gpu: nvgpu: vgpu: profiler reservation support

Support for hwpm reservations in the virtual case:

- Add session ops for checking and setting global and context reservations, and
  releasing reservations
- in the native case, these just update reservation counts and flags
- in the vgpu case, when the reservation count is 0, check with the RM server
  that a reservation is possible: for global reservations, no other guest
  can have a reservation; for context reservations, no other guest can have
  a global reservation
- in the vgpu case, when the reservation count is decremented to 0, notify
  the RM server that the guest no longer has any reservations

Bug 1775465
JIRA VFND-3428

Change-Id: Idf115b730e465e35d0745c96a8f8ab6b645c7cae
Signed-off-by: Peter Daifuku <pdaifuku@nvidia.com>
Reviewed-on: http://git-master/r/1323375
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>

4 files changed, 183 insertions, 34 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/dbg_gpu_gk20a.c b/drivers/gpu/nvgpu/gk20a/dbg_gpu_gk20a.c
index 0efbcf06..db534318 100644
--- a/drivers/gpu/nvgpu/gk20a/dbg_gpu_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/dbg_gpu_gk20a.c
@@ -431,8 +431,8 @@ int dbg_unbind_single_channel_gk20a(struct dbg_session_gk20a *dbg_s,
                 if ((prof_obj->session_id == dbg_s->id) &&
                         (prof_obj->ch->hw_chid == chid)) {
                         if (prof_obj->has_reservation) {
-                                g->profiler_reservation_count--;
-                                dbg_s->has_profiler_reservation = false;
+                                g->ops.dbg_session_ops.
+                                  release_profiler_reservation(dbg_s, prof_obj);
                         }
                         list_del(&prof_obj->prof_obj_entry);
                         kfree(prof_obj);
@@ -533,10 +533,9 @@ int gk20a_dbg_gpu_dev_release(struct inode *inode, struct file *filp)
         list_for_each_entry_safe(prof_obj, tmp_obj, &g->profiler_objects,
                                                         prof_obj_entry) {
                 if (prof_obj->session_id == dbg_s->id) {
-                        if (prof_obj->has_reservation) {
-                                g->global_profiler_reservation_held = false;
-                                g->profiler_reservation_count--;
-                        }
+                        if (prof_obj->has_reservation)
+                                g->ops.dbg_session_ops.
+                                  release_profiler_reservation(dbg_s, prof_obj);
                         list_del(&prof_obj->prof_obj_entry);
                         kfree(prof_obj);
                 }
@@ -1583,12 +1582,9 @@ static int nvgpu_ioctl_free_profiler_object(
                                 err = -EINVAL;
                                 break;
                         }
-                        if (prof_obj->has_reservation) {
-                                if (prof_obj->ch == NULL)
-                                        g->global_profiler_reservation_held = false;
-                                g->profiler_reservation_count--;
-                                dbg_s->has_profiler_reservation = false;
-                        }
+                        if (prof_obj->has_reservation)
+                                g->ops.dbg_session_ops.
+                                  release_profiler_reservation(dbg_s, prof_obj);
                         list_del(&prof_obj->prof_obj_entry);
                         kfree(prof_obj);
                         obj_found = true;
@@ -1626,6 +1622,51 @@ static struct dbg_profiler_object_data *find_matching_prof_obj(
         return NULL;
 }
 
+static bool nvgpu_check_and_set_global_reservation(
+                                struct dbg_session_gk20a *dbg_s,
+                                struct dbg_profiler_object_data *prof_obj)
+{
+        struct gk20a *g = dbg_s->g;
+
+        if (g->profiler_reservation_count == 0) {
+                g->global_profiler_reservation_held = true;
+                g->profiler_reservation_count = 1;
+                dbg_s->has_profiler_reservation = true;
+                prof_obj->has_reservation = true;
+                return true;
+        }
+        return false;
+}
+
+static bool nvgpu_check_and_set_context_reservation(
+                                struct dbg_session_gk20a *dbg_s,
+                                struct dbg_profiler_object_data *prof_obj)
+{
+        struct gk20a *g = dbg_s->g;
+
+        /* Assumes that we've already checked that no global reservation
+         * is in effect.
+         */
+        g->profiler_reservation_count++;
+        dbg_s->has_profiler_reservation = true;
+        prof_obj->has_reservation = true;
+        return true;
+}
+
+static void nvgpu_release_profiler_reservation(struct dbg_session_gk20a *dbg_s,
+                                struct dbg_profiler_object_data *prof_obj)
+{
+        struct gk20a *g = dbg_s->g;
+
+        g->profiler_reservation_count--;
+        if (g->profiler_reservation_count < 0)
+                gk20a_err(dev_from_gk20a(g), "Negative reservation count!");
+        dbg_s->has_profiler_reservation = false;
+        prof_obj->has_reservation = false;
+        if (prof_obj->ch == NULL)
+                g->global_profiler_reservation_held = false;
+}
+
 static int nvgpu_profiler_reserve_acquire(struct dbg_session_gk20a *dbg_s,
                                                                 u32 profiler_handle)
 {
@@ -1661,17 +1702,12 @@ static int nvgpu_profiler_reserve_acquire(struct dbg_session_gk20a *dbg_s,
                 /* Global reservations are only allowed if there are no other
                  * global or per-context reservations currently held
                  */
-                if (g->profiler_reservation_count > 0) {
+                if (!g->ops.dbg_session_ops.check_and_set_global_reservation(
+                                                        dbg_s, my_prof_obj)) {
                         gk20a_err(dev_from_gk20a(g),
                                 "global reserve: have existing reservation");
                         err =  -EBUSY;
-                        goto exit;
                 }
-
-                my_prof_obj->has_reservation = true;
-                g->global_profiler_reservation_held = true;
-                g->profiler_reservation_count = 1;
-                dbg_s->has_profiler_reservation = true;
         } else if (g->global_profiler_reservation_held) {
                 /* If there's a global reservation,
                  * we can't take a per-context one.
@@ -1679,7 +1715,6 @@ static int nvgpu_profiler_reserve_acquire(struct dbg_session_gk20a *dbg_s,
                 gk20a_err(dev_from_gk20a(g),
                         "per-ctxt reserve: global reservation in effect");
                 err = -EBUSY;
-                goto exit;
         } else if (gk20a_is_channel_marked_as_tsg(my_prof_obj->ch)) {
                 /* TSG: check that another channel in the TSG
                  * doesn't already have the reservation
@@ -1697,9 +1732,13 @@ static int nvgpu_profiler_reserve_acquire(struct dbg_session_gk20a *dbg_s,
                         }
                 }
 
-                my_prof_obj->has_reservation = true;
-                g->profiler_reservation_count++;
-                dbg_s->has_profiler_reservation = true;
+                if (!g->ops.dbg_session_ops.check_and_set_context_reservation(
+                                                        dbg_s, my_prof_obj)) {
+                        /* Another guest OS has the global reservation */
+                        gk20a_err(dev_from_gk20a(g),
+                                "per-ctxt reserve: global reservation in effect");
+                        err = -EBUSY;
+                }
         } else {
                 /* channel: check that some other profiler object doesn't
                  * already have the reservation.
@@ -1717,9 +1756,13 @@ static int nvgpu_profiler_reserve_acquire(struct dbg_session_gk20a *dbg_s,
                         }
                 }
 
-                my_prof_obj->has_reservation = true;
-                g->profiler_reservation_count++;
-                dbg_s->has_profiler_reservation = true;
+                if (!g->ops.dbg_session_ops.check_and_set_context_reservation(
+                                                        dbg_s, my_prof_obj)) {
+                        /* Another guest OS has the global reservation */
+                        gk20a_err(dev_from_gk20a(g),
+                                "per-ctxt reserve: global reservation in effect");
+                        err = -EBUSY;
+                }
         }
 exit:
         nvgpu_mutex_release(&g->dbg_sessions_lock);
@@ -1746,13 +1789,9 @@ static int nvgpu_profiler_reserve_release(struct dbg_session_gk20a *dbg_s,
                 goto exit;
         }
 
-        if (prof_obj->has_reservation) {
-                prof_obj->has_reservation = false;
-                if (prof_obj->ch == NULL)
-                        g->global_profiler_reservation_held = false;
-                g->profiler_reservation_count--;
-                dbg_s->has_profiler_reservation = false;
-        } else {
+        if (prof_obj->has_reservation)
+                g->ops.dbg_session_ops.release_profiler_reservation(dbg_s, prof_obj);
+        else {
                 gk20a_err(dev_from_gk20a(g), "No reservation found");
                 err = -EINVAL;
                 goto exit;
@@ -1874,4 +1913,10 @@ void gk20a_init_dbg_session_ops(struct gpu_ops *gops)
 {
         gops->dbg_session_ops.exec_reg_ops = exec_regops_gk20a;
         gops->dbg_session_ops.dbg_set_powergate = dbg_set_powergate;
+        gops->dbg_session_ops.check_and_set_global_reservation =
+                                        nvgpu_check_and_set_global_reservation;
+        gops->dbg_session_ops.check_and_set_context_reservation =
+                                        nvgpu_check_and_set_context_reservation;
+        gops->dbg_session_ops.release_profiler_reservation =
+                                        nvgpu_release_profiler_reservation;
 };
diff --git a/drivers/gpu/nvgpu/gk20a/gk20a.h b/drivers/gpu/nvgpu/gk20a/gk20a.h
index 25bcd895..30f1b371 100644
--- a/drivers/gpu/nvgpu/gk20a/gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/gk20a.h
@@ -28,6 +28,7 @@ struct gk20a_fecs_trace;
 struct gk20a_ctxsw_trace;
 struct acr_desc;
 struct nvgpu_mem_alloc_tracker;
+struct dbg_profiler_object_data;
 
 #include <linux/sched.h>
 #include <nvgpu/lock.h>
@@ -758,6 +759,15 @@ struct gpu_ops {
                             u64 num_ops);
                 int (*dbg_set_powergate)(struct dbg_session_gk20a *dbg_s,
                                         u32 mode);
+                bool (*check_and_set_global_reservation)(
+                                struct dbg_session_gk20a *dbg_s,
+                                struct dbg_profiler_object_data *prof_obj);
+                bool (*check_and_set_context_reservation)(
+                                struct dbg_session_gk20a *dbg_s,
+                                struct dbg_profiler_object_data *prof_obj);
+                void (*release_profiler_reservation)(
+                                struct dbg_session_gk20a *dbg_s,
+                                struct dbg_profiler_object_data *prof_obj);
         } dbg_session_ops;
         struct {
                 void (*get_program_numbers)(struct gk20a *g,
diff --git a/drivers/gpu/nvgpu/vgpu/dbg_vgpu.c b/drivers/gpu/nvgpu/vgpu/dbg_vgpu.c
index 609b497a..a2e57ed5 100644
--- a/drivers/gpu/nvgpu/vgpu/dbg_vgpu.c
+++ b/drivers/gpu/nvgpu/vgpu/dbg_vgpu.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2015-2016, NVIDIA CORPORATION.  All rights reserved.
+ * Copyright (c) 2015-2017, NVIDIA CORPORATION.  All rights reserved.
  *
  * This program is free software; you can redistribute it and/or modify it
  * under the terms and conditions of the GNU General Public License,
@@ -100,8 +100,90 @@ static int vgpu_dbg_set_powergate(struct dbg_session_gk20a *dbg_s, __u32 mode)
         return err;
 }
 
+static int vgpu_sendrecv_prof_cmd(struct dbg_session_gk20a *dbg_s, u32 mode)
+{
+        struct tegra_vgpu_cmd_msg msg;
+        struct tegra_vgpu_prof_mgt_params *p = &msg.params.prof_management;
+        int err = 0;
+
+        msg.cmd = TEGRA_VGPU_CMD_PROF_MGT;
+        msg.handle = vgpu_get_handle(dbg_s->g);
+
+        p->mode = mode;
+
+        err = vgpu_comm_sendrecv(&msg, sizeof(msg), sizeof(msg));
+        err = err ? err : msg.ret;
+        return err;
+}
+
+static bool vgpu_check_and_set_global_reservation(
+                                struct dbg_session_gk20a *dbg_s,
+                                struct dbg_profiler_object_data *prof_obj)
+{
+        struct gk20a *g = dbg_s->g;
+
+        if (g->profiler_reservation_count > 0)
+                return false;
+
+        /* Check that another guest OS doesn't already have a reservation */
+        if (!vgpu_sendrecv_prof_cmd(dbg_s, TEGRA_VGPU_PROF_GET_GLOBAL)) {
+                g->global_profiler_reservation_held = true;
+                g->profiler_reservation_count = 1;
+                dbg_s->has_profiler_reservation = true;
+                prof_obj->has_reservation = true;
+                return true;
+        }
+        return false;
+}
+
+static bool vgpu_check_and_set_context_reservation(
+                                struct dbg_session_gk20a *dbg_s,
+                                struct dbg_profiler_object_data *prof_obj)
+{
+        struct gk20a *g = dbg_s->g;
+
+        /* Assumes that we've already checked that no global reservation
+         * is in effect for this guest.
+         *
+         * If our reservation count is non-zero, then no other guest has the
+         * global reservation; if it is zero, need to check with RM server.
+         *
+         */
+        if ((g->profiler_reservation_count != 0) ||
+                !vgpu_sendrecv_prof_cmd(dbg_s, TEGRA_VGPU_PROF_GET_CONTEXT)) {
+                g->profiler_reservation_count++;
+                dbg_s->has_profiler_reservation = true;
+                prof_obj->has_reservation = true;
+                return true;
+        }
+        return false;
+}
+
+static void vgpu_release_profiler_reservation(
+                                struct dbg_session_gk20a *dbg_s,
+                                struct dbg_profiler_object_data *prof_obj)
+{
+        struct gk20a *g = dbg_s->g;
+
+        dbg_s->has_profiler_reservation = false;
+        prof_obj->has_reservation = false;
+        if (prof_obj->ch == NULL)
+                g->global_profiler_reservation_held = false;
+
+        /* If new reservation count is zero, notify server */
+        g->profiler_reservation_count--;
+        if (g->profiler_reservation_count == 0)
+                vgpu_sendrecv_prof_cmd(dbg_s, TEGRA_VGPU_PROF_RELEASE);
+}
+
 void vgpu_init_dbg_session_ops(struct gpu_ops *gops)
 {
         gops->dbg_session_ops.exec_reg_ops = vgpu_exec_regops;
         gops->dbg_session_ops.dbg_set_powergate = vgpu_dbg_set_powergate;
+        gops->dbg_session_ops.check_and_set_global_reservation =
+                                        vgpu_check_and_set_global_reservation;
+        gops->dbg_session_ops.check_and_set_context_reservation =
+                                        vgpu_check_and_set_context_reservation;
+        gops->dbg_session_ops.release_profiler_reservation =
+                                        vgpu_release_profiler_reservation;
 }
diff --git a/include/linux/tegra_vgpu.h b/include/linux/tegra_vgpu.h
index c0964083..bb9bdcab 100644
--- a/include/linux/tegra_vgpu.h
+++ b/include/linux/tegra_vgpu.h
@@ -102,6 +102,7 @@ enum {
         TEGRA_VGPU_CMD_SUSPEND_CONTEXTS = 66,
         TEGRA_VGPU_CMD_RESUME_CONTEXTS = 67,
         TEGRA_VGPU_CMD_CLEAR_SM_ERROR_STATE = 68,
+        TEGRA_VGPU_CMD_PROF_MGT = 72,
 };
 
 struct tegra_vgpu_connect_params {
@@ -469,6 +470,16 @@ struct tegra_vgpu_clear_sm_error_state {
         u32 sm_id;
 };
 
+enum {
+        TEGRA_VGPU_PROF_GET_GLOBAL = 0,
+        TEGRA_VGPU_PROF_GET_CONTEXT,
+        TEGRA_VGPU_PROF_RELEASE
+};
+
+struct tegra_vgpu_prof_mgt_params {
+        u32 mode;
+};
+
 struct tegra_vgpu_cmd_msg {
         u32 cmd;
         int ret;
@@ -518,6 +529,7 @@ struct tegra_vgpu_cmd_msg {
                 struct tegra_vgpu_suspend_resume_contexts suspend_contexts;
                 struct tegra_vgpu_suspend_resume_contexts resume_contexts;
                 struct tegra_vgpu_clear_sm_error_state clear_sm_error_state;
+                struct tegra_vgpu_prof_mgt_params prof_management;
                 char padding[192];
         } params;
 };