drm/vc4: Add support for drawing 3D frames.

The user submission is basically a pointer to a command list and a
pointer to uniforms.  We copy those in to the kernel, validate and
relocate them, and store the result in a GPU BO which we queue for
execution.

v2: Drop support for NV shader recs (not necessary for GL), simplify
    vc4_use_bo(), improve bin flush/semaphore checks, use __u32 style
    types.
Signed-off-by: Eric Anholt <eric@anholt.net>

drivers/gpu/drm/vc4/Makefile

@@ -8,12 +8,19 @@ vc4-y := \
 	vc4_crtc.o \
 	vc4_drv.o \
 	vc4_kms.o \
+	vc4_gem.o \
 	vc4_hdmi.o \
 	vc4_hvs.o \
+	vc4_irq.o \
 	vc4_plane.o \
+	vc4_render_cl.o \
+	vc4_trace_points.o \
 	vc4_v3d.o \
+	vc4_validate.o \
 	vc4_validate_shaders.o
 
 vc4-$(CONFIG_DEBUG_FS) += vc4_debugfs.o
 
 obj-$(CONFIG_DRM_VC4)  += vc4.o
+
+CFLAGS_vc4_trace_points.o := -I$(src)

drivers/gpu/drm/vc4/vc4_drv.c

@@ -74,6 +74,9 @@ static const struct file_operations vc4_drm_fops = {
 };
 
 static const struct drm_ioctl_desc vc4_drm_ioctls[] = {
+	DRM_IOCTL_DEF_DRV(VC4_SUBMIT_CL, vc4_submit_cl_ioctl, 0),
+	DRM_IOCTL_DEF_DRV(VC4_WAIT_SEQNO, vc4_wait_seqno_ioctl, 0),
+	DRM_IOCTL_DEF_DRV(VC4_WAIT_BO, vc4_wait_bo_ioctl, 0),
 	DRM_IOCTL_DEF_DRV(VC4_CREATE_BO, vc4_create_bo_ioctl, 0),
 	DRM_IOCTL_DEF_DRV(VC4_MMAP_BO, vc4_mmap_bo_ioctl, 0),
 	DRM_IOCTL_DEF_DRV(VC4_CREATE_SHADER_BO, vc4_create_shader_bo_ioctl, 0),
@@ -83,10 +86,16 @@ static struct drm_driver vc4_drm_driver = {
 	.driver_features = (DRIVER_MODESET |
 			    DRIVER_ATOMIC |
 			    DRIVER_GEM |
+			    DRIVER_HAVE_IRQ |
 			    DRIVER_PRIME),
 	.lastclose = vc4_lastclose,
 	.preclose = vc4_drm_preclose,
 
+	.irq_handler = vc4_irq,
+	.irq_preinstall = vc4_irq_preinstall,
+	.irq_postinstall = vc4_irq_postinstall,
+	.irq_uninstall = vc4_irq_uninstall,
+
 	.enable_vblank = vc4_enable_vblank,
 	.disable_vblank = vc4_disable_vblank,
 	.get_vblank_counter = drm_vblank_count,
@@ -181,9 +190,11 @@ static int vc4_drm_bind(struct device *dev)
 	if (ret)
 		goto unref;
 
+	vc4_gem_init(drm);
+
 	ret = component_bind_all(dev, drm);
 	if (ret)
-		goto unref;
+		goto gem_destroy;
 
 	ret = drm_dev_register(drm, 0);
 	if (ret < 0)
@@ -207,6 +218,8 @@ static int vc4_drm_bind(struct device *dev)
 	drm_dev_unregister(drm);
 unbind_all:
 	component_unbind_all(dev, drm);
+gem_destroy:
+	vc4_gem_destroy(drm);
 unref:
 	drm_dev_unref(drm);
 	vc4_bo_cache_destroy(drm);

drivers/gpu/drm/vc4/vc4_drv.h

@@ -49,6 +49,48 @@ struct vc4_dev {
 
 	/* Protects bo_cache and the BO stats. */
 	struct mutex bo_lock;
+
+	/* Sequence number for the last job queued in job_list.
+	 * Starts at 0 (no jobs emitted).
+	 */
+	uint64_t emit_seqno;
+
+	/* Sequence number for the last completed job on the GPU.
+	 * Starts at 0 (no jobs completed).
+	 */
+	uint64_t finished_seqno;
+
+	/* List of all struct vc4_exec_info for jobs to be executed.
+	 * The first job in the list is the one currently programmed
+	 * into ct0ca/ct1ca for execution.
+	 */
+	struct list_head job_list;
+	/* List of the finished vc4_exec_infos waiting to be freed by
+	 * job_done_work.
+	 */
+	struct list_head job_done_list;
+	/* Spinlock used to synchronize the job_list and seqno
+	 * accesses between the IRQ handler and GEM ioctls.
+	 */
+	spinlock_t job_lock;
+	wait_queue_head_t job_wait_queue;
+	struct work_struct job_done_work;
+
+	/* The binner overflow memory that's currently set up in
+	 * BPOA/BPOS registers.  When overflow occurs and a new one is
+	 * allocated, the previous one will be moved to
+	 * vc4->current_exec's free list.
+	 */
+	struct vc4_bo *overflow_mem;
+	struct work_struct overflow_mem_work;
+
+	struct {
+		uint32_t last_ct0ca, last_ct1ca;
+		struct timer_list timer;
+		struct work_struct reset_work;
+	} hangcheck;
+
+	struct semaphore async_modeset;
 };
 
 static inline struct vc4_dev *
@@ -60,6 +102,9 @@ to_vc4_dev(struct drm_device *dev)
 struct vc4_bo {
 	struct drm_gem_cma_object base;
 
+	/* seqno of the last job to render to this BO. */
+	uint64_t seqno;
+
 	/* List entry for the BO's position in either
 	 * vc4_exec_info->unref_list or vc4_dev->bo_cache.time_list
 	 */
@@ -130,6 +175,101 @@ to_vc4_encoder(struct drm_encoder *encoder)
 #define HVS_READ(offset) readl(vc4->hvs->regs + offset)
 #define HVS_WRITE(offset, val) writel(val, vc4->hvs->regs + offset)
 
+struct vc4_exec_info {
+	/* Sequence number for this bin/render job. */
+	uint64_t seqno;
+
+	/* Kernel-space copy of the ioctl arguments */
+	struct drm_vc4_submit_cl *args;
+
+	/* This is the array of BOs that were looked up at the start of exec.
+	 * Command validation will use indices into this array.
+	 */
+	struct drm_gem_cma_object **bo;
+	uint32_t bo_count;
+
+	/* Pointers for our position in vc4->job_list */
+	struct list_head head;
+
+	/* List of other BOs used in the job that need to be released
+	 * once the job is complete.
+	 */
+	struct list_head unref_list;
+
+	/* Current unvalidated indices into @bo loaded by the non-hardware
+	 * VC4_PACKET_GEM_HANDLES.
+	 */
+	uint32_t bo_index[2];
+
+	/* This is the BO where we store the validated command lists, shader
+	 * records, and uniforms.
+	 */
+	struct drm_gem_cma_object *exec_bo;
+
+	/**
+	 * This tracks the per-shader-record state (packet 64) that
+	 * determines the length of the shader record and the offset
+	 * it's expected to be found at.  It gets read in from the
+	 * command lists.
+	 */
+	struct vc4_shader_state {
+		uint32_t addr;
+		/* Maximum vertex index referenced by any primitive using this
+		 * shader state.
+		 */
+		uint32_t max_index;
+	} *shader_state;
+
+	/** How many shader states the user declared they were using. */
+	uint32_t shader_state_size;
+	/** How many shader state records the validator has seen. */
+	uint32_t shader_state_count;
+
+	bool found_tile_binning_mode_config_packet;
+	bool found_start_tile_binning_packet;
+	bool found_increment_semaphore_packet;
+	bool found_flush;
+	uint8_t bin_tiles_x, bin_tiles_y;
+	struct drm_gem_cma_object *tile_bo;
+	uint32_t tile_alloc_offset;
+
+	/**
+	 * Computed addresses pointing into exec_bo where we start the
+	 * bin thread (ct0) and render thread (ct1).
+	 */
+	uint32_t ct0ca, ct0ea;
+	uint32_t ct1ca, ct1ea;
+
+	/* Pointer to the unvalidated bin CL (if present). */
+	void *bin_u;
+
+	/* Pointers to the shader recs.  These paddr gets incremented as CL
+	 * packets are relocated in validate_gl_shader_state, and the vaddrs
+	 * (u and v) get incremented and size decremented as the shader recs
+	 * themselves are validated.
+	 */
+	void *shader_rec_u;
+	void *shader_rec_v;
+	uint32_t shader_rec_p;
+	uint32_t shader_rec_size;
+
+	/* Pointers to the uniform data.  These pointers are incremented, and
+	 * size decremented, as each batch of uniforms is uploaded.
+	 */
+	void *uniforms_u;
+	void *uniforms_v;
+	uint32_t uniforms_p;
+	uint32_t uniforms_size;
+};
+
+static inline struct vc4_exec_info *
+vc4_first_job(struct vc4_dev *vc4)
+{
+	if (list_empty(&vc4->job_list))
+		return NULL;
+	return list_first_entry(&vc4->job_list, struct vc4_exec_info, head);
+}
+
 /**
  * struct vc4_texture_sample_info - saves the offsets into the UBO for texture
  * setup parameters.
@@ -231,10 +371,31 @@ void vc4_debugfs_cleanup(struct drm_minor *minor);
 /* vc4_drv.c */
 void __iomem *vc4_ioremap_regs(struct platform_device *dev, int index);
 
+/* vc4_gem.c */
+void vc4_gem_init(struct drm_device *dev);
+void vc4_gem_destroy(struct drm_device *dev);
+int vc4_submit_cl_ioctl(struct drm_device *dev, void *data,
+			struct drm_file *file_priv);
+int vc4_wait_seqno_ioctl(struct drm_device *dev, void *data,
+			 struct drm_file *file_priv);
+int vc4_wait_bo_ioctl(struct drm_device *dev, void *data,
+		      struct drm_file *file_priv);
+void vc4_submit_next_job(struct drm_device *dev);
+int vc4_wait_for_seqno(struct drm_device *dev, uint64_t seqno,
+		       uint64_t timeout_ns, bool interruptible);
+void vc4_job_handle_completed(struct vc4_dev *vc4);
+
 /* vc4_hdmi.c */
 extern struct platform_driver vc4_hdmi_driver;
 int vc4_hdmi_debugfs_regs(struct seq_file *m, void *unused);
 
+/* vc4_irq.c */
+irqreturn_t vc4_irq(int irq, void *arg);
+void vc4_irq_preinstall(struct drm_device *dev);
+int vc4_irq_postinstall(struct drm_device *dev);
+void vc4_irq_uninstall(struct drm_device *dev);
+void vc4_irq_reset(struct drm_device *dev);
+
 /* vc4_hvs.c */
 extern struct platform_driver vc4_hvs_driver;
 void vc4_hvs_dump_state(struct drm_device *dev);
@@ -253,6 +414,27 @@ u32 vc4_plane_dlist_size(struct drm_plane_state *state);
 extern struct platform_driver vc4_v3d_driver;
 int vc4_v3d_debugfs_ident(struct seq_file *m, void *unused);
 int vc4_v3d_debugfs_regs(struct seq_file *m, void *unused);
+int vc4_v3d_set_power(struct vc4_dev *vc4, bool on);
+
+/* vc4_validate.c */
+int
+vc4_validate_bin_cl(struct drm_device *dev,
+		    void *validated,
+		    void *unvalidated,
+		    struct vc4_exec_info *exec);
+
+int
+vc4_validate_shader_recs(struct drm_device *dev, struct vc4_exec_info *exec);
+
+struct drm_gem_cma_object *vc4_use_bo(struct vc4_exec_info *exec,
+				      uint32_t hindex);
+
+int vc4_get_rcl(struct drm_device *dev, struct vc4_exec_info *exec);
+
+bool vc4_check_tex_size(struct vc4_exec_info *exec,
+			struct drm_gem_cma_object *fbo,
+			uint32_t offset, uint8_t tiling_format,
+			uint32_t width, uint32_t height, uint8_t cpp);
 
 /* vc4_validate_shader.c */
 struct vc4_validated_shader_info *

drivers/gpu/drm/vc4/vc4_irq.c

+/*
+ * Copyright © 2014 Broadcom
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ */
+
+/** DOC: Interrupt management for the V3D engine.
+ *
+ * We have an interrupt status register (V3D_INTCTL) which reports
+ * interrupts, and where writing 1 bits clears those interrupts.
+ * There are also a pair of interrupt registers
+ * (V3D_INTENA/V3D_INTDIS) where writing a 1 to their bits enables or
+ * disables that specific interrupt, and 0s written are ignored
+ * (reading either one returns the set of enabled interrupts).
+ *
+ * When we take a render frame interrupt, we need to wake the
+ * processes waiting for some frame to be done, and get the next frame
+ * submitted ASAP (so the hardware doesn't sit idle when there's work
+ * to do).
+ *
+ * When we take the binner out of memory interrupt, we need to
+ * allocate some new memory and pass it to the binner so that the
+ * current job can make progress.
+ */
+
+#include "vc4_drv.h"
+#include "vc4_regs.h"
+
+#define V3D_DRIVER_IRQS (V3D_INT_OUTOMEM | \
+			 V3D_INT_FRDONE)
+
+DECLARE_WAIT_QUEUE_HEAD(render_wait);
+
+static void
+vc4_overflow_mem_work(struct work_struct *work)
+{
+	struct vc4_dev *vc4 =
+		container_of(work, struct vc4_dev, overflow_mem_work);
+	struct drm_device *dev = vc4->dev;
+	struct vc4_bo *bo;
+
+	bo = vc4_bo_create(dev, 256 * 1024, true);
+	if (!bo) {
+		DRM_ERROR("Couldn't allocate binner overflow mem\n");
+		return;
+	}
+
+	/* If there's a job executing currently, then our previous
+	 * overflow allocation is getting used in that job and we need
+	 * to queue it to be released when the job is done.  But if no
+	 * job is executing at all, then we can free the old overflow
+	 * object direcctly.
+	 *
+	 * No lock necessary for this pointer since we're the only
+	 * ones that update the pointer, and our workqueue won't
+	 * reenter.
+	 */
+	if (vc4->overflow_mem) {
+		struct vc4_exec_info *current_exec;
+		unsigned long irqflags;
+
+		spin_lock_irqsave(&vc4->job_lock, irqflags);
+		current_exec = vc4_first_job(vc4);
+		if (current_exec) {
+			vc4->overflow_mem->seqno = vc4->finished_seqno + 1;
+			list_add_tail(&vc4->overflow_mem->unref_head,
+				      &current_exec->unref_list);
+			vc4->overflow_mem = NULL;
+		}
+		spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+	}
+
+	if (vc4->overflow_mem)
+		drm_gem_object_unreference_unlocked(&vc4->overflow_mem->base.base);
+	vc4->overflow_mem = bo;
+
+	V3D_WRITE(V3D_BPOA, bo->base.paddr);
+	V3D_WRITE(V3D_BPOS, bo->base.base.size);
+	V3D_WRITE(V3D_INTCTL, V3D_INT_OUTOMEM);
+	V3D_WRITE(V3D_INTENA, V3D_INT_OUTOMEM);
+}
+
+static void
+vc4_irq_finish_job(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_exec_info *exec = vc4_first_job(vc4);
+
+	if (!exec)
+		return;
+
+	vc4->finished_seqno++;
+	list_move_tail(&exec->head, &vc4->job_done_list);
+	vc4_submit_next_job(dev);
+
+	wake_up_all(&vc4->job_wait_queue);
+	schedule_work(&vc4->job_done_work);
+}
+
+irqreturn_t
+vc4_irq(int irq, void *arg)
+{
+	struct drm_device *dev = arg;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	uint32_t intctl;
+	irqreturn_t status = IRQ_NONE;
+
+	barrier();
+	intctl = V3D_READ(V3D_INTCTL);
+
+	/* Acknowledge the interrupts we're handling here. The render
+	 * frame done interrupt will be cleared, while OUTOMEM will
+	 * stay high until the underlying cause is cleared.
+	 */
+	V3D_WRITE(V3D_INTCTL, intctl);
+
+	if (intctl & V3D_INT_OUTOMEM) {
+		/* Disable OUTOMEM until the work is done. */
+		V3D_WRITE(V3D_INTDIS, V3D_INT_OUTOMEM);
+		schedule_work(&vc4->overflow_mem_work);
+		status = IRQ_HANDLED;
+	}
+
+	if (intctl & V3D_INT_FRDONE) {
+		spin_lock(&vc4->job_lock);
+		vc4_irq_finish_job(dev);
+		spin_unlock(&vc4->job_lock);
+		status = IRQ_HANDLED;
+	}
+
+	return status;
+}
+
+void
+vc4_irq_preinstall(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+
+	init_waitqueue_head(&vc4->job_wait_queue);
+	INIT_WORK(&vc4->overflow_mem_work, vc4_overflow_mem_work);
+
+	/* Clear any pending interrupts someone might have left around
+	 * for us.
+	 */
+	V3D_WRITE(V3D_INTCTL, V3D_DRIVER_IRQS);
+}
+
+int
+vc4_irq_postinstall(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+
+	/* Enable both the render done and out of memory interrupts. */
+	V3D_WRITE(V3D_INTENA, V3D_DRIVER_IRQS);
+
+	return 0;
+}
+
+void
+vc4_irq_uninstall(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+
+	/* Disable sending interrupts for our driver's IRQs. */
+	V3D_WRITE(V3D_INTDIS, V3D_DRIVER_IRQS);
+
+	/* Clear any pending interrupts we might have left. */
+	V3D_WRITE(V3D_INTCTL, V3D_DRIVER_IRQS);
+
+	cancel_work_sync(&vc4->overflow_mem_work);
+}
+
+/** Reinitializes interrupt registers when a GPU reset is performed. */
+void vc4_irq_reset(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	unsigned long irqflags;
+
+	/* Acknowledge any stale IRQs. */
+	V3D_WRITE(V3D_INTCTL, V3D_DRIVER_IRQS);
+
+	/*
+	 * Turn all our interrupts on.  Binner out of memory is the
+	 * only one we expect to trigger at this point, since we've
+	 * just come from poweron and haven't supplied any overflow
+	 * memory yet.
+	 */
+	V3D_WRITE(V3D_INTENA, V3D_DRIVER_IRQS);
+
+	spin_lock_irqsave(&vc4->job_lock, irqflags);
+	vc4_irq_finish_job(dev);
+	spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+}

drivers/gpu/drm/vc4/vc4_trace.h

+/*
+ * Copyright (C) 2015 Broadcom
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#if !defined(_VC4_TRACE_H_) || defined(TRACE_HEADER_MULTI_READ)
+#define _VC4_TRACE_H_
+
+#include <linux/stringify.h>
+#include <linux/types.h>
+#include <linux/tracepoint.h>
+
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM vc4
+#define TRACE_INCLUDE_FILE vc4_trace
+
+TRACE_EVENT(vc4_wait_for_seqno_begin,
+	    TP_PROTO(struct drm_device *dev, uint64_t seqno, uint64_t timeout),
+	    TP_ARGS(dev, seqno, timeout),
+
+	    TP_STRUCT__entry(
+			     __field(u32, dev)
+			     __field(u64, seqno)
+			     __field(u64, timeout)
+			     ),
+
+	    TP_fast_assign(
+			   __entry->dev = dev->primary->index;
+			   __entry->seqno = seqno;
+			   __entry->timeout = timeout;
+			   ),
+
+	    TP_printk("dev=%u, seqno=%llu, timeout=%llu",
+		      __entry->dev, __entry->seqno, __entry->timeout)
+);
+
+TRACE_EVENT(vc4_wait_for_seqno_end,
+	    TP_PROTO(struct drm_device *dev, uint64_t seqno),
+	    TP_ARGS(dev, seqno),
+
+	    TP_STRUCT__entry(
+			     __field(u32, dev)
+			     __field(u64, seqno)
+			     ),
+
+	    TP_fast_assign(
+			   __entry->dev = dev->primary->index;
+			   __entry->seqno = seqno;
+			   ),
+
+	    TP_printk("dev=%u, seqno=%llu",
+		      __entry->dev, __entry->seqno)
+);
+
+#endif /* _VC4_TRACE_H_ */
+
+/* This part must be outside protection */
+#undef TRACE_INCLUDE_PATH
+#define TRACE_INCLUDE_PATH .
+#include <trace/define_trace.h>

drivers/gpu/drm/vc4/vc4_trace_points.c

+/*
+ * Copyright (C) 2015 Broadcom
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "vc4_drv.h"
+
+#ifndef __CHECKER__
+#define CREATE_TRACE_POINTS
+#include "vc4_trace.h"
+#endif

drivers/gpu/drm/vc4/vc4_v3d.c

@@ -144,6 +144,21 @@ int vc4_v3d_debugfs_ident(struct seq_file *m, void *unused)
 }
 #endif /* CONFIG_DEBUG_FS */
 
+/*
+ * Asks the firmware to turn on power to the V3D engine.
+ *
+ * This may be doable with just the clocks interface, though this
+ * packet does some other register setup from the firmware, too.
+ */
+int
+vc4_v3d_set_power(struct vc4_dev *vc4, bool on)
+{
+	if (on)
+		return pm_generic_poweroff(&vc4->v3d->pdev->dev);
+	else
+		return pm_generic_resume(&vc4->v3d->pdev->dev);
+}
+
 static void vc4_v3d_init_hw(struct drm_device *dev)
 {
 	struct vc4_dev *vc4 = to_vc4_dev(dev);
@@ -161,6 +176,7 @@ static int vc4_v3d_bind(struct device *dev, struct device *master, void *data)
 	struct drm_device *drm = dev_get_drvdata(master);
 	struct vc4_dev *vc4 = to_vc4_dev(drm);
 	struct vc4_v3d *v3d = NULL;
+	int ret;
 
 	v3d = devm_kzalloc(&pdev->dev, sizeof(*v3d), GFP_KERNEL);
 	if (!v3d)
@@ -180,8 +196,20 @@ static int vc4_v3d_bind(struct device *dev, struct device *master, void *data)
 		return -EINVAL;
 	}
 
+	/* Reset the binner overflow address/size at setup, to be sure
+	 * we don't reuse an old one.
+	 */
+	V3D_WRITE(V3D_BPOA, 0);
+	V3D_WRITE(V3D_BPOS, 0);
+
 	vc4_v3d_init_hw(drm);
 
+	ret = drm_irq_install(drm, platform_get_irq(pdev, 0));
+	if (ret) {
+		DRM_ERROR("Failed to install IRQ handler\n");
+		return ret;
+	}
+
 	return 0;
 }
 
@@ -191,6 +219,15 @@ static void vc4_v3d_unbind(struct device *dev, struct device *master,
 	struct drm_device *drm = dev_get_drvdata(master);
 	struct vc4_dev *vc4 = to_vc4_dev(drm);
 
+	drm_irq_uninstall(drm);
+
+	/* Disable the binner's overflow memory address, so the next
+	 * driver probe (if any) doesn't try to reuse our old
+	 * allocation.
+	 */
+	V3D_WRITE(V3D_BPOA, 0);
+	V3D_WRITE(V3D_BPOS, 0);
+
 	vc4->v3d = NULL;
 }
 

include/uapi/drm/vc4_drm.h

@@ -26,14 +26,155 @@
 
 #include "drm.h"
 
+#define DRM_VC4_SUBMIT_CL                         0x00
+#define DRM_VC4_WAIT_SEQNO                        0x01
+#define DRM_VC4_WAIT_BO                           0x02
 #define DRM_VC4_CREATE_BO                         0x03
 #define DRM_VC4_MMAP_BO                           0x04
 #define DRM_VC4_CREATE_SHADER_BO                  0x05
 
+#define DRM_IOCTL_VC4_SUBMIT_CL           DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_SUBMIT_CL, struct drm_vc4_submit_cl)
+#define DRM_IOCTL_VC4_WAIT_SEQNO          DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_WAIT_SEQNO, struct drm_vc4_wait_seqno)
+#define DRM_IOCTL_VC4_WAIT_BO             DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_WAIT_BO, struct drm_vc4_wait_bo)
 #define DRM_IOCTL_VC4_CREATE_BO           DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_CREATE_BO, struct drm_vc4_create_bo)
 #define DRM_IOCTL_VC4_MMAP_BO             DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_MMAP_BO, struct drm_vc4_mmap_bo)
 #define DRM_IOCTL_VC4_CREATE_SHADER_BO    DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_CREATE_SHADER_BO, struct drm_vc4_create_shader_bo)
 
+struct drm_vc4_submit_rcl_surface {
+	__u32 hindex; /* Handle index, or ~0 if not present. */
+	__u32 offset; /* Offset to start of buffer. */
+	/*
+	 * Bits for either render config (color_write) or load/store packet.
+	 * Bits should all be 0 for MSAA load/stores.
+	 */
+	__u16 bits;
+
+#define VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES		(1 << 0)
+	__u16 flags;
+};
+
+/**
+ * struct drm_vc4_submit_cl - ioctl argument for submitting commands to the 3D
+ * engine.
+ *
+ * Drivers typically use GPU BOs to store batchbuffers / command lists and
+ * their associated state.  However, because the VC4 lacks an MMU, we have to
+ * do validation of memory accesses by the GPU commands.  If we were to store
+ * our commands in BOs, we'd need to do uncached readback from them to do the
+ * validation process, which is too expensive.  Instead, userspace accumulates
+ * commands and associated state in plain memory, then the kernel copies the
+ * data to its own address space, and then validates and stores it in a GPU
+ * BO.
+ */
+struct drm_vc4_submit_cl {
+	/* Pointer to the binner command list.
+	 *
+	 * This is the first set of commands executed, which runs the
+	 * coordinate shader to determine where primitives land on the screen,
+	 * then writes out the state updates and draw calls necessary per tile
+	 * to the tile allocation BO.
+	 */
+	__u64 bin_cl;
+
+	/* Pointer to the shader records.
+	 *
+	 * Shader records are the structures read by the hardware that contain
+	 * pointers to uniforms, shaders, and vertex attributes.  The
+	 * reference to the shader record has enough information to determine
+	 * how many pointers are necessary (fixed number for shaders/uniforms,
+	 * and an attribute count), so those BO indices into bo_handles are
+	 * just stored as __u32s before each shader record passed in.
+	 */
+	__u64 shader_rec;
+
+	/* Pointer to uniform data and texture handles for the textures
+	 * referenced by the shader.
+	 *
+	 * For each shader state record, there is a set of uniform data in the
+	 * order referenced by the record (FS, VS, then CS).  Each set of
+	 * uniform data has a __u32 index into bo_handles per texture
+	 * sample operation, in the order the QPU_W_TMUn_S writes appear in
+	 * the program.  Following the texture BO handle indices is the actual
+	 * uniform data.
+	 *
+	 * The individual uniform state blocks don't have sizes passed in,
+	 * because the kernel has to determine the sizes anyway during shader
+	 * code validation.
+	 */
+	__u64 uniforms;
+	__u64 bo_handles;
+
+	/* Size in bytes of the binner command list. */
+	__u32 bin_cl_size;
+	/* Size in bytes of the set of shader records. */
+	__u32 shader_rec_size;
+	/* Number of shader records.
+	 *
+	 * This could just be computed from the contents of shader_records and
+	 * the address bits of references to them from the bin CL, but it
+	 * keeps the kernel from having to resize some allocations it makes.
+	 */
+	__u32 shader_rec_count;
+	/* Size in bytes of the uniform state. */
+	__u32 uniforms_size;
+
+	/* Number of BO handles passed in (size is that times 4). */
+	__u32 bo_handle_count;
+
+	/* RCL setup: */
+	__u16 width;
+	__u16 height;
+	__u8 min_x_tile;
+	__u8 min_y_tile;
+	__u8 max_x_tile;
+	__u8 max_y_tile;
+	struct drm_vc4_submit_rcl_surface color_read;
+	struct drm_vc4_submit_rcl_surface color_write;
+	struct drm_vc4_submit_rcl_surface zs_read;
+	struct drm_vc4_submit_rcl_surface zs_write;
+	struct drm_vc4_submit_rcl_surface msaa_color_write;
+	struct drm_vc4_submit_rcl_surface msaa_zs_write;
+	__u32 clear_color[2];
+	__u32 clear_z;
+	__u8 clear_s;
+
+	__u32 pad:24;
+
+#define VC4_SUBMIT_CL_USE_CLEAR_COLOR			(1 << 0)
+	__u32 flags;
+
+	/* Returned value of the seqno of this render job (for the
+	 * wait ioctl).
+	 */
+	__u64 seqno;
+};
+
+/**
+ * struct drm_vc4_wait_seqno - ioctl argument for waiting for
+ * DRM_VC4_SUBMIT_CL completion using its returned seqno.
+ *
+ * timeout_ns is the timeout in nanoseconds, where "0" means "don't
+ * block, just return the status."
+ */
+struct drm_vc4_wait_seqno {
+	__u64 seqno;
+	__u64 timeout_ns;
+};
+
+/**
+ * struct drm_vc4_wait_bo - ioctl argument for waiting for
+ * completion of the last DRM_VC4_SUBMIT_CL on a BO.
+ *
+ * This is useful for cases where multiple processes might be
+ * rendering to a BO and you want to wait for all rendering to be
+ * completed.
+ */
+struct drm_vc4_wait_bo {
+	__u32 handle;
+	__u32 pad;
+	__u64 timeout_ns;
+};
+
 /**
  * struct drm_vc4_create_bo - ioctl argument for creating VC4 BOs.
  *