15 files changed, 480 insertions, 50 deletions

diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/Kbuild b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/Kbuild
index e37b6e45eaa2..a76c2a7bd696 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/Kbuild
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/Kbuild
@@ -12,3 +12,4 @@ nvkm-y += nvkm/subdev/pmu/gm107.o
 nvkm-y += nvkm/subdev/pmu/gm20b.o
 nvkm-y += nvkm/subdev/pmu/gp100.o
 nvkm-y += nvkm/subdev/pmu/gp102.o
+nvkm-y += nvkm/subdev/pmu/gp10b.o
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/base.c
index ea2e11771bca..a0fe607c9c07 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/base.c
@@ -23,7 +23,7 @@
  */
 #include "priv.h"
 
-#include <core/msgqueue.h>
+#include <core/firmware.h>
 #include <subdev/timer.h>
 
 bool
@@ -85,6 +85,12 @@ nvkm_pmu_fini(struct nvkm_subdev *subdev, bool suspend)
 		pmu->func->fini(pmu);
 
 	flush_work(&pmu->recv.work);
+
+	reinit_completion(&pmu->wpr_ready);
+
+	nvkm_falcon_cmdq_fini(pmu->lpq);
+	nvkm_falcon_cmdq_fini(pmu->hpq);
+	pmu->initmsg_received = false;
 	return 0;
 }
 
@@ -133,19 +139,15 @@ nvkm_pmu_init(struct nvkm_subdev *subdev)
 	return ret;
 }
 
-static int
-nvkm_pmu_oneinit(struct nvkm_subdev *subdev)
-{
-	struct nvkm_pmu *pmu = nvkm_pmu(subdev);
-	return nvkm_falcon_v1_new(&pmu->subdev, "PMU", 0x10a000, &pmu->falcon);
-}
-
 static void *
 nvkm_pmu_dtor(struct nvkm_subdev *subdev)
 {
 	struct nvkm_pmu *pmu = nvkm_pmu(subdev);
-	nvkm_msgqueue_del(&pmu->queue);
-	nvkm_falcon_del(&pmu->falcon);
+	nvkm_falcon_msgq_del(&pmu->msgq);
+	nvkm_falcon_cmdq_del(&pmu->lpq);
+	nvkm_falcon_cmdq_del(&pmu->hpq);
+	nvkm_falcon_qmgr_del(&pmu->qmgr);
+	nvkm_falcon_dtor(&pmu->falcon);
 	return nvkm_pmu(subdev);
 }
 
@@ -153,29 +155,50 @@ static const struct nvkm_subdev_func
 nvkm_pmu = {
 	.dtor = nvkm_pmu_dtor,
 	.preinit = nvkm_pmu_preinit,
-	.oneinit = nvkm_pmu_oneinit,
 	.init = nvkm_pmu_init,
 	.fini = nvkm_pmu_fini,
 	.intr = nvkm_pmu_intr,
 };
 
 int
-nvkm_pmu_ctor(const struct nvkm_pmu_func *func, struct nvkm_device *device,
+nvkm_pmu_ctor(const struct nvkm_pmu_fwif *fwif, struct nvkm_device *device,
 	      int index, struct nvkm_pmu *pmu)
 {
+	int ret;
+
 	nvkm_subdev_ctor(&nvkm_pmu, device, index, &pmu->subdev);
-	pmu->func = func;
+
 	INIT_WORK(&pmu->recv.work, nvkm_pmu_recv);
 	init_waitqueue_head(&pmu->recv.wait);
+
+	fwif = nvkm_firmware_load(&pmu->subdev, fwif, "Pmu", pmu);
+	if (IS_ERR(fwif))
+		return PTR_ERR(fwif);
+
+	pmu->func = fwif->func;
+
+	ret = nvkm_falcon_ctor(pmu->func->flcn, &pmu->subdev,
+			       nvkm_subdev_name[pmu->subdev.index], 0x10a000,
+			       &pmu->falcon);
+	if (ret)
+		return ret;
+
+	if ((ret = nvkm_falcon_qmgr_new(&pmu->falcon, &pmu->qmgr)) ||
+	    (ret = nvkm_falcon_cmdq_new(pmu->qmgr, "hpq", &pmu->hpq)) ||
+	    (ret = nvkm_falcon_cmdq_new(pmu->qmgr, "lpq", &pmu->lpq)) ||
+	    (ret = nvkm_falcon_msgq_new(pmu->qmgr, "msgq", &pmu->msgq)))
+		return ret;
+
+	init_completion(&pmu->wpr_ready);
 	return 0;
 }
 
 int
-nvkm_pmu_new_(const struct nvkm_pmu_func *func, struct nvkm_device *device,
+nvkm_pmu_new_(const struct nvkm_pmu_fwif *fwif, struct nvkm_device *device,
 	      int index, struct nvkm_pmu **ppmu)
 {
 	struct nvkm_pmu *pmu;
 	if (!(pmu = *ppmu = kzalloc(sizeof(*pmu), GFP_KERNEL)))
 		return -ENOMEM;
-	return nvkm_pmu_ctor(func, device, index, *ppmu);
+	return nvkm_pmu_ctor(fwif, device, index, *ppmu);
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gf100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gf100.c
index 0b458656e870..3ecb3d9cbcf2 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gf100.c
@@ -42,6 +42,7 @@ gf100_pmu_enabled(struct nvkm_pmu *pmu)
 
 static const struct nvkm_pmu_func
 gf100_pmu = {
+	.flcn = &gt215_pmu_flcn,
 	.code.data = gf100_pmu_code,
 	.code.size = sizeof(gf100_pmu_code),
 	.data.data = gf100_pmu_data,
@@ -56,7 +57,19 @@ gf100_pmu = {
 };
 
 int
+gf100_pmu_nofw(struct nvkm_pmu *pmu, int ver, const struct nvkm_pmu_fwif *fwif)
+{
+	return 0;
+}
+
+static const struct nvkm_pmu_fwif
+gf100_pmu_fwif[] = {
+	{ -1, gf100_pmu_nofw, &gf100_pmu },
+	{}
+};
+
+int
 gf100_pmu_new(struct nvkm_device *device, int index, struct nvkm_pmu **ppmu)
 {
-	return nvkm_pmu_new_(&gf100_pmu, device, index, ppmu);
+	return nvkm_pmu_new_(gf100_pmu_fwif, device, index, ppmu);
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gf119.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gf119.c
index 3dfa79d4fb13..8dd0271aaaee 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gf119.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gf119.c
@@ -26,6 +26,7 @@
 
 static const struct nvkm_pmu_func
 gf119_pmu = {
+	.flcn = &gt215_pmu_flcn,
 	.code.data = gf119_pmu_code,
 	.code.size = sizeof(gf119_pmu_code),
 	.data.data = gf119_pmu_data,
@@ -39,8 +40,14 @@ gf119_pmu = {
 	.recv = gt215_pmu_recv,
 };
 
+static const struct nvkm_pmu_fwif
+gf119_pmu_fwif[] = {
+	{ -1, gf100_pmu_nofw, &gf119_pmu },
+	{}
+};
+
 int
 gf119_pmu_new(struct nvkm_device *device, int index, struct nvkm_pmu **ppmu)
 {
-	return nvkm_pmu_new_(&gf119_pmu, device, index, ppmu);
+	return nvkm_pmu_new_(gf119_pmu_fwif, device, index, ppmu);
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gk104.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gk104.c
index 8f7ec10fd2a4..8b70cc17a634 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gk104.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gk104.c
@@ -105,6 +105,7 @@ gk104_pmu_pgob(struct nvkm_pmu *pmu, bool enable)
 
 static const struct nvkm_pmu_func
 gk104_pmu = {
+	.flcn = &gt215_pmu_flcn,
 	.code.data = gk104_pmu_code,
 	.code.size = sizeof(gk104_pmu_code),
 	.data.data = gk104_pmu_data,
@@ -119,8 +120,14 @@ gk104_pmu = {
 	.pgob = gk104_pmu_pgob,
 };
 
+static const struct nvkm_pmu_fwif
+gk104_pmu_fwif[] = {
+	{ -1, gf100_pmu_nofw, &gk104_pmu },
+	{}
+};
+
 int
 gk104_pmu_new(struct nvkm_device *device, int index, struct nvkm_pmu **ppmu)
 {
-	return nvkm_pmu_new_(&gk104_pmu, device, index, ppmu);
+	return nvkm_pmu_new_(gk104_pmu_fwif, device, index, ppmu);
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gk110.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gk110.c
index 345741d55a56..0081f2141b10 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gk110.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gk110.c
@@ -84,6 +84,7 @@ gk110_pmu_pgob(struct nvkm_pmu *pmu, bool enable)
 
 static const struct nvkm_pmu_func
 gk110_pmu = {
+	.flcn = &gt215_pmu_flcn,
 	.code.data = gk110_pmu_code,
 	.code.size = sizeof(gk110_pmu_code),
 	.data.data = gk110_pmu_data,
@@ -98,8 +99,14 @@ gk110_pmu = {
 	.pgob = gk110_pmu_pgob,
 };
 
+static const struct nvkm_pmu_fwif
+gk110_pmu_fwif[] = {
+	{ -1, gf100_pmu_nofw, &gk110_pmu },
+	{}
+};
+
 int
 gk110_pmu_new(struct nvkm_device *device, int index, struct nvkm_pmu **ppmu)
 {
-	return nvkm_pmu_new_(&gk110_pmu, device, index, ppmu);
+	return nvkm_pmu_new_(gk110_pmu_fwif, device, index, ppmu);
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gk208.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gk208.c
index e4acf7876ea1..b227c701a5e7 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gk208.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gk208.c
@@ -26,6 +26,7 @@
 
 static const struct nvkm_pmu_func
 gk208_pmu = {
+	.flcn = &gt215_pmu_flcn,
 	.code.data = gk208_pmu_code,
 	.code.size = sizeof(gk208_pmu_code),
 	.data.data = gk208_pmu_data,
@@ -40,8 +41,14 @@ gk208_pmu = {
 	.pgob = gk110_pmu_pgob,
 };
 
+static const struct nvkm_pmu_fwif
+gk208_pmu_fwif[] = {
+	{ -1, gf100_pmu_nofw, &gk208_pmu },
+	{}
+};
+
 int
 gk208_pmu_new(struct nvkm_device *device, int index, struct nvkm_pmu **ppmu)
 {
-	return nvkm_pmu_new_(&gk208_pmu, device, index, ppmu);
+	return nvkm_pmu_new_(gk208_pmu_fwif, device, index, ppmu);
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gk20a.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gk20a.c
index 05e81855c367..26c1adf8f44c 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gk20a.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gk20a.c
@@ -95,7 +95,7 @@ static void
 gk20a_pmu_dvfs_get_dev_status(struct gk20a_pmu *pmu,
 			      struct gk20a_pmu_dvfs_dev_status *status)
 {
-	struct nvkm_falcon *falcon = pmu->base.falcon;
+	struct nvkm_falcon *falcon = &pmu->base.falcon;
 
 	status->busy = nvkm_falcon_rd32(falcon, 0x508 + (BUSY_SLOT * 0x10));
 	status->total= nvkm_falcon_rd32(falcon, 0x508 + (CLK_SLOT * 0x10));
@@ -104,7 +104,7 @@ gk20a_pmu_dvfs_get_dev_status(struct gk20a_pmu *pmu,
 static void
 gk20a_pmu_dvfs_reset_dev_status(struct gk20a_pmu *pmu)
 {
-	struct nvkm_falcon *falcon = pmu->base.falcon;
+	struct nvkm_falcon *falcon = &pmu->base.falcon;
 
 	nvkm_falcon_wr32(falcon, 0x508 + (BUSY_SLOT * 0x10), 0x80000000);
 	nvkm_falcon_wr32(falcon, 0x508 + (CLK_SLOT * 0x10), 0x80000000);
@@ -160,7 +160,7 @@ gk20a_pmu_fini(struct nvkm_pmu *pmu)
 	struct gk20a_pmu *gpmu = gk20a_pmu(pmu);
 	nvkm_timer_alarm(pmu->subdev.device->timer, 0, &gpmu->alarm);
 
-	nvkm_falcon_put(pmu->falcon, &pmu->subdev);
+	nvkm_falcon_put(&pmu->falcon, &pmu->subdev);
 }
 
 static int
@@ -169,7 +169,7 @@ gk20a_pmu_init(struct nvkm_pmu *pmu)
 	struct gk20a_pmu *gpmu = gk20a_pmu(pmu);
 	struct nvkm_subdev *subdev = &pmu->subdev;
 	struct nvkm_device *device = pmu->subdev.device;
-	struct nvkm_falcon *falcon = pmu->falcon;
+	struct nvkm_falcon *falcon = &pmu->falcon;
 	int ret;
 
 	ret = nvkm_falcon_get(falcon, subdev);
@@ -196,25 +196,34 @@ gk20a_dvfs_data= {
 
 static const struct nvkm_pmu_func
 gk20a_pmu = {
+	.flcn = &gt215_pmu_flcn,
 	.enabled = gf100_pmu_enabled,
 	.init = gk20a_pmu_init,
 	.fini = gk20a_pmu_fini,
 	.reset = gf100_pmu_reset,
 };
 
+static const struct nvkm_pmu_fwif
+gk20a_pmu_fwif[] = {
+	{ -1, gf100_pmu_nofw, &gk20a_pmu },
+	{}
+};
+
 int
 gk20a_pmu_new(struct nvkm_device *device, int index, struct nvkm_pmu **ppmu)
 {
 	struct gk20a_pmu *pmu;
+	int ret;
 
 	if (!(pmu = kzalloc(sizeof(*pmu), GFP_KERNEL)))
 		return -ENOMEM;
 	*ppmu = &pmu->base;
 
-	nvkm_pmu_ctor(&gk20a_pmu, device, index, &pmu->base);
+	ret = nvkm_pmu_ctor(gk20a_pmu_fwif, device, index, &pmu->base);
+	if (ret)
+		return ret;
 
 	pmu->data = &gk20a_dvfs_data;
 	nvkm_alarm_init(&pmu->alarm, gk20a_pmu_dvfs_work);
-
 	return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gm107.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gm107.c
index 459df1ef9e70..5afb55e58b51 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gm107.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gm107.c
@@ -28,6 +28,7 @@
 
 static const struct nvkm_pmu_func
 gm107_pmu = {
+	.flcn = &gt215_pmu_flcn,
 	.code.data = gm107_pmu_code,
 	.code.size = sizeof(gm107_pmu_code),
 	.data.data = gm107_pmu_data,
@@ -41,8 +42,14 @@ gm107_pmu = {
 	.recv = gt215_pmu_recv,
 };
 
+static const struct nvkm_pmu_fwif
+gm107_pmu_fwif[] = {
+	{ -1, gf100_pmu_nofw, &gm107_pmu },
+	{}
+};
+
 int
 gm107_pmu_new(struct nvkm_device *device, int index, struct nvkm_pmu **ppmu)
 {
-	return nvkm_pmu_new_(&gm107_pmu, device, index, ppmu);
+	return nvkm_pmu_new_(gm107_pmu_fwif, device, index, ppmu);
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gm20b.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gm20b.c
index 31c843145c7a..82571032a07d 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gm20b.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gm20b.c
@@ -19,38 +19,224 @@
  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  * DEALINGS IN THE SOFTWARE.
  */
-
-#include <engine/falcon.h>
-#include <core/msgqueue.h>
 #include "priv.h"
 
-static void
+#include <core/memory.h>
+#include <subdev/acr.h>
+
+#include <nvfw/flcn.h>
+#include <nvfw/pmu.h>
+
+static int
+gm20b_pmu_acr_bootstrap_falcon_cb(void *priv, struct nv_falcon_msg *hdr)
+{
+	struct nv_pmu_acr_bootstrap_falcon_msg *msg =
+		container_of(hdr, typeof(*msg), msg.hdr);
+	return msg->falcon_id;
+}
+
+int
+gm20b_pmu_acr_bootstrap_falcon(struct nvkm_falcon *falcon,
+			       enum nvkm_acr_lsf_id id)
+{
+	struct nvkm_pmu *pmu = container_of(falcon, typeof(*pmu), falcon);
+	struct nv_pmu_acr_bootstrap_falcon_cmd cmd = {
+		.cmd.hdr.unit_id = NV_PMU_UNIT_ACR,
+		.cmd.hdr.size = sizeof(cmd),
+		.cmd.cmd_type = NV_PMU_ACR_CMD_BOOTSTRAP_FALCON,
+		.flags = NV_PMU_ACR_BOOTSTRAP_FALCON_FLAGS_RESET_YES,
+		.falcon_id = id,
+	};
+	int ret;
+
+	ret = nvkm_falcon_cmdq_send(pmu->hpq, &cmd.cmd.hdr,
+				    gm20b_pmu_acr_bootstrap_falcon_cb,
+				    &pmu->subdev, msecs_to_jiffies(1000));
+	if (ret >= 0) {
+		if (ret != cmd.falcon_id)
+			ret = -EIO;
+		else
+			ret = 0;
+	}
+
+	return ret;
+}
+
+int
+gm20b_pmu_acr_boot(struct nvkm_falcon *falcon)
+{
+	struct nv_pmu_args args = { .secure_mode = true };
+	const u32 addr_args = falcon->data.limit - sizeof(struct nv_pmu_args);
+	nvkm_falcon_load_dmem(falcon, &args, addr_args, sizeof(args), 0);
+	nvkm_falcon_start(falcon);
+	return 0;
+}
+
+void
+gm20b_pmu_acr_bld_patch(struct nvkm_acr *acr, u32 bld, s64 adjust)
+{
+	struct loader_config hdr;
+	u64 addr;
+
+	nvkm_robj(acr->wpr, bld, &hdr, sizeof(hdr));
+	addr = ((u64)hdr.code_dma_base1 << 40 | hdr.code_dma_base << 8);
+	hdr.code_dma_base  = lower_32_bits((addr + adjust) >> 8);
+	hdr.code_dma_base1 = upper_32_bits((addr + adjust) >> 8);
+	addr = ((u64)hdr.data_dma_base1 << 40 | hdr.data_dma_base << 8);
+	hdr.data_dma_base  = lower_32_bits((addr + adjust) >> 8);
+	hdr.data_dma_base1 = upper_32_bits((addr + adjust) >> 8);
+	addr = ((u64)hdr.overlay_dma_base1 << 40 | hdr.overlay_dma_base << 8);
+	hdr.overlay_dma_base  = lower_32_bits((addr + adjust) << 8);
+	hdr.overlay_dma_base1 = upper_32_bits((addr + adjust) << 8);
+	nvkm_wobj(acr->wpr, bld, &hdr, sizeof(hdr));
+
+	loader_config_dump(&acr->subdev, &hdr);
+}
+
+void
+gm20b_pmu_acr_bld_write(struct nvkm_acr *acr, u32 bld,
+			struct nvkm_acr_lsfw *lsfw)
+{
+	const u64 base = lsfw->offset.img + lsfw->app_start_offset;
+	const u64 code = (base + lsfw->app_resident_code_offset) >> 8;
+	const u64 data = (base + lsfw->app_resident_data_offset) >> 8;
+	const struct loader_config hdr = {
+		.dma_idx = FALCON_DMAIDX_UCODE,
+		.code_dma_base = lower_32_bits(code),
+		.code_size_total = lsfw->app_size,
+		.code_size_to_load = lsfw->app_resident_code_size,
+		.code_entry_point = lsfw->app_imem_entry,
+		.data_dma_base = lower_32_bits(data),
+		.data_size = lsfw->app_resident_data_size,
+		.overlay_dma_base = lower_32_bits(code),
+		.argc = 1,
+		.argv = lsfw->falcon->data.limit - sizeof(struct nv_pmu_args),
+		.code_dma_base1 = upper_32_bits(code),
+		.data_dma_base1 = upper_32_bits(data),
+		.overlay_dma_base1 = upper_32_bits(code),
+	};
+
+	nvkm_wobj(acr->wpr, bld, &hdr, sizeof(hdr));
+}
+
+static const struct nvkm_acr_lsf_func
+gm20b_pmu_acr = {
+	.flags = NVKM_ACR_LSF_DMACTL_REQ_CTX,
+	.bld_size = sizeof(struct loader_config),
+	.bld_write = gm20b_pmu_acr_bld_write,
+	.bld_patch = gm20b_pmu_acr_bld_patch,
+	.boot = gm20b_pmu_acr_boot,
+	.bootstrap_falcon = gm20b_pmu_acr_bootstrap_falcon,
+};
+
+static int
+gm20b_pmu_acr_init_wpr_callback(void *priv, struct nv_falcon_msg *hdr)
+{
+	struct nv_pmu_acr_init_wpr_region_msg *msg =
+		container_of(hdr, typeof(*msg), msg.hdr);
+	struct nvkm_pmu *pmu = priv;
+	struct nvkm_subdev *subdev = &pmu->subdev;
+
+	if (msg->error_code) {
+		nvkm_error(subdev, "ACR WPR init failure: %d\n",
+			   msg->error_code);
+		return -EINVAL;
+	}
+
+	nvkm_debug(subdev, "ACR WPR init complete\n");
+	complete_all(&pmu->wpr_ready);
+	return 0;
+}
+
+static int
+gm20b_pmu_acr_init_wpr(struct nvkm_pmu *pmu)
+{
+	struct nv_pmu_acr_init_wpr_region_cmd cmd = {
+		.cmd.hdr.unit_id = NV_PMU_UNIT_ACR,
+		.cmd.hdr.size = sizeof(cmd),
+		.cmd.cmd_type = NV_PMU_ACR_CMD_INIT_WPR_REGION,
+		.region_id = 1,
+		.wpr_offset = 0,
+	};
+
+	return nvkm_falcon_cmdq_send(pmu->hpq, &cmd.cmd.hdr,
+				     gm20b_pmu_acr_init_wpr_callback, pmu, 0);
+}
+
+int
+gm20b_pmu_initmsg(struct nvkm_pmu *pmu)
+{
+	struct nv_pmu_init_msg msg;
+	int ret;
+
+	ret = nvkm_falcon_msgq_recv_initmsg(pmu->msgq, &msg, sizeof(msg));
+	if (ret)
+		return ret;
+
+	if (msg.hdr.unit_id != NV_PMU_UNIT_INIT ||
+	    msg.msg_type != NV_PMU_INIT_MSG_INIT)
+		return -EINVAL;
+
+	nvkm_falcon_cmdq_init(pmu->hpq, msg.queue_info[0].index,
+					msg.queue_info[0].offset,
+					msg.queue_info[0].size);
+	nvkm_falcon_cmdq_init(pmu->lpq, msg.queue_info[1].index,
+					msg.queue_info[1].offset,
+					msg.queue_info[1].size);
+	nvkm_falcon_msgq_init(pmu->msgq, msg.queue_info[4].index,
+					 msg.queue_info[4].offset,
+					 msg.queue_info[4].size);
+	return gm20b_pmu_acr_init_wpr(pmu);
+}
+
+void
 gm20b_pmu_recv(struct nvkm_pmu *pmu)
 {
-	if (!pmu->queue) {
-		nvkm_warn(&pmu->subdev,
-			  "recv function called while no firmware set!\n");
-		return;
+	if (!pmu->initmsg_received) {
+		int ret = pmu->func->initmsg(pmu);
+		if (ret) {
+			nvkm_error(&pmu->subdev,
+				   "error parsing init message: %d\n", ret);
+			return;
+		}
+
+		pmu->initmsg_received = true;
 	}
 
-	nvkm_msgqueue_recv(pmu->queue);
+	nvkm_falcon_msgq_recv(pmu->msgq);
 }
 
 static const struct nvkm_pmu_func
 gm20b_pmu = {
+	.flcn = &gt215_pmu_flcn,
 	.enabled = gf100_pmu_enabled,
 	.intr = gt215_pmu_intr,
 	.recv = gm20b_pmu_recv,
+	.initmsg = gm20b_pmu_initmsg,
 };
 
+#if IS_ENABLED(CONFIG_ARCH_TEGRA_210_SOC)
+MODULE_FIRMWARE("nvidia/gm20b/pmu/desc.bin");
+MODULE_FIRMWARE("nvidia/gm20b/pmu/image.bin");
+MODULE_FIRMWARE("nvidia/gm20b/pmu/sig.bin");
+#endif
+
 int
-gm20b_pmu_new(struct nvkm_device *device, int index, struct nvkm_pmu **ppmu)
+gm20b_pmu_load(struct nvkm_pmu *pmu, int ver, const struct nvkm_pmu_fwif *fwif)
 {
-	int ret;
+	return nvkm_acr_lsfw_load_sig_image_desc(&pmu->subdev, &pmu->falcon,
+						 NVKM_ACR_LSF_PMU, "pmu/",
+						 ver, fwif->acr);
+}
 
-	ret = nvkm_pmu_new_(&gm20b_pmu, device, index, ppmu);
-	if (ret)
-		return ret;
+static const struct nvkm_pmu_fwif
+gm20b_pmu_fwif[] = {
+	{ 0, gm20b_pmu_load, &gm20b_pmu, &gm20b_pmu_acr },
+	{}
+};
 
-	return 0;
+int
+gm20b_pmu_new(struct nvkm_device *device, int index, struct nvkm_pmu **ppmu)
+{
+	return nvkm_pmu_new_(gm20b_pmu_fwif, device, index, ppmu);
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gp100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gp100.c
index e210cd6af816..09e05db21ff5 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gp100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gp100.c
@@ -25,12 +25,19 @@
 
 static const struct nvkm_pmu_func
 gp100_pmu = {
+	.flcn = &gt215_pmu_flcn,
 	.enabled = gf100_pmu_enabled,
 	.reset = gf100_pmu_reset,
 };
 
+static const struct nvkm_pmu_fwif
+gp100_pmu_fwif[] = {
+	{ -1, gf100_pmu_nofw, &gp100_pmu },
+	{}
+};
+
 int
 gp100_pmu_new(struct nvkm_device *device, int index, struct nvkm_pmu **ppmu)
 {
-	return nvkm_pmu_new_(&gp100_pmu, device, index, ppmu);
+	return nvkm_pmu_new_(gp100_pmu_fwif, device, index, ppmu);
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gp102.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gp102.c
index 98c7a2a8afc4..262b8a3dd507 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gp102.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gp102.c
@@ -39,12 +39,19 @@ gp102_pmu_enabled(struct nvkm_pmu *pmu)
 
 static const struct nvkm_pmu_func
 gp102_pmu = {
+	.flcn = &gt215_pmu_flcn,
 	.enabled = gp102_pmu_enabled,
 	.reset = gp102_pmu_reset,
 };
 
+static const struct nvkm_pmu_fwif
+gp102_pmu_fwif[] = {
+	{ -1, gf100_pmu_nofw, &gp102_pmu },
+	{}
+};
+
 int
 gp102_pmu_new(struct nvkm_device *device, int index, struct nvkm_pmu **ppmu)
 {
-	return nvkm_pmu_new_(&gp102_pmu, device, index, ppmu);
+	return nvkm_pmu_new_(gp102_pmu_fwif, device, index, ppmu);
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gp10b.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gp10b.c
new file mode 100644
index 000000000000..5b81c7320479
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gp10b.c
@@ -0,0 +1,101 @@
+/*
+ * Copyright (c) 2016, NVIDIA CORPORATION. All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+#include "priv.h"
+
+#include <subdev/acr.h>
+
+#include <nvfw/flcn.h>
+#include <nvfw/pmu.h>
+
+static int
+gp10b_pmu_acr_bootstrap_multiple_falcons_cb(void *priv,
+					    struct nv_falcon_msg *hdr)
+{
+	struct nv_pmu_acr_bootstrap_multiple_falcons_msg *msg =
+		container_of(hdr, typeof(*msg), msg.hdr);
+	return msg->falcon_mask;
+}
+static int
+gp10b_pmu_acr_bootstrap_multiple_falcons(struct nvkm_falcon *falcon, u32 mask)
+{
+	struct nvkm_pmu *pmu = container_of(falcon, typeof(*pmu), falcon);
+	struct nv_pmu_acr_bootstrap_multiple_falcons_cmd cmd = {
+		.cmd.hdr.unit_id = NV_PMU_UNIT_ACR,
+		.cmd.hdr.size = sizeof(cmd),
+		.cmd.cmd_type = NV_PMU_ACR_CMD_BOOTSTRAP_MULTIPLE_FALCONS,
+		.flags = NV_PMU_ACR_BOOTSTRAP_MULTIPLE_FALCONS_FLAGS_RESET_YES,
+		.falcon_mask = mask,
+		.wpr_lo = 0, /*XXX*/
+		.wpr_hi = 0, /*XXX*/
+	};
+	int ret;
+
+	ret = nvkm_falcon_cmdq_send(pmu->hpq, &cmd.cmd.hdr,
+				    gp10b_pmu_acr_bootstrap_multiple_falcons_cb,
+				    &pmu->subdev, msecs_to_jiffies(1000));
+	if (ret >= 0) {
+		if (ret != cmd.falcon_mask)
+			ret = -EIO;
+		else
+			ret = 0;
+	}
+
+	return ret;
+}
+
+static const struct nvkm_acr_lsf_func
+gp10b_pmu_acr = {
+	.flags = NVKM_ACR_LSF_DMACTL_REQ_CTX,
+	.bld_size = sizeof(struct loader_config),
+	.bld_write = gm20b_pmu_acr_bld_write,
+	.bld_patch = gm20b_pmu_acr_bld_patch,
+	.boot = gm20b_pmu_acr_boot,
+	.bootstrap_falcon = gm20b_pmu_acr_bootstrap_falcon,
+	.bootstrap_multiple_falcons = gp10b_pmu_acr_bootstrap_multiple_falcons,
+};
+
+static const struct nvkm_pmu_func
+gp10b_pmu = {
+	.flcn = &gt215_pmu_flcn,
+	.enabled = gf100_pmu_enabled,
+	.intr = gt215_pmu_intr,
+	.recv = gm20b_pmu_recv,
+	.initmsg = gm20b_pmu_initmsg,
+};
+
+#if IS_ENABLED(CONFIG_ARCH_TEGRA_210_SOC)
+MODULE_FIRMWARE("nvidia/gp10b/pmu/desc.bin");
+MODULE_FIRMWARE("nvidia/gp10b/pmu/image.bin");
+MODULE_FIRMWARE("nvidia/gp10b/pmu/sig.bin");
+#endif
+
+static const struct nvkm_pmu_fwif
+gp10b_pmu_fwif[] = {
+	{ 0, gm20b_pmu_load, &gp10b_pmu, &gp10b_pmu_acr },
+	{}
+};
+
+int
+gp10b_pmu_new(struct nvkm_device *device, int index, struct nvkm_pmu **ppmu)
+{
+	return nvkm_pmu_new_(gp10b_pmu_fwif, device, index, ppmu);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gt215.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gt215.c
index e04216daea58..88b909913ff9 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gt215.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gt215.c
@@ -241,8 +241,27 @@ gt215_pmu_init(struct nvkm_pmu *pmu)
 	return 0;
 }
 
+const struct nvkm_falcon_func
+gt215_pmu_flcn = {
+	.debug = 0xc08,
+	.fbif = 0xe00,
+	.load_imem = nvkm_falcon_v1_load_imem,
+	.load_dmem = nvkm_falcon_v1_load_dmem,
+	.read_dmem = nvkm_falcon_v1_read_dmem,
+	.bind_context = nvkm_falcon_v1_bind_context,
+	.wait_for_halt = nvkm_falcon_v1_wait_for_halt,
+	.clear_interrupt = nvkm_falcon_v1_clear_interrupt,
+	.set_start_addr = nvkm_falcon_v1_set_start_addr,
+	.start = nvkm_falcon_v1_start,
+	.enable = nvkm_falcon_v1_enable,
+	.disable = nvkm_falcon_v1_disable,
+	.cmdq = { 0x4a0, 0x4b0, 4 },
+	.msgq = { 0x4c8, 0x4cc, 0 },
+};
+
 static const struct nvkm_pmu_func
 gt215_pmu = {
+	.flcn = &gt215_pmu_flcn,
 	.code.data = gt215_pmu_code,
 	.code.size = sizeof(gt215_pmu_code),
 	.data.data = gt215_pmu_data,
@@ -256,8 +275,14 @@ gt215_pmu = {
 	.recv = gt215_pmu_recv,
 };
 
+static const struct nvkm_pmu_fwif
+gt215_pmu_fwif[] = {
+	{ -1, gf100_pmu_nofw, &gt215_pmu },
+	{}
+};
+
 int
 gt215_pmu_new(struct nvkm_device *device, int index, struct nvkm_pmu **ppmu)
 {
-	return nvkm_pmu_new_(&gt215_pmu, device, index, ppmu);
+	return nvkm_pmu_new_(gt215_pmu_fwif, device, index, ppmu);
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/priv.h b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/priv.h
index 26d73f9cd6d3..f470859244de 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/priv.h
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/priv.h
@@ -4,13 +4,12 @@
 #define nvkm_pmu(p) container_of((p), struct nvkm_pmu, subdev)
 #include <subdev/pmu.h>
 #include <subdev/pmu/fuc/os.h>
-
-int nvkm_pmu_ctor(const struct nvkm_pmu_func *, struct nvkm_device *,
-		  int index, struct nvkm_pmu *);
-int nvkm_pmu_new_(const struct nvkm_pmu_func *, struct nvkm_device *,
-		  int index, struct nvkm_pmu **);
+enum nvkm_acr_lsf_id;
+struct nvkm_acr_lsfw;
 
 struct nvkm_pmu_func {
+	const struct nvkm_falcon_func *flcn;
+
 	struct {
 		u32 *data;
 		u32  size;
@@ -29,9 +28,11 @@ struct nvkm_pmu_func {
 	int (*send)(struct nvkm_pmu *, u32 reply[2], u32 process,
 		    u32 message, u32 data0, u32 data1);
 	void (*recv)(struct nvkm_pmu *);
+	int (*initmsg)(struct nvkm_pmu *);
 	void (*pgob)(struct nvkm_pmu *, bool);
 };
 
+extern const struct nvkm_falcon_func gt215_pmu_flcn;
 int gt215_pmu_init(struct nvkm_pmu *);
 void gt215_pmu_fini(struct nvkm_pmu *);
 void gt215_pmu_intr(struct nvkm_pmu *);
@@ -42,4 +43,26 @@ bool gf100_pmu_enabled(struct nvkm_pmu *);
 void gf100_pmu_reset(struct nvkm_pmu *);
 
 void gk110_pmu_pgob(struct nvkm_pmu *, bool);
+
+void gm20b_pmu_acr_bld_patch(struct nvkm_acr *, u32, s64);
+void gm20b_pmu_acr_bld_write(struct nvkm_acr *, u32, struct nvkm_acr_lsfw *);
+int gm20b_pmu_acr_boot(struct nvkm_falcon *);
+int gm20b_pmu_acr_bootstrap_falcon(struct nvkm_falcon *, enum nvkm_acr_lsf_id);
+void gm20b_pmu_recv(struct nvkm_pmu *);
+int gm20b_pmu_initmsg(struct nvkm_pmu *);
+
+struct nvkm_pmu_fwif {
+	int version;
+	int (*load)(struct nvkm_pmu *, int ver, const struct nvkm_pmu_fwif *);
+	const struct nvkm_pmu_func *func;
+	const struct nvkm_acr_lsf_func *acr;
+};
+
+int gf100_pmu_nofw(struct nvkm_pmu *, int, const struct nvkm_pmu_fwif *);
+int gm20b_pmu_load(struct nvkm_pmu *, int, const struct nvkm_pmu_fwif *);
+
+int nvkm_pmu_ctor(const struct nvkm_pmu_fwif *, struct nvkm_device *,
+		  int index, struct nvkm_pmu *);
+int nvkm_pmu_new_(const struct nvkm_pmu_fwif *, struct nvkm_device *,
+		  int index, struct nvkm_pmu **);
 #endif