gpu: nvgpu: vgpu: debugger interface fixes

To run CUDA apps, the following minimal changes have been
made:
- power-gating is disabled for vgpu
- regop rd/wr returns -ENOSYS

Tools (debugger/profiler) support is known to not work and
not needed at this time.

Bug 200043227

Change-Id: I923caad78450e72d310fb9290cf2849ed5460ad5
Signed-off-by: Aingara Paramakuru <aparamakuru@nvidia.com>
Reviewed-on: http://git-master/r/592878
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>

5 files changed, 36 insertions, 11 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/ctrl_gk20a.c b/drivers/gpu/nvgpu/gk20a/ctrl_gk20a.c
index 3bcbdfd9..34351f93 100644
--- a/drivers/gpu/nvgpu/gk20a/ctrl_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/ctrl_gk20a.c
@@ -258,7 +258,6 @@ long gk20a_ctrl_dev_ioctl(struct file *filp, unsigned int cmd, unsigned long arg
         struct zbc_entry *zbc_val;
         struct zbc_query_params *zbc_tbl;
         int i, err = 0;
-        struct gk20a_platform *platform = platform_get_drvdata(dev);
 
         gk20a_dbg_fn("");
 
@@ -320,7 +319,8 @@ long gk20a_ctrl_dev_ioctl(struct file *filp, unsigned int cmd, unsigned long arg
         case NVGPU_GPU_IOCTL_ZBC_SET_TABLE:
                 set_table_args = (struct nvgpu_gpu_zbc_set_table_args *)buf;
 
-                if (platform->virtual_dev)
+                /* not supported for vgpu */
+                if (gk20a_gpu_is_virtual(dev))
                         return -ENOMEM;
 
                 zbc_val = kzalloc(sizeof(struct zbc_entry), GFP_KERNEL);
diff --git a/drivers/gpu/nvgpu/gk20a/dbg_gpu_gk20a.c b/drivers/gpu/nvgpu/gk20a/dbg_gpu_gk20a.c
index bd24e269..72998280 100644
--- a/drivers/gpu/nvgpu/gk20a/dbg_gpu_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/dbg_gpu_gk20a.c
@@ -90,6 +90,11 @@ static int gk20a_dbg_gpu_do_dev_open(struct inode *inode,
         dbg_session->g     = g;
         dbg_session->is_profiler = is_profiler;
         dbg_session->is_pg_disabled = false;
+        /* For vgpu, all power-gating features are currently disabled
+         * in the server. Set is_pg_disable to true to reflect this
+         * on the client side. */
+        if (gk20a_gpu_is_virtual(pdev))
+                dbg_session->is_pg_disabled = true;
 
         INIT_LIST_HEAD(&dbg_session->dbg_s_list_node);
         init_waitqueue_head(&dbg_session->dbg_events.wait_queue);
@@ -492,7 +497,8 @@ static int nvgpu_ioctl_channel_reg_ops(struct dbg_session_gk20a *dbg_s,
         }
 
         /* be sure that ctx info is in place */
-        if (!gr_context_info_available(dbg_s, &g->gr)) {
+        if (!gk20a_gpu_is_virtual(dbg_s->pdev) &&
+                !gr_context_info_available(dbg_s, &g->gr)) {
                 gk20a_err(dev, "gr context data not available\n");
                 return -ENODEV;
         }
diff --git a/drivers/gpu/nvgpu/gk20a/gk20a.c b/drivers/gpu/nvgpu/gk20a/gk20a.c
index 7b0db427..419b8675 100644
--- a/drivers/gpu/nvgpu/gk20a/gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/gk20a.c
@@ -1299,7 +1299,7 @@ static int gk20a_probe(struct platform_device *dev)
 
         platform_set_drvdata(dev, platform);
 
-        if (platform->virtual_dev)
+        if (gk20a_gpu_is_virtual(dev))
                 return vgpu_probe(dev);
 
         gk20a = kzalloc(sizeof(struct gk20a), GFP_KERNEL);
@@ -1449,7 +1449,7 @@ static int __exit gk20a_remove(struct platform_device *dev)
 
         gk20a_dbg_fn("");
 
-        if (platform->virtual_dev)
+        if (gk20a_gpu_is_virtual(dev))
                 return vgpu_remove(dev);
 
         if (platform->has_cde)
@@ -1527,7 +1527,9 @@ int gk20a_busy(struct platform_device *pdev)
 {
         int ret = 0;
         struct gk20a *g = get_gk20a(pdev);
+#ifdef CONFIG_PM_RUNTIME
         struct gk20a_platform *platform = gk20a_get_platform(pdev);
+#endif
 
         down_read(&g->busy_lock);
 
@@ -1549,7 +1551,7 @@ int gk20a_busy(struct platform_device *pdev)
         }
 #else
         if (!g->power_on) {
-                ret = platform->virtual_dev ?
+                ret = gk20a_gpu_is_virtual(pdev) ?
                         vgpu_pm_finalize_poweron(&pdev->dev)
                         : gk20a_pm_finalize_poweron(&pdev->dev);
                 if (ret)
diff --git a/drivers/gpu/nvgpu/gk20a/gk20a.h b/drivers/gpu/nvgpu/gk20a/gk20a.h
index afe1447e..ca6a8ae0 100644
--- a/drivers/gpu/nvgpu/gk20a/gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/gk20a.h
@@ -713,15 +713,21 @@ enum {
         KEPLER_CHANNEL_GPFIFO_C   = 0xA26F,
 };
 
+static inline bool gk20a_gpu_is_virtual(struct platform_device *dev)
+{
+        struct gk20a_platform *platform = gk20a_get_platform(dev);
+
+        return platform->virtual_dev;
+}
+
 static inline int support_gk20a_pmu(struct platform_device *dev)
 {
         if (IS_ENABLED(CONFIG_GK20A_PMU)) {
-                struct gk20a_platform *platform = gk20a_get_platform(dev);
+                /* gPMU is not supported for vgpu */
+                return !gk20a_gpu_is_virtual(dev);
+        }
 
-                /* we have not supported GPU PMU for virtualization now */
-                return !platform->virtual_dev;
-        } else
-                return 0;
+        return 0;
 }
 
 void gk20a_create_sysfs(struct platform_device *dev);
diff --git a/drivers/gpu/nvgpu/gk20a/regops_gk20a.c b/drivers/gpu/nvgpu/gk20a/regops_gk20a.c
index fe444c78..cbdd1a55 100644
--- a/drivers/gpu/nvgpu/gk20a/regops_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/regops_gk20a.c
@@ -392,6 +392,17 @@ int exec_regops_gk20a(struct dbg_session_gk20a *dbg_s,
 
         ch = dbg_s->ch;
 
+        /* For vgpu, the regops routines need to be handled in the
+         * context of the server and support for that does not exist.
+         *
+         * The two users of the regops interface are the compute driver
+         * and tools. The compute driver will work without a functional
+         * regops implementation, so we return -ENOSYS. This will allow
+         * compute apps to run with vgpu. Tools will not work in this
+         * configuration and are not required to work at this time. */
+        if (gk20a_gpu_is_virtual(dbg_s->pdev))
+                return -ENOSYS;
+
         ok = g->allow_all || validate_reg_ops(dbg_s,
                               &ctx_rd_count, &ctx_wr_count,
                               ops, num_ops);