drm/vmwgfx: Add basic support for SVGA3

SVGA3 is the next version of our PCI device. Some of the changes
include using MMIO for register accesses instead of ioports,
deprecating the FIFO MMIO and removing a lot of the old and
legacy functionality. SVGA3 doesn't support guest backed
objects right now so everything except 3D is working.

v2: Fixes all the static analyzer warnings
Signed-off-by: Zack Rusin <zackr@vmware.com>
Cc: Martin Krastev <krastevm@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20210505191007.305872-1-zackr@vmware.com

drivers/gpu/drm/vmwgfx/device_include/svga_reg.h

 /* SPDX-License-Identifier: GPL-2.0 OR MIT */
 /**********************************************************
- * Copyright 1998-2015 VMware, Inc.
+ * Copyright 1998-2021 VMware, Inc.
  *
  * Permission is hereby granted, free of charge, to any person
  * obtaining a copy of this software and associated documentation
@@ -98,6 +98,10 @@ typedef uint32 SVGAMobId;
 #define SVGA_MAGIC         0x900000UL
 #define SVGA_MAKE_ID(ver)  (SVGA_MAGIC << 8 | (ver))
 
+/* Version 3 has the control bar instead of the FIFO */
+#define SVGA_VERSION_3     3
+#define SVGA_ID_3          SVGA_MAKE_ID(SVGA_VERSION_3)
+
 /* Version 2 let the address of the frame buffer be unsigned on Win32 */
 #define SVGA_VERSION_2     2
 #define SVGA_ID_2          SVGA_MAKE_ID(SVGA_VERSION_2)
@@ -129,11 +133,12 @@ typedef uint32 SVGAMobId;
  * Interrupts are only supported when the
  * SVGA_CAP_IRQMASK capability is present.
  */
-#define SVGA_IRQFLAG_ANY_FENCE            0x1    /* Any fence was passed */
-#define SVGA_IRQFLAG_FIFO_PROGRESS        0x2    /* Made forward progress in the FIFO */
-#define SVGA_IRQFLAG_FENCE_GOAL           0x4    /* SVGA_FIFO_FENCE_GOAL reached */
-#define SVGA_IRQFLAG_COMMAND_BUFFER       0x8    /* Command buffer completed */
-#define SVGA_IRQFLAG_ERROR                0x10   /* Error while processing commands */
+#define SVGA_IRQFLAG_ANY_FENCE            (1 << 0) /* Any fence was passed */
+#define SVGA_IRQFLAG_FIFO_PROGRESS        (1 << 1) /* Made forward progress in the FIFO */
+#define SVGA_IRQFLAG_FENCE_GOAL           (1 << 2) /* SVGA_FIFO_FENCE_GOAL reached */
+#define SVGA_IRQFLAG_COMMAND_BUFFER       (1 << 3) /* Command buffer completed */
+#define SVGA_IRQFLAG_ERROR                (1 << 4) /* Error while processing commands */
+#define SVGA_IRQFLAG_MAX                  (1 << 5)
 
 /*
  * The byte-size is the size of the actual cursor data,
@@ -286,7 +291,32 @@ enum {
     */
    SVGA_REG_GBOBJECT_MEM_SIZE_KB = 76,
 
-   SVGA_REG_TOP = 77,               /* Must be 1 more than the last register */
+   /*
+    +    * These registers are for the addresses of the memory BARs for SVGA3
+    */
+   SVGA_REG_REGS_START_HIGH32 = 77,
+   SVGA_REG_REGS_START_LOW32 = 78,
+   SVGA_REG_FB_START_HIGH32 = 79,
+   SVGA_REG_FB_START_LOW32 = 80,
+
+   /*
+    * A hint register that recommends which quality level the guest should
+    * currently use to define multisample surfaces.
+    *
+    * If the register is SVGA_REG_MSHINT_DISABLED,
+    * the guest is only allowed to use SVGA3D_MS_QUALITY_FULL.
+    *
+    * Otherwise, this is a live value that can change while the VM is
+    * powered on with the hint suggestion for which quality level the guest
+    * should be using.  Guests are free to ignore the hint and use either
+    * RESOLVE or FULL quality.
+    */
+   SVGA_REG_MSHINT = 81,
+
+   SVGA_REG_IRQ_STATUS = 82,
+   SVGA_REG_DIRTY_TRACKING = 83,
+
+   SVGA_REG_TOP = 84,               /* Must be 1 more than the last register */
 
    SVGA_PALETTE_BASE = 1024,        /* Base of SVGA color map */
    /* Next 768 (== 256*3) registers exist for colormap */
@@ -310,6 +340,17 @@ typedef enum SVGARegGuestDriverId {
    SVGA_REG_GUEST_DRIVER_ID_SUBMIT  = MAX_UINT32,
 } SVGARegGuestDriverId;
 
+typedef enum SVGARegMSHint {
+   SVGA_REG_MSHINT_DISABLED = 0,
+   SVGA_REG_MSHINT_FULL     = 1,
+   SVGA_REG_MSHINT_RESOLVED = 2,
+} SVGARegMSHint;
+
+typedef enum SVGARegDirtyTracking {
+   SVGA_REG_DIRTY_TRACKING_PER_IMAGE = 0,
+   SVGA_REG_DIRTY_TRACKING_PER_SURFACE = 1,
+} SVGARegDirtyTracking;
+
 
 /*
  * Guest memory regions (GMRs):

drivers/gpu/drm/vmwgfx/vmwgfx_binding.c

@@ -788,7 +788,7 @@ static void vmw_collect_dirty_view_ids(struct vmw_ctx_binding_state *cbs,
 }
 
 /**
- * vmw_binding_emit_set_sr - Issue delayed DX shader resource binding commands
+ * vmw_emit_set_sr - Issue delayed DX shader resource binding commands
  *
  * @cbs: Pointer to the context's struct vmw_ctx_binding_state
  * @shader_slot: The shader slot of the binding.
@@ -832,7 +832,7 @@ static int vmw_emit_set_sr(struct vmw_ctx_binding_state *cbs,
 }
 
 /**
- * vmw_binding_emit_set_rt - Issue delayed DX rendertarget binding commands
+ * vmw_emit_set_rt - Issue delayed DX rendertarget binding commands
  *
  * @cbs: Pointer to the context's struct vmw_ctx_binding_state
  */
@@ -1024,7 +1024,7 @@ static void vmw_collect_dirty_vbs(struct vmw_ctx_binding_state *cbs,
 }
 
 /**
- * vmw_binding_emit_set_vb - Issue delayed vertex buffer binding commands
+ * vmw_emit_set_vb - Issue delayed vertex buffer binding commands
  *
  * @cbs: Pointer to the context's struct vmw_ctx_binding_state
  *
@@ -1394,7 +1394,7 @@ struct list_head *vmw_binding_state_list(struct vmw_ctx_binding_state *cbs)
 }
 
 /**
- * vmwgfx_binding_state_reset - clear a struct vmw_ctx_binding_state
+ * vmw_binding_state_reset - clear a struct vmw_ctx_binding_state
  *
  * @cbs: Pointer to the struct vmw_ctx_binding_state to be cleared
  *

drivers/gpu/drm/vmwgfx/vmwgfx_blit.c

@@ -421,7 +421,7 @@ static int vmw_bo_cpu_blit_line(struct vmw_bo_blit_line_data *d,
 }
 
 /**
- * ttm_bo_cpu_blit - in-kernel cpu blit.
+ * vmw_bo_cpu_blit - in-kernel cpu blit.
  *
  * @dst: Destination buffer object.
  * @dst_offset: Destination offset of blit start in bytes.

drivers/gpu/drm/vmwgfx/vmwgfx_bo.c

@@ -600,7 +600,7 @@ static void vmw_user_bo_release(struct ttm_base_object **p_base)
 
 
 /**
- * vmw_user_bo_ref_obj-release - TTM synccpu reference object release callback
+ * vmw_user_bo_ref_obj_release - TTM synccpu reference object release callback
  * for vmw user buffer objects
  *
  * @base: Pointer to the TTM base object

drivers/gpu/drm/vmwgfx/vmwgfx_cmd.c

@@ -31,15 +31,10 @@
 
 #include "vmwgfx_drv.h"
 
-struct vmw_temp_set_context {
-	SVGA3dCmdHeader header;
-	SVGA3dCmdDXTempSetContext body;
-};
-
 bool vmw_supports_3d(struct vmw_private *dev_priv)
 {
 	uint32_t fifo_min, hwversion;
-	const struct vmw_fifo_state *fifo = &dev_priv->fifo;
+	const struct vmw_fifo_state *fifo = dev_priv->fifo;
 
 	if (!(dev_priv->capabilities & SVGA_CAP_3D))
 		return false;
@@ -61,6 +56,8 @@ bool vmw_supports_3d(struct vmw_private *dev_priv)
 	if (!(dev_priv->capabilities & SVGA_CAP_EXTENDED_FIFO))
 		return false;
 
+	BUG_ON(vmw_is_svga_v3(dev_priv));
+
 	fifo_min = vmw_fifo_mem_read(dev_priv, SVGA_FIFO_MIN);
 	if (fifo_min <= SVGA_FIFO_3D_HWVERSION * sizeof(unsigned int))
 		return false;
@@ -98,16 +95,20 @@ bool vmw_fifo_have_pitchlock(struct vmw_private *dev_priv)
 	return false;
 }
 
-int vmw_fifo_init(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
+struct vmw_fifo_state *vmw_fifo_create(struct vmw_private *dev_priv)
 {
+	struct vmw_fifo_state *fifo;
 	uint32_t max;
 	uint32_t min;
 
-	fifo->dx = false;
+	if (!dev_priv->fifo_mem)
+		return NULL;
+
+	fifo = kzalloc(sizeof(*fifo), GFP_KERNEL);
 	fifo->static_buffer_size = VMWGFX_FIFO_STATIC_SIZE;
 	fifo->static_buffer = vmalloc(fifo->static_buffer_size);
 	if (unlikely(fifo->static_buffer == NULL))
-		return -ENOMEM;
+		return ERR_PTR(-ENOMEM);
 
 	fifo->dynamic_buffer = NULL;
 	fifo->reserved_size = 0;
@@ -115,20 +116,6 @@ int vmw_fifo_init(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 
 	mutex_init(&fifo->fifo_mutex);
 	init_rwsem(&fifo->rwsem);
-
-	DRM_INFO("width %d\n", vmw_read(dev_priv, SVGA_REG_WIDTH));
-	DRM_INFO("height %d\n", vmw_read(dev_priv, SVGA_REG_HEIGHT));
-	DRM_INFO("bpp %d\n", vmw_read(dev_priv, SVGA_REG_BITS_PER_PIXEL));
-
-	dev_priv->enable_state = vmw_read(dev_priv, SVGA_REG_ENABLE);
-	dev_priv->config_done_state = vmw_read(dev_priv, SVGA_REG_CONFIG_DONE);
-	dev_priv->traces_state = vmw_read(dev_priv, SVGA_REG_TRACES);
-
-	vmw_write(dev_priv, SVGA_REG_ENABLE, SVGA_REG_ENABLE_ENABLE |
-		  SVGA_REG_ENABLE_HIDE);
-
-	vmw_write(dev_priv, SVGA_REG_TRACES, 0);
-
 	min = 4;
 	if (dev_priv->capabilities & SVGA_CAP_EXTENDED_FIFO)
 		min = vmw_read(dev_priv, SVGA_REG_MEM_REGS);
@@ -155,35 +142,23 @@ int vmw_fifo_init(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 		 (unsigned int) max,
 		 (unsigned int) min,
 		 (unsigned int) fifo->capabilities);
-
-	atomic_set(&dev_priv->marker_seq, dev_priv->last_read_seqno);
-	vmw_fifo_mem_write(dev_priv, SVGA_FIFO_FENCE, dev_priv->last_read_seqno);
-
-	return 0;
+	return fifo;
 }
 
 void vmw_fifo_ping_host(struct vmw_private *dev_priv, uint32_t reason)
 {
 	u32 *fifo_mem = dev_priv->fifo_mem;
-
-	if (cmpxchg(fifo_mem + SVGA_FIFO_BUSY, 0, 1) == 0)
+	if (fifo_mem && cmpxchg(fifo_mem + SVGA_FIFO_BUSY, 0, 1) == 0)
 		vmw_write(dev_priv, SVGA_REG_SYNC, reason);
+
 }
 
-void vmw_fifo_release(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
+void vmw_fifo_destroy(struct vmw_private *dev_priv)
 {
-	vmw_write(dev_priv, SVGA_REG_SYNC, SVGA_SYNC_GENERIC);
-	while (vmw_read(dev_priv, SVGA_REG_BUSY) != 0)
-		;
-
-	dev_priv->last_read_seqno = vmw_fifo_mem_read(dev_priv, SVGA_FIFO_FENCE);
+	struct vmw_fifo_state *fifo = dev_priv->fifo;
 
-	vmw_write(dev_priv, SVGA_REG_CONFIG_DONE,
-		  dev_priv->config_done_state);
-	vmw_write(dev_priv, SVGA_REG_ENABLE,
-		  dev_priv->enable_state);
-	vmw_write(dev_priv, SVGA_REG_TRACES,
-		  dev_priv->traces_state);
+	if (!fifo)
+		return;
 
 	if (likely(fifo->static_buffer != NULL)) {
 		vfree(fifo->static_buffer);
@@ -194,6 +169,8 @@ void vmw_fifo_release(struct vmw_private *dev_priv, struct vmw_fifo_state *fifo)
 		vfree(fifo->dynamic_buffer);
 		fifo->dynamic_buffer = NULL;
 	}
+	kfree(fifo);
+	dev_priv->fifo = NULL;
 }
 
 static bool vmw_fifo_is_full(struct vmw_private *dev_priv, uint32_t bytes)
@@ -289,7 +266,7 @@ static int vmw_fifo_wait(struct vmw_private *dev_priv,
 static void *vmw_local_fifo_reserve(struct vmw_private *dev_priv,
 				    uint32_t bytes)
 {
-	struct vmw_fifo_state *fifo_state = &dev_priv->fifo;
+	struct vmw_fifo_state *fifo_state = dev_priv->fifo;
 	u32  *fifo_mem = dev_priv->fifo_mem;
 	uint32_t max;
 	uint32_t min;
@@ -438,16 +415,12 @@ static void vmw_fifo_slow_copy(struct vmw_fifo_state *fifo_state,
 
 static void vmw_local_fifo_commit(struct vmw_private *dev_priv, uint32_t bytes)
 {
-	struct vmw_fifo_state *fifo_state = &dev_priv->fifo;
+	struct vmw_fifo_state *fifo_state = dev_priv->fifo;
 	uint32_t next_cmd = vmw_fifo_mem_read(dev_priv, SVGA_FIFO_NEXT_CMD);
 	uint32_t max = vmw_fifo_mem_read(dev_priv, SVGA_FIFO_MAX);
 	uint32_t min = vmw_fifo_mem_read(dev_priv, SVGA_FIFO_MIN);
 	bool reserveable = fifo_state->capabilities & SVGA_FIFO_CAP_RESERVE;
 
-	if (fifo_state->dx)
-		bytes += sizeof(struct vmw_temp_set_context);
-
-	fifo_state->dx = false;
 	BUG_ON((bytes & 3) != 0);
 	BUG_ON(bytes > fifo_state->reserved_size);
 
@@ -495,7 +468,7 @@ void vmw_cmd_commit(struct vmw_private *dev_priv, uint32_t bytes)
 
 
 /**
- * vmw_fifo_commit_flush - Commit fifo space and flush any buffered commands.
+ * vmw_cmd_commit_flush - Commit fifo space and flush any buffered commands.
  *
  * @dev_priv: Pointer to device private structure.
  * @bytes: Number of bytes to commit.
@@ -509,7 +482,7 @@ void vmw_cmd_commit_flush(struct vmw_private *dev_priv, uint32_t bytes)
 }
 
 /**
- * vmw_fifo_flush - Flush any buffered commands and make sure command processing
+ * vmw_cmd_flush - Flush any buffered commands and make sure command processing
  * starts.
  *
  * @dev_priv: Pointer to device private structure.
@@ -527,7 +500,6 @@ int vmw_cmd_flush(struct vmw_private *dev_priv, bool interruptible)
 
 int vmw_cmd_send_fence(struct vmw_private *dev_priv, uint32_t *seqno)
 {
-	struct vmw_fifo_state *fifo_state = &dev_priv->fifo;
 	struct svga_fifo_cmd_fence *cmd_fence;
 	u32 *fm;
 	int ret = 0;
@@ -546,7 +518,7 @@ int vmw_cmd_send_fence(struct vmw_private *dev_priv, uint32_t *seqno)
 		*seqno = atomic_add_return(1, &dev_priv->marker_seq);
 	} while (*seqno == 0);
 
-	if (!(fifo_state->capabilities & SVGA_FIFO_CAP_FENCE)) {
+	if (!(vmw_fifo_caps(dev_priv) & SVGA_FIFO_CAP_FENCE)) {
 
 		/*
 		 * Don't request hardware to send a fence. The
@@ -561,22 +533,22 @@ int vmw_cmd_send_fence(struct vmw_private *dev_priv, uint32_t *seqno)
 	cmd_fence = (struct svga_fifo_cmd_fence *) fm;
 	cmd_fence->fence = *seqno;
 	vmw_cmd_commit_flush(dev_priv, bytes);
-	vmw_update_seqno(dev_priv, fifo_state);
+	vmw_update_seqno(dev_priv);
 
 out_err:
 	return ret;
 }
 
 /**
- * vmw_fifo_emit_dummy_legacy_query - emits a dummy query to the fifo using
+ * vmw_cmd_emit_dummy_legacy_query - emits a dummy query to the fifo using
  * legacy query commands.
  *
  * @dev_priv: The device private structure.
  * @cid: The hardware context id used for the query.
  *
- * See the vmw_fifo_emit_dummy_query documentation.
+ * See the vmw_cmd_emit_dummy_query documentation.
  */
-static int vmw_fifo_emit_dummy_legacy_query(struct vmw_private *dev_priv,
+static int vmw_cmd_emit_dummy_legacy_query(struct vmw_private *dev_priv,
 					    uint32_t cid)
 {
 	/*
@@ -614,16 +586,16 @@ static int vmw_fifo_emit_dummy_legacy_query(struct vmw_private *dev_priv,
 }
 
 /**
- * vmw_fifo_emit_dummy_gb_query - emits a dummy query to the fifo using
+ * vmw_cmd_emit_dummy_gb_query - emits a dummy query to the fifo using
  * guest-backed resource query commands.
  *
  * @dev_priv: The device private structure.
  * @cid: The hardware context id used for the query.
  *
- * See the vmw_fifo_emit_dummy_query documentation.
+ * See the vmw_cmd_emit_dummy_query documentation.
  */
-static int vmw_fifo_emit_dummy_gb_query(struct vmw_private *dev_priv,
-					uint32_t cid)
+static int vmw_cmd_emit_dummy_gb_query(struct vmw_private *dev_priv,
+				       uint32_t cid)
 {
 	/*
 	 * A query wait without a preceding query end will
@@ -656,7 +628,7 @@ static int vmw_fifo_emit_dummy_gb_query(struct vmw_private *dev_priv,
 
 
 /**
- * vmw_fifo_emit_dummy_gb_query - emits a dummy query to the fifo using
+ * vmw_cmd_emit_dummy_query - emits a dummy query to the fifo using
  * appropriate resource query commands.
  *
  * @dev_priv: The device private structure.
@@ -677,7 +649,27 @@ int vmw_cmd_emit_dummy_query(struct vmw_private *dev_priv,
 			      uint32_t cid)
 {
 	if (dev_priv->has_mob)
-		return vmw_fifo_emit_dummy_gb_query(dev_priv, cid);
+		return vmw_cmd_emit_dummy_gb_query(dev_priv, cid);
 
-	return vmw_fifo_emit_dummy_legacy_query(dev_priv, cid);
+	return vmw_cmd_emit_dummy_legacy_query(dev_priv, cid);
+}
+
+
+/**
+ * vmw_cmd_supported - returns true if the given device supports
+ * command queues.
+ *
+ * @vmw: The device private structure.
+ *
+ * Returns true if we can issue commands.
+ */
+bool vmw_cmd_supported(struct vmw_private *vmw)
+{
+	if ((vmw->capabilities & (SVGA_CAP_COMMAND_BUFFERS |
+				  SVGA_CAP_CMD_BUFFERS_2)) != 0)
+		return true;
+	/*
+	 * We have FIFO cmd's
+	 */
+	return vmw->fifo_mem != NULL;
 }

drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c

@@ -295,7 +295,7 @@ void vmw_cmdbuf_header_free(struct vmw_cmdbuf_header *header)
 
 
 /**
- * vmw_cmbuf_header_submit: Submit a command buffer to hardware.
+ * vmw_cmdbuf_header_submit: Submit a command buffer to hardware.
  *
  * @header: The header of the buffer to submit.
  */
@@ -620,7 +620,7 @@ static void vmw_cmdbuf_work_func(struct work_struct *work)
 }
 
 /**
- * vmw_cmdbuf_man idle - Check whether the command buffer manager is idle.
+ * vmw_cmdbuf_man_idle - Check whether the command buffer manager is idle.
  *
  * @man: The command buffer manager.
  * @check_preempted: Check also the preempted queue for pending command buffers.

drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c

@@ -653,7 +653,7 @@ int vmw_cotable_notify(struct vmw_resource *res, int id)
 }
 
 /**
- * vmw_cotable_add_view - add a view to the cotable's list of active views.
+ * vmw_cotable_add_resource - add a view to the cotable's list of active views.
  *
  * @res: pointer struct vmw_resource representing the cotable.
  * @head: pointer to the struct list_head member of the resource, dedicated

drivers/gpu/drm/vmwgfx/vmwgfx_drv.c

@@ -50,7 +50,7 @@
 #define VMWGFX_VALIDATION_MEM_GRAN (16*PAGE_SIZE)
 
 
-/**
+/*
  * Fully encoded drm commands. Might move to vmw_drm.h
  */
 
@@ -246,6 +246,7 @@ static const struct drm_ioctl_desc vmw_ioctls[] = {
 
 static const struct pci_device_id vmw_pci_id_list[] = {
 	{ PCI_DEVICE(0x15ad, VMWGFX_PCI_ID_SVGA2) },
+	{ PCI_DEVICE(0x15ad, VMWGFX_PCI_ID_SVGA3) },
 	{ }
 };
 MODULE_DEVICE_TABLE(pci, vmw_pci_id_list);
@@ -393,6 +394,60 @@ static int vmw_dummy_query_bo_create(struct vmw_private *dev_priv)
 	return ret;
 }
 
+static int vmw_device_init(struct vmw_private *dev_priv)
+{
+	bool uses_fb_traces = false;
+
+	DRM_INFO("width %d\n", vmw_read(dev_priv, SVGA_REG_WIDTH));
+	DRM_INFO("height %d\n", vmw_read(dev_priv, SVGA_REG_HEIGHT));
+	DRM_INFO("bpp %d\n", vmw_read(dev_priv, SVGA_REG_BITS_PER_PIXEL));
+
+	dev_priv->enable_state = vmw_read(dev_priv, SVGA_REG_ENABLE);
+	dev_priv->config_done_state = vmw_read(dev_priv, SVGA_REG_CONFIG_DONE);
+	dev_priv->traces_state = vmw_read(dev_priv, SVGA_REG_TRACES);
+
+	vmw_write(dev_priv, SVGA_REG_ENABLE, SVGA_REG_ENABLE_ENABLE |
+		  SVGA_REG_ENABLE_HIDE);
+
+	uses_fb_traces = !vmw_cmd_supported(dev_priv) &&
+			 (dev_priv->capabilities & SVGA_CAP_TRACES) != 0;
+
+	vmw_write(dev_priv, SVGA_REG_TRACES, uses_fb_traces);
+	dev_priv->fifo = vmw_fifo_create(dev_priv);
+	if (IS_ERR(dev_priv->fifo)) {
+		int err = PTR_ERR(dev_priv->fifo);
+		dev_priv->fifo = NULL;
+		return err;
+	} else if (!dev_priv->fifo) {
+		vmw_write(dev_priv, SVGA_REG_CONFIG_DONE, 1);
+	}
+
+	dev_priv->last_read_seqno = vmw_fence_read(dev_priv);
+	atomic_set(&dev_priv->marker_seq, dev_priv->last_read_seqno);
+	return 0;
+}
+
+static void vmw_device_fini(struct vmw_private *vmw)
+{
+	/*
+	 * Legacy sync
+	 */
+	vmw_write(vmw, SVGA_REG_SYNC, SVGA_SYNC_GENERIC);
+	while (vmw_read(vmw, SVGA_REG_BUSY) != 0)
+		;
+
+	vmw->last_read_seqno = vmw_fence_read(vmw);
+
+	vmw_write(vmw, SVGA_REG_CONFIG_DONE,
+		  vmw->config_done_state);
+	vmw_write(vmw, SVGA_REG_ENABLE,
+		  vmw->enable_state);
+	vmw_write(vmw, SVGA_REG_TRACES,
+		  vmw->traces_state);
+
+	vmw_fifo_destroy(vmw);
+}
+
 /**
  * vmw_request_device_late - Perform late device setup
  *
@@ -433,9 +488,9 @@ static int vmw_request_device(struct vmw_private *dev_priv)
 {
 	int ret;
 
-	ret = vmw_fifo_init(dev_priv, &dev_priv->fifo);
+	ret = vmw_device_init(dev_priv);
 	if (unlikely(ret != 0)) {
-		DRM_ERROR("Unable to initialize FIFO.\n");
+		DRM_ERROR("Unable to initialize the device.\n");
 		return ret;
 	}
 	vmw_fence_fifo_up(dev_priv->fman);
@@ -469,7 +524,7 @@ static int vmw_request_device(struct vmw_private *dev_priv)
 		vmw_cmdbuf_man_destroy(dev_priv->cman);
 out_no_mob:
 	vmw_fence_fifo_down(dev_priv->fman);
-	vmw_fifo_release(dev_priv, &dev_priv->fifo);
+	vmw_device_fini(dev_priv);
 	return ret;
 }
 
@@ -517,7 +572,7 @@ static void vmw_release_device_late(struct vmw_private *dev_priv)
 	if (dev_priv->cman)
 		vmw_cmdbuf_man_destroy(dev_priv->cman);
 
-	vmw_fifo_release(dev_priv, &dev_priv->fifo);
+	vmw_device_fini(dev_priv);
 }
 
 /*
@@ -638,6 +693,8 @@ static void vmw_vram_manager_fini(struct vmw_private *dev_priv)
 static int vmw_setup_pci_resources(struct vmw_private *dev,
 				   unsigned long pci_id)
 {
+	resource_size_t rmmio_start;
+	resource_size_t rmmio_size;
 	resource_size_t fifo_start;
 	resource_size_t fifo_size;
 	int ret;
@@ -649,23 +706,45 @@ static int vmw_setup_pci_resources(struct vmw_private *dev,
 	if (ret)
 		return ret;
 
-	dev->io_start = pci_resource_start(pdev, 0);
-	dev->vram_start = pci_resource_start(pdev, 1);
-	dev->vram_size = pci_resource_len(pdev, 1);
-	fifo_start = pci_resource_start(pdev, 2);
-	fifo_size = pci_resource_len(pdev, 2);
-
-	DRM_INFO("FIFO at %pa size is %llu kiB\n",
-		 &fifo_start, (uint64_t)fifo_size / 1024);
-	dev->fifo_mem = devm_memremap(dev->drm.dev,
-				      fifo_start,
-				      fifo_size,
-				      MEMREMAP_WB);
-
-	if (IS_ERR(dev->fifo_mem)) {
-		DRM_ERROR("Failed mapping FIFO memory.\n");
+	dev->pci_id = pci_id;
+	if (pci_id == VMWGFX_PCI_ID_SVGA3) {
+		rmmio_start = pci_resource_start(pdev, 0);
+		rmmio_size = pci_resource_len(pdev, 0);
+		dev->vram_start = pci_resource_start(pdev, 2);
+		dev->vram_size = pci_resource_len(pdev, 2);
+
+		DRM_INFO("Register MMIO at 0x%pa size is %llu kiB\n",
+			 &rmmio_start, (uint64_t)rmmio_size / 1024);
+		dev->rmmio = devm_ioremap(dev->drm.dev,
+					  rmmio_start,
+					  rmmio_size);
+		if (IS_ERR(dev->rmmio)) {
+			DRM_ERROR("Failed mapping registers mmio memory.\n");
+			pci_release_regions(pdev);
+			return PTR_ERR(dev->rmmio);
+		}
+	} else if (pci_id == VMWGFX_PCI_ID_SVGA2) {
+		dev->io_start = pci_resource_start(pdev, 0);
+		dev->vram_start = pci_resource_start(pdev, 1);
+		dev->vram_size = pci_resource_len(pdev, 1);
+		fifo_start = pci_resource_start(pdev, 2);
+		fifo_size = pci_resource_len(pdev, 2);
+
+		DRM_INFO("FIFO at %pa size is %llu kiB\n",
+			 &fifo_start, (uint64_t)fifo_size / 1024);
+		dev->fifo_mem = devm_memremap(dev->drm.dev,
+					      fifo_start,
+					      fifo_size,
+					      MEMREMAP_WB);
+
+		if (IS_ERR(dev->fifo_mem)) {
+			DRM_ERROR("Failed mapping FIFO memory.\n");
+			pci_release_regions(pdev);
+			return PTR_ERR(dev->fifo_mem);
+		}
+	} else {
 		pci_release_regions(pdev);
-		return PTR_ERR(dev->fifo_mem);
+		return -EINVAL;
 	}
 
 	/*
@@ -684,13 +763,16 @@ static int vmw_detect_version(struct vmw_private *dev)
 {
 	uint32_t svga_id;
 
-	vmw_write(dev, SVGA_REG_ID, SVGA_ID_2);
+	vmw_write(dev, SVGA_REG_ID, vmw_is_svga_v3(dev) ?
+			  SVGA_ID_3 : SVGA_ID_2);
 	svga_id = vmw_read(dev, SVGA_REG_ID);
-	if (svga_id != SVGA_ID_2) {
+	if (svga_id != SVGA_ID_2 && svga_id != SVGA_ID_3) {
 		DRM_ERROR("Unsupported SVGA ID 0x%x on chipset 0x%x\n",
 			  svga_id, dev->vmw_chipset);
 		return -ENOSYS;
 	}
+	BUG_ON(vmw_is_svga_v3(dev) && (svga_id != SVGA_ID_3));
+	DRM_INFO("Running on SVGA version %d.\n", (svga_id & 0xff));
 	return 0;
 }
 
@@ -703,7 +785,6 @@ static int vmw_driver_load(struct vmw_private *dev_priv, u32 pci_id)
 	struct pci_dev *pdev = to_pci_dev(dev_priv->drm.dev);
 
 	dev_priv->vmw_chipset = pci_id;
-	dev_priv->last_read_seqno = (uint32_t) -100;
 	dev_priv->drm.dev_private = dev_priv;
 
 	mutex_init(&dev_priv->cmdbuf_mutex);
@@ -824,6 +905,8 @@ static int vmw_driver_load(struct vmw_private *dev_priv, u32 pci_id)
 	vmw_print_capabilities(dev_priv->capabilities);
 	if (dev_priv->capabilities & SVGA_CAP_CAP2_REGISTER)
 		vmw_print_capabilities2(dev_priv->capabilities2);
+	DRM_INFO("Supports command queues = %d\n",
+		 vmw_cmd_supported((dev_priv)));
 
 	ret = vmw_dma_masks(dev_priv);
 	if (unlikely(ret != 0))
@@ -1390,8 +1473,7 @@ static int vmw_pm_restore(struct device *kdev)
 	struct vmw_private *dev_priv = vmw_priv(dev);
 	int ret;
 
-	vmw_write(dev_priv, SVGA_REG_ID, SVGA_ID_2);
-	(void) vmw_read(dev_priv, SVGA_REG_ID);
+	vmw_detect_version(dev_priv);
 
 	if (dev_priv->enable_fb)
 		vmw_fifo_resource_inc(dev_priv);
@@ -1428,8 +1510,8 @@ static const struct file_operations vmwgfx_driver_fops = {
 	.release = drm_release,
 	.unlocked_ioctl = vmw_unlocked_ioctl,
 	.mmap = vmw_mmap,
-	.poll = vmw_fops_poll,
-	.read = vmw_fops_read,
+	.poll = drm_poll,
+	.read = drm_read,
 #if defined(CONFIG_COMPAT)
 	.compat_ioctl = vmw_compat_ioctl,
 #endif

drivers/gpu/drm/vmwgfx/vmwgfx_drv.h

@@ -66,6 +66,7 @@
 #define VMWGFX_ENABLE_SCREEN_TARGET_OTABLE 1
 
 #define VMWGFX_PCI_ID_SVGA2              0x0405
+#define VMWGFX_PCI_ID_SVGA3              0x0406
 
 /*
  * Perhaps we should have sysfs entries for these.
@@ -284,7 +285,6 @@ struct vmw_fifo_state {
 	uint32_t capabilities;
 	struct mutex fifo_mutex;
 	struct rw_semaphore rwsem;
-	bool dx;
 };
 
 /**
@@ -485,14 +485,14 @@ struct vmw_private {
 	struct drm_device drm;
 	struct ttm_device bdev;
 
-	struct vmw_fifo_state fifo;
-
 	struct drm_vma_offset_manager vma_manager;
+	unsigned long pci_id;
 	u32 vmw_chipset;
 	resource_size_t io_start;
 	resource_size_t vram_start;
 	resource_size_t vram_size;
 	resource_size_t prim_bb_mem;
+	void __iomem *rmmio;
 	u32 *fifo_mem;
 	resource_size_t fifo_mem_size;
 	uint32_t fb_max_width;
@@ -623,6 +623,7 @@ struct vmw_private {
 	 */
 	struct vmw_otable_batch otable_batch;
 
+	struct vmw_fifo_state *fifo;
 	struct vmw_cmdbuf_man *cman;
 	DECLARE_BITMAP(irqthread_pending, VMW_IRQTHREAD_MAX);
 
@@ -645,6 +646,14 @@ static inline struct vmw_fpriv *vmw_fpriv(struct drm_file *file_priv)
 	return (struct vmw_fpriv *)file_priv->driver_priv;
 }
 
+/*
+ * SVGA v3 has mmio register access and lacks fifo cmds
+ */
+static inline bool vmw_is_svga_v3(const struct vmw_private *dev)
+{
+	return dev->pci_id == VMWGFX_PCI_ID_SVGA3;
+}
+
 /*
  * The locking here is fine-grained, so that it is performed once
  * for every read- and write operation. This is of course costly, but we
@@ -655,10 +664,14 @@ static inline struct vmw_fpriv *vmw_fpriv(struct drm_file *file_priv)
 static inline void vmw_write(struct vmw_private *dev_priv,
 			     unsigned int offset, uint32_t value)
 {
-	spin_lock(&dev_priv->hw_lock);
-	outl(offset, dev_priv->io_start + VMWGFX_INDEX_PORT);
-	outl(value, dev_priv->io_start + VMWGFX_VALUE_PORT);
-	spin_unlock(&dev_priv->hw_lock);
+	if (vmw_is_svga_v3(dev_priv)) {
+		iowrite32(value, dev_priv->rmmio + offset);
+	} else {
+		spin_lock(&dev_priv->hw_lock);
+		outl(offset, dev_priv->io_start + SVGA_INDEX_PORT);
+		outl(value, dev_priv->io_start + SVGA_VALUE_PORT);
+		spin_unlock(&dev_priv->hw_lock);
+	}
 }
 
 static inline uint32_t vmw_read(struct vmw_private *dev_priv,
@@ -666,10 +679,14 @@ static inline uint32_t vmw_read(struct vmw_private *dev_priv,
 {
 	u32 val;
 
-	spin_lock(&dev_priv->hw_lock);
-	outl(offset, dev_priv->io_start + VMWGFX_INDEX_PORT);
-	val = inl(dev_priv->io_start + VMWGFX_VALUE_PORT);
-	spin_unlock(&dev_priv->hw_lock);
+	if (vmw_is_svga_v3(dev_priv)) {
+		val = ioread32(dev_priv->rmmio + offset);
+	} else {
+		spin_lock(&dev_priv->hw_lock);
+		outl(offset, dev_priv->io_start + SVGA_INDEX_PORT);
+		val = inl(dev_priv->io_start + SVGA_VALUE_PORT);
+		spin_unlock(&dev_priv->hw_lock);
+	}
 
 	return val;
 }
@@ -932,19 +949,14 @@ extern int vmw_present_ioctl(struct drm_device *dev, void *data,
 			     struct drm_file *file_priv);
 extern int vmw_present_readback_ioctl(struct drm_device *dev, void *data,
 				      struct drm_file *file_priv);
-extern __poll_t vmw_fops_poll(struct file *filp,
-				  struct poll_table_struct *wait);
-extern ssize_t vmw_fops_read(struct file *filp, char __user *buffer,
-			     size_t count, loff_t *offset);
 
 /**
  * Fifo utilities - vmwgfx_fifo.c
  */
 
-extern int vmw_fifo_init(struct vmw_private *dev_priv,
-			 struct vmw_fifo_state *fifo);
-extern void vmw_fifo_release(struct vmw_private *dev_priv,
-			     struct vmw_fifo_state *fifo);
+extern struct vmw_fifo_state *vmw_fifo_create(struct vmw_private *dev_priv);
+extern void vmw_fifo_destroy(struct vmw_private *dev_priv);
+extern bool vmw_cmd_supported(struct vmw_private *vmw);
 extern void *
 vmw_cmd_ctx_reserve(struct vmw_private *dev_priv, uint32_t bytes, int ctx_id);
 extern void vmw_cmd_commit(struct vmw_private *dev_priv, uint32_t bytes);
@@ -970,6 +982,31 @@ extern int vmw_cmd_flush(struct vmw_private *dev_priv,
 #define VMW_CMD_RESERVE(__priv, __bytes)                                     \
 	VMW_CMD_CTX_RESERVE(__priv, __bytes, SVGA3D_INVALID_ID)
 
+
+/**
+ * vmw_fifo_caps - Returns the capabilities of the FIFO command
+ * queue or 0 if fifo memory isn't present.
+ * @dev_priv: The device private context
+ */
+static inline uint32_t vmw_fifo_caps(const struct vmw_private *dev_priv)
+{
+	if (!dev_priv->fifo_mem || !dev_priv->fifo)
+		return 0;
+	return dev_priv->fifo->capabilities;
+}
+
+
+/**
+ * vmw_is_cursor_bypass3_enabled - Returns TRUE iff Cursor Bypass 3
+ * is enabled in the FIFO.
+ * @dev_priv: The device private context
+ */
+static inline bool
+vmw_is_cursor_bypass3_enabled(const struct vmw_private *dev_priv)
+{
+	return (vmw_fifo_caps(dev_priv) & SVGA_FIFO_CAP_CURSOR_BYPASS_3) != 0;
+}
+
 /**
  * TTM glue - vmwgfx_ttm_glue.c
  */
@@ -1079,9 +1116,6 @@ bool vmw_cmd_describe(const void *buf, u32 *size, char const **cmd);
  * IRQs and wating - vmwgfx_irq.c
  */
 
-extern int vmw_wait_seqno(struct vmw_private *dev_priv, bool lazy,
-			  uint32_t seqno, bool interruptible,
-			  unsigned long timeout);
 extern int vmw_irq_install(struct drm_device *dev, int irq);
 extern void vmw_irq_uninstall(struct drm_device *dev);
 extern bool vmw_seqno_passed(struct vmw_private *dev_priv,
@@ -1092,8 +1126,7 @@ extern int vmw_fallback_wait(struct vmw_private *dev_priv,
 			     uint32_t seqno,
 			     bool interruptible,
 			     unsigned long timeout);
-extern void vmw_update_seqno(struct vmw_private *dev_priv,
-				struct vmw_fifo_state *fifo_state);
+extern void vmw_update_seqno(struct vmw_private *dev_priv);
 extern void vmw_seqno_waiter_add(struct vmw_private *dev_priv);
 extern void vmw_seqno_waiter_remove(struct vmw_private *dev_priv);
 extern void vmw_goal_waiter_add(struct vmw_private *dev_priv);
@@ -1572,6 +1605,7 @@ static inline void vmw_fifo_resource_dec(struct vmw_private *dev_priv)
  */
 static inline u32 vmw_fifo_mem_read(struct vmw_private *vmw, uint32 fifo_reg)
 {
+	BUG_ON(vmw_is_svga_v3(vmw));
 	return READ_ONCE(*(vmw->fifo_mem + fifo_reg));
 }
 
@@ -1586,6 +1620,44 @@ static inline u32 vmw_fifo_mem_read(struct vmw_private *vmw, uint32 fifo_reg)
 static inline void vmw_fifo_mem_write(struct vmw_private *vmw, u32 fifo_reg,
 				      u32 value)
 {
+	BUG_ON(vmw_is_svga_v3(vmw));
 	WRITE_ONCE(*(vmw->fifo_mem + fifo_reg), value);
 }
+
+static inline u32 vmw_fence_read(struct vmw_private *dev_priv)
+{
+	u32 fence;
+	if (vmw_is_svga_v3(dev_priv))
+		fence = vmw_read(dev_priv, SVGA_REG_FENCE);
+	else
+		fence = vmw_fifo_mem_read(dev_priv, SVGA_FIFO_FENCE);
+	return fence;
+}
+
+static inline void vmw_fence_write(struct vmw_private *dev_priv,
+				  u32 fence)
+{
+	BUG_ON(vmw_is_svga_v3(dev_priv));
+	vmw_fifo_mem_write(dev_priv, SVGA_FIFO_FENCE, fence);
+}
+
+static inline u32 vmw_irq_status_read(struct vmw_private *vmw)
+{
+	u32 status;
+	if (vmw_is_svga_v3(vmw))
+		status = vmw_read(vmw, SVGA_REG_IRQ_STATUS);
+	else
+		status = inl(vmw->io_start + SVGA_IRQSTATUS_PORT);
+	return status;
+}
+
+static inline void vmw_irq_status_write(struct vmw_private *vmw,
+					uint32 status)
+{
+	if (vmw_is_svga_v3(vmw))
+		vmw_write(vmw, SVGA_REG_IRQ_STATUS, status);
+	else
+		outl(status, vmw->io_start + SVGA_IRQSTATUS_PORT);
+}
+
 #endif

drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c

@@ -711,7 +711,7 @@ vmw_cmd_res_check(struct vmw_private *dev_priv,
 }
 
 /**
- * vmw_rebind_dx_query - Rebind DX query associated with the context
+ * vmw_rebind_all_dx_query - Rebind DX query associated with the context
  *
  * @ctx_res: context the query belongs to
  *
@@ -1140,7 +1140,7 @@ static void vmw_query_bo_switch_commit(struct vmw_private *dev_priv,
 }
 
 /**
- * vmw_translate_mob_pointer - Prepare to translate a user-space buffer handle
+ * vmw_translate_mob_ptr - Prepare to translate a user-space buffer handle
  * to a MOB id.
  *
  * @dev_priv: Pointer to a device private structure.
@@ -1195,7 +1195,7 @@ static int vmw_translate_mob_ptr(struct vmw_private *dev_priv,
 }
 
 /**
- * vmw_translate_guest_pointer - Prepare to translate a user-space buffer handle
+ * vmw_translate_guest_ptr - Prepare to translate a user-space buffer handle
  * to a valid SVGAGuestPtr
  *
  * @dev_priv: Pointer to a device private structure.
@@ -2308,7 +2308,7 @@ static int vmw_cmd_dx_set_vertex_buffers(struct vmw_private *dev_priv,
 }
 
 /**
- * vmw_cmd_dx_ia_set_vertex_buffers - Validate
+ * vmw_cmd_dx_set_index_buffer - Validate
  * SVGA_3D_CMD_DX_IA_SET_INDEX_BUFFER command.
  *
  * @dev_priv: Pointer to a device private struct.
@@ -2347,7 +2347,7 @@ static int vmw_cmd_dx_set_index_buffer(struct vmw_private *dev_priv,
 }
 
 /**
- * vmw_cmd_dx_set_rendertarget - Validate SVGA_3D_CMD_DX_SET_RENDERTARGETS
+ * vmw_cmd_dx_set_rendertargets - Validate SVGA_3D_CMD_DX_SET_RENDERTARGETS
  * command
  *
  * @dev_priv: Pointer to a device private struct.
@@ -2402,7 +2402,7 @@ static int vmw_cmd_dx_clear_rendertarget_view(struct vmw_private *dev_priv,
 }
 
 /**
- * vmw_cmd_dx_clear_rendertarget_view - Validate
+ * vmw_cmd_dx_clear_depthstencil_view - Validate
  * SVGA_3D_CMD_DX_CLEAR_DEPTHSTENCIL_VIEW command
  *
  * @dev_priv: Pointer to a device private struct.
@@ -3841,7 +3841,7 @@ vmw_execbuf_copy_fence_user(struct vmw_private *dev_priv,
 
 		fence_rep.handle = fence_handle;
 		fence_rep.seqno = fence->base.seqno;
-		vmw_update_seqno(dev_priv, &dev_priv->fifo);
+		vmw_update_seqno(dev_priv);
 		fence_rep.passed_seqno = dev_priv->last_read_seqno;
 	}
 

drivers/gpu/drm/vmwgfx/vmwgfx_fence.c

@@ -139,12 +139,10 @@ static bool vmw_fence_enable_signaling(struct dma_fence *f)
 	struct vmw_fence_manager *fman = fman_from_fence(fence);
 	struct vmw_private *dev_priv = fman->dev_priv;
 
-	u32 seqno = vmw_fifo_mem_read(dev_priv, SVGA_FIFO_FENCE);
+	u32 seqno = vmw_fence_read(dev_priv);
 	if (seqno - fence->base.seqno < VMW_FENCE_WRAP)
 		return false;
 
-	vmw_fifo_ping_host(dev_priv, SVGA_SYNC_GENERIC);
-
 	return true;
 }
 
@@ -177,7 +175,6 @@ static long vmw_fence_wait(struct dma_fence *f, bool intr, signed long timeout)
 	if (likely(vmw_fence_obj_signaled(fence)))
 		return timeout;
 
-	vmw_fifo_ping_host(dev_priv, SVGA_SYNC_GENERIC);
 	vmw_seqno_waiter_add(dev_priv);
 
 	spin_lock(f->lock);
@@ -464,7 +461,7 @@ static void __vmw_fences_update(struct vmw_fence_manager *fman)
 	bool needs_rerun;
 	uint32_t seqno, new_seqno;
 
-	seqno = vmw_fifo_mem_read(fman->dev_priv, SVGA_FIFO_FENCE);
+	seqno = vmw_fence_read(fman->dev_priv);
 rerun:
 	list_for_each_entry_safe(fence, next_fence, &fman->fence_list, head) {
 		if (seqno - fence->base.seqno < VMW_FENCE_WRAP) {
@@ -486,7 +483,7 @@ static void __vmw_fences_update(struct vmw_fence_manager *fman)
 
 	needs_rerun = vmw_fence_goal_new_locked(fman, seqno);
 	if (unlikely(needs_rerun)) {
-		new_seqno = vmw_fifo_mem_read(fman->dev_priv, SVGA_FIFO_FENCE);
+		new_seqno = vmw_fence_read(fman->dev_priv);
 		if (new_seqno != seqno) {
 			seqno = new_seqno;
 			goto rerun;
@@ -529,13 +526,6 @@ int vmw_fence_obj_wait(struct vmw_fence_obj *fence, bool lazy,
 		return ret;
 }
 
-void vmw_fence_obj_flush(struct vmw_fence_obj *fence)
-{
-	struct vmw_private *dev_priv = fman_from_fence(fence)->dev_priv;
-
-	vmw_fifo_ping_host(dev_priv, SVGA_SYNC_GENERIC);
-}
-
 static void vmw_fence_destroy(struct vmw_fence_obj *fence)
 {
 	dma_fence_free(&fence->base);
@@ -992,7 +982,7 @@ static void vmw_fence_obj_add_action(struct vmw_fence_obj *fence,
 }
 
 /**
- * vmw_event_fence_action_create - Post an event for sending when a fence
+ * vmw_event_fence_action_queue - Post an event for sending when a fence
  * object seqno has passed.
  *
  * @file_priv: The file connection on which the event should be posted.

drivers/gpu/drm/vmwgfx/vmwgfx_fence.h

@@ -94,8 +94,6 @@ extern int vmw_fence_obj_wait(struct vmw_fence_obj *fence,
 			      bool lazy,
 			      bool interruptible, unsigned long timeout);
 
-extern void vmw_fence_obj_flush(struct vmw_fence_obj *fence);
-
 extern int vmw_fence_create(struct vmw_fence_manager *fman,
 			    uint32_t seqno,
 			    struct vmw_fence_obj **p_fence);

drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c

@@ -60,15 +60,13 @@ int vmw_getparam_ioctl(struct drm_device *dev, void *data,
 		param->value = dev_priv->capabilities2;
 		break;
 	case DRM_VMW_PARAM_FIFO_CAPS:
-		param->value = dev_priv->fifo.capabilities;
+		param->value = vmw_fifo_caps(dev_priv);
 		break;
 	case DRM_VMW_PARAM_MAX_FB_SIZE:
 		param->value = dev_priv->prim_bb_mem;
 		break;
 	case DRM_VMW_PARAM_FIFO_HW_VERSION:
 	{
-		const struct vmw_fifo_state *fifo = &dev_priv->fifo;
-
 		if ((dev_priv->capabilities & SVGA_CAP_GBOBJECTS)) {
 			param->value = SVGA3D_HWVERSION_WS8_B1;
 			break;
@@ -76,7 +74,7 @@ int vmw_getparam_ioctl(struct drm_device *dev, void *data,
 
 		param->value =
 			vmw_fifo_mem_read(dev_priv,
-					  ((fifo->capabilities &
+					  ((vmw_fifo_caps(dev_priv) &
 					    SVGA_FIFO_CAP_3D_HWVERSION_REVISED) ?
 						   SVGA_FIFO_3D_HWVERSION_REVISED :
 						   SVGA_FIFO_3D_HWVERSION));
@@ -398,46 +396,3 @@ int vmw_present_readback_ioctl(struct drm_device *dev, void *data,
 out_clips:
 	return ret;
 }
-
-
-/**
- * vmw_fops_poll - wrapper around the drm_poll function
- *
- * @filp: See the linux fops poll documentation.
- * @wait: See the linux fops poll documentation.
- *
- * Wrapper around the drm_poll function that makes sure the device is
- * processing the fifo if drm_poll decides to wait.
- */
-__poll_t vmw_fops_poll(struct file *filp, struct poll_table_struct *wait)
-{
-	struct drm_file *file_priv = filp->private_data;
-	struct vmw_private *dev_priv =
-		vmw_priv(file_priv->minor->dev);
-
-	vmw_fifo_ping_host(dev_priv, SVGA_SYNC_GENERIC);
-	return drm_poll(filp, wait);
-}
-
-
-/**
- * vmw_fops_read - wrapper around the drm_read function
- *
- * @filp: See the linux fops read documentation.
- * @buffer: See the linux fops read documentation.
- * @count: See the linux fops read documentation.
- * @offset: See the linux fops read documentation.
- *
- * Wrapper around the drm_read function that makes sure the device is
- * processing the fifo if drm_read decides to wait.
- */
-ssize_t vmw_fops_read(struct file *filp, char __user *buffer,
-		      size_t count, loff_t *offset)
-{
-	struct drm_file *file_priv = filp->private_data;
-	struct vmw_private *dev_priv =
-		vmw_priv(file_priv->minor->dev);
-
-	vmw_fifo_ping_host(dev_priv, SVGA_SYNC_GENERIC);
-	return drm_read(filp, buffer, count, offset);
-}

drivers/gpu/drm/vmwgfx/vmwgfx_irq.c

@@ -65,7 +65,7 @@ static irqreturn_t vmw_thread_fn(int irq, void *arg)
 }
 
 /**
- * vmw_irq_handler irq handler
+ * vmw_irq_handler: irq handler
  *
  * @irq: irq number
  * @arg: Closure argument. Pointer to a struct drm_device cast to void *
@@ -82,11 +82,11 @@ static irqreturn_t vmw_irq_handler(int irq, void *arg)
 	uint32_t status, masked_status;
 	irqreturn_t ret = IRQ_HANDLED;
 
-	status = inl(dev_priv->io_start + VMWGFX_IRQSTATUS_PORT);
+	status = vmw_irq_status_read(dev_priv);
 	masked_status = status & READ_ONCE(dev_priv->irq_mask);
 
 	if (likely(status))
-		outl(status, dev_priv->io_start + VMWGFX_IRQSTATUS_PORT);
+		vmw_irq_status_write(dev_priv, status);
 
 	if (!status)
 		return IRQ_NONE;
@@ -114,10 +114,9 @@ static bool vmw_fifo_idle(struct vmw_private *dev_priv, uint32_t seqno)
 	return (vmw_read(dev_priv, SVGA_REG_BUSY) == 0);
 }
 
-void vmw_update_seqno(struct vmw_private *dev_priv,
-			 struct vmw_fifo_state *fifo_state)
+void vmw_update_seqno(struct vmw_private *dev_priv)
 {
-	uint32_t seqno = vmw_fifo_mem_read(dev_priv, SVGA_FIFO_FENCE);
+	uint32_t seqno = vmw_fence_read(dev_priv);
 
 	if (dev_priv->last_read_seqno != seqno) {
 		dev_priv->last_read_seqno = seqno;
@@ -128,18 +127,16 @@ void vmw_update_seqno(struct vmw_private *dev_priv,
 bool vmw_seqno_passed(struct vmw_private *dev_priv,
 			 uint32_t seqno)
 {
-	struct vmw_fifo_state *fifo_state;
 	bool ret;
 
 	if (likely(dev_priv->last_read_seqno - seqno < VMW_FENCE_WRAP))
 		return true;
 
-	fifo_state = &dev_priv->fifo;
-	vmw_update_seqno(dev_priv, fifo_state);
+	vmw_update_seqno(dev_priv);
 	if (likely(dev_priv->last_read_seqno - seqno < VMW_FENCE_WRAP))
 		return true;
 
-	if (!(fifo_state->capabilities & SVGA_FIFO_CAP_FENCE) &&
+	if (!(vmw_fifo_caps(dev_priv) & SVGA_FIFO_CAP_FENCE) &&
 	    vmw_fifo_idle(dev_priv, seqno))
 		return true;
 
@@ -161,7 +158,7 @@ int vmw_fallback_wait(struct vmw_private *dev_priv,
 		      bool interruptible,
 		      unsigned long timeout)
 {
-	struct vmw_fifo_state *fifo_state = &dev_priv->fifo;
+	struct vmw_fifo_state *fifo_state = dev_priv->fifo;
 
 	uint32_t count = 0;
 	uint32_t signal_seq;
@@ -221,7 +218,7 @@ int vmw_fallback_wait(struct vmw_private *dev_priv,
 	}
 	finish_wait(&dev_priv->fence_queue, &__wait);
 	if (ret == 0 && fifo_idle)
-		vmw_fifo_mem_write(dev_priv, SVGA_FIFO_FENCE, signal_seq);
+		vmw_fence_write(dev_priv, signal_seq);
 
 	wake_up_all(&dev_priv->fence_queue);
 out_err:
@@ -236,7 +233,7 @@ void vmw_generic_waiter_add(struct vmw_private *dev_priv,
 {
 	spin_lock_bh(&dev_priv->waiter_lock);
 	if ((*waiter_count)++ == 0) {
-		outl(flag, dev_priv->io_start + VMWGFX_IRQSTATUS_PORT);
+		vmw_irq_status_write(dev_priv, flag);
 		dev_priv->irq_mask |= flag;
 		vmw_write(dev_priv, SVGA_REG_IRQMASK, dev_priv->irq_mask);
 	}
@@ -278,59 +275,13 @@ void vmw_goal_waiter_remove(struct vmw_private *dev_priv)
 				  &dev_priv->goal_queue_waiters);
 }
 
-int vmw_wait_seqno(struct vmw_private *dev_priv,
-		      bool lazy, uint32_t seqno,
-		      bool interruptible, unsigned long timeout)
-{
-	long ret;
-	struct vmw_fifo_state *fifo = &dev_priv->fifo;
-
-	if (likely(dev_priv->last_read_seqno - seqno < VMW_FENCE_WRAP))
-		return 0;
-
-	if (likely(vmw_seqno_passed(dev_priv, seqno)))
-		return 0;
-
-	vmw_fifo_ping_host(dev_priv, SVGA_SYNC_GENERIC);
-
-	if (!(fifo->capabilities & SVGA_FIFO_CAP_FENCE))
-		return vmw_fallback_wait(dev_priv, lazy, true, seqno,
-					 interruptible, timeout);
-
-	if (!(dev_priv->capabilities & SVGA_CAP_IRQMASK))
-		return vmw_fallback_wait(dev_priv, lazy, false, seqno,
-					 interruptible, timeout);
-
-	vmw_seqno_waiter_add(dev_priv);
-
-	if (interruptible)
-		ret = wait_event_interruptible_timeout
-		    (dev_priv->fence_queue,
-		     vmw_seqno_passed(dev_priv, seqno),
-		     timeout);
-	else
-		ret = wait_event_timeout
-		    (dev_priv->fence_queue,
-		     vmw_seqno_passed(dev_priv, seqno),
-		     timeout);
-
-	vmw_seqno_waiter_remove(dev_priv);
-
-	if (unlikely(ret == 0))
-		ret = -EBUSY;
-	else if (likely(ret > 0))
-		ret = 0;
-
-	return ret;
-}
-
 static void vmw_irq_preinstall(struct drm_device *dev)
 {
 	struct vmw_private *dev_priv = vmw_priv(dev);
 	uint32_t status;
 
-	status = inl(dev_priv->io_start + VMWGFX_IRQSTATUS_PORT);
-	outl(status, dev_priv->io_start + VMWGFX_IRQSTATUS_PORT);
+	status = vmw_irq_status_read(dev_priv);
+	vmw_irq_status_write(dev_priv, status);
 }
 
 void vmw_irq_uninstall(struct drm_device *dev)
@@ -346,8 +297,8 @@ void vmw_irq_uninstall(struct drm_device *dev)
 
 	vmw_write(dev_priv, SVGA_REG_IRQMASK, 0);
 
-	status = inl(dev_priv->io_start + VMWGFX_IRQSTATUS_PORT);
-	outl(status, dev_priv->io_start + VMWGFX_IRQSTATUS_PORT);
+	status = vmw_irq_status_read(dev_priv);
+	vmw_irq_status_write(dev_priv, status);
 
 	dev->irq_enabled = false;
 	free_irq(dev->irq, dev);

drivers/gpu/drm/vmwgfx/vmwgfx_kms.c

@@ -38,8 +38,10 @@
 
 void vmw_du_cleanup(struct vmw_display_unit *du)
 {
+	struct vmw_private *dev_priv = vmw_priv(du->primary.dev);
 	drm_plane_cleanup(&du->primary);
-	drm_plane_cleanup(&du->cursor);
+	if (vmw_cmd_supported(dev_priv))
+		drm_plane_cleanup(&du->cursor);
 
 	drm_connector_unregister(&du->connector);
 	drm_crtc_cleanup(&du->crtc);
@@ -128,11 +130,17 @@ static void vmw_cursor_update_position(struct vmw_private *dev_priv,
 	uint32_t count;
 
 	spin_lock(&dev_priv->cursor_lock);
-	vmw_fifo_mem_write(dev_priv, SVGA_FIFO_CURSOR_ON, show ? 1 : 0);
-	vmw_fifo_mem_write(dev_priv, SVGA_FIFO_CURSOR_X, x);
-	vmw_fifo_mem_write(dev_priv, SVGA_FIFO_CURSOR_Y, y);
-	count = vmw_fifo_mem_read(dev_priv, SVGA_FIFO_CURSOR_COUNT);
-	vmw_fifo_mem_write(dev_priv, SVGA_FIFO_CURSOR_COUNT, ++count);
+	if (vmw_is_cursor_bypass3_enabled(dev_priv)) {
+		vmw_fifo_mem_write(dev_priv, SVGA_FIFO_CURSOR_ON, show ? 1 : 0);
+		vmw_fifo_mem_write(dev_priv, SVGA_FIFO_CURSOR_X, x);
+		vmw_fifo_mem_write(dev_priv, SVGA_FIFO_CURSOR_Y, y);
+		count = vmw_fifo_mem_read(dev_priv, SVGA_FIFO_CURSOR_COUNT);
+		vmw_fifo_mem_write(dev_priv, SVGA_FIFO_CURSOR_COUNT, ++count);
+	} else {
+		vmw_write(dev_priv, SVGA_REG_CURSOR_X, x);
+		vmw_write(dev_priv, SVGA_REG_CURSOR_Y, y);
+		vmw_write(dev_priv, SVGA_REG_CURSOR_ON, show ? 1 : 0);
+	}
 	spin_unlock(&dev_priv->cursor_lock);
 }
 
@@ -289,7 +297,7 @@ void vmw_du_primary_plane_destroy(struct drm_plane *plane)
 
 
 /**
- * vmw_du_vps_unpin_surf - unpins resource associated with a framebuffer surface
+ * vmw_du_plane_unpin_surf - unpins resource associated with a framebuffer surface
  *
  * @vps: plane state associated with the display surface
  * @unreference: true if we also want to unreference the display.
@@ -474,7 +482,7 @@ int vmw_du_primary_plane_atomic_check(struct drm_plane *plane,
  * vmw_du_cursor_plane_atomic_check - check if the new state is okay
  *
  * @plane: cursor plane
- * @new_state: info on the new plane state
+ * @state: info on the new plane state
  *
  * This is a chance to fail if the new cursor state does not fit
  * our requirements.
@@ -1045,7 +1053,8 @@ static int vmw_framebuffer_bo_dirty_ext(struct drm_framebuffer *framebuffer,
 {
 	struct vmw_private *dev_priv = vmw_priv(framebuffer->dev);
 
-	if (dev_priv->active_display_unit == vmw_du_legacy)
+	if (dev_priv->active_display_unit == vmw_du_legacy &&
+	    vmw_cmd_supported(dev_priv))
 		return vmw_framebuffer_bo_dirty(framebuffer, file_priv, flags,
 						color, clips, num_clips);
 
@@ -2633,7 +2642,7 @@ int vmw_kms_fbdev_init_data(struct vmw_private *dev_priv,
 }
 
 /**
- * vmw_kms_create_implicit_placement_proparty - Set up the implicit placement
+ * vmw_kms_create_implicit_placement_property - Set up the implicit placement
  * property.
  *
  * @dev_priv: Pointer to a device private struct.

drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c

@@ -404,19 +404,24 @@ static int vmw_ldu_init(struct vmw_private *dev_priv, unsigned unit)
 
 	drm_plane_helper_add(primary, &vmw_ldu_primary_plane_helper_funcs);
 
-	/* Initialize cursor plane */
-	ret = drm_universal_plane_init(dev, &ldu->base.cursor,
-			0, &vmw_ldu_cursor_funcs,
-			vmw_cursor_plane_formats,
-			ARRAY_SIZE(vmw_cursor_plane_formats),
-			NULL, DRM_PLANE_TYPE_CURSOR, NULL);
-	if (ret) {
-		DRM_ERROR("Failed to initialize cursor plane");
-		drm_plane_cleanup(&ldu->base.primary);
-		goto err_free;
-	}
+	/*
+	 * We're going to be using traces and software cursors
+	 */
+	if (vmw_cmd_supported(dev_priv)) {
+		/* Initialize cursor plane */
+		ret = drm_universal_plane_init(dev, &ldu->base.cursor,
+					       0, &vmw_ldu_cursor_funcs,
+					       vmw_cursor_plane_formats,
+					       ARRAY_SIZE(vmw_cursor_plane_formats),
+					       NULL, DRM_PLANE_TYPE_CURSOR, NULL);
+		if (ret) {
+			DRM_ERROR("Failed to initialize cursor plane");
+			drm_plane_cleanup(&ldu->base.primary);
+			goto err_free;
+		}
 
-	drm_plane_helper_add(cursor, &vmw_ldu_cursor_plane_helper_funcs);
+		drm_plane_helper_add(cursor, &vmw_ldu_cursor_plane_helper_funcs);
+	}
 
 	ret = drm_connector_init(dev, connector, &vmw_legacy_connector_funcs,
 				 DRM_MODE_CONNECTOR_VIRTUAL);
@@ -445,9 +450,10 @@ static int vmw_ldu_init(struct vmw_private *dev_priv, unsigned unit)
 		goto err_free_encoder;
 	}
 
-	ret = drm_crtc_init_with_planes(dev, crtc, &ldu->base.primary,
-					&ldu->base.cursor,
-					&vmw_legacy_crtc_funcs, NULL);
+	ret = drm_crtc_init_with_planes(
+		      dev, crtc, &ldu->base.primary,
+		      vmw_cmd_supported(dev_priv) ? &ldu->base.cursor : NULL,
+		      &vmw_legacy_crtc_funcs, NULL);
 	if (ret) {
 		DRM_ERROR("Failed to initialize CRTC\n");
 		goto err_free_unregister;

drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c

@@ -421,7 +421,7 @@ int vmw_overlay_pause_all(struct vmw_private *dev_priv)
 static bool vmw_overlay_available(const struct vmw_private *dev_priv)
 {
 	return (dev_priv->overlay_priv != NULL &&
-		((dev_priv->fifo.capabilities & VMW_OVERLAY_CAP_MASK) ==
+		((vmw_fifo_caps(dev_priv) & VMW_OVERLAY_CAP_MASK) ==
 		 VMW_OVERLAY_CAP_MASK));
 }
 

drivers/gpu/drm/vmwgfx/vmwgfx_reg.h

@@ -34,10 +34,6 @@
 
 #include <linux/types.h>
 
-#define VMWGFX_INDEX_PORT     0x0
-#define VMWGFX_VALUE_PORT     0x1
-#define VMWGFX_IRQSTATUS_PORT 0x8
-
 struct svga_guest_mem_descriptor {
 	u32 ppn;
 	u32 num_pages;

drivers/gpu/drm/vmwgfx/vmwgfx_resource.c

@@ -280,7 +280,7 @@ int vmw_user_resource_lookup_handle(struct vmw_private *dev_priv,
 }
 
 /**
- * vmw_user_resource_lookup_handle - lookup a struct resource from a
+ * vmw_user_resource_noref_lookup_handle - lookup a struct resource from a
  * TTM user-space handle and perform basic type checks
  *
  * @dev_priv:     Pointer to a device private struct
@@ -1075,7 +1075,7 @@ enum vmw_res_type vmw_res_type(const struct vmw_resource *res)
 }
 
 /**
- * vmw_resource_update_dirty - Update a resource's dirty tracker with a
+ * vmw_resource_dirty_update - Update a resource's dirty tracker with a
  * sequential range of touched backing store memory.
  * @res: The resource.
  * @start: The first page touched.

drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c

@@ -1222,7 +1222,7 @@ static void vmw_sou_bo_clip(struct vmw_kms_dirty *dirty)
 }
 
 /**
- * vmw_kms_do_bo_dirty - Dirty part of a buffer-object backed framebuffer
+ * vmw_kms_sou_do_bo_dirty - Dirty part of a buffer-object backed framebuffer
  *
  * @dev_priv: Pointer to the device private structure.
  * @framebuffer: Pointer to the buffer-object backed framebuffer.

drivers/gpu/drm/vmwgfx/vmwgfx_so.c

@@ -90,7 +90,7 @@ static const struct vmw_res_func vmw_view_func = {
 };
 
 /**
- * struct vmw_view - view define command body stub
+ * struct vmw_view_define - view define command body stub
  *
  * @view_id: The device id of the view being defined
  * @sid: The surface id of the view being defined

drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c

@@ -742,7 +742,7 @@ int vmw_kms_stdu_dma(struct vmw_private *dev_priv,
 }
 
 /**
- * vmw_stdu_surface_clip - Callback to encode a surface copy command cliprect
+ * vmw_kms_stdu_surface_clip - Callback to encode a surface copy command cliprect
  *
  * @dirty: The closure structure.
  *
@@ -780,7 +780,7 @@ static void vmw_kms_stdu_surface_clip(struct vmw_kms_dirty *dirty)
 }
 
 /**
- * vmw_stdu_surface_fifo_commit - Callback to fill in and submit a surface
+ * vmw_kms_stdu_surface_fifo_commit - Callback to fill in and submit a surface
  * copy command.
  *
  * @dirty: The closure structure.
@@ -1571,7 +1571,7 @@ static int vmw_stdu_plane_update_surface(struct vmw_private *dev_priv,
 /**
  * vmw_stdu_primary_plane_atomic_update - formally switches STDU to new plane
  * @plane: display plane
- * @old_state: Only used to get crtc info
+ * @state: Only used to get crtc info
  *
  * Formally update stdu->display_srf to the new plane, and bind the new
  * plane STDU.  This function is called during the commit phase when

drivers/gpu/drm/vmwgfx/vmwgfx_surface.c

@@ -680,7 +680,7 @@ static void vmw_user_surface_free(struct vmw_resource *res)
 }
 
 /**
- * vmw_user_surface_free - User visible surface TTM base object destructor
+ * vmw_user_surface_base_release - User visible surface TTM base object destructor
  *
  * @p_base:         Pointer to a pointer to a TTM base object
  *                  embedded in a struct vmw_user_surface.
@@ -702,7 +702,7 @@ static void vmw_user_surface_base_release(struct ttm_base_object **p_base)
 }
 
 /**
- * vmw_user_surface_destroy_ioctl - Ioctl function implementing
+ * vmw_surface_destroy_ioctl - Ioctl function implementing
  *                                  the user surface destroy functionality.
  *
  * @dev:            Pointer to a struct drm_device.
@@ -719,7 +719,7 @@ int vmw_surface_destroy_ioctl(struct drm_device *dev, void *data,
 }
 
 /**
- * vmw_user_surface_define_ioctl - Ioctl function implementing
+ * vmw_surface_define_ioctl - Ioctl function implementing
  *                                  the user surface define functionality.
  *
  * @dev:            Pointer to a struct drm_device.
@@ -1001,7 +1001,7 @@ vmw_surface_handle_reference(struct vmw_private *dev_priv,
 }
 
 /**
- * vmw_user_surface_define_ioctl - Ioctl function implementing
+ * vmw_surface_reference_ioctl - Ioctl function implementing
  *                                  the user surface reference functionality.
  *
  * @dev:            Pointer to a struct drm_device.
@@ -1055,7 +1055,7 @@ int vmw_surface_reference_ioctl(struct drm_device *dev, void *data,
 }
 
 /**
- * vmw_surface_define_encode - Encode a surface_define command.
+ * vmw_gb_surface_create - Encode a surface_define command.
  *
  * @res:        Pointer to a struct vmw_resource embedded in a struct
  *              vmw_surface.

drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c

@@ -200,7 +200,8 @@ struct vmw_ttm_tt {
 const size_t vmw_tt_size = sizeof(struct vmw_ttm_tt);
 
 /**
- * Helper functions to advance a struct vmw_piter iterator.
+ * __vmw_piter_non_sg_next: Helper functions to advance
+ * a struct vmw_piter iterator.
  *
  * @viter: Pointer to the iterator.
  *
@@ -222,7 +223,8 @@ static bool __vmw_piter_sg_next(struct vmw_piter *viter)
 
 
 /**
- * Helper functions to return a pointer to the current page.
+ * __vmw_piter_non_sg_page: Helper functions to return a pointer
+ * to the current page.
  *
  * @viter: Pointer to the iterator
  *
@@ -236,7 +238,8 @@ static struct page *__vmw_piter_non_sg_page(struct vmw_piter *viter)
 }
 
 /**
- * Helper functions to return the DMA address of the current page.
+ * __vmw_piter_phys_addr: Helper functions to return the DMA
+ * address of the current page.
  *
  * @viter: Pointer to the iterator
  *

drivers/gpu/drm/vmwgfx/vmwgfx_validation.c

@@ -809,7 +809,7 @@ void vmw_validation_revert(struct vmw_validation_context *ctx)
 }
 
 /**
- * vmw_validation_cone - Commit validation actions after command submission
+ * vmw_validation_done - Commit validation actions after command submission
  * success.
  * @ctx: The validation context.
  * @fence: Fence with which to fence all buffer objects taking part in the