Module Name: src
Committed By: reinoud
Date: Sat Jan 14 17:42:52 UTC 2012
Modified Files:
src/sys/arch/usermode/dev: cpu.c
src/sys/arch/usermode/include: pcb.h
src/sys/arch/usermode/target/i386: cpu_i386.c
src/sys/arch/usermode/target/x86_64: cpu_x86_64.c
src/sys/arch/usermode/usermode: copy.c pmap.c syscall.c thunk.c trap.c
Log Message:
Revamp the NetBSD/usermode pagefault and illegal instruction handing. It now
can handle recursive entry and is a lot more memory tight compared to the old
implementation. Performance wise:
* slightly less number of syscalls/sec possible though could be optimized
* a lot faster context creation / destruction making overall operation faster.
To generate a diff of this commit:
cvs rdiff -u -r1.63 -r1.64 src/sys/arch/usermode/dev/cpu.c
cvs rdiff -u -r1.16 -r1.17 src/sys/arch/usermode/include/pcb.h
cvs rdiff -u -r1.1 -r1.2 src/sys/arch/usermode/target/i386/cpu_i386.c
cvs rdiff -u -r1.1 -r1.2 src/sys/arch/usermode/target/x86_64/cpu_x86_64.c
cvs rdiff -u -r1.6 -r1.7 src/sys/arch/usermode/usermode/copy.c
cvs rdiff -u -r1.101 -r1.102 src/sys/arch/usermode/usermode/pmap.c
cvs rdiff -u -r1.20 -r1.21 src/sys/arch/usermode/usermode/syscall.c
cvs rdiff -u -r1.77 -r1.78 src/sys/arch/usermode/usermode/thunk.c
cvs rdiff -u -r1.55 -r1.56 src/sys/arch/usermode/usermode/trap.c
Please note that diffs are not public domain; they are subject to the
copyright notices on the relevant files.
Modified files:
Index: src/sys/arch/usermode/dev/cpu.c
diff -u src/sys/arch/usermode/dev/cpu.c:1.63 src/sys/arch/usermode/dev/cpu.c:1.64
--- src/sys/arch/usermode/dev/cpu.c:1.63 Thu Jan 12 13:28:54 2012
+++ src/sys/arch/usermode/dev/cpu.c Sat Jan 14 17:42:51 2012
@@ -1,4 +1,4 @@
-/* $NetBSD: cpu.c,v 1.63 2012/01/12 13:28:54 reinoud Exp $ */
+/* $NetBSD: cpu.c,v 1.64 2012/01/14 17:42:51 reinoud Exp $ */
/*-
* Copyright (c) 2007 Jared D. McNeill <[email protected]>
@@ -30,7 +30,7 @@
#include "opt_hz.h"
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: cpu.c,v 1.63 2012/01/12 13:28:54 reinoud Exp $");
+__KERNEL_RCSID(0, "$NetBSD: cpu.c,v 1.64 2012/01/14 17:42:51 reinoud Exp $");
#include <sys/param.h>
#include <sys/conf.h>
@@ -78,9 +78,13 @@ typedef struct cpu_softc {
struct cpu_info *sc_ci;
} cpu_softc_t;
+
+/* statics */
static struct pcb lwp0pcb;
static void *um_msgbuf;
+
+/* attachment */
CFATTACH_DECL_NEW(cpu, sizeof(cpu_softc_t), cpu_match, cpu_attach, NULL, NULL);
static int
@@ -115,6 +119,8 @@ cpu_configure(void)
spl0();
}
+
+/* main guts */
void
cpu_reboot(int howto, char *bootstr)
{
@@ -226,7 +232,7 @@ cpu_getmcontext(struct lwp *l, mcontext_
{
struct pcb *pcb = lwp_getpcb(l);
ucontext_t *ucp = &pcb->pcb_userret_ucp;
-
+
#ifdef CPU_DEBUG
thunk_printf_debug("cpu_getmcontext\n");
#endif
@@ -277,25 +283,7 @@ cpu_lwp_free2(struct lwp *l)
if (pcb == NULL)
return;
-
- if (pcb->pcb_stack_userland) {
- free(pcb->pcb_stack_userland, M_TEMP);
- pcb->pcb_stack_userland = NULL;
- pcb->pcb_ucp.uc_stack.ss_sp = NULL;
- pcb->pcb_ucp.uc_stack.ss_size = 0;
- }
-
- if (pcb->pcb_stack_syscall) {
- free(pcb->pcb_stack_syscall, M_TEMP);
- pcb->pcb_syscall_ucp.uc_stack.ss_sp = NULL;
- pcb->pcb_syscall_ucp.uc_stack.ss_size = 0;
- }
-
- if (pcb->pcb_stack_pagefault) {
- free(pcb->pcb_stack_pagefault, M_TEMP);
- pcb->pcb_pagefault_ucp.uc_stack.ss_sp = NULL;
- pcb->pcb_pagefault_ucp.uc_stack.ss_size = 0;
- }
+ /* XXX nothing to do? */
}
static void
@@ -332,42 +320,21 @@ cpu_lwp_fork(struct lwp *l1, struct lwp
/* copy the PCB and its switchframes from parent */
memcpy(pcb2, pcb1, sizeof(struct pcb));
- stacksize = 2*PAGE_SIZE;
- pcb2->pcb_stack_userland = malloc(stacksize, M_TEMP, M_WAITOK | M_ZERO);
- pcb2->pcb_stack_syscall = malloc(stacksize, M_TEMP, M_WAITOK | M_ZERO);
- pcb2->pcb_stack_pagefault = malloc(stacksize, M_TEMP, M_WAITOK | M_ZERO);
-
- KASSERT(pcb2->pcb_stack_userland);
- KASSERT(pcb2->pcb_stack_syscall);
- KASSERT(pcb2->pcb_stack_pagefault);
-
+ /* refresh context */
if (thunk_getcontext(&pcb2->pcb_ucp))
panic("getcontext failed");
- /* set up the ucontext for the userland switch */
- pcb2->pcb_ucp.uc_stack.ss_sp = pcb2->pcb_stack_userland;
- pcb2->pcb_ucp.uc_stack.ss_size = stacksize;
+ /* recalculate the system stack top */
+ pcb2->sys_stack_top = pcb2->sys_stack + TRAPSTACKSIZE;
+
+ /* get l2 its own stack */
+ pcb2->pcb_ucp.uc_stack.ss_sp = pcb2->sys_stack;
+ pcb2->pcb_ucp.uc_stack.ss_size = pcb2->sys_stack_top - pcb2->sys_stack;
pcb2->pcb_ucp.uc_flags = _UC_STACK | _UC_CPU;
pcb2->pcb_ucp.uc_link = &pcb2->pcb_userret_ucp;
thunk_makecontext(&pcb2->pcb_ucp,
(void (*)(void)) cpu_lwp_trampoline,
3, &pcb2->pcb_ucp, func, arg);
-
- /* set up the ucontext for the syscall */
- pcb2->pcb_syscall_ucp.uc_stack.ss_sp = pcb2->pcb_stack_syscall;
- pcb2->pcb_syscall_ucp.uc_stack.ss_size = stacksize;
- pcb2->pcb_syscall_ucp.uc_flags = _UC_STACK | _UC_CPU;
- pcb2->pcb_syscall_ucp.uc_link = &pcb2->pcb_userret_ucp;
- thunk_makecontext(&pcb2->pcb_syscall_ucp, (void (*)(void)) syscall,
- 0, NULL, NULL, NULL);
-
- /* set up the ucontext for the pagefault */
- pcb2->pcb_pagefault_ucp.uc_stack.ss_sp = pcb2->pcb_stack_pagefault;
- pcb2->pcb_pagefault_ucp.uc_stack.ss_size = stacksize;
- pcb2->pcb_pagefault_ucp.uc_flags = _UC_STACK | _UC_CPU;
- pcb2->pcb_pagefault_ucp.uc_link = &pcb2->pcb_trapret_ucp;
- thunk_makecontext(&pcb2->pcb_pagefault_ucp, (void (*)(void)) pagefault,
- 0, NULL, NULL, NULL);
}
void
@@ -382,7 +349,7 @@ void
cpu_startup(void)
{
vaddr_t minaddr, maxaddr;
- size_t stacksize, msgbufsize = 32 * 1024;
+ size_t msgbufsize = 32 * 1024;
/* get ourself a message buffer */
um_msgbuf = kmem_zalloc(msgbufsize, KM_SLEEP);
@@ -400,26 +367,12 @@ cpu_startup(void)
/* init lwp0 */
memset(&lwp0pcb, 0, sizeof(lwp0pcb));
- if (thunk_getcontext(&lwp0pcb.pcb_ucp))
- panic("getcontext failed");
- uvm_lwp_setuarea(&lwp0, (vaddr_t)&lwp0pcb);
+ thunk_getcontext(&lwp0pcb.pcb_ucp);
+ uvm_lwp_setuarea(&lwp0, (vaddr_t) &lwp0pcb);
+ memcpy(&lwp0pcb.pcb_userret_ucp, &lwp0pcb.pcb_ucp, sizeof(ucontext_t));
- /* init trapframes (going nowhere!), maybe a panic func? */
- memcpy(&lwp0pcb.pcb_syscall_ucp, &lwp0pcb.pcb_ucp, sizeof(ucontext_t));
- memcpy(&lwp0pcb.pcb_userret_ucp, &lwp0pcb.pcb_ucp, sizeof(ucontext_t));
- memcpy(&lwp0pcb.pcb_pagefault_ucp, &lwp0pcb.pcb_ucp, sizeof(ucontext_t));
- memcpy(&lwp0pcb.pcb_trapret_ucp, &lwp0pcb.pcb_ucp, sizeof(ucontext_t));
-
- /* set up the ucontext for the pagefault */
- stacksize = 2*PAGE_SIZE;
- lwp0pcb.pcb_stack_pagefault = malloc(stacksize, M_TEMP, M_WAITOK | M_ZERO);
-
- lwp0pcb.pcb_pagefault_ucp.uc_stack.ss_sp = lwp0pcb.pcb_stack_pagefault;
- lwp0pcb.pcb_pagefault_ucp.uc_stack.ss_size = stacksize;
- lwp0pcb.pcb_pagefault_ucp.uc_flags = _UC_STACK | _UC_CPU;
- lwp0pcb.pcb_pagefault_ucp.uc_link = &lwp0pcb.pcb_userret_ucp;
- thunk_makecontext(&lwp0pcb.pcb_pagefault_ucp, (void (*)(void)) pagefault,
- 0, NULL, NULL, NULL);
+ /* set stack top */
+ lwp0pcb.sys_stack_top = lwp0pcb.sys_stack + TRAPSTACKSIZE;
}
void
Index: src/sys/arch/usermode/include/pcb.h
diff -u src/sys/arch/usermode/include/pcb.h:1.16 src/sys/arch/usermode/include/pcb.h:1.17
--- src/sys/arch/usermode/include/pcb.h:1.16 Fri Jan 6 12:54:59 2012
+++ src/sys/arch/usermode/include/pcb.h Sat Jan 14 17:42:51 2012
@@ -1,4 +1,4 @@
-/* $NetBSD: pcb.h,v 1.16 2012/01/06 12:54:59 reinoud Exp $ */
+/* $NetBSD: pcb.h,v 1.17 2012/01/14 17:42:51 reinoud Exp $ */
/*-
* Copyright (c) 2007 Jared D. McNeill <[email protected]>
@@ -31,30 +31,20 @@
#include <sys/cdefs.h>
#include <sys/ucontext.h>
+#include <sys/queue.h>
-/*
- * Trap frame. Pushed onto the kernel stack on a trap (synchronous exception).
- * XXX move to frame.h?
- */
-
-//typedef ucontext_t trapframe;
+#define TRAPSTACKSIZE (USPACE -2*sizeof(ucontext_t) - 3*sizeof(register_t))
struct pcb {
- void *pcb_stack_userland;
- void *pcb_stack_syscall;
- void *pcb_stack_pagefault;
-
- ucontext_t pcb_ucp; /* lwp switchframe */
- ucontext_t pcb_syscall_ucp; /* syscall context */
- ucontext_t pcb_userret_ucp; /* return to userland context */
- ucontext_t pcb_pagefault_ucp; /* pagefault context */
- ucontext_t pcb_trapret_ucp;
-
- void * pcb_onfault; /* on fault handler */
-
- int pcb_errno; /* save/restore place */
- vaddr_t pcb_fault_addr; /* save place for fault addr */
- vaddr_t pcb_fault_pc; /* save place for fault PC */
+ ucontext_t pcb_ucp; /* switchframe */
+ ucontext_t pcb_userret_ucp;
+
+ uint8_t *sys_stack_top; /* points at free point in sys_stack */
+ uint8_t sys_stack[TRAPSTACKSIZE];
+
+ void *pcb_onfault; /* on fault handler */
+
+ int pcb_errno; /* save/restore place */
};
#endif /* !_ARCH_USERMODE_INCLUDE_PCB_H */
Index: src/sys/arch/usermode/target/i386/cpu_i386.c
diff -u src/sys/arch/usermode/target/i386/cpu_i386.c:1.1 src/sys/arch/usermode/target/i386/cpu_i386.c:1.2
--- src/sys/arch/usermode/target/i386/cpu_i386.c:1.1 Sat Jan 7 20:07:01 2012
+++ src/sys/arch/usermode/target/i386/cpu_i386.c Sat Jan 14 17:42:52 2012
@@ -1,4 +1,4 @@
-/* $NetBSD: cpu_i386.c,v 1.1 2012/01/07 20:07:01 reinoud Exp $ */
+/* $NetBSD: cpu_i386.c,v 1.2 2012/01/14 17:42:52 reinoud Exp $ */
/*-
* Copyright (c) 2011 Reinoud Zandijk <[email protected]>
@@ -34,7 +34,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: cpu_i386.c,v 1.1 2012/01/07 20:07:01 reinoud Exp $");
+__KERNEL_RCSID(0, "$NetBSD: cpu_i386.c,v 1.2 2012/01/14 17:42:52 reinoud Exp $");
#include <sys/types.h>
#include <sys/systm.h>
@@ -55,6 +55,8 @@ __KERNEL_RCSID(0, "$NetBSD: cpu_i386.c,v
#include <machine/machdep.h>
#include <machine/thunk.h>
+#include "opt_exec.h"
+
#if 0
static void dump_regs(register_t *reg);;
@@ -108,16 +110,18 @@ sendsig_siginfo(const ksiginfo_t *ksi, c
struct sigframe_siginfo *fp, frame;
int sig = ksi->ksi_signo;
sig_t catcher = SIGACTION(p, sig).sa_handler;
- ucontext_t *ucp = &pcb->pcb_userret_ucp;
- register_t *reg = (register_t *) &ucp->uc_mcontext;
+ ucontext_t *ucp;
+ register_t *reg;
int onstack, error;
KASSERT(mutex_owned(p->p_lock));
+ ucp = &pcb->pcb_userret_ucp;
+ reg = (register_t *) &ucp->uc_mcontext;
#if 0
- printf("%s: ", __func__);
- printf("flags %d, ", (int) ksi->ksi_flags);
- printf("to lwp %d, signo %d, code %d, errno %d\n",
+ thunk_printf("%s: ", __func__);
+ thunk_printf("flags %d, ", (int) ksi->ksi_flags);
+ thunk_printf("to lwp %d, signo %d, code %d, errno %d\n",
(int) ksi->ksi_lid,
ksi->ksi_signo,
ksi->ksi_code,
@@ -179,7 +183,7 @@ void
setregs(struct lwp *l, struct exec_package *pack, vaddr_t stack)
{
struct pcb *pcb = lwp_getpcb(l);
- ucontext_t *ucp = &pcb->pcb_userret_ucp;
+ ucontext_t *ucp;
uint *reg, i;
#ifdef DEBUG_EXEC
@@ -190,16 +194,15 @@ setregs(struct lwp *l, struct exec_packa
pcb->pcb_ucp.uc_stack.ss_sp);
printf("\tpcb->pcb_ucp.uc_stack.ss_size = %d\n",
(int) pcb->pcb_ucp.uc_stack.ss_size);
- printf("\tpcb->pcb_userret_ucp.uc_stack.ss_sp = %p\n",
- pcb->pcb_userret_ucp.uc_stack.ss_sp);
- printf("\tpcb->pcb_userret_ucp.uc_stack.ss_size = %d\n",
- (int) pcb->pcb_userret_ucp.uc_stack.ss_size);
#endif
+ /* set up the user context */
+ ucp = &pcb->pcb_userret_ucp;
reg = (int *) &ucp->uc_mcontext;
for (i = 4; i < 11; i++)
reg[i] = 0;
+ /* use given stack */
ucp->uc_stack.ss_sp = (void *) (stack-4); /* to prevent clearing */
ucp->uc_stack.ss_size = 0; //pack->ep_ssize;
thunk_makecontext(ucp, (void *) pack->ep_entry, 0, NULL, NULL, NULL);
@@ -216,10 +219,6 @@ setregs(struct lwp *l, struct exec_packa
pcb->pcb_ucp.uc_stack.ss_sp);
printf("\tpcb->pcb_ucp.uc_stack.ss_size = %d\n",
(int) pcb->pcb_ucp.uc_stack.ss_size);
- printf("\tpcb->pcb_userret_ucp.uc_stack.ss_sp = %p\n",
- pcb->pcb_userret_ucp.uc_stack.ss_sp);
- printf("\tpcb->pcb_userret_ucp.uc_stack.ss_size = %d\n",
- (int) pcb->pcb_userret_ucp.uc_stack.ss_size);
printf("\tpack->ep_entry = %p\n",
(void *) pack->ep_entry);
#endif
@@ -274,6 +273,14 @@ md_get_pc(ucontext_t *ucp)
return reg[14]; /* EIP */
}
+register_t
+md_get_sp(ucontext_t *ucp)
+{
+ register_t *reg = (register_t *) &ucp->uc_mcontext;
+
+ return reg[17]; /* ESP */
+}
+
int
md_syscall_check_opcode(ucontext_t *ucp)
{
Index: src/sys/arch/usermode/target/x86_64/cpu_x86_64.c
diff -u src/sys/arch/usermode/target/x86_64/cpu_x86_64.c:1.1 src/sys/arch/usermode/target/x86_64/cpu_x86_64.c:1.2
--- src/sys/arch/usermode/target/x86_64/cpu_x86_64.c:1.1 Sat Jan 7 20:44:42 2012
+++ src/sys/arch/usermode/target/x86_64/cpu_x86_64.c Sat Jan 14 17:42:52 2012
@@ -1,4 +1,4 @@
-/* $NetBSD: cpu_x86_64.c,v 1.1 2012/01/07 20:44:42 reinoud Exp $ */
+/* $NetBSD: cpu_x86_64.c,v 1.2 2012/01/14 17:42:52 reinoud Exp $ */
/*-
* Copyright (c) 2011 Reinoud Zandijk <[email protected]>
@@ -35,7 +35,7 @@
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: cpu_x86_64.c,v 1.1 2012/01/07 20:44:42 reinoud Exp $");
+__KERNEL_RCSID(0, "$NetBSD: cpu_x86_64.c,v 1.2 2012/01/14 17:42:52 reinoud Exp $");
#include <sys/types.h>
#include <sys/systm.h>
@@ -110,13 +110,15 @@ sendsig_siginfo(const ksiginfo_t *ksi, c
struct sigframe_siginfo *fp, frame;
int sig = ksi->ksi_signo;
sig_t catcher = SIGACTION(p, sig).sa_handler;
- ucontext_t *ucp = &pcb->pcb_userret_ucp;
- register_t *reg = (register_t *) &ucp->uc_mcontext;
+ ucontext_t *ucp;
+ register_t *reg;
int onstack, error;
char *sp;
KASSERT(mutex_owned(p->p_lock));
+ ucp = &pcb->pcb_userret_ucp;
+ reg = (register_t *) &ucp->uc_mcontext;
#if 0
thunk_printf("%s: ", __func__);
thunk_printf("flags %d, ", (int) ksi->ksi_flags);
@@ -187,10 +189,12 @@ void
setregs(struct lwp *l, struct exec_package *pack, vaddr_t stack)
{
struct pcb *pcb = lwp_getpcb(l);
- ucontext_t *ucp = &pcb->pcb_userret_ucp;
+ ucontext_t *ucp;
register_t *reg;
int i;
+ /* set up the user context */
+ ucp = &pcb->pcb_userret_ucp;
reg = (register_t *) &ucp->uc_mcontext;
for (i = 0; i < 15; i++)
reg[i] = 0;
@@ -199,6 +203,7 @@ setregs(struct lwp *l, struct exec_packa
reg[21] = pack->ep_entry; /* RIP */
reg[24] = (register_t) stack; /* RSP */
+ /* use given stack */
ucp->uc_stack.ss_sp = (void *) stack;
ucp->uc_stack.ss_size = pack->ep_ssize;
@@ -270,6 +275,14 @@ md_get_pc(ucontext_t *ucp)
return reg[21]; /* RIP */
}
+register_t
+md_get_sp(ucontext_t *ucp)
+{
+ register_t *reg = (register_t *) &ucp->uc_mcontext;
+
+ return reg[24]; /* RSP */
+}
+
int
md_syscall_check_opcode(ucontext_t *ucp)
{
Index: src/sys/arch/usermode/usermode/copy.c
diff -u src/sys/arch/usermode/usermode/copy.c:1.6 src/sys/arch/usermode/usermode/copy.c:1.7
--- src/sys/arch/usermode/usermode/copy.c:1.6 Fri Dec 9 17:24:25 2011
+++ src/sys/arch/usermode/usermode/copy.c Sat Jan 14 17:42:52 2012
@@ -1,4 +1,4 @@
-/* $NetBSD: copy.c,v 1.6 2011/12/09 17:24:25 reinoud Exp $ */
+/* $NetBSD: copy.c,v 1.7 2012/01/14 17:42:52 reinoud Exp $ */
/*-
* Copyright (c) 2007 Jared D. McNeill <[email protected]>
@@ -27,10 +27,11 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: copy.c,v 1.6 2011/12/09 17:24:25 reinoud Exp $");
+__KERNEL_RCSID(0, "$NetBSD: copy.c,v 1.7 2012/01/14 17:42:52 reinoud Exp $");
#include <sys/types.h>
#include <sys/systm.h>
+#include <machine/thunk.h>
/* XXX until strnlen(3) has been added to the kernel, we *could* panic on it */
#define strnlen(str, maxlen) min(strlen((str)), maxlen)
@@ -38,7 +39,7 @@ __KERNEL_RCSID(0, "$NetBSD: copy.c,v 1.6
int
copyin(const void *uaddr, void *kaddr, size_t len)
{
-// aprint_debug("copyin uaddr %p, kaddr %p, len %d\n", uaddr, kaddr, (int) len);
+// thunk_printf("copyin uaddr %p, kaddr %p, len %d\n", uaddr, kaddr, (int) len);
memcpy(kaddr, uaddr, len);
return 0;
}
@@ -46,7 +47,7 @@ copyin(const void *uaddr, void *kaddr, s
int
copyout(const void *kaddr, void *uaddr, size_t len)
{
-// aprint_debug("copyout kaddr %p, uaddr %p, len %d\n", kaddr, uaddr, (int) len);
+// thunk_printf("copyout kaddr %p, uaddr %p, len %d\n", kaddr, uaddr, (int) len);
memcpy(uaddr, kaddr, len);
return 0;
}
Index: src/sys/arch/usermode/usermode/pmap.c
diff -u src/sys/arch/usermode/usermode/pmap.c:1.101 src/sys/arch/usermode/usermode/pmap.c:1.102
--- src/sys/arch/usermode/usermode/pmap.c:1.101 Tue Jan 10 12:07:17 2012
+++ src/sys/arch/usermode/usermode/pmap.c Sat Jan 14 17:42:52 2012
@@ -1,4 +1,4 @@
-/* $NetBSD: pmap.c,v 1.101 2012/01/10 12:07:17 reinoud Exp $ */
+/* $NetBSD: pmap.c,v 1.102 2012/01/14 17:42:52 reinoud Exp $ */
/*-
* Copyright (c) 2011 Reinoud Zandijk <[email protected]>
@@ -27,7 +27,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.101 2012/01/10 12:07:17 reinoud Exp $");
+__KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.102 2012/01/14 17:42:52 reinoud Exp $");
#include "opt_memsize.h"
#include "opt_kmempages.h"
@@ -41,6 +41,7 @@ __KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.1
#include <sys/pool.h>
#include <machine/thunk.h>
#include <machine/machdep.h>
+#include <machine/pcb.h>
#include <uvm/uvm.h>
@@ -194,6 +195,16 @@ pmap_bootstrap(void)
aprint_verbose("\n\n");
+ /* make critical assertions before modifying anything */
+ if (sizeof(struct pcb) > USPACE) {
+ panic("sizeof(struct pcb) is %d bytes too big for USPACE. "
+ "Please adjust TRAPSTACKSIZE calculation",
+ (int) (USPACE - sizeof(struct pcb)));
+ }
+ if (TRAPSTACKSIZE < 4*PAGE_SIZE) {
+ panic("TRAPSTACKSIZE is too small, please increase UPAGES");
+ }
+
/* protect user memory UVM area (---) */
err = thunk_munmap((void *) kmem_user_start,
kmem_k_start - kmem_user_start);
@@ -625,11 +636,8 @@ pmap_lookup_pv(pmap_t pmap, uintptr_t lp
int l1 = lpn / PMAP_L2_NENTRY;
int l2 = lpn % PMAP_L2_NENTRY;
-#ifdef DIAGNOSTIC
if (lpn >= pm_nentries)
- panic("peeing outside box : addr in page around %"PRIx64"\n",
- (uint64_t) lpn*PAGE_SIZE);
-#endif
+ return NULL;
l2tbl = pmap->pm_l1[l1];
if (l2tbl)
Index: src/sys/arch/usermode/usermode/syscall.c
diff -u src/sys/arch/usermode/usermode/syscall.c:1.20 src/sys/arch/usermode/usermode/syscall.c:1.21
--- src/sys/arch/usermode/usermode/syscall.c:1.20 Tue Jan 3 12:05:00 2012
+++ src/sys/arch/usermode/usermode/syscall.c Sat Jan 14 17:42:52 2012
@@ -1,4 +1,4 @@
-/* $NetBSD: syscall.c,v 1.20 2012/01/03 12:05:00 reinoud Exp $ */
+/* $NetBSD: syscall.c,v 1.21 2012/01/14 17:42:52 reinoud Exp $ */
/*-
* Copyright (c) 2007 Jared D. McNeill <[email protected]>
@@ -27,7 +27,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: syscall.c,v 1.20 2012/01/03 12:05:00 reinoud Exp $");
+__KERNEL_RCSID(0, "$NetBSD: syscall.c,v 1.21 2012/01/14 17:42:52 reinoud Exp $");
#include <sys/types.h>
#include <sys/param.h>
@@ -61,11 +61,12 @@ child_return(void *arg)
lwp_t *l = arg;
register_t rval[2];
struct pcb *pcb = lwp_getpcb(l);
+ ucontext_t *ucp = &pcb->pcb_userret_ucp;
/* return value zero */
rval[0] = 0;
rval[1] = 0;
- md_syscall_set_returnargs(l, &pcb->pcb_userret_ucp, 0, rval);
+ md_syscall_set_returnargs(l, ucp, 0, rval);
aprint_debug("child return! lwp %p\n", l);
userret(l);
@@ -99,8 +100,6 @@ syscall(void)
curcpu()->ci_data.cpu_nsyscall++;
LWP_CACHE_CREDS(l, l->l_proc);
- /* TODO are we allowed to execute system calls in this memory space? */
-
/* XXX do we want do do emulation? */
md_syscall_get_opcode(ucp, &opcode);
md_syscall_get_syscallnumber(ucp, &code);
@@ -161,7 +160,7 @@ syscall(void)
break;
}
//thunk_printf_debug("end of syscall : return to userland\n");
-//if (code != 4) printf("userret() code %d\n", code);
+//if (code != 4) thunk_printf("userret() code %d\n", code);
userret(l);
}
@@ -173,54 +172,54 @@ syscall_args_print(lwp_t *l, int code, i
return;
if (code != 4) {
- printf("lwp %p, code %3d, nargs %d, argsize %3d\t%s(",
+ thunk_printf("lwp %p, code %3d, nargs %d, argsize %3d\t%s(",
l, code, nargs, argsize, syscallnames[code]);
switch (code) {
case 5:
- printf("\"%s\", %"PRIx32", %"PRIx32"", (char *) (args[0]), (uint) args[1], (uint) args[2]);
+ thunk_printf("\"%s\", %"PRIx32", %"PRIx32"", (char *) (args[0]), (uint) args[1], (uint) args[2]);
break;
case 33:
- printf("\"%s\", %"PRIx32"", (char *) (args[0]), (uint) args[1]);
+ thunk_printf("\"%s\", %"PRIx32"", (char *) (args[0]), (uint) args[1]);
break;
case 50:
- printf("\"%s\"", (char *) (args[0]));
+ thunk_printf("\"%s\"", (char *) (args[0]));
break;
case 58:
- printf("\"%s\", %"PRIx32", %"PRIx32"", (char *) (args[0]), (uint) (args[1]), (uint) args[2]);
+ thunk_printf("\"%s\", %"PRIx32", %"PRIx32"", (char *) (args[0]), (uint) (args[1]), (uint) args[2]);
break;
case 59:
- printf("\"%s\", [", (char *) (args[0]));
+ thunk_printf("\"%s\", [", (char *) (args[0]));
argv = (char **) (args[1]);
if (*argv) {
while (*argv) {
- printf("\"%s\", ", *argv);
+ thunk_printf("\"%s\", ", *argv);
argv++;
}
- printf("\b\b");
+ thunk_printf("\b\b");
}
- printf("], [");
+ thunk_printf("], [");
envp = (char **) (args[2]);
if (*envp) {
while (*envp) {
- printf("\"%s\", ", *envp);
+ thunk_printf("\"%s\", ", *envp);
envp++;
}
- printf("\b\b");
+ thunk_printf("\b\b");
}
- printf("]");
+ thunk_printf("]");
break;
default:
for (int i = 0; i < nargs; i++)
- printf("%"PRIx32", ", (uint) args[i]);
+ thunk_printf("%"PRIx32", ", (uint) args[i]);
if (nargs)
- printf("\b\b");
+ thunk_printf("\b\b");
}
- printf(") ");
+ thunk_printf(") ");
}
#if 0
if ((code == 4)) {
// thunk_printf_debug("[us] %s", (char *) args[1]);
- printf("[us] %s", (char *) args[1]);
+ thunk_printf("[us] %s", (char *) args[1]);
}
#endif
}
@@ -243,7 +242,7 @@ return;
errstr = "OK";
}
if (code != 4)
- printf("=> %s: %d, (%"PRIx32", %"PRIx32")\n",
+ thunk_printf("=> %s: %d, (%"PRIx32", %"PRIx32")\n",
errstr, error, (uint) (rval[0]), (uint) (rval[1]));
}
Index: src/sys/arch/usermode/usermode/thunk.c
diff -u src/sys/arch/usermode/usermode/thunk.c:1.77 src/sys/arch/usermode/usermode/thunk.c:1.78
--- src/sys/arch/usermode/usermode/thunk.c:1.77 Wed Jan 11 12:37:50 2012
+++ src/sys/arch/usermode/usermode/thunk.c Sat Jan 14 17:42:52 2012
@@ -1,4 +1,4 @@
-/* $NetBSD: thunk.c,v 1.77 2012/01/11 12:37:50 reinoud Exp $ */
+/* $NetBSD: thunk.c,v 1.78 2012/01/14 17:42:52 reinoud Exp $ */
/*-
* Copyright (c) 2011 Jared D. McNeill <[email protected]>
@@ -28,7 +28,7 @@
#include <sys/cdefs.h>
#ifdef __NetBSD__
-__RCSID("$NetBSD: thunk.c,v 1.77 2012/01/11 12:37:50 reinoud Exp $");
+__RCSID("$NetBSD: thunk.c,v 1.78 2012/01/14 17:42:52 reinoud Exp $");
#endif
#include <sys/types.h>
@@ -960,7 +960,7 @@ thunk_rfb_open(thunk_rfb_t *rfb, uint16_
}
static ssize_t
-safe_send(int s, const void *msg, size_t len)
+safe_send(int s, const void *msg, int len)
{
const uint8_t *p;
ssize_t sent_len;
@@ -982,7 +982,7 @@ safe_send(int s, const void *msg, size_t
}
static ssize_t
-safe_recv(int s, void *buf, size_t len)
+safe_recv(int s, void *buf, int len)
{
uint8_t *p;
int recv_len;
Index: src/sys/arch/usermode/usermode/trap.c
diff -u src/sys/arch/usermode/usermode/trap.c:1.55 src/sys/arch/usermode/usermode/trap.c:1.56
--- src/sys/arch/usermode/usermode/trap.c:1.55 Wed Jan 11 12:40:26 2012
+++ src/sys/arch/usermode/usermode/trap.c Sat Jan 14 17:42:52 2012
@@ -1,4 +1,4 @@
-/* $NetBSD: trap.c,v 1.55 2012/01/11 12:40:26 reinoud Exp $ */
+/* $NetBSD: trap.c,v 1.56 2012/01/14 17:42:52 reinoud Exp $ */
/*-
* Copyright (c) 2011 Reinoud Zandijk <[email protected]>
@@ -27,7 +27,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: trap.c,v 1.55 2012/01/11 12:40:26 reinoud Exp $");
+__KERNEL_RCSID(0, "$NetBSD: trap.c,v 1.56 2012/01/14 17:42:52 reinoud Exp $");
#include <sys/types.h>
#include <sys/param.h>
@@ -35,6 +35,7 @@ __KERNEL_RCSID(0, "$NetBSD: trap.c,v 1.5
#include <sys/proc.h>
#include <sys/systm.h>
#include <sys/userret.h>
+#include <sys/kauth.h>
#include <sys/errno.h>
#include <uvm/uvm_extern.h>
@@ -54,6 +55,9 @@ static void mem_access_handler(int sig,
static void illegal_instruction_handler(int sig, siginfo_t *info, void *ctx);
extern int errno;
+static void pagefault(vaddr_t pc, vaddr_t va);
+static void illegal_instruction(void);
+
bool pmap_fault(pmap_t pmap, vaddr_t va, vm_prot_t *atype);
static stack_t sigstk;
@@ -79,7 +83,8 @@ setup_signal_handlers(void)
panic("can't set alternate stacksize: %d",
thunk_geterrno());
- sa.sa_flags = SA_RESTART | SA_SIGINFO | SA_ONSTACK;
+ /* SIGBUS and SIGSEGV need to be reentrant hence the SA_NODEFER */
+ sa.sa_flags = SA_RESTART | SA_SIGINFO | SA_ONSTACK | SA_NODEFER;
sa.sa_sigaction = mem_access_handler;
thunk_sigemptyset(&sa.sa_mask);
if (thunk_sigaction(SIGSEGV, &sa, NULL) == -1)
@@ -113,13 +118,14 @@ stop_all_signal_handlers(void)
}
+/* signal handler switching to a pagefault context */
static void
mem_access_handler(int sig, siginfo_t *info, void *ctx)
{
- ucontext_t *uct = ctx;
+ ucontext_t *ucp = ctx;
struct lwp *l;
struct pcb *pcb;
- vaddr_t va, pc;
+ vaddr_t va, sp, pc, fp;
assert((info->si_signo == SIGSEGV) || (info->si_signo == SIGBUS));
@@ -156,29 +162,55 @@ mem_access_handler(int sig, siginfo_t *i
/* get PC address of faulted memory instruction */
pc = md_get_pc(ctx);
+ /* setup for pagefault context */
+ sp = md_get_sp(ctx);
+
#if 0
- thunk_printf("memaccess error : pc = %p, va = %p\n",
- (void *) pc, (void *) va);
+ printf("memaccess error, pc %p, va %p, "
+ "sys_stack %p, sp %p, stack top %p\n",
+ (void *) pc, (void *) va,
+ (void *) pcb->sys_stack, (void *) sp,
+ (void *) pcb->sys_stack_top);
#endif
- /* copy this state to return to */
- memcpy(&pcb->pcb_trapret_ucp, uct, sizeof(ucontext_t));
+ /* if we're running on a stack of our own, use the system stack */
+ if ((sp < (vaddr_t) pcb->sys_stack) || (sp > (vaddr_t) pcb->sys_stack_top)) {
+ sp = (vaddr_t) pcb->sys_stack_top - sizeof(register_t);
+ fp = (vaddr_t) &pcb->pcb_userret_ucp;
+ } else {
+ /* stack grows down */
+ fp = sp - sizeof(ucontext_t) - sizeof(register_t); /* slack */
+ sp = fp - sizeof(register_t); /* slack */
+
+ /* sanity check before copying */
+ if (fp - 2*PAGE_SIZE < (vaddr_t) pcb->sys_stack)
+ panic("%s: out of system stack", __func__);
+ }
+
+ memcpy((void *) fp, ucp, sizeof(ucontext_t));
- /* remember our parameters */
- pcb->pcb_fault_addr = va;
- pcb->pcb_fault_pc = pc;
+ /* create context for pagefault */
+ pcb->pcb_ucp.uc_stack.ss_sp = (void *) pcb->sys_stack;
+ pcb->pcb_ucp.uc_stack.ss_size = sp - (vaddr_t) pcb->sys_stack;
+ pcb->pcb_ucp.uc_link = (void *) fp; /* link to old frame on stack */
+
+ pcb->pcb_ucp.uc_flags = _UC_STACK | _UC_CPU;
+ thunk_makecontext(&pcb->pcb_ucp, (void (*)(void)) pagefault,
+ 2, (void *) pc, (void *) va, NULL);
- /* switch to the pagefault entry on return from signal */
- memcpy(uct, &pcb->pcb_pagefault_ucp, sizeof(ucontext_t));
+ /* switch to the new pagefault entry on return from signal */
+ memcpy(ctx, &pcb->pcb_ucp, sizeof(ucontext_t));
}
+/* signal handler switching to a illegal instruction context */
static void
illegal_instruction_handler(int sig, siginfo_t *info, void *ctx)
{
- ucontext_t *uct = ctx;
+ ucontext_t *ucp = ctx;
struct lwp *l;
struct pcb *pcb;
+ vaddr_t sp, fp;
assert(info->si_signo == SIGILL);
#if 0
@@ -214,27 +246,49 @@ illegal_instruction_handler(int sig, sig
l = curlwp;
pcb = lwp_getpcb(l);
- /* copy this state to return to */
- memcpy(&pcb->pcb_userret_ucp, uct, sizeof(ucontext_t));
+ /* setup for illegal_instruction context */
+ sp = md_get_sp(ctx);
- /* if its a syscall ... */
- if (md_syscall_check_opcode(uct)) {
- /* switch to the syscall entry on return from signal */
- memcpy(uct, &pcb->pcb_syscall_ucp, sizeof(ucontext_t));
- return;
+ /* if we're running on a stack of our own, use the system stack */
+ if ((sp < (vaddr_t) pcb->sys_stack) ||
+ (sp > (vaddr_t) pcb->sys_stack_top)) {
+ sp = (vaddr_t) pcb->sys_stack_top - sizeof(register_t);
+ fp = (vaddr_t) &pcb->pcb_userret_ucp;
+ } else {
+ panic("illegal instruction inside kernel?");
+#if 0
+ /* stack grows down */
+ fp = sp - sizeof(ucontext_t) - sizeof(register_t); /* slack */
+ sp = fp - sizeof(register_t); /* slack */
+
+ /* sanity check before copying */
+ if (fp - 2*PAGE_SIZE < (vaddr_t) pcb->sys_stack)
+ panic("%s: out of system stack", __func__);
+#endif
}
- panic("should deliver a trap to the process : illegal instruction "
- "encountered\n");
+ memcpy((void *) fp, ucp, sizeof(ucontext_t));
+
+ /* create context for illegal instruction */
+ pcb->pcb_ucp.uc_stack.ss_sp = (void *) pcb->sys_stack;
+ pcb->pcb_ucp.uc_stack.ss_size = sp - (vaddr_t) pcb->sys_stack;
+ pcb->pcb_ucp.uc_link = (void *) fp; /* link to old frame on stack */
+
+ pcb->pcb_ucp.uc_flags = _UC_STACK | _UC_CPU;
+ thunk_makecontext(&pcb->pcb_ucp, (void (*)(void)) illegal_instruction,
+ 0, NULL, NULL, NULL);
+
+ /* switch to the new illegal instruction entry on return from signal */
+ memcpy(ctx, &pcb->pcb_ucp, sizeof(ucontext_t));
}
/*
- * Entry point from the segv handler; check if its a pmap reference fault or
- * let uvm handle it.
+ * Context for handing page faults from the sigsegv handler; check if its a
+ * pmap reference fault or let uvm handle it.
*/
-void
-pagefault(void)
+static void
+pagefault(vaddr_t pc, vaddr_t va)
{
struct proc *p;
struct lwp *l;
@@ -242,16 +296,14 @@ pagefault(void)
struct vmspace *vm;
struct vm_map *vm_map;
vm_prot_t atype;
- vaddr_t va, pc;
void *onfault;
- int from_kernel, lwp_errno, rv;
+ int from_kernel, lwp_errno, error;
+ ksiginfo_t ksi;
- l = curlwp;
+ l = curlwp; KASSERT(l);
pcb = lwp_getpcb(l);
p = l->l_proc;
vm = p->p_vmspace;
- va = pcb->pcb_fault_addr;
- pc = pcb->pcb_fault_pc;
lwp_errno = thunk_geterrno();
@@ -261,53 +313,120 @@ pagefault(void)
vm_map = kernel_map;
#if 0
- thunk_printf_debug("pagefault : pc %p, va %p\n",
+ thunk_printf("pagefault : pc %p, va %p\n",
(void *) pc, (void *) va);
#endif
/* can pmap handle it? on its own? (r/m) */
+ if (pmap_fault(vm_map->pmap, va, &atype)) {
+// thunk_printf("pagefault leave (pmap)\n");
+ goto out;
+ }
+
+ /* ask UVM */
+ thunk_printf_debug("pmap fault couldn't handle it! : "
+ "derived atype %d\n", atype);
+
onfault = pcb->pcb_onfault;
- rv = 0;
- if (!pmap_fault(vm_map->pmap, va, &atype)) {
- thunk_printf_debug("pmap fault couldn't handle it! : "
- "derived atype %d\n", atype);
-
- /* extra debug for now */
- if (pcb == 0)
- panic("NULL pcb!\n");
-
- pcb->pcb_onfault = NULL;
- rv = uvm_fault(vm_map, va, atype);
- pcb->pcb_onfault = onfault;
+ pcb->pcb_onfault = NULL;
+ error = uvm_fault(vm_map, va, atype);
+ pcb->pcb_onfault = onfault;
+
+ if (vm_map != kernel_map) {
+ if (error == 0)
+ uvm_grow(l->l_proc, va);
}
+ if (error == EACCES)
+ error = EFAULT;
- if (rv) {
- thunk_printf_debug("uvm_fault returned error %d\n", rv);
+ /* if uvm handled it, return */
+ if (error == 0) {
+// thunk_printf("pagefault leave (uvm)\n");
+ goto out;
+ }
- /* something got wrong */
- if (from_kernel) {
- /* copyin / copyout */
- if (!onfault)
- panic("kernel fault");
- panic("%s: can't call onfault yet\n", __func__);
- /* XXX implement me ? */
- /* jump to given onfault */
- // tf = &kernel_tf;
- // memset(tf, 0, sizeof(struct trapframe));
- // tf->tf_pc = onfault;
- // tf->tf_io[0] = (rv == EACCES) ? EFAULT : rv;
- return;
- }
- panic("%s: should deliver a trap to the process for va %p", __func__, (void *) va);
- /* XXX HOWTO see arm/arm/syscall.c illegal instruction signal */
+ /* something got wrong */
+ thunk_printf("%s: uvm fault %d, pc %p, from_kernel %d\n",
+ __func__, error, (void *) pc, from_kernel);
+
+ /* check if its from copyin/copyout */
+ if (onfault) {
+ panic("%s: can't call onfault yet\n", __func__);
+ /* XXX implement me ? */
+ /* jump to given onfault */
+ // tf = &kernel_tf;
+ // memset(tf, 0, sizeof(struct trapframe));
+ // tf->tf_pc = onfault;
+ // tf->tf_io[0] = (rv == EACCES) ? EFAULT : rv;
+ goto out;
}
+ if (from_kernel)
+ panic("Unhandled page fault in kernel mode");
+
+ /* send signal */
+ thunk_printf("giving signal to userland\n");
+
+ KSI_INIT_TRAP(&ksi);
+ ksi.ksi_signo = SIGSEGV;
+ ksi.ksi_trap = 0; /* XXX */
+ ksi.ksi_code = (error == EPERM) ? SEGV_ACCERR : SEGV_MAPERR;
+ ksi.ksi_addr = (void *) va;
+
+ if (error == ENOMEM) {
+ printf("UVM: pid %d.%d (%s), uid %d killed: "
+ "out of swap\n",
+ p->p_pid, l->l_lid, p->p_comm,
+ l->l_cred ? kauth_cred_geteuid(l->l_cred) : -1);
+ ksi.ksi_signo = SIGKILL;
+ }
+
+#if 0
+ p->p_emul->e_trapsignal(l, &ksi);
+#else
+ trapsignal(l, &ksi);
+#endif
+ mi_userret(l);
+
+// thunk_printf("pagefault leave\n");
+out:
thunk_seterrno(lwp_errno);
pcb->pcb_errno = lwp_errno;
}
-stack_t *
-usermode_signal_stack(void)
+/*
+ * Context for handing illegal instruction from the sigill handler
+ */
+static void
+illegal_instruction(void)
{
- return &sigstk;
+ struct lwp *l = curlwp;
+ struct pcb *pcb = lwp_getpcb(l);
+ ucontext_t *ucp = &pcb->pcb_userret_ucp;
+ ksiginfo_t ksi;
+
+// thunk_printf("illegal instruction\n");
+ /* if its a syscall ... */
+ if (md_syscall_check_opcode(ucp)) {
+ syscall();
+// thunk_printf("illegal instruction leave\n");
+ return;
+ }
+
+ thunk_printf("%s: giving SIGILL (TRAP)\n", __func__);
+
+ KSI_INIT_TRAP(&ksi);
+ ksi.ksi_signo = SIGILL;
+ ksi.ksi_trap = 0; /* XXX */
+ ksi.ksi_errno = 0; // info->si_errno;
+ ksi.ksi_code = 0; // info->si_code;
+ ksi.ksi_addr = (void *) md_get_pc(ucp); /* only relyable source */
+
+#if 0
+ p->p_emul->e_trapsignal(l, &ksi);
+#else
+ trapsignal(l, &ksi);
+#endif
+ mi_userret(l);
}
+
