@@ -261,6 +261,7 @@ int main(int argc, char *argv[])
{
struct trace_reader *trs;
struct trace_info *ti;
+ bool no_mmap = false;
struct stat sb;
int fd, opt, i;
@@ -296,13 +297,26 @@ int main(int argc, char *argv[])
err(1, "Stating %s", argv[1]);
ti = mmap(NULL, sb.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
- if (ti == MAP_FAILED)
- err(1, "Mmaping %s", argv[i]);
+ if (ti == MAP_FAILED) {
+ no_mmap = true;
+
+ ti = ezalloc(sb.st_size);
+ if (!ti)
+ err(1, "allocating memory for %s", argv[i]);
+
+ if (read(fd, ti, sb.st_size) == -1)
+ err(1, "reading from %s", argv[i]);
+ }
trs[i].tb = &ti->tb;
list_head_init(&trs[i].traces);
}
+ if (no_mmap) {
+ fprintf(stderr, "disabling follow mode: can't mmap() OPAL export files\n");
+ follow = 0;
+ }
+
do {
load_traces(trs, argc);
display_traces(trs, argc);
Not all kernels support mmap() on an OPAL exported memory range. Fall back to allocating a buffer and using the normal file IO system calls to read the contents of the trace buffer in those cases. This does mean we can't use "follow" mode since we can't monitor the raw trace data effectively, but otherwise it works fine. Signed-off-by: Oliver O'Halloran <oohall@gmail.com> --- external/trace/dump_trace.c | 18 ++++++++++++++++-- 1 file changed, 16 insertions(+), 2 deletions(-)