@ -110,9 +110,18 @@ static int intel_gt_probe_lmem(struct intel_gt *gt)
>>     int intel_gt_assign_ggtt(struct intel_gt *gt)
>>   {
>> -    gt->ggtt = drmm_kzalloc(&gt->i915->drm, sizeof(*gt->ggtt),
>> GFP_KERNEL);
>> +    /* Media GT shares primary GT's GGTT */
>> +    if (gt->type == GT_MEDIA) {
>> +        gt->ggtt = to_gt(gt->i915)->ggtt;
> 
> AFAICT this creates two implicit assumptions: 1) That for_each_gt
> iterates in a certain order (primary always first), when it is calling
> in here; and 2) That the primary tile is not media. Ideally a
> GEM_BUG_ON(!gt->ggtt) would cover for it, since I am not sure the
> list_add_tail below is guaranteed to explode or not.

Ok makes sense.
> 
>> +    } else {
>> +        gt->ggtt = i915_ggtt_create(gt->i915);
>> +        if (IS_ERR(gt->ggtt))
>> +            return PTR_ERR(gt->ggtt);
>> +    }
>>   -    return gt->ggtt ? 0 : -ENOMEM;
>> +    list_add_tail(&gt->ggtt_link, &gt->ggtt->gt_list);
>> +
>> +    return 0;
>>   }
>>     int intel_gt_init_mmio(struct intel_gt *gt)
>> diff --git a/drivers/gpu/drm/i915/gt/intel_gt_types.h
>> b/drivers/gpu/drm/i915/gt/intel_gt_types.h
>> index c1d9cd255e06..8d915640914b 100644
>> --- a/drivers/gpu/drm/i915/gt/intel_gt_types.h
>> +++ b/drivers/gpu/drm/i915/gt/intel_gt_types.h
>> @@ -277,6 +277,9 @@ struct intel_gt {
>>       struct kobject *sysfs_defaults;
>>         struct i915_perf_gt perf;
>> +
>> +    /** link: &ggtt.gt_list */
>> +    struct list_head ggtt_link;
>>   };
>>     struct intel_gt_definition {
>> diff --git a/drivers/gpu/drm/i915/gt/intel_gtt.h
>> b/drivers/gpu/drm/i915/gt/intel_gtt.h
>> index 4d75ba4bb41d..d1900fec6cd1 100644
>> --- a/drivers/gpu/drm/i915/gt/intel_gtt.h
>> +++ b/drivers/gpu/drm/i915/gt/intel_gtt.h
>> @@ -390,6 +390,9 @@ struct i915_ggtt {
>>       struct mutex error_mutex;
>>       struct drm_mm_node error_capture;
>>       struct drm_mm_node uc_fw;
>> +
>> +    /** List of GTs mapping this GGTT */
>> +    struct list_head gt_list;
>>   };
>>     struct i915_ppgtt {
>> @@ -584,6 +587,7 @@ void i915_ggtt_disable_guc(struct i915_ggtt *ggtt);
>>   int i915_init_ggtt(struct drm_i915_private *i915);
>>   void i915_ggtt_driver_release(struct drm_i915_private *i915);
>>   void i915_ggtt_driver_late_release(struct drm_i915_private *i915);
>> +struct i915_ggtt *i915_ggtt_create(struct drm_i915_private *i915);
>>     static inline bool i915_ggtt_has_aperture(const struct i915_ggtt
>> *ggtt)
>>   {
>> diff --git a/drivers/gpu/drm/i915/i915_driver.c
>> b/drivers/gpu/drm/i915/i915_driver.c
>> index 69103ae37779..4e1bb3c23c63 100644
>> --- a/drivers/gpu/drm/i915/i915_driver.c
>> +++ b/drivers/gpu/drm/i915/i915_driver.c
>> @@ -612,10 +612,6 @@ static int i915_driver_hw_probe(struct
>> drm_i915_private *dev_priv)
>>         i915_perf_init(dev_priv);
>>   -    ret = intel_gt_assign_ggtt(to_gt(dev_priv));
>> -    if (ret)
>> -        goto err_perf;
>> -
>>       ret = i915_ggtt_probe_hw(dev_priv);
>>       if (ret)
>>           goto err_perf;
>> @@ -1316,7 +1312,8 @@ int i915_driver_suspend_switcheroo(struct
>> drm_i915_private *i915,
>>   static int i915_drm_resume(struct drm_device *dev)
>>   {
>>       struct drm_i915_private *dev_priv = to_i915(dev);
>> -    int ret;
>> +    struct intel_gt *gt;
>> +    int ret, i;
>>         disable_rpm_wakeref_asserts(&dev_priv->runtime_pm);
>>   @@ -1331,6 +1328,11 @@ static int i915_drm_resume(struct drm_device
>> *dev)
>>           drm_err(&dev_priv->drm, "failed to re-enable GGTT\n");
>>         i915_ggtt_resume(to_gt(dev_priv)->ggtt);
>> +
>> +    for_each_gt(gt, dev_priv, i)
>> +        if (GRAPHICS_VER(gt->i915) >= 8)
>> +            setup_private_pat(gt);
> 
> If this is now called from i915_driver.c, the >= 8 check should probably
> go into setup_private_pat. And exported function renamed to
> intel_gt_setup_private_pat. Otherwise it feels like top level code has a
> little bit of too much deep knowledge of things.

got it.
> 
>> +
>>       /* Must be called after GGTT is resumed. */
>>       intel_dpt_resume(dev_priv);
>>   diff --git a/drivers/gpu/drm/i915/i915_gem.c
>> b/drivers/gpu/drm/i915/i915_gem.c
>> index 8468ca9885fd..086c4702e1bf 100644
>> --- a/drivers/gpu/drm/i915/i915_gem.c
>> +++ b/drivers/gpu/drm/i915/i915_gem.c
>> @@ -1143,6 +1143,8 @@ int i915_gem_init(struct drm_i915_private
>> *dev_priv)
>>       for_each_gt(gt, dev_priv, i) {
>>           intel_uc_fetch_firmwares(&gt->uc);
>>           intel_wopcm_init(&gt->wopcm);
>> +        if (GRAPHICS_VER(dev_priv) >= 8)
>> +            setup_private_pat(gt);
>>       }
>>         ret = i915_init_ggtt(dev_priv);
>> diff --git a/drivers/gpu/drm/i915/i915_gem_evict.c
>> b/drivers/gpu/drm/i915/i915_gem_evict.c
>> index f025ee4fa526..4cfe36b0366b 100644
>> --- a/drivers/gpu/drm/i915/i915_gem_evict.c
>> +++ b/drivers/gpu/drm/i915/i915_gem_evict.c
>> @@ -43,16 +43,25 @@ static bool dying_vma(struct i915_vma *vma)
>>       return !kref_read(&vma->obj->base.refcount);
>>   }
>>   -static int ggtt_flush(struct intel_gt *gt)
>> +static int ggtt_flush(struct i915_address_space *vm)
>>   {
>> -    /*
>> -     * Not everything in the GGTT is tracked via vma (otherwise we
>> -     * could evict as required with minimal stalling) so we are forced
>> -     * to idle the GPU and explicitly retire outstanding requests in
>> -     * the hopes that we can then remove contexts and the like only
>> -     * bound by their active reference.
>> -     */
>> -    return intel_gt_wait_for_idle(gt, MAX_SCHEDULE_TIMEOUT);
>> +    struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm);
>> +    struct intel_gt *gt;
>> +    int ret = 0;
>> +
>> +    list_for_each_entry(gt, &ggtt->gt_list, ggtt_link) {
>> +        /*
>> +         * Not everything in the GGTT is tracked via vma (otherwise we
>> +         * could evict as required with minimal stalling) so we are
>> forced
>> +         * to idle the GPU and explicitly retire outstanding requests in
>> +         * the hopes that we can then remove contexts and the like only
>> +         * bound by their active reference.
>> +         */
>> +        ret = intel_gt_wait_for_idle(gt, MAX_SCHEDULE_TIMEOUT);
>> +        if (ret)
>> +            return ret;
>> +    }
>> +    return ret;
>>   }
>>     static bool grab_vma(struct i915_vma *vma, struct i915_gem_ww_ctx
>> *ww)
>> @@ -149,6 +158,7 @@ i915_gem_evict_something(struct i915_address_space
>> *vm,
>>       struct drm_mm_node *node;
>>       enum drm_mm_insert_mode mode;
>>       struct i915_vma *active;
>> +    struct intel_gt *gt;
> 
> Declare where it is used, like in i915_gem_evict_for_node? Or maybe add
> a local helper like vm_retire_requests, not sure?
sure.

Thanks for your review Tvrtko, I would be doing these changes a little
later. Hope it should be fine.

Thanks,
Aravind.
> 
> Regards,
> 
> Tvrtko
> 
>>       int ret;
>>         lockdep_assert_held(&vm->mutex);
>> @@ -174,7 +184,14 @@ i915_gem_evict_something(struct
>> i915_address_space *vm,
>>                       min_size, alignment, color,
>>                       start, end, mode);
>>   -    intel_gt_retire_requests(vm->gt);
>> +    if (i915_is_ggtt(vm)) {
>> +        struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm);
>> +
>> +        list_for_each_entry(gt, &ggtt->gt_list, ggtt_link)
>> +            intel_gt_retire_requests(gt);
>> +    } else {
>> +        intel_gt_retire_requests(vm->gt);
>> +    }
>>     search_again:
>>       active = NULL;
>> @@ -246,7 +263,7 @@ i915_gem_evict_something(struct i915_address_space
>> *vm,
>>       if (I915_SELFTEST_ONLY(igt_evict_ctl.fail_if_busy))
>>           return -EBUSY;
>>   -    ret = ggtt_flush(vm->gt);
>> +    ret = ggtt_flush(vm);
>>       if (ret)
>>           return ret;
>>   @@ -332,7 +349,15 @@ int i915_gem_evict_for_node(struct
>> i915_address_space *vm,
>>        * a stray pin (preventing eviction) that can only be resolved by
>>        * retiring.
>>        */
>> -    intel_gt_retire_requests(vm->gt);
>> +    if (i915_is_ggtt(vm)) {
>> +        struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm);
>> +        struct intel_gt *gt;
>> +
>> +        list_for_each_entry(gt, &ggtt->gt_list, ggtt_link)
>> +            intel_gt_retire_requests(gt);
>> +    } else {
>> +        intel_gt_retire_requests(vm->gt);
>> +    }
>>         if (i915_vm_has_cache_coloring(vm)) {
>>           /* Expand search to cover neighbouring guard pages (or
>> lack!) */
>> @@ -438,7 +463,7 @@ int i915_gem_evict_vm(struct i915_address_space
>> *vm, struct i915_gem_ww_ctx *ww)
>>        * switch otherwise is ineffective.
>>        */
>>       if (i915_is_ggtt(vm)) {
>> -        ret = ggtt_flush(vm->gt);
>> +        ret = ggtt_flush(vm);
>>           if (ret)
>>               return ret;
>>       }
>> diff --git a/drivers/gpu/drm/i915/i915_vma.c
>> b/drivers/gpu/drm/i915/i915_vma.c
>> index 703fee6b5f75..726705b10637 100644
>> --- a/drivers/gpu/drm/i915/i915_vma.c
>> +++ b/drivers/gpu/drm/i915/i915_vma.c
>> @@ -1544,6 +1544,8 @@ static int __i915_ggtt_pin(struct i915_vma *vma,
>> struct i915_gem_ww_ctx *ww,
>>                  u32 align, unsigned int flags)
>>   {
>>       struct i915_address_space *vm = vma->vm;
>> +    struct intel_gt *gt;
>> +    struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm);
>>       int err;
>>         do {
>> @@ -1559,7 +1561,8 @@ static int __i915_ggtt_pin(struct i915_vma *vma,
>> struct i915_gem_ww_ctx *ww,
>>           }
>>             /* Unlike i915_vma_pin, we don't take no for an answer! */
>> -        flush_idle_contexts(vm->gt);
>> +        list_for_each_entry(gt, &ggtt->gt_list, ggtt_link)
>> +            flush_idle_contexts(gt);
>>           if (mutex_lock_interruptible(&vm->mutex) == 0) {
>>               /*
>>                * We pass NULL ww here, as we don't want to unbind
>> diff --git a/drivers/gpu/drm/i915/selftests/i915_gem.c
>> b/drivers/gpu/drm/i915/selftests/i915_gem.c
>> index e5dd82e7e480..2535b9684bd1 100644
>> --- a/drivers/gpu/drm/i915/selftests/i915_gem.c
>> +++ b/drivers/gpu/drm/i915/selftests/i915_gem.c
>> @@ -127,6 +127,8 @@ static void igt_pm_resume(struct drm_i915_private
>> *i915)
>>        */
>>       with_intel_runtime_pm(&i915->runtime_pm, wakeref) {
>>           i915_ggtt_resume(to_gt(i915)->ggtt);
>> +        if (GRAPHICS_VER(i915) >= 8)
>> +            setup_private_pat(to_gt(i915));
>>           i915_gem_resume(i915);
>>       }
>>   }


More information about the dri-devel mailing list