IB/hfi1: Fix bar0 mapping to use write combining

When the debugpat kernel boot flag is turned on the following
traces are printed:

[ 1884.793168] x86/PAT: Overlap at 0x90000000-0x92000000
[ 1884.803510] x86/PAT: reserve_memtype added [mem 0x91200000-0x9127ffff],
track uncached-minus, req write-combining, ret uncached-minus
[ 1884.818167] hfi1 0000:05:00.0: hfi1_0: WC Remapped RcvArray:
ffffc9000a980000

The ioremap_wc() clearly is not returning a write combining mapping due
to an overlap where the RcvArray is mapped in a uncached mapping prior
to creating the proposed write combining mapping.

The patch replaces the single base register for uncached CSRs that
used to overlap the RcvArray with two mappings.   One, kregbase1, from the
bar0 up to the RcvArray and another, kregbase2, from the end of the
RcvArray to the pio send buffer space.  A new dd field, base2_start,
is used to convert the zero-based offset in the CSR routines to the
correct kregbase1/kregbase2 mapping.  A single direct write of the
RcvArray CSRs is replaced with hfi1_put_tid() to insure correct access
using the new disjoint mapping.

Additionally, the kregend field is deleted since it is only ever written.

patdebug now shows the RcvArray as write combining:
[   35.688990] x86/PAT: reserve_memtype added [mem 0x91200000-0x9127ffff],
track write-combining, req write-combining, ret write-combining

To insulate from any potential issues with write combining, all
writeq are now flushed in hfi1_put_tid() and rcv_array_wc_fill().

Reviewed-by: Mitko Haralanov <mitko.haralanov@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

1 files changed, 62 insertions, 17 deletions

diff --git a/drivers/infiniband/hw/hfi1/chip.c b/drivers/infiniband/hw/hfi1/chip.c
index efc84810e821..4fce173c8667 100644
--- a/drivers/infiniband/hw/hfi1/chip.c
+++ b/drivers/infiniband/hw/hfi1/chip.c
@@ -1297,25 +1297,71 @@ CNTR_ELEM(#name, \
 	  CNTR_SYNTH, \
 	  access_ibp_##cntr)
 
+/**
+ * hfi_addr_from_offset - return addr for readq/writeq
+ * @dd - the dd device
+ * @offset - the offset of the CSR within bar0
+ *
+ * This routine selects the appropriate base address
+ * based on the indicated offset.
+ */
+static inline void __iomem *hfi1_addr_from_offset(
+	const struct hfi1_devdata *dd,
+	u32 offset)
+{
+	if (offset >= dd->base2_start)
+		return dd->kregbase2 + (offset - dd->base2_start);
+	return dd->kregbase1 + offset;
+}
+
+/**
+ * read_csr - read CSR at the indicated offset
+ * @dd - the dd device
+ * @offset - the offset of the CSR within bar0
+ *
+ * Return: the value read or all FF's if there
+ * is no mapping
+ */
 u64 read_csr(const struct hfi1_devdata *dd, u32 offset)
 {
-	if (dd->flags & HFI1_PRESENT) {
-		return readq((void __iomem *)dd->kregbase + offset);
-	}
+	if (dd->flags & HFI1_PRESENT)
+		return readq(hfi1_addr_from_offset(dd, offset));
 	return -1;
 }
 
+/**
+ * write_csr - write CSR at the indicated offset
+ * @dd - the dd device
+ * @offset - the offset of the CSR within bar0
+ * @value - value to write
+ */
 void write_csr(const struct hfi1_devdata *dd, u32 offset, u64 value)
 {
-	if (dd->flags & HFI1_PRESENT)
-		writeq(value, (void __iomem *)dd->kregbase + offset);
+	if (dd->flags & HFI1_PRESENT) {
+		void __iomem *base = hfi1_addr_from_offset(dd, offset);
+
+		/* avoid write to RcvArray */
+		if (WARN_ON(offset >= RCV_ARRAY && offset < dd->base2_start))
+			return;
+		writeq(value, base);
+	}
 }
 
+/**
+ * get_csr_addr - return te iomem address for offset
+ * @dd - the dd device
+ * @offset - the offset of the CSR within bar0
+ *
+ * Return: The iomem address to use in subsequent
+ * writeq/readq operations.
+ */
 void __iomem *get_csr_addr(
-	struct hfi1_devdata *dd,
+	const struct hfi1_devdata *dd,
 	u32 offset)
 {
-	return (void __iomem *)dd->kregbase + offset;
+	if (dd->flags & HFI1_PRESENT)
+		return hfi1_addr_from_offset(dd, offset);
+	return NULL;
 }
 
 static inline u64 read_write_csr(const struct hfi1_devdata *dd, u32 csr,
@@ -9752,14 +9798,13 @@ void hfi1_put_tid(struct hfi1_devdata *dd, u32 index,
 		  u32 type, unsigned long pa, u16 order)
 {
 	u64 reg;
-	void __iomem *base = (dd->rcvarray_wc ? dd->rcvarray_wc :
-			      (dd->kregbase + RCV_ARRAY));
 
 	if (!(dd->flags & HFI1_PRESENT))
 		goto done;
 
-	if (type == PT_INVALID) {
+	if (type == PT_INVALID || type == PT_INVALID_FLUSH) {
 		pa = 0;
+		order = 0;
 	} else if (type > PT_INVALID) {
 		dd_dev_err(dd,
 			   "unexpected receive array type %u for index %u, not handled\n",
@@ -9773,13 +9818,14 @@ void hfi1_put_tid(struct hfi1_devdata *dd, u32 index,
 		| (u64)order << RCV_ARRAY_RT_BUF_SIZE_SHIFT
 		| ((pa >> RT_ADDR_SHIFT) & RCV_ARRAY_RT_ADDR_MASK)
 					<< RCV_ARRAY_RT_ADDR_SHIFT;
-	trace_hfi1_write_rcvarray(base + (index * 8), reg);
-	writeq(reg, base + (index * 8));
+	trace_hfi1_write_rcvarray(dd->rcvarray_wc + (index * 8), reg);
+	writeq(reg, dd->rcvarray_wc + (index * 8));
 
-	if (type == PT_EAGER)
+	if (type == PT_EAGER || type == PT_INVALID_FLUSH || (index & 3) == 3)
 		/*
-		 * Eager entries are written one-by-one so we have to push them
-		 * after we write the entry.
+		 * Eager entries are written and flushed
+		 *
+		 * Expected entries are flushed every 4 writes
 		 */
 		flush_wc();
 done:
@@ -13411,8 +13457,7 @@ static void write_uninitialized_csrs_and_memories(struct hfi1_devdata *dd)
 
 	/* RcvArray */
 	for (i = 0; i < dd->chip_rcv_array_count; i++)
-		write_csr(dd, RCV_ARRAY + (8 * i),
-			  RCV_ARRAY_RT_WRITE_ENABLE_SMASK);
+		hfi1_put_tid(dd, i, PT_INVALID_FLUSH, 0, 0);
 
 	/* RcvQPMapTable */
 	for (i = 0; i < 32; i++)