[Xen-devel] [PATCH 4/6] xen/arm: Implement VLPT for guest p2m mapping in live migration

[Wei Huang <w1.huang@xxxxxxxxxxx>]

From: Jaeyong Yoo <jaeyong.yoo@xxxxxxxxxxx>

Thsi patch implements VLPT (virtual-linear page table) for fast accessing
of 3rd PTE of guest P2M. For more info about VLPT, please see
http://www.technovelty.org/linux/virtual-linear-page-table.html.

When creating a mapping for VLPT, just copy the 1st level PTE of guest p2m
to xen's 2nd level PTE. Then the mapping becomes the following:
      xen's 1st PTE -->
      xen's 2nd PTE (which is the same as 1st PTE of guest p2m) -->
      guest p2m's 2nd PTE -->
      guest p2m's 3rd PTE (the memory contents where the vlpt points)

This function is used in dirty-page tracing. When domU write-fault is
trapped by xen, xen can immediately locate the 3rd PTE of guest p2m.

The following link shows the performance comparison for handling a
dirty-page between vlpt and typical page table walking.
http://lists.xen.org/archives/html/xen-devel/2013-08/msg01503.html

Signed-off-by: Jaeyong Yoo <jaeyong.yoo@xxxxxxxxxxx>
---
 xen/arch/arm/domain.c            |   5 ++
 xen/arch/arm/mm.c                | 116 +++++++++++++++++++++++++++++++++++++++
 xen/include/asm-arm/arm32/page.h |  23 ++++----
 xen/include/asm-arm/config.h     |   9 +++
 xen/include/asm-arm/domain.h     |   7 +++
 xen/include/asm-arm/mm.h         |  16 ++++++
 6 files changed, 166 insertions(+), 10 deletions(-)

diff --git a/xen/arch/arm/domain.c b/xen/arch/arm/domain.c
index b125857..3f04a77 100644
--- a/xen/arch/arm/domain.c
+++ b/xen/arch/arm/domain.c
@@ -502,6 +502,11 @@ 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;
 
+    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;
+
     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/mm.c b/xen/arch/arm/mm.c
index 14b4686..df9d428 100644
--- a/xen/arch/arm/mm.c
+++ b/xen/arch/arm/mm.c
@@ -1252,6 +1252,122 @@ void get_gma_start_end(struct domain *d, paddr_t 
*start, paddr_t *end)
         *end = GUEST_RAM_BASE + ((paddr_t) d->max_pages << PAGE_SHIFT);
 }
 
+/* flush the vlpt area */
+void flush_vlpt(struct domain *d)
+{
+    int flush_size;
+    flush_size = (d->arch.dirty.second_lvl_end -
+                  d->arch.dirty.second_lvl_start) << SECOND_SHIFT;
+
+    /* flushing the 3rd level mapping */
+    flush_xen_data_tlb_range_va(d->arch.dirty.second_lvl_start << SECOND_SHIFT,
+                                flush_size);
+}
+
+/* restore the xen page table for vlpt mapping for domain d */
+void restore_vlpt(struct domain *d)
+{
+    int i;
+
+    dsb(sy);
+
+    for ( i = d->arch.dirty.second_lvl_start;
+          i < d->arch.dirty.second_lvl_end;
+          ++i )
+    {
+        int k = i % LPAE_ENTRIES;
+        int l = i / LPAE_ENTRIES;
+
+        if ( xen_second[i].bits != d->arch.dirty.second_lvl[l][k].bits )
+        {
+            write_pte(&xen_second[i], d->arch.dirty.second_lvl[l][k]);
+            flush_xen_data_tlb_range_va(i << SECOND_SHIFT, 1 << SECOND_SHIFT);
+        }
+    }
+
+    dsb(sy);
+    isb();
+}
+
+/* setting up the xen page table for vlpt mapping for domain d */
+int prepare_vlpt(struct domain *d)
+{
+    int xen_second_linear_base;
+    int gp2m_start_index, gp2m_end_index;
+    struct p2m_domain *p2m = &d->arch.p2m;
+    struct page_info *second_lvl_page;
+    paddr_t gma_start = 0;
+    paddr_t gma_end = 0;
+    lpae_t *first[2];
+    int i;
+    uint64_t required, avail = VIRT_LIN_P2M_END - VIRT_LIN_P2M_START;
+
+    get_gma_start_end(d, &gma_start, &gma_end);
+    required = (gma_end - gma_start) >> LPAE_SHIFT;
+
+    if ( required > avail )
+    {
+        dprintk(XENLOG_ERR, "Available VLPT is small for domU guest"
+                "(avail: %llx, required: %llx)\n", (unsigned long long)avail,
+                (unsigned long long)required);
+        return -ENOMEM;
+    }
+
+    xen_second_linear_base = second_linear_offset(VIRT_LIN_P2M_START);
+
+    gp2m_start_index = gma_start >> FIRST_SHIFT;
+    gp2m_end_index = (gma_end >> FIRST_SHIFT) + 1;
+
+    if ( xen_second_linear_base + gp2m_end_index >= LPAE_ENTRIES * 2 )
+    {
+        dprintk(XENLOG_ERR, "xen second page is small for VLPT for domU");
+        return -ENOMEM;
+    }
+
+    second_lvl_page = alloc_domheap_pages(NULL, 1, 0);
+    if ( second_lvl_page == NULL )
+        return -ENOMEM;
+
+    /* First level p2m is 2 consecutive pages */
+    d->arch.dirty.second_lvl[0] = map_domain_page_global(
+        page_to_mfn(second_lvl_page) );
+    d->arch.dirty.second_lvl[1] = map_domain_page_global(
+        page_to_mfn(second_lvl_page+1) );
+
+    first[0] = __map_domain_page(p2m->first_level);
+    first[1] = __map_domain_page(p2m->first_level+1);
+
+    for ( i = gp2m_start_index; i < gp2m_end_index; ++i )
+    {
+        int k = i % LPAE_ENTRIES;
+        int l = i / LPAE_ENTRIES;
+        int k2 = (xen_second_linear_base + i) % LPAE_ENTRIES;
+        int l2 = (xen_second_linear_base + i) / LPAE_ENTRIES;
+
+        write_pte(&xen_second[xen_second_linear_base+i], first[l][k]);
+
+        /* we copy the mapping into domain''s structure as a reference
+         * in case of the context switch (used in restore_vlpt) */
+        d->arch.dirty.second_lvl[l2][k2] = first[l][k];
+    }
+    unmap_domain_page(first[0]);
+    unmap_domain_page(first[1]);
+
+    /* storing the start and end index */
+    d->arch.dirty.second_lvl_start = xen_second_linear_base + gp2m_start_index;
+    d->arch.dirty.second_lvl_end = xen_second_linear_base + gp2m_end_index;
+
+    flush_vlpt(d);
+
+    return 0;
+}
+
+void cleanup_vlpt(struct domain *d)
+{
+    /* First level p2m is 2 consecutive pages */
+    unmap_domain_page_global(d->arch.dirty.second_lvl[0]);
+    unmap_domain_page_global(d->arch.dirty.second_lvl[1]);
+}
 /*
  * Local variables:
  * mode: C
diff --git a/xen/include/asm-arm/arm32/page.h b/xen/include/asm-arm/arm32/page.h
index 4abb281..feca06c 100644
--- a/xen/include/asm-arm/arm32/page.h
+++ b/xen/include/asm-arm/arm32/page.h
@@ -3,22 +3,25 @@
 
 #ifndef __ASSEMBLY__
 
-/* Write a pagetable entry.
- *
- * If the table entry is changing a text mapping, it is responsibility
- * of the caller to issue an ISB after write_pte.
- */
-static inline void write_pte(lpae_t *p, lpae_t pte)
+/* Write a pagetable entry. All necessary barriers are responsibility of
+ * the caller */
+static inline void __write_pte(lpae_t *p, lpae_t pte)
 {
     asm volatile (
-        /* Ensure any writes have completed with the old mappings. */
-        "dsb;"
-        /* Safely write the entry (STRD is atomic on CPUs that support LPAE) */
+        /* safely write the entry (STRD is atomic on CPUs that support LPAE) */
         "strd %0, %H0, [%1];"
-        "dsb;"
         : : "r" (pte.bits), "r" (p) : "memory");
 }
 
+/* Write a pagetable entry with dsb barriers. All necessary barriers are
+ * responsibility of the caller. */
+static inline void write_pte(lpae_t *p, lpae_t pte)
+{
+    dsb();
+    __write_pte(p, pte);
+    dsb();
+}
+
 /* Inline ASM to flush dcache on register R (may be an inline asm operand) */
 #define __clean_xen_dcache_one(R) STORE_CP32(R, DCCMVAC)
 
diff --git a/xen/include/asm-arm/config.h b/xen/include/asm-arm/config.h
index 5b7b1a8..95e84bd 100644
--- a/xen/include/asm-arm/config.h
+++ b/xen/include/asm-arm/config.h
@@ -87,6 +87,7 @@
  *   0  -   8M   <COMMON>
  *
  *  32M - 128M   Frametable: 24 bytes per page for 16GB of RAM
+ * 128M - 256M   Virtual-linear mapping to P2M table
  * 256M -   1G   VMAP: ioremap and early_ioremap use this virtual address
  *                    space
  *
@@ -124,7 +125,9 @@
 #define CONFIG_SEPARATE_XENHEAP 1
 
 #define FRAMETABLE_VIRT_START  _AT(vaddr_t,0x02000000)
+#define VIRT_LIN_P2M_START     _AT(vaddr_t,0x08000000)
 #define VMAP_VIRT_START  _AT(vaddr_t,0x10000000)
+#define VIRT_LIN_P2M_END       VMAP_VIRT_START
 #define XENHEAP_VIRT_START     _AT(vaddr_t,0x40000000)
 #define XENHEAP_VIRT_END       _AT(vaddr_t,0x7fffffff)
 #define DOMHEAP_VIRT_START     _AT(vaddr_t,0x80000000)
@@ -157,6 +160,12 @@
 
 #define HYPERVISOR_VIRT_END    DIRECTMAP_VIRT_END
 
+/* Temporary definition for VIRT_LIN_P2M_START and VIRT_LIN_P2M_END
+ * TODO: Needs evaluation!!!!
+ */
+#define VIRT_LIN_P2M_START     _AT(vaddr_t, 0x08000000)
+#define VIRT_LIN_P2M_END       VMAP_VIRT_START
+
 #endif
 
 /* Fixmap slots */
diff --git a/xen/include/asm-arm/domain.h b/xen/include/asm-arm/domain.h
index 28c359a..5321bd6 100644
--- a/xen/include/asm-arm/domain.h
+++ b/xen/include/asm-arm/domain.h
@@ -161,6 +161,13 @@ struct arch_domain
         spinlock_t                  lock;
     } vuart;
 
+    /* dirty-page tracing */
+    struct {
+        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;
+
     unsigned int evtchn_irq;
 }  __cacheline_aligned;
 
diff --git a/xen/include/asm-arm/mm.h b/xen/include/asm-arm/mm.h
index 341493a..75c27fb 100644
--- a/xen/include/asm-arm/mm.h
+++ b/xen/include/asm-arm/mm.h
@@ -4,6 +4,7 @@
 #include <xen/config.h>
 #include <xen/kernel.h>
 #include <asm/page.h>
+#include <asm/config.h>
 #include <public/xen.h>
 
 /* Align Xen to a 2 MiB boundary. */
@@ -342,6 +343,21 @@ static inline void put_page_and_type(struct page_info 
*page)
 }
 
 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);
+
+/* 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)
+{
+    lpae_t *table = (lpae_t *)VIRT_LIN_P2M_START;
+
+    /* Since we slotted the guest''s first p2m page table to xen''s
+     * second page table, one shift is enough for calculating the
+     * index of guest p2m table entry */
+    return &table[addr >> PAGE_SHIFT];
+}
 
 #endif /*  __ARCH_ARM_MM__ */
 /*
-- 
1.8.3.2


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel