drm/i915: Add the support of dual BSD rings on BDW GT3
Subsystems (2)
v2 cover drm/i915: Add the support of dual BSD rings on BDW GT3
V1->V2: Follow Daniel's comment to do the following update:
a. consider the stolen check for BDW in kernel/early-quirks.c in patch 01 b. update the comment in Patch 04
c. use the simple ping-pong mechanism to add the support of dual BSD rings.
The further optimization will be considered in another patch set.
This is the patch set that tries to add the support of dual BSD rings on BDW GT3. Based on hardware spec, the BDW GT3 has two independent BSD rings, which can be used to process the video commands. To be simpler, it is transparent to user-space driver/middleware. In such case the kernel driver will decide which ring is to dispatch the BSD video command.
As every BSD ring is powerful, it is enough to dispatch the BSD video command based on the drm fd. In such case the different BSD ring is used for video playing back and encoding.
Zhao Yakui (6):
drm/i915: Split the BDW device definition to prepare for dual BSD
rings on BDW GT3
drm/i915:Initialize the second BSD ring on BDW GT3 machine drm/i915:Handle the irq interrupt for the second BSD ring drm/i915:Add the VCS2 switch in Intel_ring_setup_status_page for Gen7 to
remove the switch check warning
drm/i915:Update the restrict check to filter out wrong Ring ID
passed by user-space
drm/i915:Use the coarse ping-pong mechanism based on drm fd to dispatch the
BSD command on BDW GT3
drivers/gpu/drm/i915/i915_dma.c | 3 ++ drivers/gpu/drm/i915/i915_drv.c | 26 +++++++++++- drivers/gpu/drm/i915/i915_drv.h | 5 +++ drivers/gpu/drm/i915/i915_gem.c | 9 ++++- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 39 +++++++++++++++++- drivers/gpu/drm/i915/i915_gpu_error.c | 1 + drivers/gpu/drm/i915/i915_irq.c | 5 ++- drivers/gpu/drm/i915/i915_reg.h | 1 + drivers/gpu/drm/i915/intel_ringbuffer.c | 59 ++++++++++++++++++++++++++++ drivers/gpu/drm/i915/intel_ringbuffer.h | 5 ++- include/drm/i915_pciids.h | 22 ++++++++--- 11 files changed, 163 insertions(+), 12 deletions(-)
Signature · 1 line
1.7.10.1
PATCH v2 1/6 drm/i915: Split the BDW device definition to prepare for dual BSD rings on BDW GT3
V1->V2: Follow Daniel's comment to consider the stolen check for BDW in kernel/early-quirks.c
Based on the hardware spec, the BDW GT3 has the different configuration with the BDW GT1/GT2. So split the BDW device info definition. This is to do the preparation for adding the Dual BSD rings on BDW GT3 machine.
--- drivers/gpu/drm/i915/i915_drv.c | 26 ++++++++++++++++++++++++-- include/drm/i915_pciids.h | 22 +++++++++++++++++----- 2 files changed, 41 insertions(+), 7 deletions(-)
drivers/gpu/drm/i915/i915_drv.c +24 -2
@@ -279,6 +279,26 @@ static const struct intel_device_info intel_broadwell_m_info = { GEN_DEFAULT_PIPEOFFSETS, }; +static const struct intel_device_info intel_broadwell_gt3d_info = { + .gen =8 , .num_pipes =3 , + .need_gfx_hws =1 , .has_hotplug =1 , + .ring_mask =RENDER_RING |BSD_RING |BLT_RING |VEBOX_RING , + .has_llc =1 , + .has_ddi =1 , + .has_fbc =1 , +GEN_DEFAULT_PIPEOFFSETS , +}; + +static const struct intel_device_info intel_broadwell_gt3m_info = { + .gen =8 , .is_mobile =1 , .num_pipes =3 , + .need_gfx_hws =1 , .has_hotplug =1 , + .ring_mask =RENDER_RING |BSD_RING |BLT_RING |VEBOX_RING , + .has_llc =1 , + .has_ddi =1 , + .has_fbc =1 , +GEN_DEFAULT_PIPEOFFSETS , +}; +/* * Make sure any device matches here are from most specific to most * general. For example, since the Quanta match is based on the subsystem
@@ -311,8 +331,10 @@ static const struct intel_device_info intel_broadwell_m_info = {INTEL_HSW_M_IDS (&intel_haswell_m_info), \INTEL_VLV_M_IDS (&intel_valleyview_m_info), \INTEL_VLV_D_IDS (&intel_valleyview_d_info), \ -INTEL_BDW_M_IDS (&intel_broadwell_m_info), \ - INTEL_BDW_D_IDS(&intel_broadwell_d_info) +INTEL_BDW_GT12M_IDS (&intel_broadwell_m_info), \ +INTEL_BDW_GT12D_IDS (&intel_broadwell_d_info), \ +INTEL_BDW_GT3M_IDS (&intel_broadwell_gt3m_info), \ + INTEL_BDW_GT3D_IDS(&intel_broadwell_gt3d_info)static const struct pci_device_id pciidlist[] = {/* aka */ INTEL_PCI_IDS ,
include/drm/i915_pciids.h +17 -5
@@ -223,14 +223,26 @@_INTEL_BDW_D (gt,0x160A , info),/* Server */ \ _INTEL_BDW_D(gt, 0x160D, info) /* Workstation */ -#define INTEL_BDW_M_IDS (info) \ +#define INTEL_BDW_GT12M_IDS (info) \ _INTEL_BDW_M_IDS(1, info), \ - _INTEL_BDW_M_IDS(2, info), \ - _INTEL_BDW_M_IDS(3, info) + _INTEL_BDW_M_IDS(2, info) -#define INTEL_BDW_D_IDS (info) \ +#define INTEL_BDW_GT12D_IDS (info) \ _INTEL_BDW_D_IDS(1, info), \ - _INTEL_BDW_D_IDS(2, info), \ + _INTEL_BDW_D_IDS(2, info) + +#define INTEL_BDW_GT3M_IDS (info) \ + _INTEL_BDW_M_IDS(3, info) + +#define INTEL_BDW_GT3D_IDS (info) \ _INTEL_BDW_D_IDS(3, info) +#define INTEL_BDW_M_IDS (info) \ + INTEL_BDW_GT12M_IDS(info), \ + INTEL_BDW_GT3M_IDS(info) + +#define INTEL_BDW_D_IDS (info) \ + INTEL_BDW_GT12D_IDS(info), \ + INTEL_BDW_GT3D_IDS(info) +#endif /* _I915_PCIIDS_H */
Signature · 1 line
1.7.10.1
Re: [PATCH V2 1/6] drm/i915: Split the BDW device definition to prepare for dual BSD rings on BDW GT3
Quoted 2 lines · Zhao Yakui
V1->V2: Follow Daniel's comment to consider the stolen check for BDW inkernel/early-quirks.c
Small style nit: We usually put the patch changelog at the end of the commit message. That way the core commit message is clearly separated from the per-patch changelog. In rare cases there's some confusion otherwise. No need to resend just for that.
-Daniel
Quoted 98 lines
Based on the hardware spec, the BDW GT3 has the different configuration with the BDW GT1/GT2. So split the BDW device info definition. This is to do the preparation for adding the Dual BSD rings on BDW GT3 machine.
Signed-off-by Zhao Yakui <[email protected]>--- drivers/gpu/drm/i915/i915_drv.c | 26 ++++++++++++++++++++++++-- include/drm/i915_pciids.h | 22 +++++++++++++++++----- 2 files changed, 41 insertions(+), 7 deletions(-)drivers/gpu/drm/i915/i915_drv.c +24
-2@@ -279,6 +279,26 @@ static const struct intel_device_info intel_broadwell_m_info = { GEN_DEFAULT_PIPEOFFSETS, }; +static const struct intel_device_info intel_broadwell_gt3d_info = { + .gen =8 , .num_pipes =3 , + .need_gfx_hws =1 , .has_hotplug =1 , + .ring_mask =RENDER_RING |BSD_RING |BLT_RING |VEBOX_RING , + .has_llc =1 , + .has_ddi =1 , + .has_fbc =1 , +GEN_DEFAULT_PIPEOFFSETS , +}; + +static const struct intel_device_info intel_broadwell_gt3m_info = { + .gen =8 , .is_mobile =1 , .num_pipes =3 , + .need_gfx_hws =1 , .has_hotplug =1 , + .ring_mask =RENDER_RING |BSD_RING |BLT_RING |VEBOX_RING , + .has_llc =1 , + .has_ddi =1 , + .has_fbc =1 , +GEN_DEFAULT_PIPEOFFSETS , +}; +/* * Make sure any device matches here are from most specific to most * general. For example, since the Quanta match is based on the subsystem @@ -311,8 +331,10 @@ static const struct intel_device_info intel_broadwell_m_info = {INTEL_HSW_M_IDS (&intel_haswell_m_info), \INTEL_VLV_M_IDS (&intel_valleyview_m_info), \INTEL_VLV_D_IDS (&intel_valleyview_d_info), \ -INTEL_BDW_M_IDS (&intel_broadwell_m_info), \ - INTEL_BDW_D_IDS(&intel_broadwell_d_info) +INTEL_BDW_GT12M_IDS (&intel_broadwell_m_info), \ +INTEL_BDW_GT12D_IDS (&intel_broadwell_d_info), \ +INTEL_BDW_GT3M_IDS (&intel_broadwell_gt3m_info), \ + INTEL_BDW_GT3D_IDS(&intel_broadwell_gt3d_info)static const struct pci_device_id pciidlist[] = {/* aka */ INTEL_PCI_IDS ,include/drm/i915_pciids.h +17
-5@@ -223,14 +223,26 @@_INTEL_BDW_D (gt,0x160A , info),/* Server */ \ _INTEL_BDW_D(gt, 0x160D, info) /* Workstation */ -#define INTEL_BDW_M_IDS (info) \ +#define INTEL_BDW_GT12M_IDS (info) \ _INTEL_BDW_M_IDS(1, info), \ - _INTEL_BDW_M_IDS(2, info), \ - _INTEL_BDW_M_IDS(3, info) + _INTEL_BDW_M_IDS(2, info) -#define INTEL_BDW_D_IDS (info) \ +#define INTEL_BDW_GT12D_IDS (info) \ _INTEL_BDW_D_IDS(1, info), \ - _INTEL_BDW_D_IDS(2, info), \ + _INTEL_BDW_D_IDS(2, info) + +#define INTEL_BDW_GT3M_IDS (info) \ + _INTEL_BDW_M_IDS(3, info) + +#define INTEL_BDW_GT3D_IDS (info) \ _INTEL_BDW_D_IDS(3, info) +#define INTEL_BDW_M_IDS (info) \ + INTEL_BDW_GT12M_IDS(info), \ + INTEL_BDW_GT3M_IDS(info) + +#define INTEL_BDW_D_IDS (info) \ + INTEL_BDW_GT12D_IDS(info), \ + INTEL_BDW_GT3D_IDS(info) +#endif /* _I915_PCIIDS_H */ Signature · 6 lines
1.7.10.1 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Re: [PATCH V2 1/6] drm/i915: Split the BDW device definition to prepare for dual BSD rings on BDW GT3
Quoted 8 lines · Daniel Vetter
Quoted 2 lines · Zhao Yakui
V1->V2: Follow Daniel's comment to consider the stolen check for BDW inkernel/early-quirks.cSmall style nit: We usually put the patch changelog at the end of the commit message. That way the core commit message is clearly separated from the per-patch changelog. In rare cases there's some confusion otherwise. No need to resend just for that.
Thanks for your advice.
I will pay attention to the style nit next time.
Thanks.
Yakui
Quoted 101 lines
-Daniel
Quoted 98 lines
Based on the hardware spec, the BDW GT3 has the different configuration with the BDW GT1/GT2. So split the BDW device info definition. This is to do the preparation for adding the Dual BSD rings on BDW GT3 machine.
Signed-off-by Zhao Yakui <[email protected]>--- drivers/gpu/drm/i915/i915_drv.c | 26 ++++++++++++++++++++++++-- include/drm/i915_pciids.h | 22 +++++++++++++++++----- 2 files changed, 41 insertions(+), 7 deletions(-)drivers/gpu/drm/i915/i915_drv.c +24
-2@@ -279,6 +279,26 @@ static const struct intel_device_info intel_broadwell_m_info = { GEN_DEFAULT_PIPEOFFSETS, }; +static const struct intel_device_info intel_broadwell_gt3d_info = { + .gen =8 , .num_pipes =3 , + .need_gfx_hws =1 , .has_hotplug =1 , + .ring_mask =RENDER_RING |BSD_RING |BLT_RING |VEBOX_RING , + .has_llc =1 , + .has_ddi =1 , + .has_fbc =1 , +GEN_DEFAULT_PIPEOFFSETS , +}; + +static const struct intel_device_info intel_broadwell_gt3m_info = { + .gen =8 , .is_mobile =1 , .num_pipes =3 , + .need_gfx_hws =1 , .has_hotplug =1 , + .ring_mask =RENDER_RING |BSD_RING |BLT_RING |VEBOX_RING , + .has_llc =1 , + .has_ddi =1 , + .has_fbc =1 , +GEN_DEFAULT_PIPEOFFSETS , +}; +/* * Make sure any device matches here are from most specific to most * general. For example, since the Quanta match is based on the subsystem @@ -311,8 +331,10 @@ static const struct intel_device_info intel_broadwell_m_info = {INTEL_HSW_M_IDS (&intel_haswell_m_info), \INTEL_VLV_M_IDS (&intel_valleyview_m_info), \INTEL_VLV_D_IDS (&intel_valleyview_d_info), \ -INTEL_BDW_M_IDS (&intel_broadwell_m_info), \ - INTEL_BDW_D_IDS(&intel_broadwell_d_info) +INTEL_BDW_GT12M_IDS (&intel_broadwell_m_info), \ +INTEL_BDW_GT12D_IDS (&intel_broadwell_d_info), \ +INTEL_BDW_GT3M_IDS (&intel_broadwell_gt3m_info), \ + INTEL_BDW_GT3D_IDS(&intel_broadwell_gt3d_info)static const struct pci_device_id pciidlist[] = {/* aka */ INTEL_PCI_IDS ,include/drm/i915_pciids.h +17
-5@@ -223,14 +223,26 @@_INTEL_BDW_D (gt,0x160A , info),/* Server */ \ _INTEL_BDW_D(gt, 0x160D, info) /* Workstation */ -#define INTEL_BDW_M_IDS (info) \ +#define INTEL_BDW_GT12M_IDS (info) \ _INTEL_BDW_M_IDS(1, info), \ - _INTEL_BDW_M_IDS(2, info), \ - _INTEL_BDW_M_IDS(3, info) + _INTEL_BDW_M_IDS(2, info) -#define INTEL_BDW_D_IDS (info) \ +#define INTEL_BDW_GT12D_IDS (info) \ _INTEL_BDW_D_IDS(1, info), \ - _INTEL_BDW_D_IDS(2, info), \ + _INTEL_BDW_D_IDS(2, info) + +#define INTEL_BDW_GT3M_IDS (info) \ + _INTEL_BDW_M_IDS(3, info) + +#define INTEL_BDW_GT3D_IDS (info) \ _INTEL_BDW_D_IDS(3, info) +#define INTEL_BDW_M_IDS (info) \ + INTEL_BDW_GT12M_IDS(info), \ + INTEL_BDW_GT3M_IDS(info) + +#define INTEL_BDW_D_IDS (info) \ + INTEL_BDW_GT12D_IDS(info), \ + INTEL_BDW_GT3D_IDS(info) +#endif /* _I915_PCIIDS_H */ Signature · 6 lines
1.7.10.1 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
PATCH v2 2/6 drm/i915:Initialize the second BSD ring on BDW GT3 machine
In reply to Zhao YakuiBased on the hardware spec, the BDW GT3 machine has two independent BSD ring that can be used to dispatch the video commands. So just initialize it.
--- drivers/gpu/drm/i915/i915_drv.c | 4 +-- drivers/gpu/drm/i915/i915_drv.h | 2 ++ drivers/gpu/drm/i915/i915_gem.c | 9 +++++- drivers/gpu/drm/i915/i915_gpu_error.c | 1 + drivers/gpu/drm/i915/i915_reg.h | 1 + drivers/gpu/drm/i915/intel_ringbuffer.c | 54 +++++++++++++++++++++++++++++++ drivers/gpu/drm/i915/intel_ringbuffer.h | 4 ++- 7 files changed, 71 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/i915_drv.c +2 -2
@@ -282,7 +282,7 @@ static const struct intel_device_info intel_broadwell_m_info = {static const struct intel_device_info intel_broadwell_gt3d_info = { .gen =8 , .num_pipes =3 , .need_gfx_hws =1 , .has_hotplug =1 , - .ring_mask =RENDER_RING |BSD_RING |BLT_RING |VEBOX_RING , + .ring_mask =RENDER_RING |BSD_RING |BLT_RING |VEBOX_RING |BSD2_RING , .has_llc =1 , .has_ddi =1 , .has_fbc =1 ,
@@ -292,7 +292,7 @@ static const struct intel_device_info intel_broadwell_gt3d_info = {static const struct intel_device_info intel_broadwell_gt3m_info = { .gen =8 , .is_mobile =1 , .num_pipes =3 , .need_gfx_hws =1 , .has_hotplug =1 , - .ring_mask =RENDER_RING |BSD_RING |BLT_RING |VEBOX_RING , + .ring_mask =RENDER_RING |BSD_RING |BLT_RING |VEBOX_RING |BSD2_RING , .has_llc =1 , .has_ddi =1 , .has_fbc =1 ,
drivers/gpu/drm/i915/i915_drv.h +2 -0
@@ -1827,7 +1827,9 @@ struct drm_i915_cmd_table {#define BSD_RING (1<<VCS)#define BLT_RING (1<<BCS)#define VEBOX_RING (1<<VECS) +#define BSD2_RING (1<<VCS2)#define HAS_BSD (dev) (INTEL_INFO(dev)->ring_mask & BSD_RING) +#define HAS_BSD2 (dev) (INTEL_INFO(dev)->ring_mask & BSD2_RING)#define HAS_BLT (dev) (INTEL_INFO(dev)->ring_mask & BLT_RING)#define HAS_VEBOX (dev) (INTEL_INFO(dev)->ring_mask & VEBOX_RING)#define HAS_LLC (dev) (INTEL_INFO(dev)->has_llc)
drivers/gpu/drm/i915/i915_gem.c +8 -1
@@ -4374,13 +4374,20 @@ static int i915_gem_init_rings(struct drm_device *dev) goto cleanup_blt_ring; } +if (HAS_BSD2 (dev)) { + ret =intel_init_bsd2_ring_buffer (dev); +if (ret) + goto cleanup_vebox_ring; + } ret =i915_gem_set_seqno (dev, ((u32 )~0 -0x1000 ));if (ret) - goto cleanup_vebox_ring; + goto cleanup_ring;return 0 ; +cleanup_ring: +intel_cleanup_ring_buffer (&dev_priv->ring[VCS2 ]); cleanup_vebox_ring:intel_cleanup_ring_buffer (&dev_priv->ring[VECS ]); cleanup_blt_ring:
drivers/gpu/drm/i915/i915_gpu_error.c +1 -0
@@ -42,6 +42,7 @@ static const char *ring_str(int ring)case VCS :return "bsd" ;case BCS :return "blt" ;case VECS :return "vebox" ; +case VCS2 :return "second bsd" ;default :return "" ; } }
drivers/gpu/drm/i915/i915_reg.h +1 -0
@@ -760,6 +760,7 @@ enum punit_power_well {#define RENDER_RING_BASE 0x02000#define BSD_RING_BASE 0x04000#define GEN6_BSD_RING_BASE 0x12000 +#define GEN8_BSD2_RING_BASE 0x1c000#define VEBOX_RING_BASE 0x1a000#define BLT_RING_BASE 0x22000#define RING_TAIL (base) ((base)+0x30)
drivers/gpu/drm/i915/intel_ringbuffer.c +54 -0
100 lines not on this page · raw message
drivers/gpu/drm/i915/intel_ringbuffer.h +3 -1
@@ -61,8 +61,9 @@ struct intel_ring_buffer {VCS ,BCS ,VECS , +VCS2 , } id; -#define I915_NUM_RINGS 4 +#define I915_NUM_RINGS 5u32 mmio_base;void __iomem *virtual_start;struct drm_device *dev;
@@ -286,6 +287,7 @@ int intel_ring_invalidate_all_caches(struct intel_ring_buffer *ring);int intel_init_render_ring_buffer (struct drm_device *dev);int intel_init_bsd_ring_buffer (struct drm_device *dev); +int intel_init_bsd2_ring_buffer (struct drm_device *dev);int intel_init_blt_ring_buffer (struct drm_device *dev);int intel_init_vebox_ring_buffer (struct drm_device *dev);
Signature · 1 line
1.7.10.1
PATCH v2 3/6 drm/i915:Handle the irq interrupt for the second BSD ring
In reply to Zhao Yakui--- drivers/gpu/drm/i915/i915_irq.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/i915_irq.c +4 -1
@@ -1347,13 +1347,16 @@ static irqreturn_t gen8_gt_irq_handler(struct drm_device *dev,DRM_ERROR ("The master control interrupt lied (GT0)!\n" ); } -if (master_ctl &GEN8_GT_VCS1_IRQ ) { +if (master_ctl & (GEN8_GT_VCS1_IRQ |GEN8_GT_VCS2_IRQ )) { tmp =I915_READ (GEN8_GT_IIR (1 ));if (tmp) { ret =IRQ_HANDLED ; vcs = tmp >>GEN8_VCS1_IRQ_SHIFT ;if (vcs >_RENDER_USER_INTERRUPT )notify_ring (dev, &dev_priv->ring[VCS ]); + vcs = tmp >>GEN8_VCS2_IRQ_SHIFT ; +if (vcs >_RENDER_USER_INTERRUPT ) +notify_ring (dev, &dev_priv->ring[VCS2 ]);I915_WRITE (GEN8_GT_IIR (1 ), tmp); }else DRM_ERROR ("The master control interrupt lied (GT1)!\n" );
Signature · 1 line
1.7.10.1
PATCH v2 4/6 drm/i915: Add the VCS2 switch in Intel_ring_setup_status_page for Gen7 to remove the switch check warning
In reply to Zhao YakuiV1->V2: Follow Daniel's comment to update the comment
The Gen7 doesn't have the second BSD ring. But it will complain the switch check warning message during compilation. So just add it to remove the switch check warning.
--- drivers/gpu/drm/i915/intel_ringbuffer.c | 5 +++++ 1 file changed, 5 insertions(+)
drivers/gpu/drm/i915/intel_ringbuffer.c +5 -0
@@ -988,6 +988,11 @@ void intel_ring_setup_status_page(struct intel_ring_buffer *ring)case BCS : mmio =BLT_HWS_PGA_GEN7 ;break ; +/* +* VCS2 actually doesn't exist on Gen7. Only shut up +* gcc switch check warning +*/ +case VCS2 :case VCS : mmio =BSD_HWS_PGA_GEN7 ;break ;
Signature · 1 line
1.7.10.1
PATCH v2 5/6 drm/i915: Update the restrict check to filter out wrong Ring ID passed by user-space
In reply to Zhao YakuiOne extra ring is added in the kernel driver but it is transparent to the user-space application/middleware. In such case the number of the rings in kernel driver is bigger than that exported to the user-space. So it needs to filter out the wrong Ring ID passed by user-space.
--- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 2 +- drivers/gpu/drm/i915/intel_ringbuffer.h | 1 + 2 files changed, 2 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/i915_gem_execbuffer.c +1 -1
@@ -1035,7 +1035,7 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data,if (args->flags &I915_EXEC_IS_PINNED ) flags |=I915_DISPATCH_PINNED ; -if ((args->flags &I915_EXEC_RING_MASK ) >I915_NUM_RINGS ) { +if ((args->flags &I915_EXEC_RING_MASK ) >LAST_USER_RING ) {DRM_DEBUG ("execbuf with unknown ring: %d\n" , (int )(args->flags &I915_EXEC_RING_MASK ));return -EINVAL ;
drivers/gpu/drm/i915/intel_ringbuffer.h +1 -0
@@ -64,6 +64,7 @@ struct intel_ring_buffer {VCS2 , } id;#define I915_NUM_RINGS 5 +#define LAST_USER_RING (VECS + 1)u32 mmio_base;void __iomem *virtual_start;struct drm_device *dev;
Signature · 1 line
1.7.10.1
PATCH v2 6/6 drm/i915:Use the coarse ping-pong mechanism based on drm fd to dispatch the BSD command on BDW GT3
In reply to Zhao YakuiV1->V2: Follow Daniel's comment and use the simple ping-pong mechanism. This is only to add the support of dual BSD rings on BDW GT3 machine. The further optimization will be considered in another patch set.
The BDW GT3 has two independent BSD rings, which can be used to process the video commands. To be simpler, it is transparent to user-space driver/middle. Instead the kernel driver will decide which ring is to dispatch the BSD video command.
As every BSD ring is powerful, it is enough to dispatch the BSD video command based on the drm fd. In such case it can play back video stream while encoding another video stream. The coarse ping-pong mechanism is used to determine which BSD ring is used to dispatch the BSD video command.
--- drivers/gpu/drm/i915/i915_dma.c | 3 +++ drivers/gpu/drm/i915/i915_drv.h | 3 +++ drivers/gpu/drm/i915/i915_gem_execbuffer.c | 37 +++++++++++++++++++++++++++- 3 files changed, 42 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/i915_dma.c +3 -0
@@ -1572,6 +1572,7 @@ int i915_driver_load(struct drm_device *dev, unsigned long flags)spin_lock_init (&dev_priv->backlight_lock);spin_lock_init (&dev_priv->uncore.lock);spin_lock_init (&dev_priv->mm.object_stat_lock); +atomic_set (&dev_priv->bsd_cmd_counter,0 );mutex_init (&dev_priv->dpio_lock);mutex_init (&dev_priv->modeset_restore_lock);
@@ -1929,6 +1930,8 @@ void i915_driver_postclose(struct drm_device *dev, struct drm_file *file) {struct drm_i915_file_private *file_priv = file->driver_priv; +if (file_priv && file_priv->bsd_ring) + file_priv->bsd_ring =NULL ;kfree (file_priv); }
drivers/gpu/drm/i915/i915_drv.h +3 -0
@@ -1466,6 +1466,8 @@ struct drm_i915_private {struct i915_dri1_state dri1;/* Old ums support infrastructure, same warning applies. */ struct i915_ums_state ums; +/* the lock for dispatch video commands on two BSD rings */ +atomic_t bsd_cmd_counter; };static inline struct drm_i915_private *to_i915 (const struct drm_device *dev)
@@ -1673,6 +1675,7 @@ struct drm_i915_file_private {struct i915_hw_context *private_default_ctx;atomic_t rps_wait_boost; +struct intel_ring_buffer *bsd_ring; };/*
drivers/gpu/drm/i915/i915_gem_execbuffer.c +36 -1
@@ -999,6 +999,34 @@ i915_reset_gen7_sol_offsets(struct drm_device *dev,return 0 ; } +/** +* Find one BSD ring to dispatch the corresponding BSD command. +* The Ring ID is returned. +*/ +static int gen8_dispatch_bsd_ring (struct drm_device *dev, +struct drm_file *file) +{ +struct drm_i915_private *dev_priv = dev->dev_private; +struct drm_i915_file_private *file_priv = file->driver_priv; + +/* Check whether the file_priv is using one ring */ +if (file_priv->bsd_ring) +return file_priv->bsd_ring->id; +else { +/* If no, use the ping-pong mechanism to select one ring */ +int counter, ring_id; +smp_mb__before_atomic_inc (); + counter =atomic_inc_return (&dev_priv->bsd_cmd_counter); +if (counter %2 ==0 ) + ring_id =VCS ; +else + ring_id =VCS2 ; + + file_priv->bsd_ring = &dev_priv->ring[ring_id]; +return ring_id; + } +} + static int i915_gem_do_execbuffer(struct drm_device *dev, void *data, struct drm_file *file,
@@ -1043,7 +1071,14 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data,if ((args->flags &I915_EXEC_RING_MASK ) ==I915_EXEC_DEFAULT ) ring = &dev_priv->ring[RCS ]; -else +else if ((args->flags &I915_EXEC_RING_MASK ) ==I915_EXEC_BSD ) { +if (HAS_BSD2 (dev)) { +int ring_id; + ring_id =gen8_dispatch_bsd_ring (dev, file); + ring = &dev_priv->ring[ring_id]; + }else + ring = &dev_priv->ring[VCS ]; + }else ring = &dev_priv->ring[(args->flags &I915_EXEC_RING_MASK ) -1 ];if (!intel_ring_initialized (ring)) {
Signature · 1 line
1.7.10.1
Re: [PATCH V2 6/6] drm/i915:Use the coarse ping-pong mechanism based on drm fd to dispatch the BSD command on BDW GT3
Quoted 47 lines · Zhao Yakui
V1->V2: Follow Daniel's comment and use the simple ping-pong mechanism. This is only to add the support of dual BSD rings on BDW GT3 machine. The further optimization will be considered in another patch set.The BDW GT3 has two independent BSD rings, which can be used to process the video commands. To be simpler, it is transparent to user-space driver/middle. Instead the kernel driver will decide which ring is to dispatch the BSD video command.
As every BSD ring is powerful, it is enough to dispatch the BSD video command based on the drm fd. In such case it can play back video stream while encoding another video stream. The coarse ping-pong mechanism is used to determine which BSD ring is used to dispatch the BSD video command.
Signed-off-by Zhao Yakui <[email protected]>--- drivers/gpu/drm/i915/i915_dma.c | 3 +++ drivers/gpu/drm/i915/i915_drv.h | 3 +++ drivers/gpu/drm/i915/i915_gem_execbuffer.c | 37 +++++++++++++++++++++++++++- 3 files changed, 42 insertions(+), 1 deletion(-)drivers/gpu/drm/i915/i915_dma.c +3
-0@@ -1572,6 +1572,7 @@ int i915_driver_load(struct drm_device *dev, unsigned long flags)spin_lock_init (&dev_priv->backlight_lock);spin_lock_init (&dev_priv->uncore.lock);spin_lock_init (&dev_priv->mm.object_stat_lock); +atomic_set (&dev_priv->bsd_cmd_counter,0 );mutex_init (&dev_priv->dpio_lock);mutex_init (&dev_priv->modeset_restore_lock);@@ -1929,6 +1930,8 @@ void i915_driver_postclose(struct drm_device *dev, struct drm_file *file) {struct drm_i915_file_private *file_priv = file->driver_priv; +if (file_priv && file_priv->bsd_ring) + file_priv->bsd_ring =NULL ;kfree (file_priv); }drivers/gpu/drm/i915/i915_drv.h +0
-0
drivers/gpu/drm/i915/i915_drv.h struct drm_i915_private { · line 1470
Quoted 5 lines
drivers/gpu/drm/i915/i915_drv.h +2
-0@@ -1466,6 +1466,8 @@ struct drm_i915_private {struct i915_dri1_state dri1;/* Old ums support infrastructure, same warning applies. */ struct i915_ums_state ums; +/* the lock for dispatch video commands on two BSD rings */ +atomic_t bsd_cmd_counter;
You're still using atomic_t for no real good reason. gen8_dispatch_bsd_ring is always called with the dev->struct_mutex lock held, so there's really no reason for it.
-Daniel
Quoted 73 lines
drivers/gpu/drm/i915/i915_drv.h +1
-0};static inline struct drm_i915_private *to_i915 (const struct drm_device *dev)@@ -1673,6 +1675,7 @@ struct drm_i915_file_private {struct i915_hw_context *private_default_ctx;atomic_t rps_wait_boost; +struct intel_ring_buffer *bsd_ring; };/* drivers/gpu/drm/i915/i915_gem_execbuffer.c +36
-1@@ -999,6 +999,34 @@ i915_reset_gen7_sol_offsets(struct drm_device *dev,return 0 ; } +/** +* Find one BSD ring to dispatch the corresponding BSD command. +* The Ring ID is returned. +*/ +static int gen8_dispatch_bsd_ring (struct drm_device *dev, +struct drm_file *file) +{ +struct drm_i915_private *dev_priv = dev->dev_private; +struct drm_i915_file_private *file_priv = file->driver_priv; + +/* Check whether the file_priv is using one ring */ +if (file_priv->bsd_ring) +return file_priv->bsd_ring->id; +else { +/* If no, use the ping-pong mechanism to select one ring */ +int counter, ring_id; +smp_mb__before_atomic_inc (); + counter =atomic_inc_return (&dev_priv->bsd_cmd_counter); +if (counter %2 ==0 ) + ring_id =VCS ; +else + ring_id =VCS2 ; + + file_priv->bsd_ring = &dev_priv->ring[ring_id]; +return ring_id; + } +} + static int i915_gem_do_execbuffer(struct drm_device *dev, void *data, struct drm_file *file,@@ -1043,7 +1071,14 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data,if ((args->flags &I915_EXEC_RING_MASK ) ==I915_EXEC_DEFAULT ) ring = &dev_priv->ring[RCS ]; -else +else if ((args->flags &I915_EXEC_RING_MASK ) ==I915_EXEC_BSD ) { +if (HAS_BSD2 (dev)) { +int ring_id; + ring_id =gen8_dispatch_bsd_ring (dev, file); + ring = &dev_priv->ring[ring_id]; + }else + ring = &dev_priv->ring[VCS ]; + }else ring = &dev_priv->ring[(args->flags &I915_EXEC_RING_MASK ) -1 ];if (!intel_ring_initialized (ring)) {Signature · 6 lines
1.7.10.1 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Re: [PATCH V2 6/6] drm/i915:Use the coarse ping-pong mechanism based on drm fd to dispatch the BSD command on BDW GT3
Quoted 43 lines · Daniel Vetter
Quoted 42 lines · Zhao Yakui
V1->V2: Follow Daniel's comment and use the simple ping-pong mechanism. This is only to add the support of dual BSD rings on BDW GT3 machine. The further optimization will be considered in another patch set.The BDW GT3 has two independent BSD rings, which can be used to process the video commands. To be simpler, it is transparent to user-space driver/middle. Instead the kernel driver will decide which ring is to dispatch the BSD video command.
As every BSD ring is powerful, it is enough to dispatch the BSD video command based on the drm fd. In such case it can play back video stream while encoding another video stream. The coarse ping-pong mechanism is used to determine which BSD ring is used to dispatch the BSD video command.
Signed-off-by Zhao Yakui <[email protected]>--- drivers/gpu/drm/i915/i915_dma.c | 3 +++ drivers/gpu/drm/i915/i915_drv.h | 3 +++ drivers/gpu/drm/i915/i915_gem_execbuffer.c | 37 +++++++++++++++++++++++++++- 3 files changed, 42 insertions(+), 1 deletion(-)drivers/gpu/drm/i915/i915_dma.c +3
-0@@ -1572,6 +1572,7 @@ int i915_driver_load(struct drm_device *dev, unsigned long flags)spin_lock_init (&dev_priv->backlight_lock);spin_lock_init (&dev_priv->uncore.lock);spin_lock_init (&dev_priv->mm.object_stat_lock); +atomic_set (&dev_priv->bsd_cmd_counter,0 );mutex_init (&dev_priv->dpio_lock);mutex_init (&dev_priv->modeset_restore_lock);@@ -1929,6 +1930,8 @@ void i915_driver_postclose(struct drm_device *dev, struct drm_file *file) {struct drm_i915_file_private *file_priv = file->driver_priv; +if (file_priv && file_priv->bsd_ring) + file_priv->bsd_ring =NULL ;kfree (file_priv); }
Quoted 14 lines
drivers/gpu/drm/i915/i915_drv.h struct drm_i915_private { · line 1470
Quoted 10 lines
drivers/gpu/drm/i915/i915_drv.h +2
-0@@ -1466,6 +1466,8 @@ struct drm_i915_private {struct i915_dri1_state dri1;/* Old ums support infrastructure, same warning applies. */ struct i915_ums_state ums; +/* the lock for dispatch video commands on two BSD rings */ +atomic_t bsd_cmd_counter;You're still using
atomic_tfor no real good reason.gen8_dispatch_bsd_ringis always called with thedev->struct_mutexlock held, so there's really no reason for it.
If the struct_mutex is used in the gen8_dispatch_bsd_ring, I can remove the atomic_t.
It seems that the struct_mutex is a big lock and it is used very frequently(i915_gem.c, i915_dma.c and so on). In my point it is a little heavier than the atomic_t if one counter is increased and returned.
If you think that the mutex is better than atomic, I will follow your advice.
Thanks.
Yakui
Quoted 76 lines
-Daniel
Quoted 73 lines
drivers/gpu/drm/i915/i915_drv.h +1
-0};static inline struct drm_i915_private *to_i915 (const struct drm_device *dev)@@ -1673,6 +1675,7 @@ struct drm_i915_file_private {struct i915_hw_context *private_default_ctx;atomic_t rps_wait_boost; +struct intel_ring_buffer *bsd_ring; };/* drivers/gpu/drm/i915/i915_gem_execbuffer.c +36
-1@@ -999,6 +999,34 @@ i915_reset_gen7_sol_offsets(struct drm_device *dev,return 0 ; } +/** +* Find one BSD ring to dispatch the corresponding BSD command. +* The Ring ID is returned. +*/ +static int gen8_dispatch_bsd_ring (struct drm_device *dev, +struct drm_file *file) +{ +struct drm_i915_private *dev_priv = dev->dev_private; +struct drm_i915_file_private *file_priv = file->driver_priv; + +/* Check whether the file_priv is using one ring */ +if (file_priv->bsd_ring) +return file_priv->bsd_ring->id; +else { +/* If no, use the ping-pong mechanism to select one ring */ +int counter, ring_id; +smp_mb__before_atomic_inc (); + counter =atomic_inc_return (&dev_priv->bsd_cmd_counter); +if (counter %2 ==0 ) + ring_id =VCS ; +else + ring_id =VCS2 ; + + file_priv->bsd_ring = &dev_priv->ring[ring_id]; +return ring_id; + } +} + static int i915_gem_do_execbuffer(struct drm_device *dev, void *data, struct drm_file *file,@@ -1043,7 +1071,14 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data,if ((args->flags &I915_EXEC_RING_MASK ) ==I915_EXEC_DEFAULT ) ring = &dev_priv->ring[RCS ]; -else +else if ((args->flags &I915_EXEC_RING_MASK ) ==I915_EXEC_BSD ) { +if (HAS_BSD2 (dev)) { +int ring_id; + ring_id =gen8_dispatch_bsd_ring (dev, file); + ring = &dev_priv->ring[ring_id]; + }else + ring = &dev_priv->ring[VCS ]; + }else ring = &dev_priv->ring[(args->flags &I915_EXEC_RING_MASK ) -1 ];if (!intel_ring_initialized (ring)) {Signature · 6 lines
1.7.10.1 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH V2 6/6] drm/i915:Use the coarse ping-pong mechanism based on drm fd to dispatch the BSD command on BDW GT3
Quoted 5 lines · Zhao Yakui
Quoted 3 lines · Daniel Vetter
You're still using
atomic_tfor no real good reason.gen8_dispatch_bsd_ringis always called with thedev->struct_mutexlock held, so there's really no reason for it.
Quoted 8 lines
If the
struct_mutexis used in thegen8_dispatch_bsd_ring, I can remove theatomic_t.
It seems that thestruct_mutexis a big lock and it is used veryfrequently(i915_gem.c, i915_dma.c and so on). In my point it is a little heavier than theatomic_tif one counter is increased and returned.If you think that the mutex is better than atomic, I will follow your advice.
You are already holding the struct_mutex whenever we touch the ring and execbuffer. Even in a fine-grained world, there will still be a mutex around all operations that touch the rings.
-Chris
Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
Re: [PATCH V2 6/6] drm/i915:Use the coarse ping-pong mechanism based on drm fd to dispatch the BSD command on BDW GT3
Quoted 12 lines · Chris Wilson
Quoted 11 lines · Zhao Yakui
Quoted 3 lines · Daniel Vetter
You're still using
atomic_tfor no real good reason.gen8_dispatch_bsd_ringis always called with thedev->struct_mutexlock held, so there's really no reason for it.If the
struct_mutexis used in thegen8_dispatch_bsd_ring, I can remove theatomic_t.
It seems that thestruct_mutexis a big lock and it is used veryfrequently(i915_gem.c, i915_dma.c and so on). In my point it is a little heavier than theatomic_tif one counter is increased and returned.
Quoted 6 lines
Quoted 2 lines
If you think that the mutex is better than atomic, I will follow your advice.
You are already holding the
struct_mutexwhenever we touch the ring and execbuffer. Even in a fine-grained world, there will still be a mutex around all operations that touch the rings.
Hi, Chris
I understand your concern. From the source code the struct_mutex
will be held when trying to do the buffer relocation and dispatch the command in one ring.
But my code is only to select one BSD ring. In such case theatomic_t usage is enough and it is unnecessary to hold the struct_mutex.
If you also think that the struct_mutex is better, I can update the
code to use the struct_mutex.
Thanks.
Yakui
Quoted 2 lines
-Chris