kernel: introduce per process FPU struct (PFPU) for more flexible machine specific fpu handling

introducing the PFPU structue which allows the machine specific
code some flexibility on how to handle the FPU process state.

for example, in the pc and pc64 kernel, the FPsave structure is
arround 512 bytes. with avx512, it could grow up to 2K. instead
of embedding that into the Proc strucutre, it is more effective
to allocate it on first use of the fpu, as most processes do not
use simd or floating point in the first place. also, the FPsave
structure has special 16 byte alignment constraint, which further
favours dynamic allocation.

this gets rid of the memmoves in pc/pc64 kernels for the aligment.

there is also devproc, which is now checking if the fpsave area
is actually valid before reading it, avoiding debuggers to see
garbage data.

the Notsave structure is gone now, as it was not used on any
machine.

4 files changed, 36 insertions, 38 deletions

diff --git a/sys/src/9/sgi/dat.h b/sys/src/9/sgi/dat.h
index 800ed7182..741ca5655 100644
--- a/sys/src/9/sgi/dat.h
+++ b/sys/src/9/sgi/dat.h
@@ -1,6 +1,7 @@
 typedef struct Conf	Conf;
 typedef struct Confmem	Confmem;
 typedef struct FPsave	FPsave;
+typedef struct PFPU	PFPU;
 typedef struct KMap	KMap;
 typedef struct Lance	Lance;
 typedef struct Lancemem	Lancemem;
@@ -8,7 +9,6 @@ typedef struct Label	Label;
 typedef struct Lock	Lock;
 typedef struct Mach	Mach;
 typedef struct MMU	MMU;
-typedef struct Notsave	Notsave;
 typedef struct PMMU	PMMU;
 typedef struct Softtlb	Softtlb;
 typedef struct Ureg	Ureg;
@@ -73,18 +73,6 @@ struct Conf
 /*
  * floating point registers
  */
-enum
-{
-	/* floating point state */
-	FPinit,
-	FPactive,
-	FPinactive,
-	FPemu,
-
-	/* bit meaning floating point illegal */
-	FPillegal= 0x100,
-};
-
 enum {
 	Nfpregs		= 32,		/* floats; half as many doubles */
 };
@@ -111,20 +99,30 @@ struct FPsave
 	int	fpcnt;			/* how many consecutive at that addr */
 };
 
-/*
- *  mmu goo in the Proc structure
- */
-struct PMMU
+struct PFPU
 {
-	int	pidonmach[MAXMACH];
+	int	fpstate;
+	FPsave	fpsave[1];
+};
+
+enum
+{
+	/* floating point state */
+	FPinit,
+	FPactive,
+	FPinactive,
+	FPemu,
+
+	/* bit meaning floating point illegal */
+	FPillegal= 0x100,
 };
 
 /*
- *  things saved in the Proc structure during a notify
+ *  mmu goo in the Proc structure
  */
-struct Notsave
+struct PMMU
 {
-	ulong	nonempty;
+	int	pidonmach[MAXMACH];
 };
 
 #include "../port/portdat.h"
diff --git a/sys/src/9/sgi/fptrap.c b/sys/src/9/sgi/fptrap.c
index 82c61bbd3..9d1c22aec 100644
--- a/sys/src/9/sgi/fptrap.c
+++ b/sys/src/9/sgi/fptrap.c
@@ -29,7 +29,7 @@ fptrap(Ureg *ur)
 {
 	ulong iw, npc;
 
-	if((up->fpsave.fpstatus&(1<<17)) == 0)
+	if((up->fpsave->fpstatus&(1<<17)) == 0)
 		return;
 
 	if(ur->cause & (1<<31))
@@ -41,7 +41,7 @@ fptrap(Ureg *ur)
 		return;
 
 	if(ur->cause & (1<<31)){
-		npc = branch(ur, up->fpsave.fpstatus);
+		npc = branch(ur, up->fpsave->fpstatus);
 		if(npc == 0)
 			return;
 		ur->pc = npc;
@@ -49,7 +49,7 @@ fptrap(Ureg *ur)
 	else
 		ur->pc += 4;
 
-	up->fpsave.fpstatus &= ~(1<<17);
+	up->fpsave->fpstatus &= ~(1<<17);
 }
 
 static void
@@ -107,8 +107,8 @@ fpunimp(ulong iw)
 	ft = (iw>>16) & ((1<<5)-1);
 	fs = (iw>>11) & ((1<<5)-1);
 	fd = (iw>>6) & ((1<<5)-1);
-	unpack(&up->fpsave, fmt, fs, &ss, &es);
-	unpack(&up->fpsave, fmt, ft, &st, &et);
+	unpack(up->fpsave, fmt, fs, &ss, &es);
+	unpack(up->fpsave, fmt, ft, &st, &et);
 	ed = 0;
 	maxe = 0;
 	maxm = 0;
@@ -124,11 +124,11 @@ fpunimp(ulong iw)
 	}
 	switch(op){
 	case ABS:
-		up->fpsave.reg[fd] &= ~0x80000000;
+		up->fpsave->reg[fd] &= ~0x80000000;
 		return 1;
 
 	case NEG:
-		up->fpsave.reg[fd] ^= 0x80000000;
+		up->fpsave->reg[fd] ^= 0x80000000;
 		return 1;
 
 	case SUB:
@@ -164,9 +164,9 @@ fpunimp(ulong iw)
 		return 0;
 	}
 	if(ed <= -(maxe-5)){	/* guess: underflow */
-		zeroreg(&up->fpsave, fmt, fd, sd);
+		zeroreg(up->fpsave, fmt, fd, sd);
 		/* Set underflow exception and sticky */
-		up->fpsave.fpstatus |= (1<<3)|(1<<13);
+		up->fpsave->fpstatus |= (1<<3)|(1<<13);
 		return 1;
 	}
 	return 0;
diff --git a/sys/src/9/sgi/main.c b/sys/src/9/sgi/main.c
index f1c239e3a..dea690f29 100644
--- a/sys/src/9/sgi/main.c
+++ b/sys/src/9/sgi/main.c
@@ -396,7 +396,7 @@ void
 procsetup(Proc *p)
 {
 	p->fpstate = FPinit;
-	p->fpsave = initfp;
+	memmove(p->fpsave, &initfp, sizeof(FPsave));
 
 	cycles(&p->kentry);
 	p->pcycles = -p->kentry;
@@ -413,11 +413,11 @@ procfork(Proc *p)
 	s = splhi();
 	switch(up->fpstate & ~FPillegal){
 	case FPactive:
-		savefpregs(&up->fpsave);
+		savefpregs(up->fpsave);
 		up->fpstate = FPinactive;
 		/* wet floor */
 	case FPinactive:
-		p->fpsave = up->fpsave;
+		memmove(p->fpsave, up->fpsave, sizeof(FPsave));
 		p->fpstate = FPinactive;
 	}
 	splx(s);
@@ -430,7 +430,7 @@ procsave(Proc *p)
 
 	if(p->fpstate == FPactive){
 		if(p->state != Moribund) {
-			savefpregs(&p->fpsave);
+			savefpregs(p->fpsave);
 			p->fpstate = FPinactive;
 		}
 	}
diff --git a/sys/src/9/sgi/trap.c b/sys/src/9/sgi/trap.c
index 4c59730f6..588e6861d 100644
--- a/sys/src/9/sgi/trap.c
+++ b/sys/src/9/sgi/trap.c
@@ -212,7 +212,7 @@ trap(Ureg *ur)
 		if(!user)
 			goto Default;
 		if(up->fpstate == FPactive){
-			savefpregs(&up->fpsave);
+			savefpregs(up->fpsave);
 			up->fpstate = FPinactive;
 		}
 		clrfpintr();
@@ -257,7 +257,7 @@ trap(Ureg *ur)
 				break;
 			}
 			if(up->fpstate == FPinit || up->fpstate == FPinactive){
-				restfpregs(&up->fpsave, up->fpsave.fpstatus&~FPEXPMASK);
+				restfpregs(up->fpsave, up->fpsave->fpstatus&~FPEXPMASK);
 				up->fpstate = FPactive;
 				ur->status |= CU1;
 				break;
@@ -289,7 +289,7 @@ trap(Ureg *ur)
 	}
 
 	if(fpchk) {
-		fpfcr31 = up->fpsave.fpstatus;
+		fpfcr31 = up->fpsave->fpstatus;
 		if((fpfcr31>>12) & ((fpfcr31>>7)|0x20) & 0x3f) {
 			spllo();
 			fpexcep	= fpexcname(ur, fpfcr31, buf1, sizeof buf1);
@@ -501,7 +501,7 @@ notify(Ureg *ur)
 		return 0;
 
 	if(up->fpstate == FPactive){
-		savefpregs(&up->fpsave);
+		savefpregs(up->fpsave);
 		up->fpstate = FPinactive;
 	}
 	up->fpstate |= FPillegal;