Get rid of non-equivalent aliases of the pcb by moving the fpu state out

of the pcb and using the p_addr member of 'struct proc' to calculate the
address of the kernel stack when switching to virtual mode after taking a trap.
Remove the now unecessary cache flushes; they're actually harmful since they
create non-equivalent aliases.  This seems to fix the memory corruption we
have been observing from time to time.

This diff does not rename fpu_curpcb, which is now somewhat incorrectly named.
I hope to change things back again as soon as we are able to map the pcb 1:1.

8 files changed, 35 insertions, 67 deletions

diff --git a/sys/arch/hppa/hppa/genassym.cf b/sys/arch/hppa/hppa/genassym.cf
index aaa4058315c..f17aecfd78a 100644
--- a/sys/arch/hppa/hppa/genassym.cf
+++ b/sys/arch/hppa/hppa/genassym.cf
@@ -1,4 +1,4 @@
-#	$OpenBSD: genassym.cf,v 1.32 2009/02/14 19:03:50 kettenis Exp $
+#	$OpenBSD: genassym.cf,v 1.33 2009/07/29 18:31:11 kettenis Exp $
 
 #
 # Copyright (c) 1982, 1990, 1993
@@ -145,7 +145,6 @@ struct	pcb
 member	pcb_fpregs
 member	pcb_onfault
 member	pcb_space
-member	pcb_uva
 member	pcb_ksp
 
 struct	user
diff --git a/sys/arch/hppa/hppa/hpux_machdep.c b/sys/arch/hppa/hppa/hpux_machdep.c
index 351c3961bd6..ffcad83594b 100644
--- a/sys/arch/hppa/hppa/hpux_machdep.c
+++ b/sys/arch/hppa/hppa/hpux_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: hpux_machdep.c,v 1.2 2005/03/26 20:37:24 mickey Exp $	*/
+/*	$OpenBSD: hpux_machdep.c,v 1.3 2009/07/29 18:31:11 kettenis Exp $	*/
 
 /*
  * Copyright (c) 2005 Michael Shalayeff
@@ -115,11 +115,10 @@ hpux_setregs(struct proc *p, struct exec_package *pack, u_long stack,
 	copyout(&zero, (caddr_t)(stack + HPPA_FRAME_CRP), sizeof(register_t));
 
 	/* reset any of the pending FPU exceptions */
-	pcb->pcb_fpregs[0] = ((u_int64_t)HPPA_FPU_INIT) << 32;
-	pcb->pcb_fpregs[1] = 0;
-	pcb->pcb_fpregs[2] = 0;
-	pcb->pcb_fpregs[3] = 0;
-	fdcache(HPPA_SID_KERNEL, (vaddr_t)pcb->pcb_fpregs, 8 * 4);
+	pcb->pcb_fpregs->fpr_regs[0] = ((u_int64_t)HPPA_FPU_INIT) << 32;
+	pcb->pcb_fpregs->fpr_regs[1] = 0;
+	pcb->pcb_fpregs->fpr_regs[2] = 0;
+	pcb->pcb_fpregs->fpr_regs[3] = 0;
 	if (tf->tf_cr30 == fpu_curpcb) {
 		fpu_curpcb = 0;
 		/* force an fpu ctxsw, we won't be hugged by the cpu_switch */
diff --git a/sys/arch/hppa/hppa/locore.S b/sys/arch/hppa/hppa/locore.S
index 217add80e3c..e7b75483035 100644
--- a/sys/arch/hppa/hppa/locore.S
+++ b/sys/arch/hppa/hppa/locore.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: locore.S,v 1.160 2009/02/14 19:03:50 kettenis Exp $	*/
+/*	$OpenBSD: locore.S,v 1.161 2009/07/29 18:31:11 kettenis Exp $	*/
 
 /*
  * Copyright (c) 1998-2004 Michael Shalayeff
@@ -198,7 +198,6 @@ $start_zero_tf
 	mtctl	arg3, cr30
 	stw	r0, U_PCB+PCB_ONFAULT(arg3)
 	stw	r0, U_PCB+PCB_SPACE(arg3)	/* XXX HPPA_SID_KERNEL == 0 */
-	stw	arg3, U_PCB+PCB_UVA(arg3)
 	ldil	L%(USPACE+NBPG), arg0		/* normal U plus red zone */
 	add	arg0, arg3, arg0
 	ldil	L%proc0paddr, t1
@@ -1628,10 +1627,11 @@ ENTRY(TLABEL(all),0)
 	comb,<>	t1, t2, $trap_from_kernel
 	dep	r0, 31, 6, sp
 
-	mfctl	cr30, t2
+	ldil	L%curproc, t2
+	ldw	R%curproc(t2), t2
 	depi	1, T_USER_POS, 1, r1
 	depi	1, TFF_LAST_POS, 1, r1
-	ldw	U_PCB+PCB_UVA(t2), sp
+	ldw	P_ADDR(t2), sp
 	ldo	NBPG(sp), sp
 
 $trap_from_kernel
diff --git a/sys/arch/hppa/hppa/machdep.c b/sys/arch/hppa/hppa/machdep.c
index 843dac586e3..d7053f805f0 100644
--- a/sys/arch/hppa/hppa/machdep.c
+++ b/sys/arch/hppa/hppa/machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: machdep.c,v 1.173 2009/06/15 17:01:25 beck Exp $	*/
+/*	$OpenBSD: machdep.c,v 1.174 2009/07/29 18:31:11 kettenis Exp $	*/
 
 /*
  * Copyright (c) 1999-2003 Michael Shalayeff
@@ -164,6 +164,7 @@ paddr_t	avail_end;
 struct user *proc0paddr;
 long mem_ex_storage[EXTENT_FIXED_STORAGE_SIZE(64) / sizeof(long)];
 struct extent *hppa_ex;
+struct pool hppa_fppl;
 
 struct vm_map *exec_map = NULL;
 struct vm_map *phys_map = NULL;
@@ -421,6 +422,8 @@ hppa_init(start)
 #endif
 	ficacheall();
 	fdcacheall();
+
+	pool_init(&hppa_fppl, sizeof(struct fpreg), 16, 0, 0, "hppafp", NULL);
 }
 
 void
@@ -1233,11 +1236,10 @@ setregs(p, pack, stack, retval)
 		fpu_exit();
 		fpu_curpcb = 0;
 	}
-	pcb->pcb_fpregs[0] = ((u_int64_t)HPPA_FPU_INIT) << 32;
-	pcb->pcb_fpregs[1] = 0;
-	pcb->pcb_fpregs[2] = 0;
-	pcb->pcb_fpregs[3] = 0;
-	fdcache(HPPA_SID_KERNEL, (vaddr_t)pcb->pcb_fpregs, 8 * 4);
+	pcb->pcb_fpregs->fpr_regs[0] = ((u_int64_t)HPPA_FPU_INIT) << 32;
+	pcb->pcb_fpregs->fpr_regs[1] = 0;
+	pcb->pcb_fpregs->fpr_regs[2] = 0;
+	pcb->pcb_fpregs->fpr_regs[3] = 0;
 
 	retval[1] = 0;
 }
@@ -1456,8 +1458,6 @@ sys_sigreturn(p, v, retval)
 	tf->tf_r31 = ksc.sc_regs[31];
 	bcopy(ksc.sc_fpregs, p->p_addr->u_pcb.pcb_fpregs,
 	    sizeof(ksc.sc_fpregs));
-	fdcache(HPPA_SID_KERNEL, (vaddr_t)p->p_addr->u_pcb.pcb_fpregs,
-	    sizeof(ksc.sc_fpregs));
 
 	tf->tf_iioq_head = ksc.sc_pcoqh | HPPA_PC_PRIV_USER;
 	tf->tf_iioq_tail = ksc.sc_pcoqt | HPPA_PC_PRIV_USER;
diff --git a/sys/arch/hppa/hppa/pmap.c b/sys/arch/hppa/hppa/pmap.c
index 70f4ff147f7..a2829a528c6 100644
--- a/sys/arch/hppa/hppa/pmap.c
+++ b/sys/arch/hppa/hppa/pmap.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: pmap.c,v 1.140 2009/07/26 15:47:23 kettenis Exp $	*/
+/*	$OpenBSD: pmap.c,v 1.141 2009/07/29 18:31:11 kettenis Exp $	*/
 
 /*
  * Copyright (c) 1998-2004 Michael Shalayeff
@@ -1151,8 +1151,6 @@ pmap_activate(struct proc *p)
 	struct pcb *pcb = &p->p_addr->u_pcb;
 
 	pcb->pcb_space = pmap->pm_space;
-	pcb->pcb_uva = (vaddr_t)p->p_addr;
-	fdcache(HPPA_SID_KERNEL, (vaddr_t)pcb, PAGE_SIZE);
 }
 
 void
diff --git a/sys/arch/hppa/hppa/process_machdep.c b/sys/arch/hppa/hppa/process_machdep.c
index ed329879051..192bf61d72c 100644
--- a/sys/arch/hppa/hppa/process_machdep.c
+++ b/sys/arch/hppa/hppa/process_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: process_machdep.c,v 1.14 2007/07/20 20:52:51 kettenis Exp $	*/
+/*	$OpenBSD: process_machdep.c,v 1.15 2009/07/29 18:31:11 kettenis Exp $	*/
 
 /*
  * Copyright (c) 1999-2004 Michael Shalayeff
@@ -160,7 +160,6 @@ process_write_fpregs(p, fpregs)
 	}
 
 	bcopy(fpregs, p->p_addr->u_pcb.pcb_fpregs, 32 * 8);
-	fdcache(HPPA_SID_KERNEL, (vaddr_t)p->p_addr->u_pcb.pcb_fpregs, 32 * 8);
 
 	return (0);
 }
diff --git a/sys/arch/hppa/hppa/vm_machdep.c b/sys/arch/hppa/hppa/vm_machdep.c
index b93d42f3102..3b67e465cd8 100644
--- a/sys/arch/hppa/hppa/vm_machdep.c
+++ b/sys/arch/hppa/hppa/vm_machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vm_machdep.c,v 1.67 2009/06/26 18:55:20 miod Exp $	*/
+/*	$OpenBSD: vm_machdep.c,v 1.68 2009/07/29 18:31:11 kettenis Exp $	*/
 
 /*
  * Copyright (c) 1999-2004 Michael Shalayeff
@@ -38,6 +38,7 @@
 #include <sys/ptrace.h>
 #include <sys/exec.h>
 #include <sys/core.h>
+#include <sys/pool.h>
 
 #include <machine/cpufunc.h>
 #include <machine/pmap.h>
@@ -45,6 +46,7 @@
 
 #include <uvm/uvm.h>
 
+extern struct pool hppa_fppl;
 
 /*
  * Dump the machine specific header information at the start of a core dump.
@@ -103,7 +105,6 @@ cpu_fork(p1, p2, stack, stacksize, func, arg)
 	struct pcb *pcbp;
 	struct trapframe *tf;
 	register_t sp, osp;
-	paddr_t pa;
 
 #ifdef DIAGNOSTIC
 	if (round_page(sizeof(struct user)) > NBPG)
@@ -115,38 +116,25 @@ cpu_fork(p1, p2, stack, stacksize, func, arg)
 		mtctl(0, CR_CCR);
 	}
 
-	/*
-	 * Stash the physical for the pcb of U for later perusal
-	 */
-	if (!pmap_extract(pmap_kernel(), (vaddr_t)p2->p_addr, &pa))
-		panic("pmap_extract(%p) failed", p2->p_addr);
-
-	/*
-	 * XXX This shouldn't be necessary, since there shouldn't be any
-	 * allocated cache lines for this new pcb.  But for some reason
-	 * that is not the case...
-	 */
-	fdcache(HPPA_SID_KERNEL, pa, PAGE_SIZE);
-	pdtlb(HPPA_SID_KERNEL, pa);
-	pitlb(HPPA_SID_KERNEL, pa);
-
 	pcbp = &p2->p_addr->u_pcb;
 	bcopy(&p1->p_addr->u_pcb, pcbp, sizeof(*pcbp));
 	/* space is cached for the copy{in,out}'s pleasure */
 	pcbp->pcb_space = p2->p_vmspace->vm_map.pmap->pm_space;
-	pcbp->pcb_uva = (vaddr_t)p2->p_addr;
+	pcbp->pcb_fpregs = pool_get(&hppa_fppl, PR_WAITOK);
+	*pcbp->pcb_fpregs = *p1->p_addr->u_pcb.pcb_fpregs;
 	/* reset any of the pending FPU exceptions from parent */
-	pcbp->pcb_fpregs[0] = HPPA_FPU_FORK(pcbp->pcb_fpregs[0]);
-	pcbp->pcb_fpregs[1] = 0;
-	pcbp->pcb_fpregs[2] = 0;
-	pcbp->pcb_fpregs[3] = 0;
+	pcbp->pcb_fpregs->fpr_regs[0] =
+	    HPPA_FPU_FORK(pcbp->pcb_fpregs->fpr_regs[0]);
+	pcbp->pcb_fpregs->fpr_regs[1] = 0;
+	pcbp->pcb_fpregs->fpr_regs[2] = 0;
+	pcbp->pcb_fpregs->fpr_regs[3] = 0;
 
 	sp = (register_t)p2->p_addr + NBPG;
 	p2->p_md.md_regs = tf = (struct trapframe *)sp;
 	sp += sizeof(struct trapframe);
 	bcopy(p1->p_md.md_regs, tf, sizeof(*tf));
 
-	tf->tf_cr30 = pa;
+	tf->tf_cr30 = (paddr_t)pcbp->pcb_fpregs;
 
 	tf->tf_sr0 = tf->tf_sr1 = tf->tf_sr2 = tf->tf_sr3 =
 	tf->tf_sr4 = tf->tf_sr5 = tf->tf_sr6 =
@@ -181,11 +169,6 @@ cpu_fork(p1, p2, stack, stacksize, func, arg)
 	*HPPA_FRAME_CARG(0, sp) = (register_t)arg;
 	*HPPA_FRAME_CARG(1, sp) = KERNMODE(func);
 	pcbp->pcb_ksp = sp;
-
-	fdcache(HPPA_SID_KERNEL, (vaddr_t)p2->p_addr, PAGE_SIZE);
-	pdtlb(HPPA_SID_KERNEL, (vaddr_t)p2->p_addr);
-	ficache(HPPA_SID_KERNEL, (vaddr_t)p2->p_addr, PAGE_SIZE);
-	pitlb(HPPA_SID_KERNEL, (vaddr_t)p2->p_addr);
 }
 
 void
@@ -194,11 +177,13 @@ cpu_exit(p)
 {
 	extern paddr_t fpu_curpcb;	/* from locore.S */
 	struct trapframe *tf = p->p_md.md_regs;
+	struct pcb *pcb = &p->p_addr->u_pcb;
 
 	if (fpu_curpcb == tf->tf_cr30) {
 		fpu_exit();
 		fpu_curpcb = 0;
 	}
+	pool_put(&hppa_fppl, pcb->pcb_fpregs);
 
 	pmap_deactivate(p);
 	sched_exit(p);
diff --git a/sys/arch/hppa/include/pcb.h b/sys/arch/hppa/include/pcb.h
index 68d7a0ba384..d6032d20412 100644
--- a/sys/arch/hppa/include/pcb.h
+++ b/sys/arch/hppa/include/pcb.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: pcb.h,v 1.11 2007/07/29 20:15:56 kettenis Exp $	*/
+/*	$OpenBSD: pcb.h,v 1.12 2009/07/29 18:31:11 kettenis Exp $	*/
 
 /*
  * Copyright (c) 1999-2004 Michael Shalayeff
@@ -33,19 +33,7 @@
 #include <machine/reg.h>
 
 struct pcb {
-	u_int64_t pcb_fpregs[HPPA_NFPREGS+1];	/* not in the trapframe */
-	vaddr_t		pcb_uva;	/* KVA for U-area */
-
-	/*
-	 * The members above are primarily accessed by there physical
-	 * address, wheras the members below are accessed exclusively
-	 * by there virtual address.  Unfortunately this structure
-	 * ends up being non-equivalently mapped, which will cause
-	 * data corruption if those members share a cache line.  Since
-	 * the maximum cache line size is 64 bytes, adding 64 bytes of
-	 * padding makes sure that will never happen.
-	 */
-	u_char		pcb_pad[64];
+	struct fpreg	*pcb_fpregs;	/* not in the trapframe */
 
 	u_int		pcb_ksp;	/* kernel sp for ctxsw */
 	u_int		pcb_onfault;	/* SW copy fault handler */