cvs commit: src/sys/pci agp.c

[Hiroki Sato]

Seigo Tanimura <tanimura at FreeBSD.org> wrote
  in <200610150504.k9F548ld008933 at repoman.freebsd.org>:

ta> tanimura    2006-10-15 05:04:07 UTC
ta>
ta>   FreeBSD src repository
ta>
ta>   Modified files:
ta>     sys/pci              agp.c
ta>   Log:
ta>   Fix the wraparound of memsize >=2GB.
ta>
ta>   Revision  Changes    Path
ta>   1.54      +3 -2      src/sys/pci/agp.c

 I have doubt about this change because int memsize->u_int memsize
 does not solve the problem directly; memsize never occurs wraparound
 actually and an implicit cast to unsigned int just makes the problem
 invisible.  The questionable code fragment in agp.c is the following:

 memsize = ptoa(Maxmem) >> 20;
 for (i = 0; i < agp_max_size; i++) {
         if (memsize <= agp_max[i][0])
                 break;
 }

 ptoa(Maxmem)>>20 will occur a wraparound problem when Maxmem>=2GB, so
 this part should be fixed instead.  BTW, this should be a problem
 only on i386 since the definition of ptoa() is "#define ptoa(x) ((x)
 << PAGE_SHIFT)".  The other platforms use a cast like "#define
 ptoa(x) ((unsigned long)(x) << PAGE_SHIFT)".

 I think it can be solved by using "ptoa((unsigned long)Maxmem)" or
 so, but I am not sure if this is reasonable because there are more
 notional types like vm_paddr_t.  If "#define ptoa(x) ((vm_paddr_t)(x)
 << PAGE_SHIFT)" works fine on all platforms, it looks more reasonable
 to me, but I have a misunderstanding?

--
| Hiroki SATO
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 187 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/cvs-src/attachments/20061026/92bf34a1/attachment.pgp