[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[xen master] x86: introduce helper for recording degree of contiguity in page tables



commit 8c6a4963f07da518a74c35344bc9b28a84a78fee
Author:     Jan Beulich <jbeulich@xxxxxxxx>
AuthorDate: Mon Jul 25 15:37:34 2022 +0200
Commit:     Jan Beulich <jbeulich@xxxxxxxx>
CommitDate: Mon Jul 25 15:37:34 2022 +0200

    x86: introduce helper for recording degree of contiguity in page tables
    
    This is a re-usable helper (kind of a template) which gets introduced
    without users so that the individual subsequent patches introducing such
    users can get committed independently of one another.
    
    See the comment at the top of the new file. To demonstrate the effect,
    if a page table had just 16 entries, this would be the set of markers
    for a page table with fully contiguous mappings:
    
    index  0 1 2 3 4 5 6 7 8 9 A B C D E F
    marker 4 0 1 0 2 0 1 0 3 0 1 0 2 0 1 0
    
    "Contiguous" here means not only present entries with successively
    increasing MFNs, each one suitably aligned for its slot, but also a
    respective number of all non-present entries.
    
    Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
    Reviewed-by: Roger Pau Monné <roger.pau@xxxxxxxxxx>
    Reviewed-by: Paul Durrant <paul@xxxxxxx>
---
 xen/arch/x86/include/asm/pt-contig-markers.h | 110 +++++++++++++++++++++++++++
 1 file changed, 110 insertions(+)

diff --git a/xen/arch/x86/include/asm/pt-contig-markers.h 
b/xen/arch/x86/include/asm/pt-contig-markers.h
new file mode 100644
index 0000000000..b3c1fe8035
--- /dev/null
+++ b/xen/arch/x86/include/asm/pt-contig-markers.h
@@ -0,0 +1,110 @@
+#ifndef __ASM_X86_PT_CONTIG_MARKERS_H
+#define __ASM_X86_PT_CONTIG_MARKERS_H
+
+/*
+ * Short of having function templates in C, the function defined below is
+ * intended to be used by multiple parties interested in recording the
+ * degree of contiguity in mappings by a single page table.
+ *
+ * Scheme: Every entry records the order of contiguous successive entries,
+ * up to the maximum order covered by that entry (which is the number of
+ * clear low bits in its index, with entry 0 being the exception using
+ * the base-2 logarithm of the number of entries in a single page table).
+ * While a few entries need touching upon update, knowing whether the
+ * table is fully contiguous (and can hence be replaced by a higher level
+ * leaf entry) is then possible by simply looking at entry 0's marker.
+ *
+ * Prereqs:
+ * - CONTIG_MASK needs to be #define-d, to a value having at least 4
+ *   contiguous bits (ignored by hardware), before including this file (or
+ *   else only CONTIG_LEVEL_SHIFT and CONTIG_NR will become available),
+ * - page tables to be passed to the helper need to be initialized with
+ *   correct markers,
+ * - not-present entries need to be entirely clear except for the marker.
+ */
+
+/* This is the same for all anticipated users, so doesn't need passing in. */
+#define CONTIG_LEVEL_SHIFT 9
+#define CONTIG_NR          (1 << CONTIG_LEVEL_SHIFT)
+
+#ifdef CONTIG_MASK
+
+#include <xen/bitops.h>
+#include <xen/lib.h>
+#include <xen/page-size.h>
+
+#define GET_MARKER(e) MASK_EXTR(e, CONTIG_MASK)
+#define SET_MARKER(e, m) \
+    ((void)((e) = ((e) & ~CONTIG_MASK) | MASK_INSR(m, CONTIG_MASK)))
+
+#define IS_CONTIG(kind, pt, i, idx, shift, b) \
+    ((kind) == PTE_kind_leaf \
+     ? (((pt)[i] ^ (pt)[idx]) & ~CONTIG_MASK) == (1ULL << ((b) + (shift))) \
+     : !((pt)[i] & ~CONTIG_MASK))
+
+enum PTE_kind {
+    PTE_kind_null,
+    PTE_kind_leaf,
+    PTE_kind_table,
+};
+
+static bool pt_update_contig_markers(uint64_t *pt, unsigned int idx,
+                                     unsigned int level, enum PTE_kind kind)
+{
+    unsigned int b, i = idx;
+    unsigned int shift = (level - 1) * CONTIG_LEVEL_SHIFT + PAGE_SHIFT;
+
+    ASSERT(idx < CONTIG_NR);
+    ASSERT(!(pt[idx] & CONTIG_MASK));
+
+    /* Step 1: Reduce markers in lower numbered entries. */
+    while ( i )
+    {
+        b = find_first_set_bit(i);
+        i &= ~(1U << b);
+        if ( GET_MARKER(pt[i]) <= b )
+            break;
+        SET_MARKER(pt[i], b);
+    }
+
+    /* An intermediate table is never contiguous with anything. */
+    if ( kind == PTE_kind_table )
+        return false;
+
+    /*
+     * Present entries need in-sync index and address to be a candidate
+     * for being contiguous: What we're after is whether ultimately the
+     * intermediate table can be replaced by a superpage.
+     */
+    if ( kind != PTE_kind_null &&
+         idx != ((pt[idx] >> shift) & (CONTIG_NR - 1)) )
+        return false;
+
+    /* Step 2: Check higher numbered entries for contiguity. */
+    for ( b = 0; b < CONTIG_LEVEL_SHIFT && !(idx & (1U << b)); ++b )
+    {
+        i = idx | (1U << b);
+        if ( !IS_CONTIG(kind, pt, i, idx, shift, b) || GET_MARKER(pt[i]) != b )
+            break;
+    }
+
+    /* Step 3: Update markers in this and lower numbered entries. */
+    for ( ; SET_MARKER(pt[idx], b), b < CONTIG_LEVEL_SHIFT; ++b )
+    {
+        i = idx ^ (1U << b);
+        if ( !IS_CONTIG(kind, pt, i, idx, shift, b) || GET_MARKER(pt[i]) != b )
+            break;
+        idx &= ~(1U << b);
+    }
+
+    return b == CONTIG_LEVEL_SHIFT;
+}
+
+#undef IS_CONTIG
+#undef SET_MARKER
+#undef GET_MARKER
+#undef CONTIG_MASK
+
+#endif /* CONTIG_MASK */
+
+#endif /* __ASM_X86_PT_CONTIG_MARKERS_H */
--
generated by git-patchbot for /home/xen/git/xen.git#master



 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.