drm/nouveau/pmu/gm20b,gp10b: boot RTOS from PMU init

Cleanup before falcon changes.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>

9 files changed, 92 insertions, 60 deletions

diff --git a/drivers/gpu/drm/nouveau/include/nvkm/engine/falcon.h b/drivers/gpu/drm/nouveau/include/nvkm/engine/falcon.h
index 176f0921d82f..780e25fdd119 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/engine/falcon.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/engine/falcon.h
@@ -16,7 +16,7 @@ enum nvkm_falcon_dmaidx {
 
 struct nvkm_falcon {
 	const struct nvkm_falcon_func *func;
-	const struct nvkm_subdev *owner;
+	struct nvkm_subdev *owner;
 	const char *name;
 	u32 addr;
 
@@ -24,7 +24,7 @@ struct nvkm_falcon {
 	struct mutex dmem_mutex;
 	bool oneinit;
 
-	const struct nvkm_subdev *user;
+	struct nvkm_subdev *user;
 
 	u8 version;
 	u8 secret;
@@ -50,8 +50,8 @@ struct nvkm_falcon {
 	struct nvkm_engine engine;
 };
 
-int nvkm_falcon_get(struct nvkm_falcon *, const struct nvkm_subdev *);
-void nvkm_falcon_put(struct nvkm_falcon *, const struct nvkm_subdev *);
+int nvkm_falcon_get(struct nvkm_falcon *, struct nvkm_subdev *);
+void nvkm_falcon_put(struct nvkm_falcon *, struct nvkm_subdev *);
 
 int nvkm_falcon_new_(const struct nvkm_falcon_func *, struct nvkm_device *,
 		     enum nvkm_subdev_type, int inst, bool enable, u32 addr, struct nvkm_engine **);
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/acr.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/acr.h
index c0b254f7f0b5..3a269bb1d41b 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/acr.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/acr.h
@@ -50,6 +50,7 @@ struct nvkm_acr {
 	struct nvkm_vmm *vmm;
 
 	bool done;
+	struct nvkm_acr_lsf *rtos;
 
 	const struct firmware *wpr_fw;
 	bool wpr_comp;
diff --git a/drivers/gpu/drm/nouveau/nvkm/falcon/base.c b/drivers/gpu/drm/nouveau/nvkm/falcon/base.c
index f3f90c1063dd..5a5b96dad640 100644
--- a/drivers/gpu/drm/nouveau/nvkm/falcon/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/falcon/base.c
@@ -169,7 +169,7 @@ nvkm_falcon_oneinit(struct nvkm_falcon *falcon)
 }
 
 void
-nvkm_falcon_put(struct nvkm_falcon *falcon, const struct nvkm_subdev *user)
+nvkm_falcon_put(struct nvkm_falcon *falcon, struct nvkm_subdev *user)
 {
 	if (unlikely(!falcon))
 		return;
@@ -183,7 +183,7 @@ nvkm_falcon_put(struct nvkm_falcon *falcon, const struct nvkm_subdev *user)
 }
 
 int
-nvkm_falcon_get(struct nvkm_falcon *falcon, const struct nvkm_subdev *user)
+nvkm_falcon_get(struct nvkm_falcon *falcon, struct nvkm_subdev *user)
 {
 	int ret = 0;
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/acr/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/acr/base.c
index af6cac696d43..777b6cb33b47 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/acr/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/acr/base.c
@@ -63,10 +63,30 @@ nvkm_acr_hsf_boot(struct nvkm_acr *acr, const char *name)
 	return 0;
 }
 
+static struct nvkm_acr_lsf *
+nvkm_acr_rtos(struct nvkm_acr *acr)
+{
+	struct nvkm_acr_lsf *lsf;
+
+	if (acr) {
+		list_for_each_entry(lsf, &acr->lsf, head) {
+			if (lsf->func->bootstrap_falcon)
+				return lsf;
+		}
+	}
+
+	return NULL;
+}
+
 static void
 nvkm_acr_unload(struct nvkm_acr *acr)
 {
 	if (acr->done) {
+		if (acr->rtos) {
+			nvkm_subdev_unref(acr->rtos->falcon->owner);
+			acr->rtos = NULL;
+		}
+
 		nvkm_acr_hsf_boot(acr, "unload");
 		acr->done = false;
 	}
@@ -76,6 +96,7 @@ static int
 nvkm_acr_load(struct nvkm_acr *acr)
 {
 	struct nvkm_subdev *subdev = &acr->subdev;
+	struct nvkm_acr_lsf *rtos = nvkm_acr_rtos(acr);
 	struct nvkm_acr_lsf *lsf;
 	u64 start, limit;
 	int ret;
@@ -100,6 +121,14 @@ nvkm_acr_load(struct nvkm_acr *acr)
 
 	acr->done = true;
 
+	if (rtos) {
+		ret = nvkm_subdev_ref(rtos->falcon->owner);
+		if (ret)
+			return ret;
+
+		acr->rtos = rtos;
+	}
+
 	list_for_each_entry(lsf, &acr->lsf, head) {
 		if (lsf->func->boot) {
 			ret = lsf->func->boot(lsf->falcon);
@@ -118,33 +147,17 @@ nvkm_acr_reload(struct nvkm_acr *acr)
 	return nvkm_acr_load(acr);
 }
 
-static struct nvkm_acr_lsf *
-nvkm_acr_falcon(struct nvkm_device *device)
-{
-	struct nvkm_acr *acr = device->acr;
-	struct nvkm_acr_lsf *lsf;
-
-	if (acr) {
-		list_for_each_entry(lsf, &acr->lsf, head) {
-			if (lsf->func->bootstrap_falcon)
-				return lsf;
-		}
-	}
-
-	return NULL;
-}
-
 int
 nvkm_acr_bootstrap_falcons(struct nvkm_device *device, unsigned long mask)
 {
-	struct nvkm_acr_lsf *acrflcn = nvkm_acr_falcon(device);
 	struct nvkm_acr *acr = device->acr;
+	struct nvkm_acr_lsf *rtos = nvkm_acr_rtos(acr);
 	unsigned long id;
 
 	/* If there's no LS FW managing bootstrapping of other LS falcons,
 	 * we depend on the HS firmware being able to do it instead.
 	 */
-	if (!acrflcn) {
+	if (!rtos) {
 		/* Which isn't possible everywhere... */
 		if ((mask & acr->func->bootstrap_falcons) == mask) {
 			int ret = nvkm_acr_reload(acr);
@@ -156,16 +169,14 @@ nvkm_acr_bootstrap_falcons(struct nvkm_device *device, unsigned long mask)
 		return -ENOSYS;
 	}
 
-	if ((mask & acrflcn->func->bootstrap_falcons) != mask)
+	if ((mask & rtos->func->bootstrap_falcons) != mask)
 		return -ENOSYS;
 
-	if (acrflcn->func->bootstrap_multiple_falcons) {
-		return acrflcn->func->
-			bootstrap_multiple_falcons(acrflcn->falcon, mask);
-	}
+	if (rtos->func->bootstrap_multiple_falcons)
+		return rtos->func->bootstrap_multiple_falcons(rtos->falcon, mask);
 
 	for_each_set_bit(id, &mask, NVKM_ACR_LSF_NUM) {
-		int ret = acrflcn->func->bootstrap_falcon(acrflcn->falcon, id);
+		int ret = rtos->func->bootstrap_falcon(rtos->falcon, id);
 		if (ret)
 			return ret;
 	}
@@ -189,6 +200,9 @@ nvkm_acr_managed_falcon(struct nvkm_device *device, enum nvkm_acr_lsf_id id)
 static int
 nvkm_acr_fini(struct nvkm_subdev *subdev, bool suspend)
 {
+	if (!subdev->use.enabled)
+		return 0;
+
 	nvkm_acr_unload(nvkm_acr(subdev));
 	return 0;
 }
@@ -196,10 +210,12 @@ nvkm_acr_fini(struct nvkm_subdev *subdev, bool suspend)
 static int
 nvkm_acr_init(struct nvkm_subdev *subdev)
 {
-	if (!nvkm_acr_falcon(subdev->device))
+	struct nvkm_acr *acr = nvkm_acr(subdev);
+
+	if (!nvkm_acr_rtos(acr))
 		return 0;
 
-	return nvkm_acr_load(nvkm_acr(subdev));
+	return nvkm_acr_load(acr);
 }
 
 static void
@@ -218,7 +234,7 @@ nvkm_acr_oneinit(struct nvkm_subdev *subdev)
 	struct nvkm_acr *acr = nvkm_acr(subdev);
 	struct nvkm_acr_hsfw *hsfw;
 	struct nvkm_acr_lsfw *lsfw, *lsft;
-	struct nvkm_acr_lsf *lsf;
+	struct nvkm_acr_lsf *lsf, *rtos;
 	u32 wpr_size = 0;
 	u64 falcons;
 	int ret, i;
@@ -260,10 +276,10 @@ nvkm_acr_oneinit(struct nvkm_subdev *subdev)
 	}
 
 	/* Ensure the falcon that'll provide ACR functions is booted first. */
-	lsf = nvkm_acr_falcon(device);
-	if (lsf) {
-		falcons = lsf->func->bootstrap_falcons;
-		list_move(&lsf->head, &acr->lsf);
+	rtos = nvkm_acr_rtos(acr);
+	if (rtos) {
+		falcons = rtos->func->bootstrap_falcons;
+		list_move(&rtos->head, &acr->lsf);
 	} else {
 		falcons = acr->func->bootstrap_falcons;
 	}
@@ -301,7 +317,7 @@ nvkm_acr_oneinit(struct nvkm_subdev *subdev)
 		nvkm_wobj(acr->wpr, 0, acr->wpr_fw->data, acr->wpr_fw->size);
 
 	if (!acr->wpr_fw || acr->wpr_comp)
-		acr->func->wpr_build(acr, nvkm_acr_falcon(device));
+		acr->func->wpr_build(acr, rtos);
 	acr->func->wpr_patch(acr, (s64)acr->wpr_start - acr->wpr_prev);
 
 	if (acr->wpr_fw && acr->wpr_comp) {
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/base.c
index 51808a4e75ac..8f2f50ad4ded 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/base.c
@@ -87,13 +87,6 @@ nvkm_pmu_fini(struct nvkm_subdev *subdev, bool suspend)
 	if (pmu->func->fini)
 		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;
 }
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gm20b.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gm20b.c
index 2b73848d4aed..a72403777329 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gm20b.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gm20b.c
@@ -62,16 +62,6 @@ gm20b_pmu_acr_bootstrap_falcon(struct nvkm_falcon *falcon,
 	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)
 {
@@ -125,7 +115,6 @@ gm20b_pmu_acr = {
 	.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_falcons = BIT_ULL(NVKM_ACR_LSF_PMU) |
 			     BIT_ULL(NVKM_ACR_LSF_FECS) |
 			     BIT_ULL(NVKM_ACR_LSF_GPCCS),
@@ -198,8 +187,7 @@ gm20b_pmu_recv(struct nvkm_pmu *pmu)
 	if (!pmu->initmsg_received) {
 		int ret = pmu->func->initmsg(pmu);
 		if (ret) {
-			nvkm_error(&pmu->subdev,
-				   "error parsing init message: %d\n", ret);
+			nvkm_error(&pmu->subdev, "error parsing init message: %d\n", ret);
 			return;
 		}
 
@@ -209,9 +197,44 @@ gm20b_pmu_recv(struct nvkm_pmu *pmu)
 	nvkm_falcon_msgq_recv(pmu->msgq);
 }
 
+static void
+gm20b_pmu_fini(struct nvkm_pmu *pmu)
+{
+	/*TODO: shutdown RTOS. */
+
+	flush_work(&pmu->recv.work);
+	nvkm_falcon_cmdq_fini(pmu->lpq);
+	nvkm_falcon_cmdq_fini(pmu->hpq);
+
+	reinit_completion(&pmu->wpr_ready);
+
+	nvkm_falcon_put(&pmu->falcon, &pmu->subdev);
+}
+
+static int
+gm20b_pmu_init(struct nvkm_pmu *pmu)
+{
+	struct nvkm_falcon *falcon = &pmu->falcon;
+	struct nv_pmu_args args = { .secure_mode = true };
+	u32 addr_args = falcon->data.limit - sizeof(args);
+	int ret;
+
+	ret = nvkm_falcon_get(&pmu->falcon, &pmu->subdev);
+	if (ret)
+		return ret;
+
+	pmu->initmsg_received = false;
+
+	nvkm_falcon_load_dmem(falcon, &args, addr_args, sizeof(args), 0);
+	nvkm_falcon_start(falcon);
+	return 0;
+}
+
 const struct nvkm_pmu_func
 gm20b_pmu = {
 	.flcn = &gm200_pmu_flcn,
+	.init = gm20b_pmu_init,
+	.fini = gm20b_pmu_fini,
 	.intr = gt215_pmu_intr,
 	.recv = gm20b_pmu_recv,
 	.initmsg = gm20b_pmu_initmsg,
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gp10b.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gp10b.c
index 9fba37ec05e0..a6f410ba60bc 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gp10b.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gp10b.c
@@ -68,7 +68,6 @@ gp10b_pmu_acr = {
 	.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_falcons = BIT_ULL(NVKM_ACR_LSF_PMU) |
 			     BIT_ULL(NVKM_ACR_LSF_FECS) |
 			     BIT_ULL(NVKM_ACR_LSF_GPCCS),
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gt215.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gt215.c
index 6d1a966be081..32cee21ed858 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gt215.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/gt215.c
@@ -178,6 +178,7 @@ void
 gt215_pmu_fini(struct nvkm_pmu *pmu)
 {
 	nvkm_wr32(pmu->subdev.device, 0x10a014, 0x00000060);
+	flush_work(&pmu->recv.work);
 }
 
 static void
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/priv.h b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/priv.h
index 1e7815b1a649..75ede265a52c 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/priv.h
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pmu/priv.h
@@ -50,7 +50,6 @@ extern const struct nvkm_falcon_func gm200_pmu_flcn;
 extern const struct nvkm_pmu_func gm20b_pmu;
 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);
 
 struct nvkm_pmu_fwif {