Significantly speed up ar(1) on UFS file systems

[emaste]

Summary

Fault in the buffer prior to writing as a workaround for poor performance due to interaction with kernel fs deadlock avoidance code. See the comment prior to vn_io_fault_doio() in vfs_vnops.c for details of the issue.

Thanks @kib for diagnosing, providing an explanation of the issue and workaround.

Test Plan

% truncate -s 16M obj.o
% ar r out.a obj.o

Total time with stock and patched ar(1):

x ar.r284891
+ ar.patched
+----------------------------------------------------------------------+   
|+                                                                     |   
|+                                                                    x|   
|+                                                                   xx|   
|A                                                                   |A|   
+----------------------------------------------------------------------+   
    N         Min          Max        Median           Avg        Stddev   
x   3       1.307        1.321         1.315     1.3143333  0.0070237692   
+   3        0.02        0.023         0.022   0.021666667  0.0015275252   
Difference at 95.0% confidence
        -1.29267 +/- 0.0115203
        -98.3515% +/- 0.876513%
        (Student's t, pooled s = 0.00508265)

[kib]

usr.bin/ar/write.c

568 ↗

(On Diff #6543)

It is simpler to do

if (s == 0)
  return;
for (p = buf; p < buf + s; p += page_size)
   *p;
*(volatile const char *)(buf + s - 1);

I do not insist.

[emaste]

• give full path to file with comment explaining the deadlock avoidance
• avoid confusing page mask manipulation and just touch the last byte explicitly