ozone: drm: Add fd number to crash keys when dmabuf mmap fails.

ui/ozone/platform/drm/common/client_native_pixmap_dmabuf.cc

Index: ui/ozone/platform/drm/common/client_native_pixmap_dmabuf.cc
diff --git a/ui/ozone/platform/drm/common/client_native_pixmap_dmabuf.cc b/ui/ozone/platform/drm/common/client_native_pixmap_dmabuf.cc
index faee338dfcefe5bbde68332aa67b8db90c1b8676..634464a81adf2829f924c7187d81d071c90f241f 100644
--- a/ui/ozone/platform/drm/common/client_native_pixmap_dmabuf.cc
+++ b/ui/ozone/platform/drm/common/client_native_pixmap_dmabuf.cc
@@ -13,6 +13,7 @@
 #include "base/debug/crash_logging.h"
 #include "base/memory/ptr_util.h"
 #include "base/process/memory.h"
+#include "base/process/process_metrics.h"
 #include "base/strings/stringprintf.h"
 #include "base/trace_event/trace_event.h"
 
@@ -91,13 +92,24 @@ ClientNativePixmapDmaBuf::ClientNativePixmapDmaBuf(
         "(addr=nullptr, length=%zu, prot=(PROT_READ | PROT_WRITE), "
         "flags=MAP_SHARED, fd=%d[valid=%d], offset=0)",
         map_size, dmabuf_fd_.get(), fd_valid);
+    std::string errno_str = logging::SystemErrorCodeToString(mmap_error);
+    std::unique_ptr<base::ProcessMetrics> process_metrics(
+        base::ProcessMetrics::CreateCurrentProcessMetrics());

[Daniele Castagna, 2017/03/02 02:05:04]

CreateCurrentProcessMetrics calls base::GetCurrentProcessHandle().
This won't probably work in the renderer since we're in a new pid namespace and
/proc is not remounted. We should access /proc/self instead of /proc/id.

[dshwang, 2017/03/02 19:59:31]

That's good point. By the way, it's base::GetCurrentProcessHandle()
implamantation issue, rather than this CL. Furthermore, even if the impl uses
"/proc/self", renderer still cannot read because of sandbox filesystem
protection. As renderer has no way, renderer will has just "-1" for it.

[Daniele Castagna, 2017/03/02 20:08:52]

Are you sure this works in the GPU process and the value you get is what you
expect?
Did you try to run this code even once in the browser/renderer/gpu process?

[dshwang, 2017/03/02 20:15:12]

I run this code in chromeos, and it works in only browser process. gpu process
never use this code, as this class is for client for gpu process.

+    std::string number_of_fds =
+        base::StringPrintf("%d", process_metrics->GetOpenFdCount());

[Daniele Castagna, 2017/03/02 02:05:04]

This would report -1 in case we're out of fds since we won't be able to open
/proc/{PID}/fd. Is that what you intended?

[dshwang, 2017/03/02 19:59:31]

OH, that's good point. I didn't think about it. Fortunately, it looks good. If
browser process report -1, we can think it reachs FD limit. so looks good to go.

[Daniele Castagna, 2017/03/02 20:08:52]

We were thinking of addding a UMA stat for openfds. As rsesek suggested, we
could open the fd for /proc/self/fd early and keep it open. In that way we
wouldn't have the problem of trying to open a dir when we might have reached the
fd limit.

[dshwang, 2017/03/02 20:15:12]

good to know. do you want to wait for UMA stat change?

     base::debug::ScopedCrashKey params_crash_key("mmap_params", mmap_params);
     base::debug::ScopedCrashKey size_crash_key("buffer_size", size.ToString());
-    base::debug::ScopedCrashKey errno_crash_key(
-        "errno", logging::SystemErrorCodeToString(mmap_error));
+    base::debug::ScopedCrashKey errno_crash_key("errno", errno_str);
+    base::debug::ScopedCrashKey number_of_fds_crash_key("number_of_fds",
+                                                        number_of_fds);
+    LOG(ERROR) << "Failed to mmap dmabuf; mmap_params: " << mmap_params
+               << ", buffer_size: (" << size.ToString()
+               << "),  errno: " << errno_str
+               << " , number_of_fds: " << number_of_fds;
     if (mmap_error == ENOMEM)
       base::TerminateBecauseOutOfMemory(map_size);
-    CHECK(false) << "Failed to mmap dmabuf.";
+    CHECK(false);
+    ;

[Daniele Castagna, 2017/03/02 02:05:04]

Why did you remove a useful log and just left a floating semi colon?

[dshwang, 2017/03/02 19:59:31]

Done.

   }
 }