From: Ben Gras <ben@minix3.org>
Date: Wed, 12 May 2010 08:31:05 +0000 (+0000)
Subject: kernel/vm: change pde table info from single buffer to explicit per-process.
X-Git-Tag: v3.1.7~63
X-Git-Url: http://zhaoyanbai.com/repos/%22http:/www.isc.org/icons/zlib_tech.html?a=commitdiff_plain;h=c5c25e7abce1b50d23aa81aa152f07c0f530f0a8;p=minix.git

kernel/vm: change pde table info from single buffer to explicit per-process.

makes code in kernel more readable, and allows better sanity checking on
using the pde info.
---

diff --git a/include/arch/i386/archtypes.h b/include/arch/i386/archtypes.h
index 3300a2ebb..7db2f481d 100644
--- a/include/arch/i386/archtypes.h
+++ b/include/arch/i386/archtypes.h
@@ -25,6 +25,7 @@ struct segdesc_s {		/* segment descriptor for protected mode */
 typedef struct segframe {
 	reg_t p_ldt_sel;    /* selector in gdt with ldt base and limit */
 	reg_t	p_cr3;		/* page table root */
+	u32_t	*p_cr3_v;
 	struct segdesc_s p_ldt[LDT_SIZE]; /* CS, DS and remote */
 } segframe_t;
 
diff --git a/include/minix/com.h b/include/minix/com.h
index 86ac88016..9d64eec8f 100644
--- a/include/minix/com.h
+++ b/include/minix/com.h
@@ -601,6 +601,8 @@
 #define	SVMCTL_MRG_ADDR2	m2_l2	/* MEMREQ_GET reply: source address */
 #define SVMCTL_MRG_REQUESTOR	m2_p1	/* MEMREQ_GET reply: requestor */
 #define SVMCTL_MAP_VIR_ADDR	m1_p1
+#define SVMCTL_PTROOT		m1_i3
+#define SVMCTL_PTROOT_V		m1_p1
 
 /* Reply message for VMCTL_KERN_PHYSMAP */
 #define SVMCTL_MAP_FLAGS	m2_i1	/* VMMF_* */
@@ -610,7 +612,6 @@
 #define VMMF_UNCACHED		(1L << 0)
 
 /* Values for SVMCTL_PARAM. */
-#define VMCTL_I386_SETCR3	10
 #define VMCTL_CLEAR_PAGEFAULT	12
 #define VMCTL_I386_GETCR3	13
 #define VMCTL_MEMREQ_GET	14
@@ -618,13 +619,13 @@
 #define VMCTL_INCSP		16
 #define VMCTL_NOPAGEZERO	18
 #define VMCTL_I386_KERNELLIMIT	19
-#define VMCTL_I386_PAGEDIRS	20
 #define VMCTL_I386_FREEPDE	23
 #define VMCTL_ENABLE_PAGING	24
 #define VMCTL_I386_INVLPG	25
 #define VMCTL_FLUSHTLB		26
 #define VMCTL_KERN_PHYSMAP	27
 #define VMCTL_KERN_MAP_REPLY	28
+#define VMCTL_SETADDRSPACE	29
 
 /* Codes and field names for SYS_SYSCTL. */
 #define SYSCTL_CODE		m1_i1	/* SYSCTL_CODE_* below */
diff --git a/include/minix/syslib.h b/include/minix/syslib.h
index 5d5954066..f09339374 100644
--- a/include/minix/syslib.h
+++ b/include/minix/syslib.h
@@ -79,6 +79,9 @@ _PROTOTYPE( int sys_sysctl_stacktrace, (endpoint_t who));
 _PROTOTYPE( int sys_vmctl_get_mapping, (int index, phys_bytes *addr,
 	phys_bytes *len, int *flags));
 _PROTOTYPE( int sys_vmctl_reply_mapping, (int index, vir_bytes addr));
+_PROTOTYPE( int sys_vmctl_set_addrspace, (endpoint_t who,
+        phys_bytes ptroot, void *ptroot_v));
+
 
 /* Shorthands for sys_sdevio() system call. */
 #define sys_insb(port, proc_ep, buffer, count) \
diff --git a/kernel/arch/i386/arch_do_vmctl.c b/kernel/arch/i386/arch_do_vmctl.c
index ca1d291b4..a1be5e39d 100644
--- a/kernel/arch/i386/arch_do_vmctl.c
+++ b/kernel/arch/i386/arch_do_vmctl.c
@@ -12,8 +12,6 @@
 
 #include "proto.h"
 
-extern u8_t *vm_pagedirs;
-
 /*===========================================================================*
  *				arch_do_vmctl				     *
  *===========================================================================*/
@@ -26,13 +24,15 @@ struct proc *p;
 		/* Get process CR3. */
 		m_ptr->SVMCTL_VALUE = p->p_seg.p_cr3;
 		return OK;
-	case VMCTL_I386_SETCR3:
+	case VMCTL_SETADDRSPACE:
 		/* Set process CR3. */
-		if(m_ptr->SVMCTL_VALUE) {
-			p->p_seg.p_cr3 = m_ptr->SVMCTL_VALUE;
+		if(m_ptr->SVMCTL_PTROOT) {
+			p->p_seg.p_cr3 = m_ptr->SVMCTL_PTROOT;
+			p->p_seg.p_cr3_v = (u32_t *) m_ptr->SVMCTL_PTROOT_V;
 			p->p_misc_flags |= MF_FULLVM;
 		} else {
 			p->p_seg.p_cr3 = 0;
+			p->p_seg.p_cr3_v = NULL;
 			p->p_misc_flags &= ~MF_FULLVM;
 		}
 		RTS_UNSET(p, RTS_VMINHIBIT);
@@ -48,11 +48,6 @@ struct proc *p;
 		r = prot_set_kern_seg_limit(m_ptr->SVMCTL_VALUE);
 		return r;
 	}
-	case VMCTL_I386_PAGEDIRS:
-	{
-		vm_pagedirs = (u8_t *) m_ptr->SVMCTL_VALUE;
-		return OK;
-	}
 	case VMCTL_I386_FREEPDE:
 	{
 		i386_freepde(m_ptr->SVMCTL_VALUE);
diff --git a/kernel/arch/i386/memory.c b/kernel/arch/i386/memory.c
index bb2f327e2..b4c6111ad 100644
--- a/kernel/arch/i386/memory.c
+++ b/kernel/arch/i386/memory.c
@@ -28,12 +28,6 @@
 
 PRIVATE int psok = 0;
 
-#define PROCPDEPTR(pr, pi) ((u32_t *) ((u8_t *) vm_pagedirs +\
-				I386_PAGE_SIZE * pr->p_nr +	\
-				I386_VM_PT_ENT_SIZE * pi))
-
-PUBLIC u8_t *vm_pagedirs = NULL;
-
 #define MAX_FREEPDES (3 * CONFIG_MAX_CPUS)
 PRIVATE int nfreepdes = 0, freepdes[MAX_FREEPDES];
 
@@ -90,6 +84,7 @@ PRIVATE phys_bytes createpde(
 
 	assert(free_pde_idx >= 0 && free_pde_idx < nfreepdes);
 	pde = freepdes[free_pde_idx];
+	assert(pde >= 0 && pde < 1024);
 
 	if(pr && ((pr == ptproc) || !HASPT(pr))) {
 		/* Process memory is requested, and
@@ -106,7 +101,8 @@ PRIVATE phys_bytes createpde(
 		 * accessible directly. Grab the PDE entry of that process'
 		 * page table that corresponds to the requested address.
 		 */
-		pdeval = *PROCPDEPTR(pr, I386_VM_PDE(linaddr));
+		assert(pr->p_seg.p_cr3_v);
+		pdeval = pr->p_seg.p_cr3_v[I386_VM_PDE(linaddr)];
 	} else {
 		/* Requested address is physical. Make up the PDE entry. */
 		pdeval = (linaddr & I386_VM_ADDR_MASK_4MB) | 
@@ -118,8 +114,9 @@ PRIVATE phys_bytes createpde(
 	 * can access, into the currently loaded page table so it becomes
 	 * visible.
 	 */
-	if(*PROCPDEPTR(ptproc, pde) != pdeval) {
-		*PROCPDEPTR(ptproc, pde) = pdeval;
+	assert(ptproc->p_seg.p_cr3_v);
+	if(ptproc->p_seg.p_cr3_v[pde] != pdeval) {
+		ptproc->p_seg.p_cr3_v[pde] = pdeval;
 		*changed = 1;
 	}
 
@@ -155,6 +152,11 @@ PRIVATE int lin_lin_copy(const struct proc *srcproc, vir_bytes srclinaddr,
 
 	assert(procslot >= 0 && procslot < I386_VM_DIR_ENTRIES);
 
+	if(srcproc) assert(!RTS_ISSET(srcproc, RTS_SLOT_FREE));
+	if(dstproc) assert(!RTS_ISSET(dstproc, RTS_SLOT_FREE));
+	assert(!RTS_ISSET(ptproc, RTS_SLOT_FREE));
+	assert(ptproc->p_seg.p_cr3_v);
+
 	while(bytes > 0) {
 		phys_bytes srcptr, dstptr;
 		vir_bytes chunk = bytes;
@@ -191,6 +193,11 @@ PRIVATE int lin_lin_copy(const struct proc *srcproc, vir_bytes srclinaddr,
 		dstlinaddr += chunk;
 	}
 
+	if(srcproc) assert(!RTS_ISSET(srcproc, RTS_SLOT_FREE));
+	if(dstproc) assert(!RTS_ISSET(dstproc, RTS_SLOT_FREE));
+	assert(!RTS_ISSET(ptproc, RTS_SLOT_FREE));
+	assert(ptproc->p_seg.p_cr3_v);
+
 	return OK;
 }
 
@@ -677,6 +684,8 @@ int vm_phys_memset(phys_bytes ph, const u8_t c, phys_bytes bytes)
 
 	assert(nfreepdes >= 3);
 
+	assert(ptproc->p_seg.p_cr3_v);
+
 	/* With VM, we have to map in the physical memory. 
 	 * We can do this 4MB at a time.
 	 */
@@ -695,6 +704,7 @@ int vm_phys_memset(phys_bytes ph, const u8_t c, phys_bytes bytes)
 		ph += chunk;
 	}
 
+	assert(ptproc->p_seg.p_cr3_v);
 
 	return OK;
 }
@@ -1004,3 +1014,8 @@ PUBLIC int arch_enable_paging(struct proc * caller, const message * m_ptr)
 
 	return OK;
 }
+
+PUBLIC void release_address_space(struct proc *pr)
+{
+	pr->p_seg.p_cr3_v = NULL;
+}
diff --git a/kernel/arch/i386/sconst.h b/kernel/arch/i386/sconst.h
index 2a5e3b667..ca59b62bf 100644
--- a/kernel/arch/i386/sconst.h
+++ b/kernel/arch/i386/sconst.h
@@ -30,7 +30,8 @@
 	FP_SAVE_AREA_P = P_STACKTOP
 	P_LDT_SEL = FP_SAVE_AREA_P + 532
 	P_CR3 = P_LDT_SEL+W
-	P_LDT = P_CR3+W
+	P_CR3_V = P_CR3+4
+	P_LDT = P_CR3_V+W
 	P_MISC_FLAGS = P_LDT + 50
 	Msize = 9	/* size of a message in 32-bit words*/
 
diff --git a/kernel/proto.h b/kernel/proto.h
index 560f10662..7f58ac821 100644
--- a/kernel/proto.h
+++ b/kernel/proto.h
@@ -182,4 +182,5 @@ _PROTOTYPE( int copy_msg_from_user, (struct proc * p, message * user_mbuf,
 _PROTOTYPE( int copy_msg_to_user, (struct proc * p, message * src,
 							message * user_mbuf));
 _PROTOTYPE(void switch_address_space, (struct proc * p));
+_PROTOTYPE(void release_address_space, (struct proc *pr));
 #endif /* PROTO_H */
diff --git a/kernel/system/do_clear.c b/kernel/system/do_clear.c
index 58819aa5c..db960cb7a 100644
--- a/kernel/system/do_clear.c
+++ b/kernel/system/do_clear.c
@@ -31,6 +31,8 @@ PUBLIC int do_clear(struct proc * caller, message * m_ptr)
   }
   rc = proc_addr(exit_p);	/* clean up */
 
+  release_address_space(rc);
+
   /* Don't clear if already cleared. */
   if(isemptyp(rc)) return OK;
 
diff --git a/kernel/system/do_fork.c b/kernel/system/do_fork.c
index 936df8f5a..11629afb9 100644
--- a/kernel/system/do_fork.c
+++ b/kernel/system/do_fork.c
@@ -117,6 +117,9 @@ PUBLIC int do_fork(struct proc * caller, message * m_ptr)
   RTS_UNSET(rpc, (RTS_SIGNALED | RTS_SIG_PENDING | RTS_P_STOP));
   sigemptyset(&rpc->p_pending);
 
+  rpc->p_seg.p_cr3 = 0;
+  rpc->p_seg.p_cr3_v = NULL;
+
   return r;
 }
 
diff --git a/lib/libsys/sys_vmctl.c b/lib/libsys/sys_vmctl.c
index 73ab87430..965671054 100644
--- a/lib/libsys/sys_vmctl.c
+++ b/lib/libsys/sys_vmctl.c
@@ -26,6 +26,21 @@ PUBLIC int sys_vmctl_get_cr3_i386(endpoint_t who, u32_t *cr3)
   return(r);
 }
 
+PUBLIC int sys_vmctl_set_addrspace(endpoint_t who,
+        phys_bytes ptroot, void *ptroot_v)
+{
+  message m;
+  int r;
+
+  m.SVMCTL_WHO = who;
+  m.SVMCTL_PARAM = VMCTL_SETADDRSPACE;
+  m.SVMCTL_PTROOT = ptroot;
+  m.SVMCTL_PTROOT_V = ptroot_v;
+  r = _kernel_call(SYS_VMCTL, &m);
+
+  return(r);
+}
+
 PUBLIC int sys_vmctl_get_memreq(endpoint_t *who, vir_bytes *mem,
         vir_bytes *len, int *wrflag, endpoint_t *who_s, vir_bytes *mem_s,
         endpoint_t *requestor)
diff --git a/servers/vm/arch/i386/pagetable.c b/servers/vm/arch/i386/pagetable.c
index 4fa248311..939322b41 100644
--- a/servers/vm/arch/i386/pagetable.c
+++ b/servers/vm/arch/i386/pagetable.c
@@ -717,7 +717,7 @@ PUBLIC void pt_init(phys_bytes usedlimit)
  */
         pt_t *newpt;
         int s, r;
-        vir_bytes v, kpagedir;
+        vir_bytes v;
         phys_bytes lo, hi; 
         vir_bytes extra_clicks;
         u32_t moveup = 0;
@@ -886,14 +886,6 @@ PUBLIC void pt_init(phys_bytes usedlimit)
 	/* first pde in use by process. */
 	proc_pde = free_pde;
 
-	kpagedir = arch_map2vir(&vmproc[VMP_SYSTEM],
-		pagedir_pde*I386_BIG_PAGE_SIZE);
-
-	/* Tell kernel how to get at the page directories. */
-	if((r=sys_vmctl(SELF, VMCTL_I386_PAGEDIRS, kpagedir)) != OK) {
-                panic("VMCTL_I386_PAGEDIRS failed: %d", r);
-	}
-       
         /* Give our process the new, copied, private page table. */
 	pt_mapkernel(newpt);	/* didn't know about vm_dir pages earlier */
         pt_bind(newpt, vmprocess);
@@ -922,12 +914,15 @@ PUBLIC int pt_bind(pt_t *pt, struct vmproc *who)
 {
 	int slot, ispt;
 	u32_t phys;
+	void *pdes;
 
 	/* Basic sanity checks. */
 	assert(who);
 	assert(who->vm_flags & VMF_INUSE);
 	assert(pt);
 
+	assert(pagedir_pde >= 0);
+
 	slot = who->vm_slot;
 	assert(slot >= 0);
 	assert(slot < ELEMENTS(vmproc));
@@ -939,12 +934,21 @@ PUBLIC int pt_bind(pt_t *pt, struct vmproc *who)
 	/* Update "page directory pagetable." */
 	page_directories[slot] = phys | I386_VM_PRESENT|I386_VM_WRITE;
 
+	/* This is where the PDE's will be visible to the kernel
+	 * in its address space.
+	 */
+	pdes = (void *) arch_map2vir(&vmproc[VMP_SYSTEM],
+		pagedir_pde*I386_BIG_PAGE_SIZE + 
+			slot * I386_PAGE_SIZE);
+
 #if 0
-	printf("VM: slot %d has pde val 0x%lx\n", slot, page_directories[slot]);
+	printf("VM: slot %d endpoint %d has pde val 0x%lx at kernel address 0x%lx\n",
+		slot, who->vm_endpoint, page_directories[slot], pdes);
 #endif
 	/* Tell kernel about new page table root. */
-	return sys_vmctl(who->vm_endpoint, VMCTL_I386_SETCR3,
-		pt ? pt->pt_dir_phys : 0);
+	return sys_vmctl_set_addrspace(who->vm_endpoint,
+		pt ? pt->pt_dir_phys : 0,
+		pt ? pdes : 0);
 }
 
 /*===========================================================================*