diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index bcef6ea..3b30bb2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -542,6 +542,9 @@ struct amdgpu_asic_funcs {
 	bool (*need_full_reset)(struct amdgpu_device *adev);
 	/* initialize doorbell layout for specific asic*/
 	void (*init_doorbell_index)(struct amdgpu_device *adev);
+	/* PCIe bandwidth usage */
+	void (*get_pcie_usage)(struct amdgpu_device *adev, uint64_t *count0,
+			       uint64_t *count1);
 };
 
 /*
@@ -1042,6 +1045,7 @@ int emu_soc_asic_init(struct amdgpu_device *adev);
 #define amdgpu_asic_invalidate_hdp(adev, r) (adev)->asic_funcs->invalidate_hdp((adev), (r))
 #define amdgpu_asic_need_full_reset(adev) (adev)->asic_funcs->need_full_reset((adev))
 #define amdgpu_asic_init_doorbell_index(adev) (adev)->asic_funcs->init_doorbell_index((adev))
+#define amdgpu_asic_get_pcie_usage(adev, cnt0, cnt1) ((adev)->asic_funcs->get_pcie_usage((adev), (cnt0), (cnt1)))
 
 /* Common functions */
 bool amdgpu_device_should_recover_gpu(struct amdgpu_device *adev);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_pm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_pm.c
index 6896dec..b38c06f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_pm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_pm.c
@@ -990,6 +990,31 @@ static ssize_t amdgpu_get_busy_percent(struct device *dev,
 	return snprintf(buf, PAGE_SIZE, "%d\n", value);
 }
 
+/**
+ * DOC: pcie_bw
+ *
+ * The amdgpu driver provides a sysfs API for estimating how much data
+ * has been received and sent by the GPU in the last second through PCIe.
+ * The file pcie_bw is used for this.
+ * The Perf counters count the number of received and sent messages and return
+ * those values, as well as the maximum payload size of a PCIe packet (mps).
+ * Note that it is not possible to easily and quickly obtain the size of each
+ * packet transmitted, so we output the max payload size (mps) to allow for
+ * quick estimation of the PCIe bandwidth usage
+ */
+static ssize_t amdgpu_get_pcie_bw(struct device *dev,
+		struct device_attribute *attr,
+		char *buf)
+{
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = ddev->dev_private;
+	uint64_t count0, count1;
+
+	amdgpu_asic_get_pcie_usage(adev, &count0, &count1);
+	return snprintf(buf, PAGE_SIZE,	"%llu %llu %i\n",
+			count0, count1, pcie_get_mps(adev->pdev));
+}
+
 static DEVICE_ATTR(power_dpm_state, S_IRUGO | S_IWUSR, amdgpu_get_dpm_state, amdgpu_set_dpm_state);
 static DEVICE_ATTR(power_dpm_force_performance_level, S_IRUGO | S_IWUSR,
 		   amdgpu_get_dpm_forced_performance_level,
@@ -1025,6 +1050,7 @@ static DEVICE_ATTR(pp_od_clk_voltage, S_IRUGO | S_IWUSR,
 		amdgpu_set_pp_od_clk_voltage);
 static DEVICE_ATTR(gpu_busy_percent, S_IRUGO,
 		amdgpu_get_busy_percent, NULL);
+static DEVICE_ATTR(pcie_bw, S_IRUGO, amdgpu_get_pcie_bw, NULL);
 
 static ssize_t amdgpu_hwmon_show_temp(struct device *dev,
 				      struct device_attribute *attr,
@@ -2108,6 +2134,14 @@ int amdgpu_pm_sysfs_init(struct amdgpu_device *adev)
 				"gpu_busy_level\n");
 		return ret;
 	}
+	/* PCIe Perf counters won't work on APU nodes */
+	if (adev->flags & !AMD_IS_APU) {
+		ret = device_create_file(adev->dev, &dev_attr_pcie_bw);
+		if (ret) {
+			DRM_ERROR("failed to create device file pcie_bw\n");
+			return ret;
+		}
+	}
 	ret = amdgpu_debugfs_pm_init(adev);
 	if (ret) {
 		DRM_ERROR("Failed to register debugfs file for dpm!\n");
@@ -2147,6 +2181,8 @@ void amdgpu_pm_sysfs_fini(struct amdgpu_device *adev)
 		device_remove_file(adev->dev,
 				&dev_attr_pp_od_clk_voltage);
 	device_remove_file(adev->dev, &dev_attr_gpu_busy_percent);
+	if (adev->flags & !AMD_IS_APU)
+		device_remove_file(adev->dev, &dev_attr_pcie_bw);
 }
 
 void amdgpu_pm_compute_clocks(struct amdgpu_device *adev)
diff --git a/drivers/gpu/drm/amd/amdgpu/cik.c b/drivers/gpu/drm/amd/amdgpu/cik.c
index 71c50d8..6277de5 100644
--- a/drivers/gpu/drm/amd/amdgpu/cik.c
+++ b/drivers/gpu/drm/amd/amdgpu/cik.c
@@ -1741,6 +1741,52 @@ static bool cik_need_full_reset(struct amdgpu_device *adev)
 	return true;
 }
 
+static void cik_get_pcie_usage(struct amdgpu_device *adev, uint64_t *count0,
+			       uint64_t *count1)
+{
+	uint32_t perfctr = 0;
+	uint64_t cnt0_of, cnt1_of;
+	int tmp;
+
+	/* This reports 0 on APUs, so return to avoid writing/reading registers
+	 * that may or may not be different from their GPU counterparts
+	 */
+	if (adev->flags & AMD_IS_APU)
+		return;
+
+	/* Set the 2 events that we wish to watch, defined above */
+	/* Reg 40 is # received msgs, Reg 104 is # of posted requests sent */
+	perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK, EVENT0_SEL, 40);
+	perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK, EVENT1_SEL, 104);
+
+	/* Write to enable desired perf counters */
+	WREG32_PCIE(ixPCIE_PERF_CNTL_TXCLK, perfctr);
+	/* Zero out and enable the perf counters
+	 * Write 0x5:
+	 * Bit 0 = Start all counters(1)
+	 * Bit 2 = Global counter reset enable(1)
+	 */
+	WREG32_PCIE(ixPCIE_PERF_COUNT_CNTL, 0x00000005);
+
+	msleep(1000);
+
+	/* Load the shadow and disable the perf counters
+	 * Write 0x2:
+	 * Bit 0 = Stop counters(0)
+	 * Bit 1 = Load the shadow counters(1)
+	 */
+	WREG32_PCIE(ixPCIE_PERF_COUNT_CNTL, 0x00000002);
+
+	/* Read register values to get any >32bit overflow */
+	tmp = RREG32_PCIE(ixPCIE_PERF_CNTL_TXCLK);
+	cnt0_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK, COUNTER0_UPPER);
+	cnt1_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK, COUNTER1_UPPER);
+
+	/* Get the values and add the overflow */
+	*count0 = RREG32_PCIE(ixPCIE_PERF_COUNT0_TXCLK) | (cnt0_of << 32);
+	*count1 = RREG32_PCIE(ixPCIE_PERF_COUNT1_TXCLK) | (cnt1_of << 32);
+}
+
 static const struct amdgpu_asic_funcs cik_asic_funcs =
 {
 	.read_disabled_bios = &cik_read_disabled_bios,
@@ -1756,6 +1802,7 @@ static const struct amdgpu_asic_funcs cik_asic_funcs =
 	.invalidate_hdp = &cik_invalidate_hdp,
 	.need_full_reset = &cik_need_full_reset,
 	.init_doorbell_index = &legacy_doorbell_index_init,
+	.get_pcie_usage = &cik_get_pcie_usage,
 };
 
 static int cik_common_early_init(void *handle)
diff --git a/drivers/gpu/drm/amd/amdgpu/si.c b/drivers/gpu/drm/amd/amdgpu/si.c
index f8408f8..7d2a487 100644
--- a/drivers/gpu/drm/amd/amdgpu/si.c
+++ b/drivers/gpu/drm/amd/amdgpu/si.c
@@ -47,6 +47,7 @@
 #include "dce/dce_6_0_d.h"
 #include "uvd/uvd_4_0_d.h"
 #include "bif/bif_3_0_d.h"
+#include "bif/bif_3_0_sh_mask.h"
 
 static const u32 tahiti_golden_registers[] =
 {
@@ -1323,6 +1324,52 @@ static void si_set_pcie_lanes(struct amdgpu_device *adev, int lanes)
 	WREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL, link_width_cntl);
 }
 
+static void si_get_pcie_usage(struct amdgpu_device *adev, uint64_t *count0,
+			      uint64_t *count1)
+{
+	uint32_t perfctr = 0;
+	uint64_t cnt0_of, cnt1_of;
+	int tmp;
+
+	/* This reports 0 on APUs, so return to avoid writing/reading registers
+	 * that may or may not be different from their GPU counterparts
+	 */
+        if (adev->flags & AMD_IS_APU)
+                return;
+
+	/* Set the 2 events that we wish to watch, defined above */
+	/* Reg 40 is # received msgs, Reg 104 is # of posted requests sent */
+	perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK, EVENT0_SEL, 40);
+	perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK, EVENT1_SEL, 104);
+
+	/* Write to enable desired perf counters */
+	WREG32_PCIE(ixPCIE_PERF_CNTL_TXCLK, perfctr);
+	/* Zero out and enable the perf counters
+	 * Write 0x5:
+	 * Bit 0 = Start all counters(1)
+	 * Bit 2 = Global counter reset enable(1)
+	 */
+	WREG32_PCIE(ixPCIE_PERF_COUNT_CNTL, 0x00000005);
+
+	msleep(1000);
+
+	/* Load the shadow and disable the perf counters
+	 * Write 0x2:
+	 * Bit 0 = Stop counters(0)
+	 * Bit 1 = Load the shadow counters(1)
+	 */
+	WREG32_PCIE(ixPCIE_PERF_COUNT_CNTL, 0x00000002);
+
+	/* Read register values to get any >32bit overflow */
+	tmp = RREG32_PCIE(ixPCIE_PERF_CNTL_TXCLK);
+	cnt0_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK, COUNTER0_UPPER);
+	cnt1_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK, COUNTER1_UPPER);
+
+	/* Get the values and add the overflow */
+	*count0 = RREG32_PCIE(ixPCIE_PERF_COUNT0_TXCLK) | (cnt0_of << 32);
+	*count1 = RREG32_PCIE(ixPCIE_PERF_COUNT1_TXCLK) | (cnt1_of << 32);
+}
+
 static const struct amdgpu_asic_funcs si_asic_funcs =
 {
 	.read_disabled_bios = &si_read_disabled_bios,
@@ -1339,6 +1386,7 @@ static const struct amdgpu_asic_funcs si_asic_funcs =
 	.flush_hdp = &si_flush_hdp,
 	.invalidate_hdp = &si_invalidate_hdp,
 	.need_full_reset = &si_need_full_reset,
+	.get_pcie_usage = &si_get_pcie_usage,
 };
 
 static uint32_t si_get_rev_id(struct amdgpu_device *adev)
diff --git a/drivers/gpu/drm/amd/amdgpu/soc15.c b/drivers/gpu/drm/amd/amdgpu/soc15.c
index 8849b74..bb89833e 100644
--- a/drivers/gpu/drm/amd/amdgpu/soc15.c
+++ b/drivers/gpu/drm/amd/amdgpu/soc15.c
@@ -43,6 +43,9 @@
 #include "hdp/hdp_4_0_sh_mask.h"
 #include "smuio/smuio_9_0_offset.h"
 #include "smuio/smuio_9_0_sh_mask.h"
+#include "nbio/nbio_7_0_default.h"
+#include "nbio/nbio_7_0_sh_mask.h"
+#include "nbio/nbio_7_0_smn.h"
 
 #include "soc15.h"
 #include "soc15_common.h"
@@ -601,6 +604,51 @@ static bool soc15_need_full_reset(struct amdgpu_device *adev)
 	/* change this when we implement soft reset */
 	return true;
 }
+static void soc15_get_pcie_usage(struct amdgpu_device *adev, uint64_t *count0,
+				 uint64_t *count1)
+{
+	uint32_t perfctr = 0;
+	uint64_t cnt0_of, cnt1_of;
+	int tmp;
+
+	/* This reports 0 on APUs, so return to avoid writing/reading registers
+	 * that may or may not be different from their GPU counterparts
+	 */
+	 if (adev->flags & AMD_IS_APU)
+		 return;
+
+	/* Set the 2 events that we wish to watch, defined above */
+	/* Reg 40 is # received msgs, Reg 104 is # of posted requests sent */
+	perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK, EVENT0_SEL, 40);
+	perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK, EVENT1_SEL, 104);
+
+	/* Write to enable desired perf counters */
+	WREG32_PCIE(smnPCIE_PERF_CNTL_TXCLK, perfctr);
+	/* Zero out and enable the perf counters
+	 * Write 0x5:
+	 * Bit 0 = Start all counters(1)
+	 * Bit 2 = Global counter reset enable(1)
+	 */
+	WREG32_PCIE(smnPCIE_PERF_COUNT_CNTL, 0x00000005);
+
+	msleep(1000);
+
+	/* Load the shadow and disable the perf counters
+	 * Write 0x2:
+	 * Bit 0 = Stop counters(0)
+	 * Bit 1 = Load the shadow counters(1)
+	 */
+	WREG32_PCIE(smnPCIE_PERF_COUNT_CNTL, 0x00000002);
+
+	/* Read register values to get any >32bit overflow */
+	tmp = RREG32_PCIE(smnPCIE_PERF_CNTL_TXCLK);
+	cnt0_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK, COUNTER0_UPPER);
+	cnt1_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK, COUNTER1_UPPER);
+
+	/* Get the values and add the overflow */
+	*count0 = RREG32_PCIE(smnPCIE_PERF_COUNT0_TXCLK) | (cnt0_of << 32);
+	*count1 = RREG32_PCIE(smnPCIE_PERF_COUNT1_TXCLK) | (cnt1_of << 32);
+}
 
 static const struct amdgpu_asic_funcs soc15_asic_funcs =
 {
@@ -617,6 +665,7 @@ static const struct amdgpu_asic_funcs soc15_asic_funcs =
 	.invalidate_hdp = &soc15_invalidate_hdp,
 	.need_full_reset = &soc15_need_full_reset,
 	.init_doorbell_index = &vega10_doorbell_index_init,
+	.get_pcie_usage = &soc15_get_pcie_usage,
 };
 
 static const struct amdgpu_asic_funcs vega20_asic_funcs =
@@ -634,6 +683,7 @@ static const struct amdgpu_asic_funcs vega20_asic_funcs =
 	.invalidate_hdp = &soc15_invalidate_hdp,
 	.need_full_reset = &soc15_need_full_reset,
 	.init_doorbell_index = &vega20_doorbell_index_init,
+	.get_pcie_usage = &soc15_get_pcie_usage,
 };
 
 static int soc15_common_early_init(void *handle)
diff --git a/drivers/gpu/drm/amd/amdgpu/vi.c b/drivers/gpu/drm/amd/amdgpu/vi.c
index 03e7be5..cdc8ab8 100644
--- a/drivers/gpu/drm/amd/amdgpu/vi.c
+++ b/drivers/gpu/drm/amd/amdgpu/vi.c
@@ -941,6 +941,52 @@ static bool vi_need_full_reset(struct amdgpu_device *adev)
 	}
 }
 
+static void vi_get_pcie_usage(struct amdgpu_device *adev, uint64_t *count0,
+			      uint64_t *count1)
+{
+	uint32_t perfctr = 0;
+	uint64_t cnt0_of, cnt1_of;
+	int tmp;
+
+	/* This reports 0 on APUs, so return to avoid writing/reading registers
+	 * that may or may not be different from their GPU counterparts
+	 */
+	if (adev->flags & AMD_IS_APU)
+		return;
+
+	/* Set the 2 events that we wish to watch, defined above */
+	/* Reg 40 is # received msgs, Reg 104 is # of posted requests sent */
+	perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK, EVENT0_SEL, 40);
+	perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK, EVENT1_SEL, 104);
+
+	/* Write to enable desired perf counters */
+	WREG32_PCIE(ixPCIE_PERF_CNTL_TXCLK, perfctr);
+	/* Zero out and enable the perf counters
+	 * Write 0x5:
+	 * Bit 0 = Start all counters(1)
+	 * Bit 2 = Global counter reset enable(1)
+	 */
+	WREG32_PCIE(ixPCIE_PERF_COUNT_CNTL, 0x00000005);
+
+	msleep(1000);
+
+	/* Load the shadow and disable the perf counters
+	 * Write 0x2:
+	 * Bit 0 = Stop counters(0)
+	 * Bit 1 = Load the shadow counters(1)
+	 */
+	WREG32_PCIE(ixPCIE_PERF_COUNT_CNTL, 0x00000002);
+
+	/* Read register values to get any >32bit overflow */
+	tmp = RREG32_PCIE(ixPCIE_PERF_CNTL_TXCLK);
+	cnt0_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK, COUNTER0_UPPER);
+	cnt1_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK, COUNTER1_UPPER);
+
+	/* Get the values and add the overflow */
+	*count0 = RREG32_PCIE(ixPCIE_PERF_COUNT0_TXCLK) | (cnt0_of << 32);
+	*count1 = RREG32_PCIE(ixPCIE_PERF_COUNT1_TXCLK) | (cnt1_of << 32);
+}
+
 static const struct amdgpu_asic_funcs vi_asic_funcs =
 {
 	.read_disabled_bios = &vi_read_disabled_bios,
@@ -956,6 +1002,7 @@ static const struct amdgpu_asic_funcs vi_asic_funcs =
 	.invalidate_hdp = &vi_invalidate_hdp,
 	.need_full_reset = &vi_need_full_reset,
 	.init_doorbell_index = &legacy_doorbell_index_init,
+	.get_pcie_usage = &vi_get_pcie_usage,
 };
 
 #define CZ_REV_BRISTOL(rev)	 \