aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/xen/enlighten.c
diff options
context:
space:
mode:
Diffstat (limited to 'arch/x86/xen/enlighten.c')
-rw-r--r--arch/x86/xen/enlighten.c118
1 files changed, 11 insertions, 107 deletions
diff --git a/arch/x86/xen/enlighten.c b/arch/x86/xen/enlighten.c
index bf4bda6d3e9a..9642d4a38602 100644
--- a/arch/x86/xen/enlighten.c
+++ b/arch/x86/xen/enlighten.c
@@ -31,7 +31,6 @@
31#include <linux/pci.h> 31#include <linux/pci.h>
32#include <linux/gfp.h> 32#include <linux/gfp.h>
33#include <linux/memblock.h> 33#include <linux/memblock.h>
34#include <linux/syscore_ops.h>
35 34
36#include <xen/xen.h> 35#include <xen/xen.h>
37#include <xen/interface/xen.h> 36#include <xen/interface/xen.h>
@@ -1470,130 +1469,38 @@ asmlinkage void __init xen_start_kernel(void)
1470#endif 1469#endif
1471} 1470}
1472 1471
1473#ifdef CONFIG_XEN_PVHVM 1472void __ref xen_hvm_init_shared_info(void)
1474/*
1475 * The pfn containing the shared_info is located somewhere in RAM. This
1476 * will cause trouble if the current kernel is doing a kexec boot into a
1477 * new kernel. The new kernel (and its startup code) can not know where
1478 * the pfn is, so it can not reserve the page. The hypervisor will
1479 * continue to update the pfn, and as a result memory corruption occours
1480 * in the new kernel.
1481 *
1482 * One way to work around this issue is to allocate a page in the
1483 * xen-platform pci device's BAR memory range. But pci init is done very
1484 * late and the shared_info page is already in use very early to read
1485 * the pvclock. So moving the pfn from RAM to MMIO is racy because some
1486 * code paths on other vcpus could access the pfn during the small
1487 * window when the old pfn is moved to the new pfn. There is even a
1488 * small window were the old pfn is not backed by a mfn, and during that
1489 * time all reads return -1.
1490 *
1491 * Because it is not known upfront where the MMIO region is located it
1492 * can not be used right from the start in xen_hvm_init_shared_info.
1493 *
1494 * To minimise trouble the move of the pfn is done shortly before kexec.
1495 * This does not eliminate the race because all vcpus are still online
1496 * when the syscore_ops will be called. But hopefully there is no work
1497 * pending at this point in time. Also the syscore_op is run last which
1498 * reduces the risk further.
1499 */
1500
1501static struct shared_info *xen_hvm_shared_info;
1502
1503static void xen_hvm_connect_shared_info(unsigned long pfn)
1504{ 1473{
1474 int cpu;
1505 struct xen_add_to_physmap xatp; 1475 struct xen_add_to_physmap xatp;
1476 static struct shared_info *shared_info_page = 0;
1506 1477
1478 if (!shared_info_page)
1479 shared_info_page = (struct shared_info *)
1480 extend_brk(PAGE_SIZE, PAGE_SIZE);
1507 xatp.domid = DOMID_SELF; 1481 xatp.domid = DOMID_SELF;
1508 xatp.idx = 0; 1482 xatp.idx = 0;
1509 xatp.space = XENMAPSPACE_shared_info; 1483 xatp.space = XENMAPSPACE_shared_info;
1510 xatp.gpfn = pfn; 1484 xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
1511 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp)) 1485 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
1512 BUG(); 1486 BUG();
1513 1487
1514} 1488 HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
1515static void xen_hvm_set_shared_info(struct shared_info *sip)
1516{
1517 int cpu;
1518
1519 HYPERVISOR_shared_info = sip;
1520 1489
1521 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info 1490 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1522 * page, we use it in the event channel upcall and in some pvclock 1491 * page, we use it in the event channel upcall and in some pvclock
1523 * related functions. We don't need the vcpu_info placement 1492 * related functions. We don't need the vcpu_info placement
1524 * optimizations because we don't use any pv_mmu or pv_irq op on 1493 * optimizations because we don't use any pv_mmu or pv_irq op on
1525 * HVM. 1494 * HVM.
1526 * When xen_hvm_set_shared_info is run at boot time only vcpu 0 is 1495 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1527 * online but xen_hvm_set_shared_info is run at resume time too and 1496 * online but xen_hvm_init_shared_info is run at resume time too and
1528 * in that case multiple vcpus might be online. */ 1497 * in that case multiple vcpus might be online. */
1529 for_each_online_cpu(cpu) { 1498 for_each_online_cpu(cpu) {
1530 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu]; 1499 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
1531 } 1500 }
1532} 1501}
1533 1502
1534/* Reconnect the shared_info pfn to a mfn */ 1503#ifdef CONFIG_XEN_PVHVM
1535void xen_hvm_resume_shared_info(void)
1536{
1537 xen_hvm_connect_shared_info(__pa(xen_hvm_shared_info) >> PAGE_SHIFT);
1538}
1539
1540#ifdef CONFIG_KEXEC
1541static struct shared_info *xen_hvm_shared_info_kexec;
1542static unsigned long xen_hvm_shared_info_pfn_kexec;
1543
1544/* Remember a pfn in MMIO space for kexec reboot */
1545void __devinit xen_hvm_prepare_kexec(struct shared_info *sip, unsigned long pfn)
1546{
1547 xen_hvm_shared_info_kexec = sip;
1548 xen_hvm_shared_info_pfn_kexec = pfn;
1549}
1550
1551static void xen_hvm_syscore_shutdown(void)
1552{
1553 struct xen_memory_reservation reservation = {
1554 .domid = DOMID_SELF,
1555 .nr_extents = 1,
1556 };
1557 unsigned long prev_pfn;
1558 int rc;
1559
1560 if (!xen_hvm_shared_info_kexec)
1561 return;
1562
1563 prev_pfn = __pa(xen_hvm_shared_info) >> PAGE_SHIFT;
1564 set_xen_guest_handle(reservation.extent_start, &prev_pfn);
1565
1566 /* Move pfn to MMIO, disconnects previous pfn from mfn */
1567 xen_hvm_connect_shared_info(xen_hvm_shared_info_pfn_kexec);
1568
1569 /* Update pointers, following hypercall is also a memory barrier */
1570 xen_hvm_set_shared_info(xen_hvm_shared_info_kexec);
1571
1572 /* Allocate new mfn for previous pfn */
1573 do {
1574 rc = HYPERVISOR_memory_op(XENMEM_populate_physmap, &reservation);
1575 if (rc == 0)
1576 msleep(123);
1577 } while (rc == 0);
1578
1579 /* Make sure the previous pfn is really connected to a (new) mfn */
1580 BUG_ON(rc != 1);
1581}
1582
1583static struct syscore_ops xen_hvm_syscore_ops = {
1584 .shutdown = xen_hvm_syscore_shutdown,
1585};
1586#endif
1587
1588/* Use a pfn in RAM, may move to MMIO before kexec. */
1589static void __init xen_hvm_init_shared_info(void)
1590{
1591 /* Remember pointer for resume */
1592 xen_hvm_shared_info = extend_brk(PAGE_SIZE, PAGE_SIZE);
1593 xen_hvm_connect_shared_info(__pa(xen_hvm_shared_info) >> PAGE_SHIFT);
1594 xen_hvm_set_shared_info(xen_hvm_shared_info);
1595}
1596
1597static void __init init_hvm_pv_info(void) 1504static void __init init_hvm_pv_info(void)
1598{ 1505{
1599 int major, minor; 1506 int major, minor;
@@ -1644,9 +1551,6 @@ static void __init xen_hvm_guest_init(void)
1644 init_hvm_pv_info(); 1551 init_hvm_pv_info();
1645 1552
1646 xen_hvm_init_shared_info(); 1553 xen_hvm_init_shared_info();
1647#ifdef CONFIG_KEXEC
1648 register_syscore_ops(&xen_hvm_syscore_ops);
1649#endif
1650 1554
1651 if (xen_feature(XENFEAT_hvm_callback_vector)) 1555 if (xen_feature(XENFEAT_hvm_callback_vector))
1652 xen_have_vector_callback = 1; 1556 xen_have_vector_callback = 1;