Fix xc_ptrace (live debug, not coredump debug) for 32-bit pae.

Clean up interfaces and add simple documentation for using
the domU debug interfaces.

Signed-off-by: Keir Fraser <keir@xensource.com>

4 files changed, 193 insertions, 94 deletions

diff --git a/tools/libxc/Makefile b/tools/libxc/Makefile
index f1ac817d1f..9c4b705641 100644
--- a/tools/libxc/Makefile
+++ b/tools/libxc/Makefile
@@ -30,8 +30,10 @@ BUILD_SRCS += xc_load_elf.c
 ifeq ($(XEN_TARGET_ARCH),ia64)
 BUILD_SRCS += xc_ia64_stubs.c
 else
+ifeq ($(XEN_TARGET_ARCH),x86_32)
 SRCS       += xc_ptrace.c
 SRCS       += xc_ptrace_core.c
+endif
 BUILD_SRCS += xc_load_aout9.c
 BUILD_SRCS += xc_linux_restore.c
 BUILD_SRCS += xc_linux_save.c
diff --git a/tools/libxc/xc_ptrace.c b/tools/libxc/xc_ptrace.c
index 1fe0fcbe54..f1170d30f0 100644
--- a/tools/libxc/xc_ptrace.c
+++ b/tools/libxc/xc_ptrace.c
@@ -1,6 +1,7 @@
 #include <sys/ptrace.h>
 #include <sys/wait.h>
 #include "xc_private.h"
+#include "xg_private.h"
 #include <time.h>
 
 #define X86_CR0_PE              0x00000001 /* Enable Protected Mode    (RW) */
@@ -10,14 +11,6 @@
 #define PSL_T  0x00000100 /* trace enable bit */
 #define VCPU            0               /* XXX */
 
-/*
- * long  
- * ptrace(enum __ptrace_request request, pid_t pid, void *addr, void *data);
- */
-
-
-int waitdomain(int domain, int *status, int options);
-
 char * ptrace_names[] = {
     "PTRACE_TRACEME",
     "PTRACE_PEEKTEXT",
@@ -122,8 +115,6 @@ struct gdb_regs {
 #define vtopti(va) (((va) >> PAGE_SHIFT) & 0x3ff)
 
 /* XXX application state */
-
-static int                      xc_handle;
 static long   nr_pages = 0;
 unsigned long   *page_array = NULL;
 static int                      regs_valid[MAX_VIRT_CPUS];
@@ -133,14 +124,60 @@ static vcpu_guest_context_t ctxt[MAX_VIRT_CPUS];
 static inline int paging_enabled(vcpu_guest_context_t *v)
 {
     unsigned long cr0 = v->ctrlreg[0];
-
     return (cr0 & X86_CR0_PE) && (cr0 & X86_CR0_PG);
 }
 
 /* --------------------- */
 
 static void *
-map_domain_va(unsigned long domid, int cpu, void * guest_va, int perm)
+map_domain_va_pae(
+    int xc_handle,
+    unsigned long domid,
+    int cpu,
+    void *guest_va,
+    int perm)
+{
+    unsigned long l2p, l1p, p, va = (unsigned long)guest_va;
+    u64 *l3, *l2, *l1;
+    static void *v;
+
+    FETCH_REGS(cpu);
+
+    l3 = xc_map_foreign_range(
+        xc_handle, domid, PAGE_SIZE, PROT_READ, cr3[cpu] >> PAGE_SHIFT);
+    if ( l3 == NULL )
+        goto error_out;
+
+    l2p = l3[l3_table_offset_pae(va)] >> PAGE_SHIFT;
+    l2 = xc_map_foreign_range(xc_handle, domid, PAGE_SIZE, PROT_READ, l2p);
+    if ( l2 == NULL )
+        goto error_out;
+
+    l1p = l2[l2_table_offset_pae(va)] >> PAGE_SHIFT;
+    l1 = xc_map_foreign_range(xc_handle, domid, PAGE_SIZE, perm, l1p);
+    if ( l1 == NULL )
+        goto error_out;
+
+    p = l1[l1_table_offset_pae(va)] >> PAGE_SHIFT;
+    if ( v != NULL )
+        munmap(v, PAGE_SIZE);
+    v = xc_map_foreign_range(xc_handle, domid, PAGE_SIZE, perm, p);
+    if ( v == NULL )
+        goto error_out;
+
+    return (void *)((unsigned long)v | (va & (PAGE_SIZE - 1)));
+
+ error_out:
+    return NULL;
+}
+
+static void *
+map_domain_va(
+    int xc_handle,
+    unsigned long domid,
+    int cpu,
+    void *guest_va,
+    int perm)
 {
     unsigned long pde, page;
     unsigned long va = (unsigned long)guest_va;
@@ -151,20 +188,35 @@ map_domain_va(unsigned long domid, int cpu, void * guest_va, int perm)
     static unsigned long  pde_phys[MAX_VIRT_CPUS];
     static unsigned long *pde_virt[MAX_VIRT_CPUS];
     static unsigned long  page_phys[MAX_VIRT_CPUS];
-    static unsigned long *page_virt[MAX_VIRT_CPUS];
-    
+    static unsigned long *page_virt[MAX_VIRT_CPUS];    
     static int            prev_perm[MAX_VIRT_CPUS];
+    static enum { MODE_UNKNOWN, MODE_32, MODE_PAE } mode;
+
+    if ( mode == MODE_UNKNOWN )
+    {
+        xen_capabilities_info_t caps;
+        (void)xc_version(xc_handle, XENVER_capabilities, caps);
+        mode = MODE_32;
+        if ( strstr(caps, "_x86_32p") )
+            mode = MODE_PAE;
+    }
+
+    if ( mode == MODE_PAE )
+        return map_domain_va_pae(xc_handle, domid, cpu, guest_va, perm);
 
-    if (nr_pages != npgs) {
-        if (nr_pages > 0)
+    if ( nr_pages != npgs )
+    {
+        if ( nr_pages > 0 )
             free(page_array);
         nr_pages = npgs;
-        if ((page_array = malloc(nr_pages * sizeof(unsigned long))) == NULL) {
+        if ( (page_array = malloc(nr_pages * sizeof(unsigned long))) == NULL )
+        {
             printf("Could not allocate memory\n");
             goto error_out;
         }
-
-        if (xc_get_pfn_list(xc_handle, domid, page_array, nr_pages) != nr_pages) {
+        if ( xc_get_pfn_list(xc_handle, domid,
+                             page_array, nr_pages) != nr_pages )
+        {
             printf("Could not get the page frame list\n");
             goto error_out;
         }
@@ -172,48 +224,52 @@ map_domain_va(unsigned long domid, int cpu, void * guest_va, int perm)
 
     FETCH_REGS(cpu);
 
-    if (cr3[cpu] != cr3_phys[cpu]) 
+    if ( cr3[cpu] != cr3_phys[cpu] )
     {
         cr3_phys[cpu] = cr3[cpu];
-        if (cr3_virt[cpu])
+        if ( cr3_virt[cpu] )
             munmap(cr3_virt[cpu], PAGE_SIZE);
-        if ((cr3_virt[cpu] = xc_map_foreign_range(xc_handle, domid, PAGE_SIZE,
-                                                  PROT_READ,
-                                                  cr3_phys[cpu] >> PAGE_SHIFT)) == NULL)
+        cr3_virt[cpu] = xc_map_foreign_range(
+            xc_handle, domid, PAGE_SIZE, PROT_READ,
+            cr3_phys[cpu] >> PAGE_SHIFT);
+        if ( cr3_virt[cpu] == NULL )
             goto error_out;
-    } 
-    if ((pde = cr3_virt[cpu][vtopdi(va)]) == 0) /* logical address */
+    }
+    if ( (pde = cr3_virt[cpu][vtopdi(va)]) == 0 )
         goto error_out;
-    if ((ctxt[cpu].flags & VGCF_VMX_GUEST) && paging_enabled(&ctxt[cpu]))
+    if ( (ctxt[cpu].flags & VGCF_VMX_GUEST) && paging_enabled(&ctxt[cpu]) )
         pde = page_array[pde >> PAGE_SHIFT] << PAGE_SHIFT;
-    if (pde != pde_phys[cpu]) 
+    if ( pde != pde_phys[cpu] )
     {
         pde_phys[cpu] = pde;
-        if (pde_virt[cpu])
+        if ( pde_virt[cpu] )
             munmap(pde_virt[cpu], PAGE_SIZE);
-        if ((pde_virt[cpu] = xc_map_foreign_range(xc_handle, domid, PAGE_SIZE,
-                                                  PROT_READ,
-                                                  pde_phys[cpu] >> PAGE_SHIFT)) == NULL)
+        pde_virt[cpu] = xc_map_foreign_range(
+            xc_handle, domid, PAGE_SIZE, PROT_READ,
+            pde_phys[cpu] >> PAGE_SHIFT);
+        if ( pde_virt[cpu] == NULL )
             goto error_out;
     }
-    if ((page = pde_virt[cpu][vtopti(va)]) == 0) /* logical address */
+    if ( (page = pde_virt[cpu][vtopti(va)]) == 0 )
         goto error_out;
-    if (ctxt[cpu].flags & VGCF_VMX_GUEST && paging_enabled(&ctxt[cpu]))
+    if ( (ctxt[cpu].flags & VGCF_VMX_GUEST) && paging_enabled(&ctxt[cpu]) )
         page = page_array[page >> PAGE_SHIFT] << PAGE_SHIFT;
-    if (page != page_phys[cpu] || perm != prev_perm[cpu]) 
+    if ( (page != page_phys[cpu]) || (perm != prev_perm[cpu]) )
     {
         page_phys[cpu] = page;
-        if (page_virt[cpu])
+        if ( page_virt[cpu] )
             munmap(page_virt[cpu], PAGE_SIZE);
-        if ((page_virt[cpu] = xc_map_foreign_range(xc_handle, domid, PAGE_SIZE,
-                                                   perm,
-                                                   page_phys[cpu] >> PAGE_SHIFT)) == NULL) {
-            printf("cr3 %lx pde %lx page %lx pti %lx\n", cr3[cpu], pde, page, vtopti(va));
+        page_virt[cpu] = xc_map_foreign_range(
+            xc_handle, domid, PAGE_SIZE, perm,
+            page_phys[cpu] >> PAGE_SHIFT);
+        if ( page_virt[cpu] == NULL )
+        {
             page_phys[cpu] = 0;
             goto error_out;
         }
         prev_perm[cpu] = perm;
     } 
+
     return (void *)(((unsigned long)page_virt[cpu]) | (va & BSD_PAGE_MASK));
 
  error_out:
@@ -221,7 +277,11 @@ map_domain_va(unsigned long domid, int cpu, void * guest_va, int perm)
 }
 
 int 
-xc_waitdomain(int domain, int *status, int options)
+xc_waitdomain(
+    int xc_handle,
+    int domain,
+    int *status,
+    int options)
 {
     dom0_op_t op;
     int retval;
@@ -229,38 +289,39 @@ xc_waitdomain(int domain, int *status, int options)
     ts.tv_sec = 0;
     ts.tv_nsec = 10*1000*1000;
 
-    if (!xc_handle)
-        if ((xc_handle = xc_interface_open()) < 0) 
-        {
-            printf("xc_interface_open failed\n");
-            return -1;
-        }
     op.cmd = DOM0_GETDOMAININFO;
     op.u.getdomaininfo.domain = domain;
- retry:
 
+ retry:
     retval = do_dom0_op(xc_handle, &op);
-    if (retval || op.u.getdomaininfo.domain != domain) {
+    if ( retval || (op.u.getdomaininfo.domain != domain) )
+    {
         printf("getdomaininfo failed\n");
         goto done;
     }
     *status = op.u.getdomaininfo.flags;
     
-    if (options & WNOHANG)
+    if ( options & WNOHANG )
         goto done;
- 
 
-    if (!(op.u.getdomaininfo.flags & DOMFLAGS_PAUSED)) { 
+    if ( !(op.u.getdomaininfo.flags & DOMFLAGS_PAUSED) )
+    {
         nanosleep(&ts,NULL);
         goto retry;
     }
+
  done:
     return retval;
 
 }
 
 long
-xc_ptrace(enum __ptrace_request request, u32 domid, long eaddr, long edata)
+xc_ptrace(
+    int xc_handle,
+    enum __ptrace_request request,
+    u32 domid,
+    long eaddr,
+    long edata)
 {
     dom0_op_t       op;
     int             status = 0;
@@ -273,44 +334,51 @@ xc_ptrace(enum __ptrace_request request, u32 domid, long eaddr, long edata)
 
     op.interface_version = DOM0_INTERFACE_VERSION;
     
-    if (!xc_handle)
-        if ((xc_handle = xc_interface_open()) < 0)
-            return -1;
-#if 0
-    printf("%20s %d, %p, %p \n", ptrace_names[request], domid, addr, data);
-#endif
-    switch (request) { 
+    switch ( request )
+    { 
     case PTRACE_PEEKTEXT:
     case PTRACE_PEEKDATA:
-        if ((guest_va = (unsigned long *)map_domain_va(domid, cpu, addr, PROT_READ)) == NULL) {
+        guest_va = (unsigned long *)map_domain_va(
+            xc_handle, domid, cpu, addr, PROT_READ);
+        if ( guest_va == NULL )
+        {
             status = EFAULT;
             goto error_out;
         }
-
         retval = *guest_va;
         break;
+
     case PTRACE_POKETEXT:
     case PTRACE_POKEDATA:
-        if ((guest_va = (unsigned long *)map_domain_va(domid, cpu, addr, PROT_READ|PROT_WRITE)) == NULL) {
+        guest_va = (unsigned long *)map_domain_va(
+            xc_handle, domid, cpu, addr, PROT_READ|PROT_WRITE);
+        if ( guest_va == NULL )
+        {
             status = EFAULT;
             goto error_out;
         }
-
         *guest_va = (unsigned long)data;
         break;
+
     case PTRACE_GETREGS:
     case PTRACE_GETFPREGS:
     case PTRACE_GETFPXREGS:
         FETCH_REGS(cpu);
-
-        if (request == PTRACE_GETREGS) {
+        if ( request == PTRACE_GETREGS )
+        {
             SET_PT_REGS(pt, ctxt[cpu].user_regs); 
             memcpy(data, &pt, sizeof(struct gdb_regs));
-        } else if (request == PTRACE_GETFPREGS)
+        }
+        else if (request == PTRACE_GETFPREGS)
+        {
             memcpy(data, &ctxt[cpu].fpu_ctxt, sizeof(ctxt[cpu].fpu_ctxt));
+        }
         else /*if (request == PTRACE_GETFPXREGS)*/
+        {
             memcpy(data, &ctxt[cpu].fpu_ctxt, sizeof(ctxt[cpu].fpu_ctxt));
+        }
         break;
+
     case PTRACE_SETREGS:
         op.cmd = DOM0_SETDOMAININFO;
         SET_XC_REGS(((struct gdb_regs *)data), ctxt[VCPU].user_regs);
@@ -321,17 +389,19 @@ xc_ptrace(enum __ptrace_request request, u32 domid, long eaddr, long edata)
         retval = do_dom0_op(xc_handle, &op);
         if (retval)
             goto error_out;
-
         break;
+
     case PTRACE_ATTACH:
         op.cmd = DOM0_GETDOMAININFO;
         op.u.getdomaininfo.domain = domid;
         retval = do_dom0_op(xc_handle, &op);
-        if (retval || op.u.getdomaininfo.domain != domid) {
+        if ( retval || (op.u.getdomaininfo.domain != domid) )
+        {
             perror("dom0 op failed");
             goto error_out;
         }
-        if (op.u.getdomaininfo.flags & DOMFLAGS_PAUSED) {
+        if ( op.u.getdomaininfo.flags & DOMFLAGS_PAUSED )
+        {
             printf("domain currently paused\n");
             goto error_out;
         }
@@ -340,6 +410,7 @@ xc_ptrace(enum __ptrace_request request, u32 domid, long eaddr, long edata)
         op.u.pausedomain.domain = domid;
         retval = do_dom0_op(xc_handle, &op);
         break;
+
     case PTRACE_SINGLESTEP:
         ctxt[VCPU].user_regs.eflags |= PSL_T;
         op.cmd = DOM0_SETDOMAININFO;
@@ -347,24 +418,29 @@ xc_ptrace(enum __ptrace_request request, u32 domid, long eaddr, long edata)
         op.u.setdomaininfo.vcpu = 0;
         op.u.setdomaininfo.ctxt = &ctxt[cpu];
         retval = do_dom0_op(xc_handle, &op); 
-        if (retval) {
+        if ( retval )
+        {
             perror("dom0 op failed");
             goto error_out;
         }
         /* FALLTHROUGH */
+
     case PTRACE_CONT:
     case PTRACE_DETACH:
-        if (request != PTRACE_SINGLESTEP) {
+        if ( request != PTRACE_SINGLESTEP )
+        {
             FETCH_REGS(cpu);
             /* Clear trace flag */
-            if (ctxt[cpu].user_regs.eflags & PSL_T) {
+            if ( ctxt[cpu].user_regs.eflags & PSL_T )
+            {
                 ctxt[cpu].user_regs.eflags &= ~PSL_T;
                 op.cmd = DOM0_SETDOMAININFO;
                 op.u.setdomaininfo.domain = domid;
                 op.u.setdomaininfo.vcpu = cpu;
                 op.u.setdomaininfo.ctxt = &ctxt[cpu];
                 retval = do_dom0_op(xc_handle, &op); 
-                if (retval) {
+                if ( retval )
+                {
                     perror("dom0 op failed");
                     goto error_out;
                 }
@@ -375,6 +451,7 @@ xc_ptrace(enum __ptrace_request request, u32 domid, long eaddr, long edata)
         op.u.unpausedomain.domain = domid > 0 ? domid : -domid;
         retval = do_dom0_op(xc_handle, &op);
         break;
+
     case PTRACE_SETFPREGS:
     case PTRACE_SETFPXREGS:
     case PTRACE_PEEKUSER:
@@ -387,15 +464,18 @@ xc_ptrace(enum __ptrace_request request, u32 domid, long eaddr, long edata)
         /* XXX not yet supported */
         status = ENOSYS;
         break;
+
     case PTRACE_TRACEME:
         printf("PTRACE_TRACEME is an invalid request under Xen\n");
         status = EINVAL;
     }
     
-    if (status) {
+    if ( status )
+    {
         errno = status;
         retval = -1;
     }
+
  error_out:
     return retval;
 }
diff --git a/tools/libxc/xc_ptrace_core.c b/tools/libxc/xc_ptrace_core.c
index ed05dc5666..ffb23cea2d 100644
--- a/tools/libxc/xc_ptrace_core.c
+++ b/tools/libxc/xc_ptrace_core.c
@@ -166,7 +166,11 @@ map_domain_va(unsigned long domfd, int cpu, void * guest_va)
 }
 
 int 
-xc_waitdomain_core(int domfd, int *status, int options)
+xc_waitdomain_core(
+    int xc_handle,
+    int domfd,
+    int *status,
+    int options)
 {
     int retval = -1;
     int nr_vcpus;
@@ -215,7 +219,12 @@ xc_waitdomain_core(int domfd, int *status, int options)
 }
 
 long
-xc_ptrace_core(enum __ptrace_request request, u32 domfd, long eaddr, long edata)
+xc_ptrace_core(
+    int xc_handle,
+    enum __ptrace_request request,
+    u32 domfd,
+    long eaddr,
+    long edata)
 {
     int             status = 0;
     struct gdb_regs pt;
diff --git a/tools/libxc/xenctrl.h b/tools/libxc/xenctrl.h
index 5b0adffe50..5094e17ba7 100644
--- a/tools/libxc/xenctrl.h
+++ b/tools/libxc/xenctrl.h
@@ -101,23 +101,31 @@ typedef struct xc_core_header {
 } xc_core_header_t;
 
 
-long xc_ptrace(enum __ptrace_request request, 
-               u32  domid,
-               long addr, 
-               long data);
-
-long xc_ptrace_core(enum __ptrace_request request, 
-                    u32 domid, 
-                    long addr, 
-                    long data);
-
-int xc_waitdomain(int domain, 
-                  int *status, 
-                  int options);
-
-int xc_waitdomain_core(int domain, 
-                       int *status, 
-                       int options);
+long xc_ptrace(
+    int xc_handle,
+    enum __ptrace_request request, 
+    u32  domid,
+    long addr, 
+    long data);
+
+long xc_ptrace_core(
+    int xc_handle,
+    enum __ptrace_request request, 
+    u32 domid, 
+    long addr, 
+    long data);
+
+int xc_waitdomain(
+    int xc_handle,
+    int domain, 
+    int *status, 
+    int options);
+
+int xc_waitdomain_core(
+    int xc_handle,
+    int domain, 
+    int *status, 
+    int options);
 
 /*
  * DOMAIN MANAGEMENT FUNCTIONS