1 files changed, 744 insertions, 0 deletions

diff --git a/include/gk20a/fecs_trace_gk20a.c b/include/gk20a/fecs_trace_gk20a.c
new file mode 100644
index 0000000..5c1c5e0
--- /dev/null
+++ b/include/gk20a/fecs_trace_gk20a.c
@@ -0,0 +1,744 @@
+/*
+ * Copyright (c) 2016-2019, 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 <nvgpu/kmem.h>
+#include <nvgpu/dma.h>
+#include <nvgpu/enabled.h>
+#include <nvgpu/bug.h>
+#include <nvgpu/hashtable.h>
+#include <nvgpu/circ_buf.h>
+#include <nvgpu/thread.h>
+#include <nvgpu/barrier.h>
+#include <nvgpu/mm.h>
+#include <nvgpu/enabled.h>
+#include <nvgpu/ctxsw_trace.h>
+#include <nvgpu/io.h>
+#include <nvgpu/utils.h>
+#include <nvgpu/timers.h>
+#include <nvgpu/channel.h>
+
+#include "fecs_trace_gk20a.h"
+#include "gk20a.h"
+#include "gr_gk20a.h"
+
+#include <nvgpu/log.h>
+#include <nvgpu/fecs_trace.h>
+
+#include <nvgpu/hw/gk20a/hw_ctxsw_prog_gk20a.h>
+#include <nvgpu/hw/gk20a/hw_gr_gk20a.h>
+
+struct gk20a_fecs_trace_hash_ent {
+        u32 context_ptr;
+        pid_t pid;
+        struct hlist_node node;
+};
+
+struct gk20a_fecs_trace {
+
+        DECLARE_HASHTABLE(pid_hash_table, GK20A_FECS_TRACE_HASH_BITS);
+        struct nvgpu_mutex hash_lock;
+        struct nvgpu_mutex poll_lock;
+        struct nvgpu_thread poll_task;
+        bool init;
+        struct nvgpu_mutex enable_lock;
+        u32 enable_count;
+};
+
+#ifdef CONFIG_GK20A_CTXSW_TRACE
+u32 gk20a_fecs_trace_record_ts_tag_invalid_ts_v(void)
+{
+        return ctxsw_prog_record_timestamp_timestamp_hi_tag_invalid_timestamp_v();
+}
+
+u32 gk20a_fecs_trace_record_ts_tag_v(u64 ts)
+{
+        return ctxsw_prog_record_timestamp_timestamp_hi_tag_v((u32) (ts >> 32));
+}
+
+u64 gk20a_fecs_trace_record_ts_timestamp_v(u64 ts)
+{
+        return ts & ~(((u64)ctxsw_prog_record_timestamp_timestamp_hi_tag_m()) << 32);
+}
+
+static u32 gk20a_fecs_trace_fecs_context_ptr(struct gk20a *g, struct channel_gk20a *ch)
+{
+        return (u32) (nvgpu_inst_block_addr(g, &ch->inst_block) >> 12LL);
+}
+
+int gk20a_fecs_trace_num_ts(void)
+{
+        return (ctxsw_prog_record_timestamp_record_size_in_bytes_v()
+                - sizeof(struct gk20a_fecs_trace_record)) / sizeof(u64);
+}
+
+struct gk20a_fecs_trace_record *gk20a_fecs_trace_get_record(
+        struct gk20a *g, int idx)
+{
+        struct nvgpu_mem *mem = &g->gr.global_ctx_buffer[FECS_TRACE_BUFFER].mem;
+
+        return (struct gk20a_fecs_trace_record *)
+                ((u8 *) mem->cpu_va
+                + (idx * ctxsw_prog_record_timestamp_record_size_in_bytes_v()));
+}
+
+bool gk20a_fecs_trace_is_valid_record(struct gk20a_fecs_trace_record *r)
+{
+        /*
+         * testing magic_hi should suffice. magic_lo is sometimes used
+         * as a sequence number in experimental ucode.
+         */
+        return (r->magic_hi
+                == ctxsw_prog_record_timestamp_magic_value_hi_v_value_v());
+}
+
+int gk20a_fecs_trace_get_read_index(struct gk20a *g)
+{
+        return gr_gk20a_elpg_protected_call(g,
+                        gk20a_readl(g, gr_fecs_mailbox1_r()));
+}
+
+int gk20a_fecs_trace_get_write_index(struct gk20a *g)
+{
+        return gr_gk20a_elpg_protected_call(g,
+                        gk20a_readl(g, gr_fecs_mailbox0_r()));
+}
+
+static int gk20a_fecs_trace_set_read_index(struct gk20a *g, int index)
+{
+        nvgpu_log(g, gpu_dbg_ctxsw, "set read=%d", index);
+        return gr_gk20a_elpg_protected_call(g,
+                        (gk20a_writel(g, gr_fecs_mailbox1_r(), index), 0));
+}
+
+void gk20a_fecs_trace_hash_dump(struct gk20a *g)
+{
+        u32 bkt;
+        struct gk20a_fecs_trace_hash_ent *ent;
+        struct gk20a_fecs_trace *trace = g->fecs_trace;
+
+        nvgpu_log(g, gpu_dbg_ctxsw, "dumping hash table");
+
+        nvgpu_mutex_acquire(&trace->hash_lock);
+        hash_for_each(trace->pid_hash_table, bkt, ent, node)
+        {
+                nvgpu_log(g, gpu_dbg_ctxsw, " ent=%p bkt=%x context_ptr=%x pid=%d",
+                        ent, bkt, ent->context_ptr, ent->pid);
+
+        }
+        nvgpu_mutex_release(&trace->hash_lock);
+}
+
+static int gk20a_fecs_trace_hash_add(struct gk20a *g, u32 context_ptr, pid_t pid)
+{
+        struct gk20a_fecs_trace_hash_ent *he;
+        struct gk20a_fecs_trace *trace = g->fecs_trace;
+
+        nvgpu_log(g, gpu_dbg_fn | gpu_dbg_ctxsw,
+                "adding hash entry context_ptr=%x -> pid=%d", context_ptr, pid);
+
+        he = nvgpu_kzalloc(g, sizeof(*he));
+        if (unlikely(!he)) {
+                nvgpu_warn(g,
+                        "can't alloc new hash entry for context_ptr=%x pid=%d",
+                        context_ptr, pid);
+                return -ENOMEM;
+        }
+
+        he->context_ptr = context_ptr;
+        he->pid = pid;
+        nvgpu_mutex_acquire(&trace->hash_lock);
+        hash_add(trace->pid_hash_table, &he->node, context_ptr);
+        nvgpu_mutex_release(&trace->hash_lock);
+        return 0;
+}
+
+static void gk20a_fecs_trace_hash_del(struct gk20a *g, u32 context_ptr)
+{
+        struct hlist_node *tmp;
+        struct gk20a_fecs_trace_hash_ent *ent;
+        struct gk20a_fecs_trace *trace = g->fecs_trace;
+
+        nvgpu_log(g, gpu_dbg_fn | gpu_dbg_ctxsw,
+                "freeing hash entry context_ptr=%x", context_ptr);
+
+        nvgpu_mutex_acquire(&trace->hash_lock);
+        hash_for_each_possible_safe(trace->pid_hash_table, ent, tmp, node,
+                context_ptr) {
+                if (ent->context_ptr == context_ptr) {
+                        hash_del(&ent->node);
+                        nvgpu_log(g, gpu_dbg_ctxsw,
+                                "freed hash entry=%p context_ptr=%x", ent,
+                                ent->context_ptr);
+                        nvgpu_kfree(g, ent);
+                        break;
+                }
+        }
+        nvgpu_mutex_release(&trace->hash_lock);
+}
+
+static void gk20a_fecs_trace_free_hash_table(struct gk20a *g)
+{
+        u32 bkt;
+        struct hlist_node *tmp;
+        struct gk20a_fecs_trace_hash_ent *ent;
+        struct gk20a_fecs_trace *trace = g->fecs_trace;
+
+        nvgpu_log(g, gpu_dbg_fn | gpu_dbg_ctxsw, "trace=%p", trace);
+
+        nvgpu_mutex_acquire(&trace->hash_lock);
+        hash_for_each_safe(trace->pid_hash_table, bkt, tmp, ent, node) {
+                hash_del(&ent->node);
+                nvgpu_kfree(g, ent);
+        }
+        nvgpu_mutex_release(&trace->hash_lock);
+
+}
+
+static pid_t gk20a_fecs_trace_find_pid(struct gk20a *g, u32 context_ptr)
+{
+        struct gk20a_fecs_trace_hash_ent *ent;