From: James Hogan Date: Wed, 22 Sep 2010 00:00:19 +0000 Subject: Re: [PATCH] fbmem: Fix fb_read, fb_write unaligned accesses. Message-Id: <201009220100.27090.james@albanarts.com> MIME-Version: 1 Content-Type: multipart/mixed; boundary="nextPart2702498.Moc59V2gJj" List-Id: References: <201009180123.48303.james@albanarts.com> <20100921161950.b7f45273.akpm@linux-foundation.org> In-Reply-To: <20100921161950.b7f45273.akpm@linux-foundation.org> To: Andrew Morton Cc: sparclinux@vger.kernel.org, linux-kernel@vger.kernel.org, linux-fbdev@vger.kernel.org, "David S. Miller" , Dave Airlie , Marcin Slusarz , Florian Tobias Schandinat , Denys Vlasenko , Jesse Barnes , James Simmons --nextPart2702498.Moc59V2gJj Content-Type: Text/Plain; charset="us-ascii" Content-Transfer-Encoding: quoted-printable On Wednesday 22 September 2010 00:19:50 Andrew Morton wrote: > On Sat, 18 Sep 2010 01:23:47 +0100 >=20 > James Hogan wrote: > > Apologies for corrupted patch. I'll try again. > > Comments? I'd also appreciate if somebody familiar with sbus on sparc > > could check this patch is sane since I know virtually nothing about sbus > > and am not in a position to compile for sparc, let alone test on it: > >=20 > > fb_{read,write} access the framebuffer using lots of fb_{read,write}l's > > but don't check that the file position is aligned which can cause > > problems on some architectures which do not support unaligned accesses. >=20 > What are these "problems"? On the arch I hit this on (which isn't in tree) I experienced a fault at th= e=20 point of the unaligned write in kernel code, but this is only because=20 unaligned access checking is switched on which isn't always possible=20 (otherwise the write would have just silently failed/done something else). = It=20 terminated the program, but didn't cause any other problems. Documentation/unaligned-memory-access.txt also mentions that some=20 architectures raise exceptions that can't be worked around, or fail silentl= y=20 and may actually perform different writes to the one requested. > I'd have thought they would be fairly fatal, in which case this is a > high-priority patch. But I'd also have thought that the problems would > have been noted before now. The common way to access the framebuffer is by mmapping it into userland, s= o I=20 don't think the read/write syscalls are commonly used on it, and in any cas= e=20 colour data tends to be naturally aligned so unaligned writes would be=20 uncommon. >=20 > So I assume that you're doing something which nobody has done before. >=20 > Confused. Help? The actualy case that hit this is admitedly rather silly. I did a quick tes= t=20 of the framebuffer by typing this at a shell: yes > /dev/fb0 I'd have to check it again to see exactly how lots of 2 byte writes ("y\n")= =20 ended up unaligned, but at the point of the crash the dest pointer was=20 definitely unaligned (the hex address ended in a 5) and the patch fixed it = for=20 me. Hope that helps. Cheers James --nextPart2702498.Moc59V2gJj Content-Type: application/pgp-signature; name=signature.asc Content-Description: This is a digitally signed message part. -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.14 (GNU/Linux) iEYEABECAAYFAkyZRxoACgkQ4hGc8zKz77AB2wCfcDWZPNY8ahrE0xeJqyhBcJQn OpwAn1q/4BPkHaTE2TlDYwCPc+jn/OQu =1C06 -----END PGP SIGNATURE----- --nextPart2702498.Moc59V2gJj--