diff mbox

[Xen-devel,5/6] xen/arm: Implement hypercall for dirty page tracing

Message ID 1397148539-19084-6-git-send-email-w1.huang@samsung.com
State New
Headers show

Commit Message

Wei Huang April 10, 2014, 4:48 p.m. UTC
From: Jaeyong Yoo <jaeyong.yoo@samsung.com>

This patch adds hypercall for shadow operations, including enable/disable
and clean/peek dirty page bitmap.

The design consists of two parts: dirty page detecting and saving. For
detecting, we setup the guest p2m's leaf PTE read-only and whenever
the guest tries to write something, permission fault happens and traps
into xen. The permission-faulted GPA should be saved for the toolstack,
which checks which pages are dirty. For this purpose, it temporarily saves
the GPAs into bitmap.

Signed-off-by: Jaeyong Yoo <jaeyong.yoo@samsung.com>
Signed-off-by: Wei Huang <w1.huang@samsung.com>
---
 xen/arch/arm/domain.c           |  14 +++
 xen/arch/arm/domctl.c           |   9 ++
 xen/arch/arm/mm.c               | 105 +++++++++++++++++++-
 xen/arch/arm/p2m.c              | 208 ++++++++++++++++++++++++++++++++++++++++
 xen/arch/arm/traps.c            |   9 ++
 xen/include/asm-arm/domain.h    |   7 ++
 xen/include/asm-arm/mm.h        |   8 ++
 xen/include/asm-arm/p2m.h       |   8 +-
 xen/include/asm-arm/processor.h |   2 +
 9 files changed, 368 insertions(+), 2 deletions(-)
diff mbox

Patch

diff --git a/xen/arch/arm/domain.c b/xen/arch/arm/domain.c
index 3f04a77..d2531ed 100644
--- a/xen/arch/arm/domain.c
+++ b/xen/arch/arm/domain.c
@@ -207,6 +207,12 @@  static void ctxt_switch_to(struct vcpu *n)
 
     isb();
 
+    /* Dirty-page tracing
+     * NB: How do we consider SMP case?
+     */
+    if ( n->domain->arch.dirty.mode )
+        restore_vlpt(n->domain);
+
     /* This is could trigger an hardware interrupt from the virtual
      * timer. The interrupt needs to be injected into the guest. */
     virt_timer_restore(n);
@@ -502,11 +508,19 @@  int arch_domain_create(struct domain *d, unsigned int domcr_flags)
     /* Default the virtual ID to match the physical */
     d->arch.vpidr = boot_cpu_data.midr.bits;
 
+    /* init for dirty-page tracing */
+    d->arch.dirty.count = 0;
+    d->arch.dirty.mode = 0;
+    spin_lock_init(&d->arch.dirty.lock);
+
     d->arch.dirty.second_lvl_start = 0;
     d->arch.dirty.second_lvl_end = 0;
     d->arch.dirty.second_lvl[0] = NULL;
     d->arch.dirty.second_lvl[1] = NULL;
 
+    memset(d->arch.dirty.bitmap, 0, sizeof(d->arch.dirty.bitmap));
+    d->arch.dirty.bitmap_pages = 0;
+
     clear_page(d->shared_info);
     share_xen_page_with_guest(
         virt_to_page(d->shared_info), d, XENSHARE_writable);
diff --git a/xen/arch/arm/domctl.c b/xen/arch/arm/domctl.c
index 914de29..e67381b 100644
--- a/xen/arch/arm/domctl.c
+++ b/xen/arch/arm/domctl.c
@@ -95,6 +95,15 @@  long arch_do_domctl(struct xen_domctl *domctl, struct domain *d,
             xfree(c.data);
     }
     break;
+    case XEN_DOMCTL_shadow_op:
+    {
+        domain_pause(d);
+        ret = dirty_mode_op(d, &domctl->u.shadow_op);
+        domain_unpause(d);
+
+        copyback = 1;
+    }
+    break;
 
     case XEN_DOMCTL_cacheflush:
     {
diff --git a/xen/arch/arm/mm.c b/xen/arch/arm/mm.c
index df9d428..ea0b051 100644
--- a/xen/arch/arm/mm.c
+++ b/xen/arch/arm/mm.c
@@ -865,7 +865,6 @@  void destroy_xen_mappings(unsigned long v, unsigned long e)
     create_xen_entries(REMOVE, v, 0, (e - v) >> PAGE_SHIFT, 0);
 }
 
-enum mg { mg_clear, mg_ro, mg_rw, mg_rx };
 static void set_pte_flags_on_range(const char *p, unsigned long l, enum mg mg)
 {
     lpae_t pte;
@@ -1240,9 +1239,64 @@  int is_iomem_page(unsigned long mfn)
     return 0;
 }
 
+
 /*
  * Return start and end addresses of guest
  */
+static inline void mark_dirty_bitmap(struct domain *d, paddr_t addr)
+{
+    paddr_t ram_base = (paddr_t) GUEST_RAM_BASE;
+    int bit_index = PFN_DOWN(addr - ram_base);
+    int page_index = bit_index >> (PAGE_SHIFT + 3);
+    int bit_index_residual = bit_index & ((1ul << (PAGE_SHIFT + 3)) - 1);
+
+    set_bit(bit_index_residual, d->arch.dirty.bitmap[page_index]);
+}
+
+/* Routine for dirty-page tracing
+ *
+ * On first write, it page faults, its entry is changed to read-write,
+ * and on retry the write succeeds. For locating p2m of the faulting entry,
+ * we use virtual-linear page table.
+ *
+ * Returns zero if addr is not valid or dirty mode is not set
+ */
+int handle_page_fault(struct domain *d, paddr_t addr)
+{
+
+    lpae_t *vlp2m_pte = 0;
+    paddr_t gma_start = 0;
+    paddr_t gma_end = 0;
+
+    if ( !d->arch.dirty.mode )
+        return 0;
+    get_gma_start_end(d, &gma_start, &gma_end);
+
+    /* Ensure that addr is inside guest's RAM */
+    if ( addr < gma_start || addr > gma_end )
+        return 0;
+
+    vlp2m_pte = get_vlpt_3lvl_pte(addr);
+    if ( vlp2m_pte->p2m.valid && vlp2m_pte->p2m.write == 0 &&
+         vlp2m_pte->p2m.type == p2m_ram_logdirty )
+    {
+        lpae_t pte = *vlp2m_pte;
+        pte.p2m.write = 1;
+        write_pte(vlp2m_pte, pte);
+        flush_tlb_local();
+
+        /* only necessary to lock between get-dirty bitmap and mark dirty
+         * bitmap. If get-dirty bitmap happens immediately before this
+         * lock, the corresponding dirty-page would be marked at the next
+         * round of get-dirty bitmap */
+        spin_lock(&d->arch.dirty.lock);
+        mark_dirty_bitmap(d, addr);
+        spin_unlock(&d->arch.dirty.lock);
+    }
+
+    return 1;
+}
+
 void get_gma_start_end(struct domain *d, paddr_t *start, paddr_t *end)
 {
     if ( start )
@@ -1368,6 +1422,55 @@  void cleanup_vlpt(struct domain *d)
     unmap_domain_page_global(d->arch.dirty.second_lvl[0]);
     unmap_domain_page_global(d->arch.dirty.second_lvl[1]);
 }
+
+int prepare_bitmap(struct domain *d)
+{
+    paddr_t gma_start = 0;
+    paddr_t gma_end = 0;
+    int nr_bytes;
+    int nr_pages;
+    int i;
+
+    get_gma_start_end(d, &gma_start, &gma_end);
+
+    nr_bytes = (PFN_DOWN(gma_end - gma_start) + 7) / 8;
+    nr_pages = (nr_bytes + PAGE_SIZE - 1) / PAGE_SIZE;
+
+    BUG_ON( nr_pages > MAX_DIRTY_BITMAP_PAGES );
+
+    for ( i = 0; i < nr_pages; ++i )
+    {
+        struct page_info *page;
+        page = alloc_domheap_page(NULL, 0);
+        if ( page == NULL )
+            goto cleanup_on_failure;
+
+        d->arch.dirty.bitmap[i] = map_domain_page_global(__page_to_mfn(page));
+        clear_page(d->arch.dirty.bitmap[i]);
+    }
+
+    d->arch.dirty.bitmap_pages = nr_pages;
+    return 0;
+
+cleanup_on_failure:
+    nr_pages = i;
+    for ( i = 0; i < nr_pages; ++i )
+    {
+        unmap_domain_page_global(d->arch.dirty.bitmap[i]);
+    }
+
+    return -ENOMEM;
+}
+
+void cleanup_bitmap(struct domain *d)
+{
+    int i;
+    for ( i = 0; i < d->arch.dirty.bitmap_pages; ++i )
+    {
+        unmap_domain_page_global(d->arch.dirty.bitmap[i]);
+    }
+}
+
 /*
  * Local variables:
  * mode: C
diff --git a/xen/arch/arm/p2m.c b/xen/arch/arm/p2m.c
index 403fd89..2925f80 100644
--- a/xen/arch/arm/p2m.c
+++ b/xen/arch/arm/p2m.c
@@ -6,6 +6,8 @@ 
 #include <xen/bitops.h>
 #include <asm/flushtlb.h>
 #include <asm/gic.h>
+#include <xen/guest_access.h>
+#include <xen/pfn.h>
 #include <asm/event.h>
 #include <asm/hardirq.h>
 #include <asm/page.h>
@@ -208,6 +210,7 @@  static lpae_t mfn_to_p2m_entry(unsigned long mfn, unsigned int mattr,
         break;
 
     case p2m_ram_ro:
+    case p2m_ram_logdirty:
         e.p2m.xn = 0;
         e.p2m.write = 0;
         break;
@@ -261,6 +264,10 @@  static int p2m_create_table(struct domain *d,
 
     pte = mfn_to_p2m_entry(page_to_mfn(page), MATTR_MEM, p2m_invalid);
 
+    /* mark the write bit (page table's case, ro bit) as 0
+     * so, it is writable in case of vlpt access */
+    pte.pt.ro = 0;
+
     write_pte(entry, pte);
 
     return 0;
@@ -697,6 +704,207 @@  unsigned long gmfn_to_mfn(struct domain *d, unsigned long gpfn)
     return p >> PAGE_SHIFT;
 }
 
+/* Change types across all p2m entries in a domain */
+void p2m_change_entry_type_global(struct domain *d, enum mg nt)
+{
+    struct p2m_domain *p2m = &d->arch.p2m;
+    paddr_t ram_base;
+    int i1, i2, i3;
+    int first_index, second_index, third_index;
+    lpae_t *first = __map_domain_page(p2m->first_level);
+    lpae_t pte, *second = NULL, *third = NULL;
+
+    get_gma_start_end(d, &ram_base, NULL);
+
+    first_index = first_table_offset((uint64_t)ram_base);
+    second_index = second_table_offset((uint64_t)ram_base);
+    third_index = third_table_offset((uint64_t)ram_base);
+
+    BUG_ON( !first && "Can't map first level p2m." );
+
+    spin_lock(&p2m->lock);
+
+    for ( i1 = first_index; i1 < LPAE_ENTRIES*2; ++i1 )
+    {
+        lpae_walk_t first_pte = first[i1].walk;
+        if ( !first_pte.valid || !first_pte.table )
+            goto out;
+
+        second = map_domain_page(first_pte.base);
+        BUG_ON( !second && "Can't map second level p2m.");
+        for ( i2 = second_index; i2 < LPAE_ENTRIES; ++i2 )
+        {
+            lpae_walk_t second_pte = second[i2].walk;
+
+            if ( !second_pte.valid || !second_pte.table )
+                goto out;
+
+            third = map_domain_page(second_pte.base);
+            BUG_ON( !third && "Can't map third level p2m.");
+
+            for ( i3 = third_index; i3 < LPAE_ENTRIES; ++i3 )
+            {
+                lpae_walk_t third_pte = third[i3].walk;
+                if ( !third_pte.valid )
+                    goto out;
+
+                pte = third[i3];
+                if ( nt == mg_ro )
+                {
+                    if ( pte.p2m.write == 1 )
+                    {
+                        pte.p2m.write = 0;
+                        pte.p2m.type = p2m_ram_logdirty;
+                    }
+                    else
+                    {
+                        /* reuse avail bit as an indicator of 'actual'
+                         * read-only */
+                        pte.p2m.type = p2m_ram_rw;
+                    }
+                }
+                else if ( nt == mg_rw )
+                {
+                    if ( pte.p2m.write == 0 &&
+                         pte.p2m.type == p2m_ram_logdirty )
+                    {
+                        pte.p2m.write = p2m_ram_rw;
+                    }
+                }
+                write_pte(&third[i3], pte);
+            }
+            unmap_domain_page(third);
+
+            third = NULL;
+            third_index = 0;
+        }
+        unmap_domain_page(second);
+
+        second = NULL;
+        second_index = 0;
+        third_index = 0;
+    }
+
+out:
+    flush_tlb_all_local();
+    if ( third ) unmap_domain_page(third);
+    if ( second ) unmap_domain_page(second);
+    if ( first ) unmap_domain_page(first);
+
+    spin_unlock(&p2m->lock);
+}
+
+/* Read a domain's log-dirty bitmap and stats.
+ * If the operation is a CLEAN, clear the bitmap and stats. */
+int log_dirty_op(struct domain *d, xen_domctl_shadow_op_t *sc)
+{
+    int peek = 1;
+    int i;
+    int bitmap_size;
+    paddr_t gma_start, gma_end;
+
+    /* this hypercall is called from domain 0, and we don't know which guest's
+     * vlpt is mapped in xen_second, so, to be sure, we restore vlpt here */
+    restore_vlpt(d);
+
+    get_gma_start_end(d, &gma_start, &gma_end);
+    bitmap_size = (gma_end - gma_start) / 8;
+
+    if ( guest_handle_is_null(sc->dirty_bitmap) )
+    {
+        peek = 0;
+    }
+    else
+    {
+        spin_lock(&d->arch.dirty.lock);
+        for ( i = 0; i < d->arch.dirty.bitmap_pages; ++i )
+        {
+            int j = 0;
+            uint8_t *bitmap;
+            copy_to_guest_offset(sc->dirty_bitmap, i * PAGE_SIZE,
+                                 d->arch.dirty.bitmap[i],
+                                 bitmap_size < PAGE_SIZE ? bitmap_size :
+                                                           PAGE_SIZE);
+            bitmap_size -= PAGE_SIZE;
+
+            /* set p2m page table read-only */
+            bitmap = d->arch.dirty.bitmap[i];
+            while ((j = find_next_bit((const long unsigned int *)bitmap,
+                                      PAGE_SIZE*8, j)) < PAGE_SIZE*8)
+            {
+                lpae_t *vlpt;
+                paddr_t addr = gma_start +
+                               (i << (2*PAGE_SHIFT+3)) +
+                               (j << PAGE_SHIFT);
+                vlpt = get_vlpt_3lvl_pte(addr);
+                vlpt->p2m.write = 0;
+                j++;
+            }
+        }
+
+        if ( sc->op == XEN_DOMCTL_SHADOW_OP_CLEAN )
+        {
+            for ( i = 0; i < d->arch.dirty.bitmap_pages; ++i )
+            {
+                clear_page(d->arch.dirty.bitmap[i]);
+            }
+        }
+
+        spin_unlock(&d->arch.dirty.lock);
+        flush_tlb_local();
+    }
+
+    sc->stats.dirty_count = d->arch.dirty.count;
+
+    return 0;
+}
+
+long dirty_mode_op(struct domain *d, xen_domctl_shadow_op_t *sc)
+{
+    long ret = 0;
+    switch (sc->op)
+    {
+        case XEN_DOMCTL_SHADOW_OP_ENABLE_LOGDIRTY:
+        case XEN_DOMCTL_SHADOW_OP_OFF:
+        {
+            enum mg nt = sc->op == XEN_DOMCTL_SHADOW_OP_OFF ? mg_rw : mg_ro;
+
+            d->arch.dirty.mode = sc->op == XEN_DOMCTL_SHADOW_OP_OFF ? 0 : 1;
+            p2m_change_entry_type_global(d, nt);
+
+            if ( sc->op == XEN_DOMCTL_SHADOW_OP_OFF )
+            {
+                cleanup_vlpt(d);
+                cleanup_bitmap(d);
+            }
+            else
+            {
+                if ( (ret = prepare_vlpt(d)) )
+                   return ret;
+
+                if ( (ret = prepare_bitmap(d)) )
+                {
+                   /* in case of failure, we have to cleanup vlpt */
+                   cleanup_vlpt(d);
+                   return ret;
+                }
+            }
+        }
+        break;
+
+        case XEN_DOMCTL_SHADOW_OP_CLEAN:
+        case XEN_DOMCTL_SHADOW_OP_PEEK:
+        {
+            ret = log_dirty_op(d, sc);
+        }
+        break;
+
+        default:
+            return -ENOSYS;
+    }
+    return ret;
+}
+
 /*
  * Local variables:
  * mode: C
diff --git a/xen/arch/arm/traps.c b/xen/arch/arm/traps.c
index a7edc4e..42c8b75 100644
--- a/xen/arch/arm/traps.c
+++ b/xen/arch/arm/traps.c
@@ -1491,6 +1491,8 @@  static void do_trap_data_abort_guest(struct cpu_user_regs *regs,
     struct hsr_dabt dabt = hsr.dabt;
     int rc;
     mmio_info_t info;
+    int page_fault = ( (dabt.dfsc & FSC_MASK) ==
+                          (FSC_FLT_PERM | FSC_3D_LEVEL) && dabt.write );
 
     if ( !check_conditional_instr(regs, hsr) )
     {
@@ -1512,6 +1514,13 @@  static void do_trap_data_abort_guest(struct cpu_user_regs *regs,
     if ( rc == -EFAULT )
         goto bad_data_abort;
 
+    /* domU page fault handling for guest live migration */
+    /* dabt.valid can be 0 here */
+    if ( page_fault && handle_page_fault(current->domain, info.gpa) )
+    {
+        /* Do not modify pc after page fault to repeat memory operation */
+        return;
+    }
     /* XXX: Decode the instruction if ISS is not valid */
     if ( !dabt.valid )
         goto bad_data_abort;
diff --git a/xen/include/asm-arm/domain.h b/xen/include/asm-arm/domain.h
index 5321bd6..99f9f51 100644
--- a/xen/include/asm-arm/domain.h
+++ b/xen/include/asm-arm/domain.h
@@ -163,9 +163,16 @@  struct arch_domain
 
     /* dirty-page tracing */
     struct {
+#define MAX_DIRTY_BITMAP_PAGES 64        /* support upto 8GB guest memory */
+        spinlock_t lock;                 /* protect list: head, mvn_head */
+        volatile int mode;               /* 1 if dirty pages tracing enabled */
+        volatile unsigned int count;     /* dirty pages counter */
         volatile int second_lvl_start;   /* for context switch */
         volatile int second_lvl_end;
         lpae_t *second_lvl[2];           /* copy of guest p2m's first */
+        /* dirty bitmap */
+        uint8_t *bitmap[MAX_DIRTY_BITMAP_PAGES];
+        int bitmap_pages;                /* number of dirty bitmap pages */
     } dirty;
 
     unsigned int evtchn_irq;
diff --git a/xen/include/asm-arm/mm.h b/xen/include/asm-arm/mm.h
index 75c27fb..6733021 100644
--- a/xen/include/asm-arm/mm.h
+++ b/xen/include/asm-arm/mm.h
@@ -342,11 +342,19 @@  static inline void put_page_and_type(struct page_info *page)
     put_page(page);
 }
 
+enum mg { mg_clear, mg_ro, mg_rw, mg_rx };
+
+/* routine for dirty-page tracing */
+int handle_page_fault(struct domain *d, paddr_t addr);
+
 void get_gma_start_end(struct domain *d, paddr_t *start, paddr_t *end);
 int prepare_vlpt(struct domain *d);
 void cleanup_vlpt(struct domain *d);
 void restore_vlpt(struct domain *d);
 
+int prepare_bitmap(struct domain *d);
+void cleanup_bitmap(struct domain *d);
+
 /* calculate the xen''s virtual address for accessing the leaf PTE of
  * a given address (GPA) */
 static inline lpae_t * get_vlpt_3lvl_pte(paddr_t addr)
diff --git a/xen/include/asm-arm/p2m.h b/xen/include/asm-arm/p2m.h
index bd71abe..0cecbe7 100644
--- a/xen/include/asm-arm/p2m.h
+++ b/xen/include/asm-arm/p2m.h
@@ -2,6 +2,7 @@ 
 #define _XEN_P2M_H
 
 #include <xen/mm.h>
+#include <public/domctl.h>
 
 struct domain;
 
@@ -41,6 +42,7 @@  typedef enum {
     p2m_invalid = 0,    /* Nothing mapped here */
     p2m_ram_rw,         /* Normal read/write guest RAM */
     p2m_ram_ro,         /* Read-only; writes are silently dropped */
+    p2m_ram_logdirty,   /* Read-only: special mode for log dirty */
     p2m_mmio_direct,    /* Read/write mapping of genuine MMIO area */
     p2m_map_foreign,    /* Ram pages from foreign domain */
     p2m_grant_map_rw,   /* Read/write grant mapping */
@@ -49,7 +51,8 @@  typedef enum {
 } p2m_type_t;
 
 #define p2m_is_foreign(_t)  ((_t) == p2m_map_foreign)
-#define p2m_is_ram(_t)      ((_t) == p2m_ram_rw || (_t) == p2m_ram_ro)
+#define p2m_is_ram(_t)      ((_t) == p2m_ram_rw || (_t) == p2m_ram_ro ||  \
+                             (_t) == p2m_ram_logdirty)
 
 /* Initialise vmid allocator */
 void p2m_vmid_allocator_init(void);
@@ -178,6 +181,9 @@  static inline int get_page_and_type(struct page_info *page,
     return rc;
 }
 
+void p2m_change_entry_type_global(struct domain *d, enum mg nt);
+long dirty_mode_op(struct domain *d, xen_domctl_shadow_op_t *sc);
+
 #endif /* _XEN_P2M_H */
 
 /*
diff --git a/xen/include/asm-arm/processor.h b/xen/include/asm-arm/processor.h
index 06e638f..9dc49c3 100644
--- a/xen/include/asm-arm/processor.h
+++ b/xen/include/asm-arm/processor.h
@@ -399,6 +399,8 @@  union hsr {
 #define FSC_CPR        (0x3a) /* Coprocossor Abort */
 
 #define FSC_LL_MASK    (_AC(0x03,U)<<0)
+#define FSC_MASK       (0x3f) /* Fault status mask */
+#define FSC_3D_LEVEL   (0x03) /* Third level fault*/
 
 /* Time counter hypervisor control register */
 #define CNTHCTL_PA      (1u<<0)  /* Kernel/user access to physical counter */