gpu: nvgpu: Dump offending push buffer fragment

When outputting debug dump, print the contents of current push buffer
segment.

Also changes the debug dump to use pr_cont when applicable, and dumps
state before recovering in case channel was not loaded to an engine.

Bug 1498688

Change-Id: I5ca12f64bae8f12333d82350278c700645d5007e
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: http://git-master/r/422198

4 files changed, 132 insertions, 23 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/channel_gk20a.c b/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
index 0f805214..486e815c 100644
--- a/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
@@ -929,6 +929,21 @@ static void channel_gk20a_free_priv_cmdbuf(struct channel_gk20a *c)
         memset(q, 0, sizeof(struct priv_cmd_queue));
 }
 
+int gk20a_find_from_priv_cmdbuf(struct channel_gk20a *c,
+                                u64 gpu_va, u32 **cpu_va)
+{
+        struct priv_cmd_queue *q = &c->priv_cmd_q;
+        int ret;
+
+        if (gpu_va >= q->base_gpuva && gpu_va < (q->base_gpuva + q->size)) {
+                *cpu_va = gpu_va - q->base_gpuva + q->mem.base_cpuva;
+                ret = 0;
+        } else
+                ret = -EINVAL;
+
+        return ret;
+}
+
 /* allocate a cmd buffer with given size. size is number of u32 entries */
 int gk20a_channel_alloc_priv_cmdbuf(struct channel_gk20a *c, u32 orig_size,
                              struct priv_cmd_entry **entry)
diff --git a/drivers/gpu/nvgpu/gk20a/channel_gk20a.h b/drivers/gpu/nvgpu/gk20a/channel_gk20a.h
index 60437e66..36697e02 100644
--- a/drivers/gpu/nvgpu/gk20a/channel_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/channel_gk20a.h
@@ -159,6 +159,8 @@ void gk20a_set_error_notifier(struct channel_gk20a *ch, __u32 error);
 void gk20a_channel_semaphore_wakeup(struct gk20a *g);
 int gk20a_channel_alloc_priv_cmdbuf(struct channel_gk20a *c, u32 size,
                              struct priv_cmd_entry **entry);
+int gk20a_find_from_priv_cmdbuf(struct channel_gk20a *c,
+                                u64 addr, u32 **cpu_va);
 
 int gk20a_channel_suspend(struct gk20a *g);
 int gk20a_channel_resume(struct gk20a *g);
diff --git a/drivers/gpu/nvgpu/gk20a/debug_gk20a.c b/drivers/gpu/nvgpu/gk20a/debug_gk20a.c
index f5e0b73d..c607b138 100644
--- a/drivers/gpu/nvgpu/gk20a/debug_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/debug_gk20a.c
@@ -19,6 +19,7 @@
 #include <linux/seq_file.h>
 
 #include <linux/io.h>
+#include <linux/dma-buf.h>
 
 #include "gk20a.h"
 #include "debug_gk20a.h"
@@ -33,6 +34,7 @@ struct platform_device *gk20a_device;
 
 struct gk20a_debug_output {
         void (*fn)(void *ctx, const char *str, size_t len);
+        void (*cont)(void *ctx, const char *str, size_t len);
         void *ctx;
         char buf[256];
 };
@@ -80,6 +82,12 @@ static inline void gk20a_debug_write_printk(void *ctx, const char *str,
         pr_info("%s", str);
 }
 
+static inline void gk20a_debug_cont_printk(void *ctx, const char *str,
+                                            size_t len)
+{
+        pr_cont("%s", str);
+}
+
 static inline void gk20a_debug_write_to_seqfile(void *ctx, const char *str,
                                                 size_t len)
 {
@@ -97,6 +105,17 @@ void gk20a_debug_output(struct gk20a_debug_output *o, const char *fmt, ...)
         o->fn(o->ctx, o->buf, len);
 }
 
+void gk20a_debug_output_cont(struct gk20a_debug_output *o, const char *fmt, ...)
+{
+        va_list args;
+        int len;
+
+        va_start(args, fmt);
+        len = vsnprintf(o->buf, sizeof(o->buf), fmt, args);
+        va_end(args);
+        o->cont(o->ctx, o->buf, len);
+}
+
 static void gk20a_debug_show_channel(struct gk20a *g,
                                      struct gk20a_debug_output *o,
                                      struct channel_gk20a *ch)
@@ -113,14 +132,14 @@ static void gk20a_debug_show_channel(struct gk20a *g,
         syncpointa = gk20a_mem_rd32(inst_ptr, ram_fc_syncpointa_w());
         syncpointb = gk20a_mem_rd32(inst_ptr, ram_fc_syncpointb_w());
 
-        gk20a_debug_output(o, "%d-%s, pid %d: ", ch->hw_chid,
+        gk20a_debug_output_cont(o, "%d-%s, pid %d: ", ch->hw_chid,
                         ch->g->dev->name,
                         ch->pid);
-        gk20a_debug_output(o, "%s in use %s %s\n",
+        gk20a_debug_output_cont(o, "%s in use %s %s\n",
                         ccsr_channel_enable_v(channel) ? "" : "not",
                         ccsr_chan_status_str[status],
                         ccsr_channel_busy_v(channel) ? "busy" : "not busy");
-        gk20a_debug_output(o, "TOP: %016llx PUT: %016llx GET: %016llx "
+        gk20a_debug_output_cont(o, "TOP: %016llx PUT: %016llx GET: %016llx "
                         "FETCH: %016llx\nHEADER: %08x COUNT: %08x\n"
                         "SYNCPOINT %08x %08x SEMAPHORE %08x %08x %08x %08x\n",
                 (u64)gk20a_mem_rd32(inst_ptr, ram_fc_pb_top_level_get_w()) +
@@ -144,14 +163,81 @@ static void gk20a_debug_show_channel(struct gk20a *g,
         if ((pbdma_syncpointb_op_v(syncpointb) == pbdma_syncpointb_op_wait_v())
                 && (pbdma_syncpointb_wait_switch_v(syncpointb) ==
                         pbdma_syncpointb_wait_switch_en_v()))
-                gk20a_debug_output(o, "%s on syncpt %u (%s) val %u\n",
+                gk20a_debug_output_cont(o, "%s on syncpt %u (%s) val %u\n",
                         (status == 3 || status == 8) ? "Waiting" : "Waited",
                         pbdma_syncpointb_syncpt_index_v(syncpointb),
                         nvhost_syncpt_get_name(g->host1x_dev,
                                 pbdma_syncpointb_syncpt_index_v(syncpointb)),
                         pbdma_syncpointa_payload_v(syncpointa));
 
-        gk20a_debug_output(o, "\n");
+        gk20a_debug_output_cont(o, "\n");
+}
+
+static void gk20a_dump_gpfifo(struct channel_gk20a *ch,
+                          struct gpfifo *g, struct gk20a_debug_output *o)
+{
+        struct dma_buf *pb = NULL;
+        u32 *pb_cpu_va = NULL;
+        u64 pb_offset = 0;
+        int i, err = 0;
+
+        u64 gpu_va = (u64)g->entry0
+                | (u64)pbdma_gp_entry1_get_hi_v(g->entry1) << 32ULL;
+        u32 length = pbdma_gp_entry1_length_v(g->entry1);
+
+        if (gk20a_find_from_priv_cmdbuf(ch, gpu_va, &pb_cpu_va)) {
+                gk20a_debug_output_cont(o, "U: ");
+                err = gk20a_vm_find_buffer(ch->vm, gpu_va, &pb, &pb_offset);
+        }
+        if (err) {
+                gk20a_debug_output_cont(o, "Couldn't find push buffer\n");
+                return;
+        }
+
+        if (pb)
+                pb_cpu_va = dma_buf_vmap(pb);
+        for (i = 0; i < length; i++) {
+                if (i && i % 8 == 0)
+                        gk20a_debug_output_cont(o, "\n");
+                gk20a_debug_output_cont(o, "%08x ", *(pb_cpu_va + (pb_offset/4) + i));
+        }
+
+        if (pb)
+                dma_buf_vunmap(pb, pb_cpu_va);
+
+        gk20a_debug_output_cont(o, "\n");
+}
+
+static void gk20a_dump_pb(struct gk20a *g,
+                          u32 pbdma_id, struct gk20a_debug_output *o)
+{
+        u32 gp_get = gk20a_readl(g, pbdma_gp_get_r(pbdma_id));
+        u32 status = gk20a_readl(g, fifo_pbdma_status_r(pbdma_id));
+        u32 chan_status = fifo_pbdma_status_chan_status_v(status);
+        u32 hw_chid = fifo_pbdma_status_id_v(status);
+        struct channel_gk20a *ch = g->fifo.channel+ hw_chid;
+
+        gk20a_debug_output_cont(o, "%s pbdma %d: ", g->dev->name, pbdma_id);
+        gk20a_debug_output_cont(o,
+                        "id: %d (%s), next_id: %d (%s) status: %s\n",
+                        fifo_pbdma_status_id_v(status),
+                        fifo_pbdma_status_id_type_v(status) ?
+                                "tsg" : "channel",
+                        fifo_pbdma_status_next_id_v(status),
+                        fifo_pbdma_status_next_id_type_v(status) ?
+                                "tsg" : "channel",
+                        chan_status_str[chan_status]);
+        gk20a_debug_output_cont(o, "PUT: %08x GET: %08x "
+                        "FETCH: %08x HEADER: %08x\n",
+                gk20a_readl(g, pbdma_gp_put_r(pbdma_id)),
+                gk20a_readl(g, pbdma_gp_get_r(pbdma_id)),
+                gk20a_readl(g, pbdma_gp_fetch_r(pbdma_id)),
+                gk20a_readl(g, pbdma_pb_header_r(pbdma_id)));
+
+        if (ch->in_use) {
+                gk20a_dump_gpfifo(ch, &ch->gpfifo.cpu_va[(gp_get-2) % ch->gpfifo.entry_num], o);
+                gk20a_dump_gpfifo(ch, &ch->gpfifo.cpu_va[(gp_get-1) % ch->gpfifo.entry_num], o);
+        }
 }
 
 void gk20a_debug_show_dump(struct platform_device *pdev,
@@ -164,12 +250,13 @@ void gk20a_debug_show_dump(struct platform_device *pdev,
         int i;
 
         gk20a_busy(g->dev);
+        gk20a_debug_output(o, "");
         for (i = 0; i < fifo_pbdma_status__size_1_v(); i++) {
                 u32 status = gk20a_readl(g, fifo_pbdma_status_r(i));
                 u32 chan_status = fifo_pbdma_status_chan_status_v(status);
 
-                gk20a_debug_output(o, "%s pbdma %d: ", g->dev->name, i);
-                gk20a_debug_output(o,
+                gk20a_debug_output_cont(o, "%s pbdma %d: ", g->dev->name, i);
+                gk20a_debug_output_cont(o,
                                 "id: %d (%s), next_id: %d (%s) status: %s\n",
                                 fifo_pbdma_status_id_v(status),
                                 fifo_pbdma_status_id_type_v(status) ?
@@ -178,23 +265,23 @@ void gk20a_debug_show_dump(struct platform_device *pdev,
                                 fifo_pbdma_status_next_id_type_v(status) ?
                                         "tsg" : "channel",
                                 chan_status_str[chan_status]);
-                gk20a_debug_output(o, "PUT: %016llx GET: %016llx "
+                gk20a_debug_output_cont(o, "PUT: %08x GET: %08x "
                                 "FETCH: %08x HEADER: %08x\n",
-                        (u64)gk20a_readl(g, pbdma_put_r(i)) +
-                        ((u64)gk20a_readl(g, pbdma_put_hi_r(i)) << 32ULL),
-                        (u64)gk20a_readl(g, pbdma_get_r(i)) +
-                        ((u64)gk20a_readl(g, pbdma_get_hi_r(i)) << 32ULL),
+                        gk20a_readl(g, pbdma_gp_put_r(i)),
+                        gk20a_readl(g, pbdma_gp_get_r(i)),
                         gk20a_readl(g, pbdma_gp_fetch_r(i)),
                         gk20a_readl(g, pbdma_pb_header_r(i)));
+
+                gk20a_dump_pb(g, i, o);
         }
-        gk20a_debug_output(o, "\n");
+        gk20a_debug_output_cont(o, "\n");
 
         for (i = 0; i < fifo_engine_status__size_1_v(); i++) {
                 u32 status = gk20a_readl(g, fifo_engine_status_r(i));
                 u32 ctx_status = fifo_engine_status_ctx_status_v(status);
 
-                gk20a_debug_output(o, "%s eng %d: ", g->dev->name, i);
-                gk20a_debug_output(o,
+                gk20a_debug_output_cont(o, "%s eng %d: ", g->dev->name, i);
+                gk20a_debug_output_cont(o,
                                 "id: %d (%s), next_id: %d (%s), ctx: %s ",
                                 fifo_engine_status_id_v(status),
                                 fifo_engine_status_id_type_v(status) ?
@@ -205,12 +292,12 @@ void gk20a_debug_show_dump(struct platform_device *pdev,
                                 ctx_status_str[ctx_status]);
 
                 if (fifo_engine_status_faulted_v(status))
-                        gk20a_debug_output(o, "faulted ");
+                        gk20a_debug_output_cont(o, "faulted ");
                 if (fifo_engine_status_engine_v(status))
-                        gk20a_debug_output(o, "busy ");
-                gk20a_debug_output(o, "\n");
+                        gk20a_debug_output_cont(o, "busy ");
+                gk20a_debug_output_cont(o, "\n");
         }
-        gk20a_debug_output(o, "\n");
+        gk20a_debug_output_cont(o, "\n");
 
         for (chid = 0; chid < f->num_channels; chid++) {
                 if (f->channel[chid].in_use) {
@@ -225,7 +312,8 @@ void gk20a_debug_dump(struct platform_device *pdev)
 {
         struct gk20a_platform *platform = gk20a_get_platform(pdev);
         struct gk20a_debug_output o = {
-                .fn = gk20a_debug_write_printk
+                .fn = gk20a_debug_write_printk,
+                .cont = gk20a_debug_cont_printk
         };
 
         if (platform->dump_platform_dependencies)
@@ -237,7 +325,8 @@ void gk20a_debug_dump(struct platform_device *pdev)
 void gk20a_debug_dump_device(struct platform_device *pdev)
 {
         struct gk20a_debug_output o = {
-                .fn = gk20a_debug_write_printk
+                .fn = gk20a_debug_write_printk,
+                .cont = gk20a_debug_cont_printk
         };
 
         /* Dump the first device if no info is provided */
@@ -253,6 +342,7 @@ static int gk20a_debug_show(struct seq_file *s, void *unused)
         struct platform_device *pdev = s->private;
         struct gk20a_debug_output o = {
                 .fn = gk20a_debug_write_to_seqfile,
+                .cont = gk20a_debug_write_to_seqfile,
                 .ctx = s,
         };
         gk20a_debug_show_dump(pdev, &o);
diff --git a/drivers/gpu/nvgpu/gk20a/fifo_gk20a.c b/drivers/gpu/nvgpu/gk20a/fifo_gk20a.c
index cba1f0fd..3c0611e6 100644
--- a/drivers/gpu/nvgpu/gk20a/fifo_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/fifo_gk20a.c
@@ -1180,13 +1180,15 @@ void gk20a_fifo_recover_ch(struct gk20a *g, u32 hw_chid, bool verbose)
                 struct channel_gk20a *ch =
                         g->fifo.channel + hw_chid;
 
+                if (verbose)
+                        gk20a_debug_dump(g->dev);
+
                 gk20a_channel_abort(ch);
                 for (i = 0; i < g->fifo.max_runlists; i++)
                         gk20a_fifo_update_runlist(g, i,
                                         hw_chid, false, false);
 
-                if (gk20a_fifo_set_ctx_mmu_error(g, ch))
-                        gk20a_debug_dump(g->dev);
+                gk20a_fifo_set_ctx_mmu_error(g, ch);
         }
 }