test: iterate both reader and writer threads starting at the actual pagesize

[rscheff]

Summary

The writer iterates from pagesize to 1M, while the reader iterates
from 4k to 1M - on systems with a pagesize other than 4k, the "jumbo"
test would likely fail because of this.

[pho]

In my opinion, any (broken) test scenario that can crash the kernel is a good stress test. :)

I applied your patch and still see the panic.

22:09 /usr/src/tools/test/stress2/misc $ git diff jumbo.sh
diff --git a/tools/test/stress2/misc/jumbo.sh b/tools/test/stress2/misc/jumbo.sh
index 8dcc5aa7aca33a..a2d09bafaa8268 100755
--- a/tools/test/stress2/misc/jumbo.sh
+++ b/tools/test/stress2/misc/jumbo.sh
@@ -131,7 +131,7 @@ reader(void) {
        if ((buf = malloc(MX)) == NULL)
                        err(1, "malloc(%d), %s:%d", MX, __FILE__, __LINE__);
        setproctitle("reader");
-       for (i = 4096; i < MX; i += 1024) {
+       for (i = sysconf(_SC_PAGESIZE); i < MX; i += 1024) {
                alarm(TIMEOUT);
                if ((n = recvfrom(msgsock, buf, i, MSG_WAITALL, NULL,
                    NULL)) < 0) {
22:09 /usr/src/tools/test/stress2/misc $ ``

[pho]

The change looks good to me.
I'm always a bit hesitant to change a test scenario which has proven to trigger multiple panics, but this looks fine to me.