Fail reads which return the flag value for a non-existent register

Also check for read success in get_runlist_iter().

2 files changed, 23 insertions, 9 deletions

diff --git a/nvdebug.h b/nvdebug.h
index 8009b84..8db07ee 100644
--- a/nvdebug.h
+++ b/nvdebug.h
@@ -960,30 +960,42 @@ static inline int file2parentgpuidx(const struct file *f) {
 // Similar to nvgpu_readl()
 // (except we don't try to resolve situations where regs is NULL)
 static inline u32 nvdebug_readl(struct nvdebug_state *s, u32 r) {
+        u32 ret;
         if (unlikely(!s->regs || (s->g && !gk20a_regs(s->g)))) {
-                printk(KERN_ERR "[nvdebug] Attempted nvgpu_readl on non-existent registers!\n");
+                printk(KERN_ERR "[nvdebug] nvdebug_readl: Unable to read; registers unavailable. Is GPU on?\n");
                 return -1;
         }
-        return readl(s->regs + r);
+        ret = readl(s->regs + r);
+        // It seems like the GPU returns this as a flag value for bad addresses
+        if (ret == 0xbadf5040) {
+                printk(KERN_ERR "[nvdebug] nvdebug_readl: Unable to read from register offset %#x; bad data\n", r);
+                return -1;
+        }
+        return ret;
 }
 
 // quadword version of nvdebug_readl()
 static inline u64 nvdebug_readq(struct nvdebug_state *s, u32 r) {
         u64 ret;
         if (unlikely(!s->regs || (s->g && !gk20a_regs(s->g)))) {
-                printk(KERN_ERR "[nvdebug] Attempted nvgpu_readl on non-existent registers!\n");
+                printk(KERN_ERR "[nvdebug] nvdebug_readq: Unable to read; registers unavailable. Is GPU on?\n");
                 return -1;
         }
         // readq seems to always return the uppermost 32 bits as 0, so workaround with readl
         ret = readl(s->regs + r);
         ret |= ((u64)readl(s->regs + r + 4)) << 32;
+        // It seems like the GPU returns this as a flag value for bad addresses
+        if ((ret & 0xffffffffull) == 0xbadf5040ull) {
+                printk(KERN_ERR "[nvdebug] nvdebug_readq: Unable to read from register offset %#x; bad data\n", r);
+                return -1;
+        }
         return ret;
 }
 
 // Similar to nvgpu_writel()
 static inline void nvdebug_writel(struct nvdebug_state *s, u32 r, u32 v) {
         if (unlikely(!s->regs || (s->g && !gk20a_regs(s->g)))) {
-                printk(KERN_ERR "[nvdebug] Attempted nvgpu_writel on non-existent registers!\n");
+                printk(KERN_ERR "[nvdebug] nvdebug_writel: Unable to write; registers unavailable. Is GPU on?\n");
                 return;
         }
         writel_relaxed(v, s->regs + r);
@@ -994,9 +1006,9 @@ static inline void nvdebug_writel(struct nvdebug_state *s, u32 r, u32 v) {
 // XXX: This probably doesn't work XXX: Untested
 static inline void nvdebug_writeq(struct nvdebug_state *s, u32 r, u64 v) {
         if (unlikely(!s->regs || (s->g && !gk20a_regs(s->g)))) {
-                printk(KERN_ERR "[nvdebug] Attempted nvgpu_writel on non-existent registers!\n");
-                return;
-        }
-        writeq_relaxed(v, s->regs + r);
-        wmb();
+                printk(KERN_ERR "[nvdebug] nvdebug_writeq: Unable to write; registers unavailable. Is GPU on?\n");
+                return;
+        }
+        writeq_relaxed(v, s->regs + r);
+        wmb();
 }
diff --git a/runlist.c b/runlist.c
index 94be18e..f6a05ac 100644
--- a/runlist.c
+++ b/runlist.c
@@ -41,6 +41,8 @@ int get_runlist_iter(struct nvdebug_state *g, int rl_id, struct runlist_iter *rl
         // before because the GPU had simply gone to sleep and invalidated its
         // register state, so nvgpu_readl() was simply returning garbage.
         rl_info.raw = nvdebug_readl(g, NV_PFIFO_ENG_RUNLIST(rl_id));
+        if (rl_info.raw == -1)
+                return -EIO;
         runlist_iova = ((u64)rl_base.ptr) << 12;
         printk(KERN_INFO "[nvdebug] Runlist %d @ %llx in %s (config raw: %x)\n",
                rl_id, runlist_iova, target_to_text(rl_base.target), rl_base.raw);