[Xen-devel] [PATCH v2] tools: libxc: flush data cache after loading images into guest memory

[Ian Campbell <ian.campbell@xxxxxxxxxx>]

On ARM guest OSes are started with MMU and Caches disables (as they are on
native) however caching is enabled in the domain running the builder and
therefore we must flush the cache as we load the blobs, otherwise when the
guest starts running it may not see them. The dom0 build in the hypervisor has
the same requirements and already does the right thing.

The mechanism for performing a cache flush from userspace is OS specific, so
implement this as a new osdep hook:

 - On 32-bit ARM Linux provides a system call to flush the cache.
 - On 64-bit ARM Linux the processor is configured to allow cache flushes
   directly from userspace.
 - Non-Linux platforms will need to provide their own implementation. If
   similar mechanisms are not available then a new privcmd ioctl should be a
   suitable alternative.

No cache maintenance is required on x86, so provide a stub for all non-Linux
platforms which returns success on x86 only and log an error otherwise.

This fixes guest building on Xgene which has a very large L3 cache and so is
particularly susceptible to this problem. It has also been observed
sporadically on midway.

Signed-off-by: Ian Campbell <ian.campbell@xxxxxxxxxx>
Cc: Andre Przywara <andre.przywara@xxxxxxxxxxx>
Cc: Pranavkumar Sawargaonkar <psawargaonkar@xxxxxxx>
Cc: Anup Patel <apatel@xxxxxxx>
---
v2: Add IPRINTF to ENOSYS debug module.
Freeze: Bugfix.
---
 tools/libxc/xc_dom_armzimageloader.c |    1 +
 tools/libxc/xc_dom_binloader.c       |    1 +
 tools/libxc/xc_dom_core.c            |    2 ++
 tools/libxc/xc_linux_osdep.c         |   39 ++++++++++++++++++++++++++++++++++
 tools/libxc/xc_minios.c              |   11 ++++++++++
 tools/libxc/xc_netbsd.c              |   12 +++++++++++
 tools/libxc/xc_private.c             |    5 +++++
 tools/libxc/xc_private.h             |    3 +++
 tools/libxc/xc_solaris.c             |   12 +++++++++++
 tools/libxc/xenctrl_osdep_ENOSYS.c   |    9 ++++++++
 tools/libxc/xenctrlosdep.h           |    1 +
 11 files changed, 96 insertions(+)

diff --git a/tools/libxc/xc_dom_armzimageloader.c 
b/tools/libxc/xc_dom_armzimageloader.c
index e6516a1..508f74b 100644
--- a/tools/libxc/xc_dom_armzimageloader.c
+++ b/tools/libxc/xc_dom_armzimageloader.c
@@ -229,6 +229,7 @@ static int xc_dom_load_zimage_kernel(struct xc_dom_image 
*dom)
               __func__, dom->kernel_size, dom->kernel_blob, dst);
 
     memcpy(dst, dom->kernel_blob, dom->kernel_size);
+    xc_cache_flush(dom->xch, dst, dom->kernel_size);
 
     return 0;
 }
diff --git a/tools/libxc/xc_dom_binloader.c b/tools/libxc/xc_dom_binloader.c
index e1de5b5..aa0463c 100644
--- a/tools/libxc/xc_dom_binloader.c
+++ b/tools/libxc/xc_dom_binloader.c
@@ -301,6 +301,7 @@ static int xc_dom_load_bin_kernel(struct xc_dom_image *dom)
 
     memcpy(dest, image + skip, text_size);
     memset(dest + text_size, 0, bss_size);
+    xc_cache_flush(dom->xch, dest, text_size+bss_size);
 
     return 0;
 }
diff --git a/tools/libxc/xc_dom_core.c b/tools/libxc/xc_dom_core.c
index 77a4e64..d46ac22 100644
--- a/tools/libxc/xc_dom_core.c
+++ b/tools/libxc/xc_dom_core.c
@@ -978,6 +978,7 @@ int xc_dom_build_image(struct xc_dom_image *dom)
         }
         else
             memcpy(ramdiskmap, dom->ramdisk_blob, dom->ramdisk_size);
+        xc_cache_flush(dom->xch, ramdiskmap, ramdisklen);
     }
 
     /* load devicetree */
@@ -997,6 +998,7 @@ int xc_dom_build_image(struct xc_dom_image *dom)
             goto err;
         }
         memcpy(devicetreemap, dom->devicetree_blob, dom->devicetree_size);
+        xc_cache_flush(dom->xch, devicetreemap, dom->devicetree_size);
     }
 
     /* allocate other pages */
diff --git a/tools/libxc/xc_linux_osdep.c b/tools/libxc/xc_linux_osdep.c
index 73860a2..8362495 100644
--- a/tools/libxc/xc_linux_osdep.c
+++ b/tools/libxc/xc_linux_osdep.c
@@ -30,6 +30,7 @@
 
 #include <sys/mman.h>
 #include <sys/ioctl.h>
+#include <sys/syscall.h>
 
 #include <xen/memory.h>
 #include <xen/sys/evtchn.h>
@@ -416,6 +417,42 @@ static void *linux_privcmd_map_foreign_ranges(xc_interface 
*xch, xc_osdep_handle
     return ret;
 }
 
+static void linux_privcmd_cache_flush(xc_interface *xch,
+                                     const void *ptr, size_t nr)
+{
+#if defined(__arm__)
+    unsigned long start = (unsigned long)ptr;
+    unsigned long end = start + nr;
+    /* cacheflush(unsigned long start, unsigned long end, int flags) */
+    int rc = syscall(__ARM_NR_cacheflush, start, end, 0);
+    if ( rc < 0 )
+           PERROR("cache flush operation failed: %d\n", errno);
+#elif defined(__aarch64__)
+    unsigned long start = (unsigned long)ptr;
+    unsigned long end = start + nr;
+    unsigned long p, ctr;
+    int stride;
+
+    /* Flush cache using direct DC CVAC instructions. This is
+     * available to EL0 when SCTLR_EL1.UCI is set, which Linux does.
+     *
+     * Bits 19:16 of CTR_EL0 are log2 of the minimum dcache line size
+     * in words, which we use as our stride length. This is readable
+     * with SCTLR_EL1.UCT is set, which Linux does.
+     */
+    asm volatile ("mrs %0, ctr_el0" : "=r" (ctr));
+
+    stride = 4 * (1 << ((ctr & 0xf0000UL) >> 16));
+
+    for ( p = start ; p < end ; p += stride )
+        asm volatile ("dc cvac, %0" :  : "r" (p));
+#elif defined(__i386__) || defined(__x86_64__)
+    /* No need for cache maintenance on x86 */
+#else
+    PERROR("No cache flush operation defined for architecture");
+#endif
+}
+
 static struct xc_osdep_ops linux_privcmd_ops = {
     .open = &linux_privcmd_open,
     .close = &linux_privcmd_close,
@@ -430,6 +467,8 @@ static struct xc_osdep_ops linux_privcmd_ops = {
         .map_foreign_bulk = &linux_privcmd_map_foreign_bulk,
         .map_foreign_range = &linux_privcmd_map_foreign_range,
         .map_foreign_ranges = &linux_privcmd_map_foreign_ranges,
+
+        .cache_flush = &linux_privcmd_cache_flush,
     },
 };
 
diff --git a/tools/libxc/xc_minios.c b/tools/libxc/xc_minios.c
index dec4d73..3b2f553 100644
--- a/tools/libxc/xc_minios.c
+++ b/tools/libxc/xc_minios.c
@@ -181,6 +181,15 @@ static void 
*minios_privcmd_map_foreign_ranges(xc_interface *xch, xc_osdep_handl
     return ret;
 }
 
+static void minios_privcmd_cache_flush(xc_interface *xch,
+                                       const void *ptr, size_t nr)
+{
+#if defined(__i386__) || defined(__x86_64__)
+    /* No need for cache maintenance on x86 */
+#else
+    PERROR("No cache flush operation defined for architecture");
+#endif
+}
 
 static struct xc_osdep_ops minios_privcmd_ops = {
     .open = &minios_privcmd_open,
@@ -196,6 +205,8 @@ static struct xc_osdep_ops minios_privcmd_ops = {
         .map_foreign_bulk = &minios_privcmd_map_foreign_bulk,
         .map_foreign_range = &minios_privcmd_map_foreign_range,
         .map_foreign_ranges = &minios_privcmd_map_foreign_ranges,
+
+        .cache_flush = &minios_privcmd_cache_flush,
     },
 };
 
diff --git a/tools/libxc/xc_netbsd.c b/tools/libxc/xc_netbsd.c
index 8a90ef3..11e1027 100644
--- a/tools/libxc/xc_netbsd.c
+++ b/tools/libxc/xc_netbsd.c
@@ -207,6 +207,16 @@ mmap_failed:
        return NULL;
 }
 
+static void netbsd_privcmd_cache_flush(xc_interface *xch,
+                                       const void *ptr, size_t nr)
+{
+#if defined(__i386__) || defined(__x86_64__)
+    /* No need for cache maintenance on x86 */
+#else
+    PERROR("No cache flush operation defined for architecture");
+#endif
+}
+
 static struct xc_osdep_ops netbsd_privcmd_ops = {
     .open = &netbsd_privcmd_open,
     .close = &netbsd_privcmd_close,
@@ -221,6 +231,8 @@ static struct xc_osdep_ops netbsd_privcmd_ops = {
         .map_foreign_bulk = &xc_map_foreign_bulk_compat,
         .map_foreign_range = &netbsd_privcmd_map_foreign_range,
         .map_foreign_ranges = &netbsd_privcmd_map_foreign_ranges,
+
+        .cache_flush = &netbsd_privcmd_cache_flush,
     },
 };
 
diff --git a/tools/libxc/xc_private.c b/tools/libxc/xc_private.c
index 838fd21..3ccee2b 100644
--- a/tools/libxc/xc_private.c
+++ b/tools/libxc/xc_private.c
@@ -249,6 +249,11 @@ int do_xen_hypercall(xc_interface *xch, 
privcmd_hypercall_t *hypercall)
     return xch->ops->u.privcmd.hypercall(xch, xch->ops_handle, hypercall);
 }
 
+void xc_cache_flush(xc_interface *xch, const void *p, size_t n)
+{
+    xch->ops->u.privcmd.cache_flush(xch, p, n);
+}
+
 xc_evtchn *xc_evtchn_open(xentoollog_logger *logger,
                              unsigned open_flags)
 {
diff --git a/tools/libxc/xc_private.h b/tools/libxc/xc_private.h
index 92271c9..50a0aa7 100644
--- a/tools/libxc/xc_private.h
+++ b/tools/libxc/xc_private.h
@@ -304,6 +304,9 @@ void bitmap_byte_to_64(uint64_t *lp, const uint8_t *bp, int 
nbits);
 /* Optionally flush file to disk and discard page cache */
 void discard_file_cache(xc_interface *xch, int fd, int flush);
 
+/* Flush data cache */
+void xc_cache_flush(xc_interface *xch, const void *p, size_t n);
+
 #define MAX_MMU_UPDATES 1024
 struct xc_mmu {
     mmu_update_t updates[MAX_MMU_UPDATES];
diff --git a/tools/libxc/xc_solaris.c b/tools/libxc/xc_solaris.c
index 7257a54..83c3777 100644
--- a/tools/libxc/xc_solaris.c
+++ b/tools/libxc/xc_solaris.c
@@ -178,6 +178,16 @@ mmap_failed:
     return NULL;
 }
 
+static void solaris_privcmd_cache_flush(xc_interface *xch,
+                                        const void *ptr, size_t nr)
+{
+#if defined(__i386__) || defined(__x86_64__)
+    /* No need for cache maintenance on x86 */
+#else
+    PERROR("No cache flush operation defined for architecture");
+#endif
+}
+
 static struct xc_osdep_ops solaris_privcmd_ops = {
     .open = &solaris_privcmd_open,
     .close = &solaris_privcmd_close,
@@ -192,6 +202,8 @@ static struct xc_osdep_ops solaris_privcmd_ops = {
         .map_foreign_bulk = &xc_map_foreign_bulk_compat,
         .map_foreign_range = &solaris_privcmd_map_foreign_range,
         .map_foreign_ranges = &solaris_privcmd_map_foreign_ranges,
+
+        .cache_flush = &solaris_privcmd_cache_flush,
     },
 };
 
diff --git a/tools/libxc/xenctrl_osdep_ENOSYS.c 
b/tools/libxc/xenctrl_osdep_ENOSYS.c
index 4821342..d911b10 100644
--- a/tools/libxc/xenctrl_osdep_ENOSYS.c
+++ b/tools/libxc/xenctrl_osdep_ENOSYS.c
@@ -63,6 +63,13 @@ static void *ENOSYS_privcmd_map_foreign_ranges(xc_interface 
*xch, xc_osdep_handl
     return MAP_FAILED;
 }
 
+static void ENOSYS_privcmd_cache_flush(xc_interface *xch, const void *p, 
size_t n)
+{
+    unsigned long start = (unsigned long)p;
+    unsigned long end = start + n;
+    IPRINTF(xch, "ENOSYS_privcmd: cache_flush: %#lx-%#lx\n", start, end);
+}
+
 static struct xc_osdep_ops ENOSYS_privcmd_ops =
 {
     .open      = &ENOSYS_privcmd_open,
@@ -74,6 +81,8 @@ static struct xc_osdep_ops ENOSYS_privcmd_ops =
         .map_foreign_bulk = &ENOSYS_privcmd_map_foreign_bulk,
         .map_foreign_range = &ENOSYS_privcmd_map_foreign_range,
         .map_foreign_ranges = &ENOSYS_privcmd_map_foreign_ranges,
+
+        .cache_flush = &ENOSYS_privcmd_cache_flush,
     }
 };
 
diff --git a/tools/libxc/xenctrlosdep.h b/tools/libxc/xenctrlosdep.h
index e610a24..6c9a005 100644
--- a/tools/libxc/xenctrlosdep.h
+++ b/tools/libxc/xenctrlosdep.h
@@ -89,6 +89,7 @@ struct xc_osdep_ops
             void *(*map_foreign_ranges)(xc_interface *xch, xc_osdep_handle h, 
uint32_t dom, size_t size, int prot,
                                         size_t chunksize, privcmd_mmap_entry_t 
entries[],
                                         int nentries);
+            void (*cache_flush)(xc_interface *xch, const void *p, size_t n);
         } privcmd;
         struct {
             int (*fd)(xc_evtchn *xce, xc_osdep_handle h);
-- 
1.7.10.4


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