Re: [Qemu-devel] [PATCH 2/8] s390: autodetect map private

[Christian Borntraeger <borntraeger@de.ibm.com>]

> Is there any way we can move the above code to target-s390x? Having the
> branch below is already invasive enough for generic code, but we really
> don't need all the special s390 quirks to live here.


Hmm, we have to have a special hook somehow.
What about this approach?

-----------------------

By default qemu will use MAP_PRIVATE for guest pages. This will write
protect pages and thus break on s390 systems that dont support this feature.
Therefore qemu has a hack to always use MAP_SHARED for s390. But MAP_SHARED
has other problems (no dirty pages tracking, a lot more swap overhead etc.)
Newer systems allow the distinction via KVM_CAP_S390_COW. With this feature
qemu can use the standard qemu alloc if available, otherwise it will use
the old s390 hack.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
---
 exec.c             |   19 ++++---------------
 kvm.h              |    1 +
 oslib-posix.c      |    3 +++
 target-s390x/kvm.c |   34 ++++++++++++++++++++++++++++++++++
 4 files changed, 42 insertions(+), 15 deletions(-)

diff --git a/exec.c b/exec.c
index a587e7a..9b9b8e1 100644
--- a/exec.c
+++ b/exec.c
@@ -2645,26 +2645,15 @@ ram_addr_t qemu_ram_alloc_from_ptr(ram_addr_t size, void *host,
             exit(1);
 #endif
         } else {
-#if defined(TARGET_S390X) && defined(CONFIG_KVM)
-            /* S390 KVM requires the topmost vma of the RAM to be smaller than
-               an system defined value, which is at least 256GB. Larger systems
-               have larger values. We put the guest between the end of data
-               segment (system break) and this value. We use 32GB as a base to
-               have enough room for the system break to grow. */
-            new_block->host = mmap((void*)0x800000000, size,
-                                   PROT_EXEC|PROT_READ|PROT_WRITE,
-                                   MAP_SHARED | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
-            if (new_block->host == MAP_FAILED) {
-                fprintf(stderr, "Allocating RAM failed\n");
-                abort();
-            }
-#else
             if (xen_enabled()) {
                 xen_ram_alloc(new_block->offset, size, mr);
+#if defined(TARGET_S390X)
+            } else if (kvm_enabled()) {
+                new_block->host = kvm_arch_alloc(size);
+#endif
             } else {
                 new_block->host = qemu_vmalloc(size);
             }
-#endif
             qemu_madvise(new_block->host, size, QEMU_MADV_MERGEABLE);
         }
     }
diff --git a/kvm.h b/kvm.h
index 9c7b0ea..9d50016 100644
--- a/kvm.h
+++ b/kvm.h
@@ -102,6 +102,7 @@ int kvm_vcpu_ioctl(CPUArchState *env, int type, ...);
 
 extern const KVMCapabilityInfo kvm_arch_required_capabilities[];
 
+void *kvm_arch_alloc(ram_addr_t size);
 void kvm_arch_pre_run(CPUArchState *env, struct kvm_run *run);
 void kvm_arch_post_run(CPUArchState *env, struct kvm_run *run);
 
diff --git a/oslib-posix.c b/oslib-posix.c
index b6a3c7f..93902ac 100644
--- a/oslib-posix.c
+++ b/oslib-posix.c
@@ -41,6 +41,9 @@ extern int daemon(int, int);
       therefore we need special code which handles running on Valgrind. */
 #  define QEMU_VMALLOC_ALIGN (512 * 4096)
 #  define CONFIG_VALGRIND
+#elif defined(__linux__) && defined(__s390x__)
+   /* Use 1 MiB (segment size) alignment so gmap can be used by KVM. */
+#  define QEMU_VMALLOC_ALIGN (256 * 4096)
 #else
 #  define QEMU_VMALLOC_ALIGN getpagesize()
 #endif
diff --git a/target-s390x/kvm.c b/target-s390x/kvm.c
index 90aad61..ccf5daa 100644
--- a/target-s390x/kvm.c
+++ b/target-s390x/kvm.c
@@ -135,6 +135,40 @@ int kvm_arch_get_registers(CPUS390XState *env)
     return 0;
 }
 
+/*
+ * Legacy layout for s390:
+ * Older S390 KVM requires the topmost vma of the RAM to be
+ * smaller than an system defined value, which is at least 256GB.
+ * Larger systems have larger values. We put the guest between
+ * the end of data segment (system break) and this value. We
+ * use 32GB as a base to have enough room for the system break
+ * to grow. We also have to use MAP parameters that avoid
+ * read-only mapping of guest pages.
+ */
+static void *legacy_s390_alloc(ram_addr_t size)
+{
+    void *mem;
+
+    mem = mmap((void *) 0x800000000ULL, size,
+               PROT_EXEC|PROT_READ|PROT_WRITE,
+               MAP_SHARED | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
+    if (mem == MAP_FAILED) {
+        fprintf(stderr, "Allocating RAM failed\n");
+        abort();
+    }
+    return mem;
+}
+
+void *kvm_arch_alloc(ram_addr_t size)
+{
+    if (kvm_check_extension(kvm_state, KVM_CAP_S390_GMAP) &&
+        kvm_check_extension(kvm_state, KVM_CAP_S390_COW)) {
+        return qemu_vmalloc(size);
+    } else {
+        return legacy_s390_alloc(size);
+    }
+}
+
 int kvm_arch_insert_sw_breakpoint(CPUS390XState *env, struct kvm_sw_breakpoint *bp)
 {
     static const uint8_t diag_501[] = {0x83, 0x24, 0x05, 0x01};