more portable fpu devprocio

on the architecture of the new sse support for the pae kernel
1.  there is no support for x87.  an early version of this work had
a FPArch structure, so that instead of fpclear, one wrote fp->clear().
while really clean, and potentially helpful for arm, this results in
a lot of churn and incompatability.  other than devarch, there should
be no dependence on sse except in fpsse.c

2.  in order to reduce churn, the assembly code does the proper
rounding to insure alignment of the fxsave area.

3.  devprocio is the only place where we care where the fxsave area
is.  the least churn was to add a default devprocio (the same as the
old one), and allow architectures to override it.  this saves us from
having to import sse-specific ideas about alignment into the portable
code.

4.  odd bit: we are clearing the xmm registers by hand when fp is
intialized to provide a consistent register state for new processes,
and to avoid leaking register information.

Reference: /n/atom/patch/applied2013/fpuprocio
Date: Mon Jun  3 18:12:05 CES 2013
Signed-off-by: quanstro@quanstro.net

--- /sys/src/9/port/devproc.c	Mon Jun  3 18:00:25 2013
+++ /sys/src/9/port/devproc.c	Mon Jun  3 18:00:26 2013
@@ -80,7 +80,7 @@
 	"args",		{Qargs},	0,			0660,
 	"ctl",		{Qctl},		0,			0000,
 	"fd",		{Qfd},		0,			0444,
-	"fpregs",	{Qfpregs},	sizeof(FPsave),		0000,
+	"fpregs",	{Qfpregs},	0,		0000,
 	"kregs",	{Qkregs},	sizeof(Ureg),		0600,
 	"mem",		{Qmem},		0,			0000,
 	"note",		{Qnote},	0,			0000,
@@ -178,6 +178,29 @@
 	}
 }
 
+static long 
+·fpudevprocio(Proc *p, void *va, long n, uvlong offset, int write)
+{
+	uchar *fp;
+	int sz;
+
+	fp = (uchar*)&p->fpsave;
+	sz = sizeof(FPsave);
+
+	if(offset >= sz)
+		n = 0;
+	else if(offset+n > sz)
+		n = sz - offset;
+
+	if(write)
+		memmove(fp+offset, va, n);
+	else
+		memmove(va, fp+offset, n);
+	return n;
+}
+
+long (*fpudevprocio)(Proc*, void*, long, uvlong, int) = ·fpudevprocio;
+
 static int
 procgen(Chan *c, char *name, Dirtab *tab, int, int s, Dir *dp)
 {
@@ -805,11 +828,6 @@
 		setkernur(&kur, p);
 		rptr = (uchar*)&kur;
 		rsize = sizeof(Ureg);
-		goto regread;
-
-	case Qfpregs:
-		rptr = (uchar*)&p->fpsave;
-		rsize = sizeof(FPsave);
 	regread:
 		if(rptr == 0)
 			error(Enoreg);
@@ -820,6 +838,9 @@
 		memmove(a, rptr+offset, n);
 		return n;
 
+	case Qfpregs:
+		return fpudevprocio(p, va, n, offset, 0);
+
 	case Qstatus:
 		if(offset >= STATSIZE)
 			return 0;
@@ -1070,11 +1091,7 @@
 		break;
 
 	case Qfpregs:
-		if(offset >= sizeof(FPsave))
-			n = 0;
-		else if(offset+n > sizeof(FPsave))
-			n = sizeof(FPsave) - offset;
-		memmove((uchar*)&p->fpsave+offset, va, n);
+		n = fpudevprocio(p, va, n, offset, 1);
 		break;
 
 	case Qctl: