perf/x86/intel/uncore: Add BW counters for GT, IA and IO breakdown

Linux only has support to read total DDR reads and writes. Here we
add support to enable bandwidth breakdown-GT, IA and IO. Breakdown
of BW is important to debug and optimize memory access. This can also
be used for telemetry and improving the system software.The offsets for
GT, IA and IO are added and these free running counters can be accessed
via MMIO space.

The BW breakdown can be measured using the following cmd:

  perf stat -e uncore_imc/gt_requests/,uncore_imc/ia_requests/,uncore_imc/io_requests/

             30.57 MiB  uncore_imc/gt_requests/
           1346.13 MiB  uncore_imc/ia_requests/
            190.97 MiB  uncore_imc/io_requests/

       5.984572733 seconds time elapsed

     BW/s = <gt,ia,io>_requests/time elapsed
Signed-off-by: NVaibhav Shankar <vaibhav.shankar@intel.com>
Signed-off-by: NIngo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20200814022234.23605-1-vaibhav.shankar@intel.com

arch/x86/events/intel/uncore_snb.c

@@ -390,6 +390,18 @@ static struct uncore_event_desc snb_uncore_imc_events[] = {
 	INTEL_UNCORE_EVENT_DESC(data_writes.scale, "6.103515625e-5"),
 	INTEL_UNCORE_EVENT_DESC(data_writes.unit, "MiB"),
 
+	INTEL_UNCORE_EVENT_DESC(gt_requests, "event=0x03"),
+	INTEL_UNCORE_EVENT_DESC(gt_requests.scale, "6.103515625e-5"),
+	INTEL_UNCORE_EVENT_DESC(gt_requests.unit, "MiB"),
+
+	INTEL_UNCORE_EVENT_DESC(ia_requests, "event=0x04"),
+	INTEL_UNCORE_EVENT_DESC(ia_requests.scale, "6.103515625e-5"),
+	INTEL_UNCORE_EVENT_DESC(ia_requests.unit, "MiB"),
+
+	INTEL_UNCORE_EVENT_DESC(io_requests, "event=0x05"),
+	INTEL_UNCORE_EVENT_DESC(io_requests.scale, "6.103515625e-5"),
+	INTEL_UNCORE_EVENT_DESC(io_requests.unit, "MiB"),
+
 	{ /* end: all zeroes */ },
 };
 
@@ -405,13 +417,35 @@ static struct uncore_event_desc snb_uncore_imc_events[] = {
 #define SNB_UNCORE_PCI_IMC_DATA_WRITES_BASE	0x5054
 #define SNB_UNCORE_PCI_IMC_CTR_BASE		SNB_UNCORE_PCI_IMC_DATA_READS_BASE
 
+/* BW break down- legacy counters */
+#define SNB_UNCORE_PCI_IMC_GT_REQUESTS		0x3
+#define SNB_UNCORE_PCI_IMC_GT_REQUESTS_BASE	0x5040
+#define SNB_UNCORE_PCI_IMC_IA_REQUESTS		0x4
+#define SNB_UNCORE_PCI_IMC_IA_REQUESTS_BASE	0x5044
+#define SNB_UNCORE_PCI_IMC_IO_REQUESTS		0x5
+#define SNB_UNCORE_PCI_IMC_IO_REQUESTS_BASE	0x5048
+
 enum perf_snb_uncore_imc_freerunning_types {
-	SNB_PCI_UNCORE_IMC_DATA		= 0,
+	SNB_PCI_UNCORE_IMC_DATA_READS		= 0,
+	SNB_PCI_UNCORE_IMC_DATA_WRITES,
+	SNB_PCI_UNCORE_IMC_GT_REQUESTS,
+	SNB_PCI_UNCORE_IMC_IA_REQUESTS,
+	SNB_PCI_UNCORE_IMC_IO_REQUESTS,
+
 	SNB_PCI_UNCORE_IMC_FREERUNNING_TYPE_MAX,
 };
 
 static struct freerunning_counters snb_uncore_imc_freerunning[] = {
-	[SNB_PCI_UNCORE_IMC_DATA]     = { SNB_UNCORE_PCI_IMC_DATA_READS_BASE, 0x4, 0x0, 2, 32 },
+	[SNB_PCI_UNCORE_IMC_DATA_READS]		= { SNB_UNCORE_PCI_IMC_DATA_READS_BASE,
+							0x0, 0x0, 1, 32 },
+	[SNB_PCI_UNCORE_IMC_DATA_READS]		= { SNB_UNCORE_PCI_IMC_DATA_WRITES_BASE,
+							0x0, 0x0, 1, 32 },
+	[SNB_PCI_UNCORE_IMC_GT_REQUESTS]	= { SNB_UNCORE_PCI_IMC_GT_REQUESTS_BASE,
+							0x0, 0x0, 1, 32 },
+	[SNB_PCI_UNCORE_IMC_IA_REQUESTS]	= { SNB_UNCORE_PCI_IMC_IA_REQUESTS_BASE,
+							0x0, 0x0, 1, 32 },
+	[SNB_PCI_UNCORE_IMC_IO_REQUESTS]	= { SNB_UNCORE_PCI_IMC_IO_REQUESTS_BASE,
+							0x0, 0x0, 1, 32 },
 };
 
 static struct attribute *snb_uncore_imc_formats_attr[] = {
@@ -525,6 +559,18 @@ static int snb_uncore_imc_event_init(struct perf_event *event)
 		base = SNB_UNCORE_PCI_IMC_DATA_WRITES_BASE;
 		idx = UNCORE_PMC_IDX_FREERUNNING;
 		break;
+	case SNB_UNCORE_PCI_IMC_GT_REQUESTS:
+		base = SNB_UNCORE_PCI_IMC_GT_REQUESTS_BASE;
+		idx = UNCORE_PMC_IDX_FREERUNNING;
+		break;
+	case SNB_UNCORE_PCI_IMC_IA_REQUESTS:
+		base = SNB_UNCORE_PCI_IMC_IA_REQUESTS_BASE;
+		idx = UNCORE_PMC_IDX_FREERUNNING;
+		break;
+	case SNB_UNCORE_PCI_IMC_IO_REQUESTS:
+		base = SNB_UNCORE_PCI_IMC_IO_REQUESTS_BASE;
+		idx = UNCORE_PMC_IDX_FREERUNNING;
+		break;
 	default:
 		return -EINVAL;
 	}
@@ -598,7 +644,7 @@ static struct intel_uncore_ops snb_uncore_imc_ops = {
 
 static struct intel_uncore_type snb_uncore_imc = {
 	.name		= "imc",
-	.num_counters   = 2,
+	.num_counters   = 5,
 	.num_boxes	= 1,
 	.num_freerunning_types	= SNB_PCI_UNCORE_IMC_FREERUNNING_TYPE_MAX,
 	.mmio_map_size	= SNB_UNCORE_PCI_IMC_MAP_SIZE,