Fix warning messages related to lcp_rtt_* functions

[enaess]

Compiling the latest mainline on Solaris adds additional warning messages, we should address these. And why use the keyword "volatile" in these circumstances?

  lcp.c: In function ‘lcp_rtt_update_buffer’:
  lcp.c:2310:15: warning: passing argument 1 of ‘msync’ discards ‘volatile’ qualifier from pointer target type [-Wdiscarded-qualifiers]
       if (msync(lcp_rtt_buffer, LCP_RTT_FILE_SIZE, MS_ASYNC) < 0)
                 ^~~~~~~~~~~~~~
  In file included from lcp.c:54:0:
  /usr/include/sys/mman.h:204:12: note: expected ‘void *’ but argument is of type ‘volatile u_int32_t * {aka volatile unsigned int *}’
   extern int msync(void *, size_t, int);
              ^~~~~
  lcp.c: In function ‘lcp_rtt_open_file’:
  lcp.c:2444:9: warning: passing argument 1 of ‘memset’ discards ‘volatile’ qualifier from pointer target type [-Wdiscarded-qualifiers]
    memset(lcp_rtt_buffer, 0, LCP_RTT_FILE_SIZE);
           ^~~~~~~~~~~~~~
  In file included from /usr/include/string.h:12:0,
                   from lcp.c:48:
  /usr/include/iso/string_iso.h:85:14: note: expected ‘void *’ but argument is of type ‘volatile u_int32_t * {aka volatile unsigned int *}’
   extern void *memset(void *, int, size_t);
                ^~~~~~
  lcp.c: In function ‘lcp_rtt_close_file’:
  lcp.c:2460:16: warning: passing argument 1 of ‘munmap’ discards ‘volatile’ qualifier from pointer target type [-Wdiscarded-qualifiers]
       if (munmap(lcp_rtt_buffer, LCP_RTT_FILE_SIZE) < 0)
                  ^~~~~~~~~~~~~~
  In file included from lcp.c:54:0:
  /usr/include/sys/mman.h:202:12: note: expected ‘void *’ but argument is of type ‘volatile u_int32_t * {aka volatile unsigned int *}’
   extern int munmap(void *, size_t);
              ^~~~~~

[enaess]

@rfc1036

[rfc1036]

I think that the purpose is to prevent reordering by the compiler.

[paulusmack]

The buffer is volatile because other programs could potentially mmap it and access it concurrently, and so we want to limit reordering of accesses by the compiler. Volatile is not a complete fix, though, because the CPU can still reorder accesses, and to fix it properly we would need barrier instructions, which tends to get machine-dependent.

It is possible to make the accesses volatile,rather than the data, i.e. use casts to volatile where the buffer entries are read or written rather than making the data structure itself volatile.

[Neustradamus]

About Solaris:

• https://github.com/ppp-project/ppp/issues/458

cc: @carlsonj, @RICCIARDI-Adrien.

[Neustradamus]

@enaess: This bug always exists?

[paulusmack]

Fixed by fafbfdf