zymq: lilu dallas, multicore

implement multiprocessor support.
This commit is contained in:
cinap_lenrek 2014-12-29 16:02:57 +01:00
parent e08cc06517
commit cb2103879e
10 changed files with 205 additions and 66 deletions

View file

@ -140,7 +140,6 @@ struct Mach
Proc* readied; /* for runproc */
ulong schedticks; /* next forced context switch */
int cputype;
ulong delayloop;
/* stats */

View file

@ -9,6 +9,7 @@ uintptr cankaddr(uintptr);
void procsave(Proc *);
void procrestore(Proc *);
void idlehands(void);
void sendevent(void);
void coherence(void);
void procfork(Proc *);
void procsetup(Proc *);
@ -37,7 +38,9 @@ uintptr getdfar(void);
void links(void);
void* vmap(uintptr, ulong);
void timerinit(void);
void synccycles(void);
void setpmcr(ulong);
void setpmcnten(ulong);
void* tmpmap(uintptr);
void tmpunmap(void*);
void flushpg(void *);

View file

@ -44,6 +44,9 @@ intrinit(void)
mpcore[ICCICR] = 7;
mpcore[ICCBPR] = 3;
mpcore[ICCPMR] = 255;
if(m->machno != 0)
return;
/* disable all irqs and clear any pending interrupts */
for(i = 0; i < NINTR/32; i++){
@ -66,7 +69,7 @@ intrenable(int irq, void (*f)(Ureg *, void *), void *arg, int type, char *name)
panic("intrenable: invalid irq %d", irq);
if(type != LEVEL && type != EDGE)
panic("intrenable: invalid type %d", type);
if(irqs[irq].f != nil)
if(irqs[irq].f != nil && irqs[irq].f != f)
panic("intrenable: handler already assigned");
if(irq >= NPRIVATE){
e = &mpcore[ICDIPTR + (irq >> 2)];

View file

@ -18,7 +18,7 @@ _start0:
PUTC('l')
MOVW $SECSZ, R0
MOVW $(CPU0L1-KZERO), R4
MOVW $(MACHL1(0)-KZERO), R4
MOVW $KZERO, R1
ADD R1>>(SECSH-2), R4, R1
MOVW $(L1SEC|L1CACHED|L1KERRW), R2
@ -42,13 +42,20 @@ _start2:
BGE _start2
MOVW $(UART_BASE|L2VALID|L2DEVICE|L2KERRW), R0
MOVW $(VMAPL2 - KZERO), R1
MOVW $(VMAPL2-KZERO), R1
MOVW R0, (R1)
PUTC('n')
MOVW $(MACH(0)-KZERO), R(Rmach)
_start3:
/* enable MMU permission checking */
MOVW $0x55555555, R0
MCR 15, 0, R0, C(3), C(0), 0
MOVW $0, R0
MCR 15, 0, R0, C(8), C(7), 0
MOVW $(CPU0L1 - KZERO | TTBATTR), R1
ADD $TTBATTR, R4, R1
MCR 15, 0, R1, C(2), C(0), 0
MOVW $0x20c5047b, R1
MOVW $_virt(SB), R2
@ -59,17 +66,18 @@ _start2:
TEXT _virt(SB), $-4
DSB
ISB
MOVW $(MACH(0) + MACHSIZE), R13
MOVW $(MACH(0) + 12), R0
ADD $KZERO, R(Rmach)
MOVW R(Rmach), R13
ADD $MACHSIZE, R13
MOVW R(Rmach), R0
ADD $12, R0
BL loadsp(SB)
MOVW $vectors(SB), R1
MCR 15, 0, R1, C(12), C(0)
/* enable MMU permission checking */
MOVW $0x55555555, R0
MCR 15, 0, R0, C(3), C(0), 0
/* enable maths coprocessors in CPACR but disable them in FPEXC */
MRC 15, 0, R0, C(1), C(0), 2
ORR $(15<<20), R0
@ -96,14 +104,25 @@ TEXT _virt(SB), $-4
MOVW $(VMAP+0x30), R8
PUTC('9')
/* kernel Mach* in TPIDRPRW */
MCR 15, 0, R(Rmach), C(13), C(0), 4
MOVW $setR12(SB), R12
MOVW $MACH(0), R(Rmach)
MOVW $0, R(Rup)
BL main(SB)
B idlehands(SB)
BL _div(SB) /* hack to load _div */
TEXT mpbootstrap(SB), $-4
MOVW $0xE0001030, R8
PUTC('M')
PUTC('P')
MOVW $(MACH(1)-KZERO), R(Rmach)
MOVW $(MACHL1(1)-KZERO), R4
B _start3
TEXT touser(SB), $-4
CPS(CPSID)
@ -238,6 +257,10 @@ TEXT idlehands(SB), $0
WFE
RET
TEXT sendevent(SB), $0
SEV
RET
TEXT ttbget(SB), $0
MRC 15, 0, R0, C(2), C(0), 0
BIC $0x7f, R0
@ -282,6 +305,10 @@ TEXT setpmcr(SB), $0
MCR 15, 0, R0, C(9), C(12), 0
RET
TEXT setpmcnten(SB), $0
MCR 15, 0, R0, C(9), C(12), 1
RET
TEXT perfticks(SB), $0
MRC 15, 0, R0, C(9), C(13), 0
RET
@ -453,3 +480,4 @@ TEXT palookur(SB), $0
DSB
MRC 15, 0, R0, C(7), C(4), 0
RET

View file

@ -39,7 +39,8 @@ _exc:
SUB $(18*4), R13
MOVM.IA [R0-R14], (R13)
MOVW $MACH(0), R(Rmach) /* FIXME */
/* get Mach* from TPIDRPRW */
MRC 15, 0, R(Rmach), C(13), C(0), 4
MOVW 8(R(Rmach)), R(Rup)
MOVW $setR12(SB), R12
@ -79,7 +80,8 @@ TEXT _vsvc(SB), $-4
MOVM.DB.S [R0-R14], (R13)
SUB $(15*4), R13
MOVW $MACH(0), R(Rmach) /* FIXME */
/* get Mach* from TPIDRPRW */
MRC 15, 0, R(Rmach), C(13), C(0), 4
MOVW 8(R(Rmach)), R(Rup)
MOVW $setR12(SB), R12

View file

@ -182,9 +182,9 @@ confinit(void)
int i;
conf.nmach = 1;
conf.nproc = 100;
conf.nproc = 2000;
conf.ialloc = 16*1024*1024;
conf.nimage = conf.nproc;
conf.nimage = 200;
conf.mem[0].base = PGROUND((ulong)end - KZERO);
conf.mem[0].limit = 1024*1024*1024;
conf.npage = 0;
@ -346,10 +346,64 @@ isaconfig(char *, int, ISAConf*)
return 1;
}
void
cpuidprint(void)
{
print("\ncpu%d: %dMHz ARM Cortex-A9\n", m->machno, m->cpumhz);
}
void
mpinit(void)
{
extern void mpbootstrap(void); /* l.s */
Mach *m1;
ulong *v;
int i;
if(getconf("*nomp"))
return;
conf.nmach = 2;
conf.copymode = 1;
m1 = MACHP(1);
memset(m1, 0, MACHSIZE);
m1->machno = 1;
m1->l1.pa = MACHL1(m1->machno)-KZERO;
m1->l1.va = KADDR(m1->l1.pa);
memset(m1->l1.va, 0, L1SZ);
for(i=0; i<L1X(VMAPSZ); i++)
m1->l1.va[L1X(VMAP)+i] = m->l1.va[L1X(VMAP)+i];
for(i=0; i<L1X(-KZERO); i++)
m1->l1.va[L1X(KZERO)+i] = m->l1.va[L1X(KZERO)+i];
coherence();
cleandse((uchar*)KZERO, (uchar*)0xFFFFFFFF);
v = tmpmap(0xFFFFF000);
v[0xFF0/4] = PADDR(mpbootstrap);
coherence();
cleandse(v, (uchar*)v+BY2PG);
tmpunmap(v);
sendevent();
synccycles();
}
void
main(void)
{
active.machs = 1;
active.machs |= (1 << m->machno);
if(m->machno != 0){
mmuinit();
intrinit();
timerinit();
cpuidprint();
synccycles();
timersinit();
schedinit();
return;
}
uartinit();
mmuinit();
l2init();
@ -361,6 +415,7 @@ main(void)
xinit();
printinit();
quotefmtinstall();
cpuidprint();
sanity();
todinit();
timersinit();
@ -376,5 +431,6 @@ main(void)
swapinit();
screeninit();
userinit();
mpinit();
schedinit();
}

View file

@ -40,8 +40,8 @@
#define MACHSIZE 8192
#define MACH(n) (KZERO+(n)*MACHSIZE)
#define MACHP(n) ((Mach *)MACH(n))
#define CPU0L1 ROUND(MACH(MAXMACH), L1SZ)
#define VMAPL2 (CPU0L1 + L1SZ)
#define MACHL1(n) (ROUND(MACH(MAXMACH), L1SZ) + (n)*L1SZ)
#define VMAPL2 MACHL1(MAXMACH)
#define VMAPL2SZ (L2SZ * (VMAPSZ / SECSZ))
#define TMAPL2(n) (VMAPL2 + VMAPL2SZ + (n) * L2SZ)
#define TMAPL2SZ (MAXMACH * L2SZ)
@ -80,6 +80,7 @@
#define DSB WORD $0xf57ff04f
#define ISB WORD $0xf57ff06f
#define WFE WORD $0xe320f002
#define SEV WORD $0xe320f004
#define CPS(m) WORD $(0xf1000000|(m))
#define CPSMODE (1<<17)
#define CPSIE (3<<6|2<<18)
@ -121,4 +122,4 @@
#define L2WRITE 0
#define L2LOCAL (1<<11)
#define TTBATTR (1<<6|1<<3)
#define TTBATTR (1<<6|1<<3|1<<1)

View file

@ -12,10 +12,13 @@ mmuinit(void)
{
m->l1.pa = ttbget();
m->l1.va = KADDR(m->l1.pa);
mpcore = vmap(MPCORE_BASE, 0x2000);
slcr = vmap(SLCR_BASE, 0x1000);
memset((uchar*)TMAPL2(m->machno), 0, TMAPL2SZ);
m->l1.va[L1X(TMAP)] = PADDR(TMAPL2(m->machno)) | L1PT;
incref(&m->l1);
if(mpcore != nil)
return;
mpcore = vmap(MPCORE_BASE, 0x2000);
slcr = vmap(SLCR_BASE, 0x1000);
}
void
@ -38,29 +41,28 @@ l1alloc(void)
int s;
s = splhi();
if(m->l1free != nil){
p = m->l1free;
p = m->l1free;
if(p != nil){
m->l1free = p->next;
p->next = nil;
m->l1free = m->l1free->next;
m->nfree--;
splx(s);
return p;
}else{
p = smalloc(sizeof(L1));
for(;;){
p->va = mallocalign(L1SZ, L1SZ, 0, 0);
if(p->va != nil)
break;
if(!waserror()){
resrcwait("no memory for L1 table");
poperror();
}
}
memmove(p->va, m->l1.va, L1SZ);
p->pa = PADDR(p->va);
splx(s);
return p;
}
splx(s);
p = smalloc(sizeof(L1));
for(;;){
p->va = mallocalign(L1SZ, L1SZ, 0, 0);
if(p->va != nil)
break;
if(!waserror()){
resrcwait("no memory for L1 table");
poperror();
}
}
p->pa = PADDR(p->va);
memmove(p->va, m->l1.va, L1SZ);
return p;
}
static void
@ -89,11 +91,9 @@ upallocl1(void)
p = l1alloc();
s = splhi();
if(up->l1 != nil)
l1free(p);
else{
up->l1 = p;
l1switch(p, 1);
}
panic("upalloc1: up->l1 != nil");
up->l1 = p;
l1switch(p, 1);
splx(s);
}
@ -114,6 +114,7 @@ l2free(Proc *proc)
*t = 0;
l = &p->next;
}
proc->l1->va[L1X(TMAP)] = 0;
*l = proc->mmufree;
proc->mmufree = proc->mmuused;
proc->mmuused = 0;
@ -139,6 +140,7 @@ putmmu(uintptr va, uintptr pa, Page *pg)
ulong *e;
ulong *l2;
PTE old;
char *ctl;
uintptr l2p;
int s;
@ -178,17 +180,17 @@ putmmu(uintptr va, uintptr pa, Page *pg)
splx(s);
if((old & L2VALID) != 0)
flushpg((void *) va);
if(pg->cachectl[0] == PG_TXTFLUSH){
ctl = &pg->cachectl[m->machno];
if(*ctl == PG_TXTFLUSH){
cleandse((void *) va, (void *) (va + BY2PG));
invalise((void *) va, (void *) (va + BY2PG));
pg->cachectl[0] = PG_NOFLUSH;
*ctl = PG_NOFLUSH;
}
}
void
checkmmu(uintptr, uintptr)
{
print("checkmmu\n");
}
void
@ -286,7 +288,7 @@ kmap(Page *page)
e = &up->l1->va[L1X(KMAP)];
if((*e & 3) == 0){
if(up->kmaptable != nil)
panic("kmaptable");
panic("kmaptable != nil");
up->kmaptable = newpage(0, 0, 0);
s = splhi();
v = tmpmap(up->kmaptable->pa);
@ -300,7 +302,7 @@ kmap(Page *page)
return (KMap *) KMAP;
}
if(up->kmaptable == nil)
panic("kmaptable");
panic("kmaptable == nil");
e = (ulong *) (KMAP + NKMAP * BY2PG);
for(i = 0; i < NKMAP; i++)
if((e[i] & 3) == 0){
@ -338,7 +340,6 @@ void *
tmpmap(ulong pa)
{
ulong *u, *ub, *ue;
void *v;
if(islo())
panic("tmpmap: islow %#p", getcallerpc(&pa));
@ -349,9 +350,13 @@ tmpmap(ulong pa)
for(u = ub; u < ue; u++)
if((*u & 3) == 0){
*u = pa | L2VALID | L2CACHED | L2KERRW;
assert(m->l1.va[L1X(TMAP)] != 0);
if(up != nil && up->l1 != nil)
up->l1->va[L1X(TMAP)] = m->l1.va[L1X(TMAP)];
coherence();
v = (void *) ((u - ub) * BY2PG + TMAP);
return v;
return (void *) ((u - ub) * BY2PG + TMAP);
}
panic("tmpmap: full (pa=%#.8lux)", pa);
return nil;
@ -361,7 +366,7 @@ void
tmpunmap(void *v)
{
ulong *u;
if(v >= (void*) KZERO)
return;
if(v < (void*)TMAP || v >= (void*)(TMAP + TMAPSZ))

View file

@ -80,12 +80,39 @@ timerirq(Ureg *u, void *)
void
timerinit(void)
{
int mhz;
mhz = PS_CLK * (slcr[ARM_PLL_CTRL] >> 12 & 0x7f) / (slcr[ARM_CLK_CTRL] >> 8 & 0x3f);
timerhz = mhz * 500000;
m->cpumhz = PS_CLK * (slcr[ARM_PLL_CTRL] >> 12 & 0x7f) / (slcr[ARM_CLK_CTRL] >> 8 & 0x3f);
m->cpuhz = m->cpumhz * 1000000;
timerhz = m->cpuhz / 2;
mpcore[GTIMERCTL] = TIMERDIV - 1 << 8 | 3;
mpcore[LTIMERCTL] = LTIMERDIV - 1 << 8 | 4;
intrenable(TIMERIRQ, timerirq, nil, EDGE, "clock");
/* enable and reset cycle counter register */
m->cyclefreq = m->cpuhz;
setpmcnten((1<<31));
coherence();
setpmcr(7);
}
/*
* synchronize all cpu's cycle counter registers
*/
void
synccycles(void)
{
static Ref r1, r2;
int s;
s = splhi();
r2.ref = 0;
incref(&r1);
while(r1.ref != conf.nmach)
;
setpmcr(7);
m->cycleshi = MACHP(0)->cycleshi;
incref(&r2);
while(r2.ref != conf.nmach)
;
r1.ref = 0;
splx(s);
}

View file

@ -20,7 +20,7 @@ _dumpstack(Ureg *ureg)
iprint("dumpstack disabled\n");
return;
}
iprint("dumpstack\n");
iprint("cpu%d: dumpstack\n", m->machno);
x = 0;
x += iprint("ktrace /arm/9zynq %.8lux %.8lux %.8lux <<EOF\n", ureg->pc, ureg->sp, ureg->r14);
@ -57,7 +57,7 @@ faultarm(Ureg *ureg, ulong fsr, uintptr addr)
{
int user, insyscall, read, n;
static char buf[ERRMAX];
read = (fsr & (1<<11)) == 0;
user = userureg(ureg);
if(!user){
@ -88,18 +88,24 @@ faultarm(Ureg *ureg, ulong fsr, uintptr addr)
static void
mathtrap(Ureg *, ulong)
{
int s;
if((up->fpstate & FPillegal) != 0){
postnote(up, 1, "sys: floating point in note handler", NDebug);
return;
}
switch(up->fpstate){
case FPinit:
s = splhi();
fpinit();
up->fpstate = FPactive;
splx(s);
break;
case FPinactive:
s = splhi();
fprestore(&up->fpsave);
up->fpstate = FPactive;
splx(s);
break;
case FPactive:
postnote(up, 1, "sys: floating point error", NDebug);
@ -138,6 +144,7 @@ trap(Ureg *ureg)
postnote(up, 1, "sys: trap: invalid opcode", NDebug);
break;
}
dumpregs(ureg);
panic("invalid opcode at pc=%#.8lux lr=%#.8lux", ureg->pc, ureg->r14);
break;
case PsrMiabt:
@ -153,7 +160,7 @@ trap(Ureg *ureg)
intr(ureg);
break;
default:
print("unknown trap type %ulx\n", ureg->type);
iprint("cpu%d: unknown trap type %ulx\n", m->machno, ureg->type);
}
splhi();
if(user){
@ -408,9 +415,17 @@ dumpstack(void)
}
void
dumpregs(Ureg *)
dumpregs(Ureg *ureg)
{
print("dumpregs\n");
iprint("trap: %lux psr %8.8lux type %2.2lux pc %8.8lux link %8.8lux\n",
ureg->type, ureg->psr, ureg->type, ureg->pc, ureg->link);
iprint("R14 %8.8lux R13 %8.8lux R12 %8.8lux R11 %8.8lux R10 %8.8lux\n",
ureg->r14, ureg->r13, ureg->r12, ureg->r11, ureg->r10);
iprint("R9 %8.8lux R8 %8.8lux R7 %8.8lux R6 %8.8lux R5 %8.8lux\n",
ureg->r9, ureg->r8, ureg->r7, ureg->r6, ureg->r5);
iprint("R4 %8.8lux R3 %8.8lux R2 %8.8lux R1 %8.8lux R0 %8.8lux\n",
ureg->r4, ureg->r3, ureg->r2, ureg->r1, ureg->r0);
iprint("pc %#lux link %#lux\n", ureg->pc, ureg->link);
}
void
@ -476,7 +491,7 @@ procsave(Proc *p)
cycles(&t);
p->kentry -= t;
p->pcycles += t;
l1switch(&m->l1, 0);
}