Comment 18 for bug 2052929

The master branch has been updated by Jakub Jelinek <email address hidden>:

https://gcc.gnu.org/g:8455d6f6cd43b7b143ab9ee19437452fceba9cc9

commit r14-9769-g8455d6f6cd43b7b143ab9ee19437452fceba9cc9
Author: Jakub Jelinek <email address hidden>
Date: Wed Apr 3 10:02:35 2024 +0200

    libquadmath: Don't assume the storage for __float128 arguments is aligned [PR114533]

    With the register_printf_type/register_printf_modifier/register_printf_specifier
    APIs the C library is just told the size of the argument and is provided with
    a callback to fetch the argument from va_list using va_arg into C library provided
    memory. The C library isn't told what alignment requirement it has, but we were
    using direct load of a __float128 value from that memory which assumes
    __alignof (__float128) alignment.

    The following patch fixes that by using memcpy instead.

    I haven't been able to reproduce an actual crash, tried
     #include <quadmath.h>
     #include <stdlib.h>
     #include <stdio.h>

    int main ()
    {
      __float128 r;
      int prec = 20;
      int width = 46;
      char buf[128];

      r = 2.0q;
      r = sqrtq (r);
      int n = quadmath_snprintf (buf, sizeof buf, "%+-#*.20Qe", width, r);
      if ((size_t) n < sizeof buf)
        printf ("%s\n", buf);
        /* Prints: +1.41421356237309504880e+00 */
      quadmath_snprintf (buf, sizeof buf, "%Qa", r);
      if ((size_t) n < sizeof buf)
        printf ("%s\n", buf);
        /* Prints: 0x1.6a09e667f3bcc908b2fb1366ea96p+0 */
      n = quadmath_snprintf (NULL, 0, "%+-#46.*Qe", prec, r);
      if (n > -1)
        {
          char *str = malloc (n + 1);
          if (str)
            {
              quadmath_snprintf (str, n + 1, "%+-#46.*Qe", prec, r);
              printf ("%s\n", str);
              /* Prints: +1.41421356237309504880e+00 */
            }
          free (str);
        }
      printf ("%+-#*.20Qe\n", width, r);
      printf ("%Qa\n", r);
      printf ("%+-#46.*Qe\n", prec, r);
      printf ("%d %Qe %d %Qe %d %Qe\n", 1, r, 2, r, 3, r);
      return 0;
    }
    In any case, I think memcpy for loading from it is right.

    2024-04-03 Simon Chopin <email address hidden>
                Jakub Jelinek <email address hidden>

            PR libquadmath/114533
            * printf/printf_fp.c (__quadmath_printf_fp): Use memcpy to copy
            __float128 out of args.
            * printf/printf_fphex.c (__quadmath_printf_fphex): Likewise.

    Signed-off-by: Simon Chopin <email address hidden>