Merge branch 'perf/core' into perf/uprobes

Merge in latest upstream (and the latest perf development tree),
to prepare for tooling changes, and also to pick up v3.4 MM
changes that the uprobes code needs to take care of.

Signed-off-by: Ingo Molnar <mingo@kernel.org>

1 files changed, 178 insertions, 63 deletions

diff --git a/drivers/usb/host/xhci-mem.c b/drivers/usb/host/xhci-mem.c
index 383fc857491..cae4c6f2845 100644
--- a/drivers/usb/host/xhci-mem.c
+++ b/drivers/usb/host/xhci-mem.c
@@ -34,10 +34,12 @@
  * Section 4.11.1.1:
  * "All components of all Command and Transfer TRBs shall be initialized to '0'"
  */
-static struct xhci_segment *xhci_segment_alloc(struct xhci_hcd *xhci, gfp_t flags)
+static struct xhci_segment *xhci_segment_alloc(struct xhci_hcd *xhci,
+					unsigned int cycle_state, gfp_t flags)
 {
 	struct xhci_segment *seg;
 	dma_addr_t	dma;
+	int		i;
 
 	seg = kzalloc(sizeof *seg, flags);
 	if (!seg)
@@ -50,6 +52,11 @@ static struct xhci_segment *xhci_segment_alloc(struct xhci_hcd *xhci, gfp_t flag
 	}
 
 	memset(seg->trbs, 0, SEGMENT_SIZE);
+	/* If the cycle state is 0, set the cycle bit to 1 for all the TRBs */
+	if (cycle_state == 0) {
+		for (i = 0; i < TRBS_PER_SEGMENT; i++)
+			seg->trbs[i].link.control |= TRB_CYCLE;
+	}
 	seg->dma = dma;
 	seg->next = NULL;
 
@@ -65,6 +72,20 @@ static void xhci_segment_free(struct xhci_hcd *xhci, struct xhci_segment *seg)
 	kfree(seg);
 }
 
+static void xhci_free_segments_for_ring(struct xhci_hcd *xhci,
+				struct xhci_segment *first)
+{
+	struct xhci_segment *seg;
+
+	seg = first->next;
+	while (seg != first) {
+		struct xhci_segment *next = seg->next;
+		xhci_segment_free(xhci, seg);
+		seg = next;
+	}
+	xhci_segment_free(xhci, first);
+}
+
 /*
  * Make the prev segment point to the next segment.
  *
@@ -73,14 +94,14 @@ static void xhci_segment_free(struct xhci_hcd *xhci, struct xhci_segment *seg)
  * related flags, such as End TRB, Toggle Cycle, and no snoop.
  */
 static void xhci_link_segments(struct xhci_hcd *xhci, struct xhci_segment *prev,
-		struct xhci_segment *next, bool link_trbs, bool isoc)
+		struct xhci_segment *next, enum xhci_ring_type type)
 {
 	u32 val;
 
 	if (!prev || !next)
 		return;
 	prev->next = next;
-	if (link_trbs) {
+	if (type != TYPE_EVENT) {
 		prev->trbs[TRBS_PER_SEGMENT-1].link.segment_ptr =
 			cpu_to_le64(next->dma);
 
@@ -91,35 +112,55 @@ static void xhci_link_segments(struct xhci_hcd *xhci, struct xhci_segment *prev,
 		/* Always set the chain bit with 0.95 hardware */
 		/* Set chain bit for isoc rings on AMD 0.96 host */
 		if (xhci_link_trb_quirk(xhci) ||
-				(isoc && (xhci->quirks & XHCI_AMD_0x96_HOST)))
+				(type == TYPE_ISOC &&
+				 (xhci->quirks & XHCI_AMD_0x96_HOST)))
 			val |= TRB_CHAIN;
 		prev->trbs[TRBS_PER_SEGMENT-1].link.control = cpu_to_le32(val);
 	}
 }
 
+/*
+ * Link the ring to the new segments.
+ * Set Toggle Cycle for the new ring if needed.
+ */
+static void xhci_link_rings(struct xhci_hcd *xhci, struct xhci_ring *ring,
+		struct xhci_segment *first, struct xhci_segment *last,
+		unsigned int num_segs)
+{
+	struct xhci_segment *next;
+
+	if (!ring || !first || !last)
+		return;
+
+	next = ring->enq_seg->next;
+	xhci_link_segments(xhci, ring->enq_seg, first, ring->type);
+	xhci_link_segments(xhci, last, next, ring->type);
+	ring->num_segs += num_segs;
+	ring->num_trbs_free += (TRBS_PER_SEGMENT - 1) * num_segs;
+
+	if (ring->type != TYPE_EVENT && ring->enq_seg == ring->last_seg) {
+		ring->last_seg->trbs[TRBS_PER_SEGMENT-1].link.control
+			&= ~cpu_to_le32(LINK_TOGGLE);
+		last->trbs[TRBS_PER_SEGMENT-1].link.control
+			|= cpu_to_le32(LINK_TOGGLE);
+		ring->last_seg = last;
+	}
+}
+
 /* XXX: Do we need the hcd structure in all these functions? */
 void xhci_ring_free(struct xhci_hcd *xhci, struct xhci_ring *ring)
 {
-	struct xhci_segment *seg;
-	struct xhci_segment *first_seg;
-
 	if (!ring)
 		return;
-	if (ring->first_seg) {
-		first_seg = ring->first_seg;
-		seg = first_seg->next;
-		while (seg != first_seg) {
-			struct xhci_segment *next = seg->next;
-			xhci_segment_free(xhci, seg);
-			seg = next;
-		}
-		xhci_segment_free(xhci, first_seg);
-		ring->first_seg = NULL;
-	}
+
+	if (ring->first_seg)
+		xhci_free_segments_for_ring(xhci, ring->first_seg);
+
 	kfree(ring);
 }
 
-static void xhci_initialize_ring_info(struct xhci_ring *ring)
+static void xhci_initialize_ring_info(struct xhci_ring *ring,
+					unsigned int cycle_state)
 {
 	/* The ring is empty, so the enqueue pointer == dequeue pointer */
 	ring->enqueue = ring->first_seg->trbs;
@@ -129,11 +170,53 @@ static void xhci_initialize_ring_info(struct xhci_ring *ring)
 	/* The ring is initialized to 0. The producer must write 1 to the cycle
 	 * bit to handover ownership of the TRB, so PCS = 1.  The consumer must
 	 * compare CCS to the cycle bit to check ownership, so CCS = 1.
+	 *
+	 * New rings are initialized with cycle state equal to 1; if we are
+	 * handling ring expansion, set the cycle state equal to the old ring.
 	 */
-	ring->cycle_state = 1;
+	ring->cycle_state = cycle_state;
 	/* Not necessary for new rings, but needed for re-initialized rings */
 	ring->enq_updates = 0;
 	ring->deq_updates = 0;
+
+	/*
+	 * Each segment has a link TRB, and leave an extra TRB for SW
+	 * accounting purpose
+	 */
+	ring->num_trbs_free = ring->num_segs * (TRBS_PER_SEGMENT - 1) - 1;
+}
+
+/* Allocate segments and link them for a ring */
+static int xhci_alloc_segments_for_ring(struct xhci_hcd *xhci,
+		struct xhci_segment **first, struct xhci_segment **last,
+		unsigned int num_segs, unsigned int cycle_state,
+		enum xhci_ring_type type, gfp_t flags)
+{
+	struct xhci_segment *prev;
+
+	prev = xhci_segment_alloc(xhci, cycle_state, flags);
+	if (!prev)
+		return -ENOMEM;
+	num_segs--;
+
+	*first = prev;
+	while (num_segs > 0) {
+		struct xhci_segment	*next;
+
+		next = xhci_segment_alloc(xhci, cycle_state, flags);
+		if (!next) {
+			xhci_free_segments_for_ring(xhci, *first);
+			return -ENOMEM;
+		}
+		xhci_link_segments(xhci, prev, next, type);
+
+		prev = next;
+		num_segs--;
+	}
+	xhci_link_segments(xhci, prev, *first, type);
+	*last = prev;
+
+	return 0;
 }
 
 /**
@@ -144,44 +227,34 @@ static void xhci_initialize_ring_info(struct xhci_ring *ring)
  * See section 4.9.1 and figures 15 and 16.
  */
 static struct xhci_ring *xhci_ring_alloc(struct xhci_hcd *xhci,
-		unsigned int num_segs, bool link_trbs, bool isoc, gfp_t flags)
+		unsigned int num_segs, unsigned int cycle_state,
+		enum xhci_ring_type type, gfp_t flags)
 {
 	struct xhci_ring	*ring;
-	struct xhci_segment	*prev;
+	int ret;
 
 	ring = kzalloc(sizeof *(ring), flags);
 	if (!ring)
 		return NULL;
 
+	ring->num_segs = num_segs;
 	INIT_LIST_HEAD(&ring->td_list);
+	ring->type = type;
 	if (num_segs == 0)
 		return ring;
 
-	ring->first_seg = xhci_segment_alloc(xhci, flags);
-	if (!ring->first_seg)
+	ret = xhci_alloc_segments_for_ring(xhci, &ring->first_seg,
+			&ring->last_seg, num_segs, cycle_state, type, flags);
+	if (ret)
 		goto fail;
-	num_segs--;
 
-	prev = ring->first_seg;
-	while (num_segs > 0) {
-		struct xhci_segment	*next;
-
-		next = xhci_segment_alloc(xhci, flags);
-		if (!next)
-			goto fail;
-		xhci_link_segments(xhci, prev, next, link_trbs, isoc);
-
-		prev = next;
-		num_segs--;
-	}
-	xhci_link_segments(xhci, prev, ring->first_seg, link_trbs, isoc);
-
-	if (link_trbs) {
+	/* Only event ring does not use link TRB */
+	if (type != TYPE_EVENT) {
 		/* See section 4.9.2.1 and 6.4.4.1 */
-		prev->trbs[TRBS_PER_SEGMENT-1].link.control |=
+		ring->last_seg->trbs[TRBS_PER_SEGMENT - 1].link.control |=
 			cpu_to_le32(LINK_TOGGLE);
 	}
-	xhci_initialize_ring_info(ring);
+	xhci_initialize_ring_info(ring, cycle_state);
 	return ring;
 
 fail:
@@ -217,23 +290,64 @@ void xhci_free_or_cache_endpoint_ring(struct xhci_hcd *xhci,
  * pointers to the beginning of the ring.
  */
 static void xhci_reinit_cached_ring(struct xhci_hcd *xhci,
-		struct xhci_ring *ring, bool isoc)
+			struct xhci_ring *ring, unsigned int cycle_state,
+			enum xhci_ring_type type)
 {
 	struct xhci_segment	*seg = ring->first_seg;
+	int i;
+
 	do {
 		memset(seg->trbs, 0,
 				sizeof(union xhci_trb)*TRBS_PER_SEGMENT);
+		if (cycle_state == 0) {
+			for (i = 0; i < TRBS_PER_SEGMENT; i++)
+				seg->trbs[i].link.control |= TRB_CYCLE;
+		}
 		/* All endpoint rings have link TRBs */
-		xhci_link_segments(xhci, seg, seg->next, 1, isoc);
+		xhci_link_segments(xhci, seg, seg->next, type);
 		seg = seg->next;
 	} while (seg != ring->first_seg);
-	xhci_initialize_ring_info(ring);
+	ring->type = type;
+	xhci_initialize_ring_info(ring, cycle_state);
 	/* td list should be empty since all URBs have been cancelled,
 	 * but just in case...
 	 */
 	INIT_LIST_HEAD(&ring->td_list);
 }
 
+/*
+ * Expand an existing ring.
+ * Look for a cached ring or allocate a new ring which has same segment numbers
+ * and link the two rings.
+ */
+int xhci_ring_expansion(struct xhci_hcd *xhci, struct xhci_ring *ring,
+				unsigned int num_trbs, gfp_t flags)
+{
+	struct xhci_segment	*first;
+	struct xhci_segment	*last;
+	unsigned int		num_segs;
+	unsigned int		num_segs_needed;
+	int			ret;
+
+	num_segs_needed = (num_trbs + (TRBS_PER_SEGMENT - 1) - 1) /
+				(TRBS_PER_SEGMENT - 1);
+
+	/* Allocate number of segments we needed, or double the ring size */
+	num_segs = ring->num_segs > num_segs_needed ?
+			ring->num_segs : num_segs_needed;
+
+	ret = xhci_alloc_segments_for_ring(xhci, &first, &last,
+			num_segs, ring->cycle_state, ring->type, flags);
+	if (ret)
+		return -ENOMEM;
+
+	xhci_link_rings(xhci, ring, first, last, num_segs);
+	xhci_dbg(xhci, "ring expansion succeed, now has %d segments\n",
+			ring->num_segs);
+
+	return 0;
+}
+
 #define CTX_SIZE(_hcc) (HCC_64BYTE_CONTEXT(_hcc) ? 64 : 32)
 
 static struct xhci_container_ctx *xhci_alloc_container_ctx(struct xhci_hcd *xhci,
@@ -528,7 +642,7 @@ struct xhci_stream_info *xhci_alloc_stream_info(struct xhci_hcd *xhci,
 	 */
 	for (cur_stream = 1; cur_stream < num_streams; cur_stream++) {
 		stream_info->stream_rings[cur_stream] =
-			xhci_ring_alloc(xhci, 1, true, false, mem_flags);
+			xhci_ring_alloc(xhci, 2, 1, TYPE_STREAM, mem_flags);
 		cur_ring = stream_info->stream_rings[cur_stream];
 		if (!cur_ring)
 			goto cleanup_rings;
@@ -862,7 +976,7 @@ int xhci_alloc_virt_device(struct xhci_hcd *xhci, int slot_id,
 	}
 
 	/* Allocate endpoint 0 ring */
-	dev->eps[0].ring = xhci_ring_alloc(xhci, 1, true, false, flags);
+	dev->eps[0].ring = xhci_ring_alloc(xhci, 2, 1, TYPE_CTRL, flags);
 	if (!dev->eps[0].ring)
 		goto fail;
 
@@ -1300,24 +1414,16 @@ int xhci_endpoint_init(struct xhci_hcd *xhci,
 	struct xhci_ring *ep_ring;
 	unsigned int max_packet;
 	unsigned int max_burst;
+	enum xhci_ring_type type;
 	u32 max_esit_payload;
 
 	ep_index = xhci_get_endpoint_index(&ep->desc);
 	ep_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, ep_index);
 
+	type = usb_endpoint_type(&ep->desc);
 	/* Set up the endpoint ring */
-	/*
-	 * Isochronous endpoint ring needs bigger size because one isoc URB
-	 * carries multiple packets and it will insert multiple tds to the
-	 * ring.
-	 * This should be replaced with dynamic ring resizing in the future.
-	 */
-	if (usb_endpoint_xfer_isoc(&ep->desc))
-		virt_dev->eps[ep_index].new_ring =
-			xhci_ring_alloc(xhci, 8, true, true, mem_flags);
-	else
-		virt_dev->eps[ep_index].new_ring =
-			xhci_ring_alloc(xhci, 1, true, false, mem_flags);
+	virt_dev->eps[ep_index].new_ring =
+		xhci_ring_alloc(xhci, 2, 1, type, mem_flags);
 	if (!virt_dev->eps[ep_index].new_ring) {
 		/* Attempt to use the ring cache */
 		if (virt_dev->num_rings_cached == 0)
@@ -1327,7 +1433,7 @@ int xhci_endpoint_init(struct xhci_hcd *xhci,
 		virt_dev->ring_cache[virt_dev->num_rings_cached] = NULL;
 		virt_dev->num_rings_cached--;
 		xhci_reinit_cached_ring(xhci, virt_dev->eps[ep_index].new_ring,
-			usb_endpoint_xfer_isoc(&ep->desc) ? true : false);
+					1, type);
 	}
 	virt_dev->eps[ep_index].skip = false;
 	ep_ring = virt_dev->eps[ep_index].new_ring;
@@ -2157,7 +2263,7 @@ int xhci_mem_init(struct xhci_hcd *xhci, gfp_t flags)
 	unsigned int	val, val2;
 	u64		val_64;
 	struct xhci_segment	*seg;
-	u32 page_size;
+	u32 page_size, temp;
 	int i;
 
 	page_size = xhci_readl(xhci, &xhci->op_regs->page_size);
@@ -2235,7 +2341,7 @@ int xhci_mem_init(struct xhci_hcd *xhci, gfp_t flags)
 		goto fail;
 
 	/* Set up the command ring to have one segments for now. */
-	xhci->cmd_ring = xhci_ring_alloc(xhci, 1, true, false, flags);
+	xhci->cmd_ring = xhci_ring_alloc(xhci, 1, 1, TYPE_COMMAND, flags);
 	if (!xhci->cmd_ring)
 		goto fail;
 	xhci_dbg(xhci, "Allocated command ring at %p\n", xhci->cmd_ring);
@@ -2266,7 +2372,7 @@ int xhci_mem_init(struct xhci_hcd *xhci, gfp_t flags)
 	 * the event ring segment table (ERST).  Section 4.9.3.
 	 */
 	xhci_dbg(xhci, "// Allocating event ring\n");
-	xhci->event_ring = xhci_ring_alloc(xhci, ERST_NUM_SEGS, false, false,
+	xhci->event_ring = xhci_ring_alloc(xhci, ERST_NUM_SEGS, 1, TYPE_EVENT,
 						flags);
 	if (!xhci->event_ring)
 		goto fail;
@@ -2340,6 +2446,15 @@ int xhci_mem_init(struct xhci_hcd *xhci, gfp_t flags)
 
 	INIT_LIST_HEAD(&xhci->lpm_failed_devs);
 
+	/* Enable USB 3.0 device notifications for function remote wake, which
+	 * is necessary for allowing USB 3.0 devices to do remote wakeup from
+	 * U3 (device suspend).
+	 */
+	temp = xhci_readl(xhci, &xhci->op_regs->dev_notification);
+	temp &= ~DEV_NOTE_MASK;
+	temp |= DEV_NOTE_FWAKE;
+	xhci_writel(xhci, temp, &xhci->op_regs->dev_notification);
+
 	return 0;
 
 fail: