Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
Merge tag 'stable/for-linus-3.6-rc3-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen
Pull three xen bug-fixes from Konrad Rzeszutek Wilk:
- Revert the kexec fix which caused on non-kexec shutdowns a race.
- Reuse existing P2M leafs - instead of requiring to allocate a large
area of bootup virtual address estate.
- Fix a one-off error when adding PFNs for balloon pages.
* tag 'stable/for-linus-3.6-rc3-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen:
xen/setup: Fix one-off error when adding for-balloon PFNs to the P2M.
xen/p2m: Reuse existing P2M leafs if they are filled with 1:1 PFNs or INVALID.
Revert "xen PVonHVM: move shared_info to MMIO before kexec"
+113
-130
+11
-107
arch/x86/xen/enlighten.c
+11
-107
arch/x86/xen/enlighten.c
···
31
31
#include <linux/pci.h>
32
32
#include <linux/gfp.h>
33
33
#include <linux/memblock.h>
34
-
#include <linux/syscore_ops.h>
35
34
36
35
#include <xen/xen.h>
37
36
#include <xen/interface/xen.h>
···
1469
1470
#endif
1470
1471
}
1471
1472
1472
-
#ifdef CONFIG_XEN_PVHVM
1473
-
/*
1474
-
* The pfn containing the shared_info is located somewhere in RAM. This
1475
-
* will cause trouble if the current kernel is doing a kexec boot into a
1476
-
* new kernel. The new kernel (and its startup code) can not know where
1477
-
* the pfn is, so it can not reserve the page. The hypervisor will
1478
-
* continue to update the pfn, and as a result memory corruption occours
1479
-
* in the new kernel.
1480
-
*
1481
-
* One way to work around this issue is to allocate a page in the
1482
-
* xen-platform pci device's BAR memory range. But pci init is done very
1483
-
* late and the shared_info page is already in use very early to read
1484
-
* the pvclock. So moving the pfn from RAM to MMIO is racy because some
1485
-
* code paths on other vcpus could access the pfn during the small
1486
-
* window when the old pfn is moved to the new pfn. There is even a
1487
-
* small window were the old pfn is not backed by a mfn, and during that
1488
-
* time all reads return -1.
1489
-
*
1490
-
* Because it is not known upfront where the MMIO region is located it
1491
-
* can not be used right from the start in xen_hvm_init_shared_info.
1492
-
*
1493
-
* To minimise trouble the move of the pfn is done shortly before kexec.
1494
-
* This does not eliminate the race because all vcpus are still online
1495
-
* when the syscore_ops will be called. But hopefully there is no work
1496
-
* pending at this point in time. Also the syscore_op is run last which
1497
-
* reduces the risk further.
1498
-
*/
1499
-
1500
-
static struct shared_info *xen_hvm_shared_info;
1501
-
1502
-
static void xen_hvm_connect_shared_info(unsigned long pfn)
1473
+
void __ref xen_hvm_init_shared_info(void)
1503
1474
{
1475
+
int cpu;
1504
1476
struct xen_add_to_physmap xatp;
1477
+
static struct shared_info *shared_info_page = 0;
1505
1478
1479
+
if (!shared_info_page)
1480
+
shared_info_page = (struct shared_info *)
1481
+
extend_brk(PAGE_SIZE, PAGE_SIZE);
1506
1482
xatp.domid = DOMID_SELF;
1507
1483
xatp.idx = 0;
1508
1484
xatp.space = XENMAPSPACE_shared_info;
1509
-
xatp.gpfn = pfn;
1485
+
xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
1510
1486
if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
1511
1487
BUG();
1512
1488
1513
-
}
1514
-
static void xen_hvm_set_shared_info(struct shared_info *sip)
1515
-
{
1516
-
int cpu;
1517
-
1518
-
HYPERVISOR_shared_info = sip;
1489
+
HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
1519
1490
1520
1491
/* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1521
1492
* page, we use it in the event channel upcall and in some pvclock
1522
1493
* related functions. We don't need the vcpu_info placement
1523
1494
* optimizations because we don't use any pv_mmu or pv_irq op on
1524
1495
* HVM.
1525
-
* When xen_hvm_set_shared_info is run at boot time only vcpu 0 is
1526
-
* online but xen_hvm_set_shared_info is run at resume time too and
1496
+
* When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1497
+
* online but xen_hvm_init_shared_info is run at resume time too and
1527
1498
* in that case multiple vcpus might be online. */
1528
1499
for_each_online_cpu(cpu) {
1529
1500
per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
1530
1501
}
1531
1502
}
1532
1503
1533
-
/* Reconnect the shared_info pfn to a mfn */
1534
-
void xen_hvm_resume_shared_info(void)
1535
-
{
1536
-
xen_hvm_connect_shared_info(__pa(xen_hvm_shared_info) >> PAGE_SHIFT);
1537
-
}
1538
-
1539
-
#ifdef CONFIG_KEXEC
1540
-
static struct shared_info *xen_hvm_shared_info_kexec;
1541
-
static unsigned long xen_hvm_shared_info_pfn_kexec;
1542
-
1543
-
/* Remember a pfn in MMIO space for kexec reboot */
1544
-
void __devinit xen_hvm_prepare_kexec(struct shared_info *sip, unsigned long pfn)
1545
-
{
1546
-
xen_hvm_shared_info_kexec = sip;
1547
-
xen_hvm_shared_info_pfn_kexec = pfn;
1548
-
}
1549
-
1550
-
static void xen_hvm_syscore_shutdown(void)
1551
-
{
1552
-
struct xen_memory_reservation reservation = {
1553
-
.domid = DOMID_SELF,
1554
-
.nr_extents = 1,
1555
-
};
1556
-
unsigned long prev_pfn;
1557
-
int rc;
1558
-
1559
-
if (!xen_hvm_shared_info_kexec)
1560
-
return;
1561
-
1562
-
prev_pfn = __pa(xen_hvm_shared_info) >> PAGE_SHIFT;
1563
-
set_xen_guest_handle(reservation.extent_start, &prev_pfn);
1564
-
1565
-
/* Move pfn to MMIO, disconnects previous pfn from mfn */
1566
-
xen_hvm_connect_shared_info(xen_hvm_shared_info_pfn_kexec);
1567
-
1568
-
/* Update pointers, following hypercall is also a memory barrier */
1569
-
xen_hvm_set_shared_info(xen_hvm_shared_info_kexec);
1570
-
1571
-
/* Allocate new mfn for previous pfn */
1572
-
do {
1573
-
rc = HYPERVISOR_memory_op(XENMEM_populate_physmap, &reservation);
1574
-
if (rc == 0)
1575
-
msleep(123);
1576
-
} while (rc == 0);
1577
-
1578
-
/* Make sure the previous pfn is really connected to a (new) mfn */
1579
-
BUG_ON(rc != 1);
1580
-
}
1581
-
1582
-
static struct syscore_ops xen_hvm_syscore_ops = {
1583
-
.shutdown = xen_hvm_syscore_shutdown,
1584
-
};
1585
-
#endif
1586
-
1587
-
/* Use a pfn in RAM, may move to MMIO before kexec. */
1588
-
static void __init xen_hvm_init_shared_info(void)
1589
-
{
1590
-
/* Remember pointer for resume */
1591
-
xen_hvm_shared_info = extend_brk(PAGE_SIZE, PAGE_SIZE);
1592
-
xen_hvm_connect_shared_info(__pa(xen_hvm_shared_info) >> PAGE_SHIFT);
1593
-
xen_hvm_set_shared_info(xen_hvm_shared_info);
1594
-
}
1595
-
1504
+
#ifdef CONFIG_XEN_PVHVM
1596
1505
static void __init init_hvm_pv_info(void)
1597
1506
{
1598
1507
int major, minor;
···
1551
1644
init_hvm_pv_info();
1552
1645
1553
1646
xen_hvm_init_shared_info();
1554
-
#ifdef CONFIG_KEXEC
1555
-
register_syscore_ops(&xen_hvm_syscore_ops);
1556
-
#endif
1557
1647
1558
1648
if (xen_feature(XENFEAT_hvm_callback_vector))
1559
1649
xen_have_vector_callback = 1;
+92
-3
arch/x86/xen/p2m.c
+92
-3
arch/x86/xen/p2m.c
···
196
196
197
197
/* When we populate back during bootup, the amount of pages can vary. The
198
198
* max we have is seen is 395979, but that does not mean it can't be more.
199
-
* But some machines can have 3GB I/O holes even. So lets reserve enough
200
-
* for 4GB of I/O and E820 holes. */
201
-
RESERVE_BRK(p2m_populated, PMD_SIZE * 4);
199
+
* Some machines can have 3GB I/O holes even. With early_can_reuse_p2m_middle
200
+
* it can re-use Xen provided mfn_list array, so we only need to allocate at
201
+
* most three P2M top nodes. */
202
+
RESERVE_BRK(p2m_populated, PAGE_SIZE * 3);
203
+
202
204
static inline unsigned p2m_top_index(unsigned long pfn)
203
205
{
204
206
BUG_ON(pfn >= MAX_P2M_PFN);
···
577
575
}
578
576
return true;
579
577
}
578
+
579
+
/*
580
+
* Skim over the P2M tree looking at pages that are either filled with
581
+
* INVALID_P2M_ENTRY or with 1:1 PFNs. If found, re-use that page and
582
+
* replace the P2M leaf with a p2m_missing or p2m_identity.
583
+
* Stick the old page in the new P2M tree location.
584
+
*/
585
+
bool __init early_can_reuse_p2m_middle(unsigned long set_pfn, unsigned long set_mfn)
586
+
{
587
+
unsigned topidx;
588
+
unsigned mididx;
589
+
unsigned ident_pfns;
590
+
unsigned inv_pfns;
591
+
unsigned long *p2m;
592
+
unsigned long *mid_mfn_p;
593
+
unsigned idx;
594
+
unsigned long pfn;
595
+
596
+
/* We only look when this entails a P2M middle layer */
597
+
if (p2m_index(set_pfn))
598
+
return false;
599
+
600
+
for (pfn = 0; pfn <= MAX_DOMAIN_PAGES; pfn += P2M_PER_PAGE) {
601
+
topidx = p2m_top_index(pfn);
602
+
603
+
if (!p2m_top[topidx])
604
+
continue;
605
+
606
+
if (p2m_top[topidx] == p2m_mid_missing)
607
+
continue;
608
+
609
+
mididx = p2m_mid_index(pfn);
610
+
p2m = p2m_top[topidx][mididx];
611
+
if (!p2m)
612
+
continue;
613
+
614
+
if ((p2m == p2m_missing) || (p2m == p2m_identity))
615
+
continue;
616
+
617
+
if ((unsigned long)p2m == INVALID_P2M_ENTRY)
618
+
continue;
619
+
620
+
ident_pfns = 0;
621
+
inv_pfns = 0;
622
+
for (idx = 0; idx < P2M_PER_PAGE; idx++) {
623
+
/* IDENTITY_PFNs are 1:1 */
624
+
if (p2m[idx] == IDENTITY_FRAME(pfn + idx))
625
+
ident_pfns++;
626
+
else if (p2m[idx] == INVALID_P2M_ENTRY)
627
+
inv_pfns++;
628
+
else
629
+
break;
630
+
}
631
+
if ((ident_pfns == P2M_PER_PAGE) || (inv_pfns == P2M_PER_PAGE))
632
+
goto found;
633
+
}
634
+
return false;
635
+
found:
636
+
/* Found one, replace old with p2m_identity or p2m_missing */
637
+
p2m_top[topidx][mididx] = (ident_pfns ? p2m_identity : p2m_missing);
638
+
/* And the other for save/restore.. */
639
+
mid_mfn_p = p2m_top_mfn_p[topidx];
640
+
/* NOTE: Even if it is a p2m_identity it should still be point to
641
+
* a page filled with INVALID_P2M_ENTRY entries. */
642
+
mid_mfn_p[mididx] = virt_to_mfn(p2m_missing);
643
+
644
+
/* Reset where we want to stick the old page in. */
645
+
topidx = p2m_top_index(set_pfn);
646
+
mididx = p2m_mid_index(set_pfn);
647
+
648
+
/* This shouldn't happen */
649
+
if (WARN_ON(p2m_top[topidx] == p2m_mid_missing))
650
+
early_alloc_p2m(set_pfn);
651
+
652
+
if (WARN_ON(p2m_top[topidx][mididx] != p2m_missing))
653
+
return false;
654
+
655
+
p2m_init(p2m);
656
+
p2m_top[topidx][mididx] = p2m;
657
+
mid_mfn_p = p2m_top_mfn_p[topidx];
658
+
mid_mfn_p[mididx] = virt_to_mfn(p2m);
659
+
660
+
return true;
661
+
}
580
662
bool __init early_set_phys_to_machine(unsigned long pfn, unsigned long mfn)
581
663
{
582
664
if (unlikely(!__set_phys_to_machine(pfn, mfn))) {
583
665
if (!early_alloc_p2m(pfn))
584
666
return false;
667
+
668
+
if (early_can_reuse_p2m_middle(pfn, mfn))
669
+
return __set_phys_to_machine(pfn, mfn);
585
670
586
671
if (!early_alloc_p2m_middle(pfn, false /* boundary crossover OK!*/))
587
672
return false;
+8
-1
arch/x86/xen/setup.c
+8
-1
arch/x86/xen/setup.c
···
78
78
memblock_reserve(start, size);
79
79
80
80
xen_max_p2m_pfn = PFN_DOWN(start + size);
81
+
for (pfn = PFN_DOWN(start); pfn < xen_max_p2m_pfn; pfn++) {
82
+
unsigned long mfn = pfn_to_mfn(pfn);
81
83
82
-
for (pfn = PFN_DOWN(start); pfn <= xen_max_p2m_pfn; pfn++)
84
+
if (WARN(mfn == pfn, "Trying to over-write 1-1 mapping (pfn: %lx)\n", pfn))
85
+
continue;
86
+
WARN(mfn != INVALID_P2M_ENTRY, "Trying to remove %lx which has %lx mfn!\n",
87
+
pfn, mfn);
88
+
83
89
__set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
90
+
}
84
91
}
85
92
86
93
static unsigned long __init xen_do_chunk(unsigned long start,
+1
-1
arch/x86/xen/suspend.c
+1
-1
arch/x86/xen/suspend.c
+1
-1
arch/x86/xen/xen-ops.h
+1
-1
arch/x86/xen/xen-ops.h
-15
drivers/xen/platform-pci.c
-15
drivers/xen/platform-pci.c
···
101
101
return 0;
102
102
}
103
103
104
-
static void __devinit prepare_shared_info(void)
105
-
{
106
-
#ifdef CONFIG_KEXEC
107
-
unsigned long addr;
108
-
struct shared_info *hvm_shared_info;
109
-
110
-
addr = alloc_xen_mmio(PAGE_SIZE);
111
-
hvm_shared_info = ioremap(addr, PAGE_SIZE);
112
-
memset(hvm_shared_info, 0, PAGE_SIZE);
113
-
xen_hvm_prepare_kexec(hvm_shared_info, addr >> PAGE_SHIFT);
114
-
#endif
115
-
}
116
-
117
104
static int __devinit platform_pci_init(struct pci_dev *pdev,
118
105
const struct pci_device_id *ent)
119
106
{
···
137
150
138
151
platform_mmio = mmio_addr;
139
152
platform_mmiolen = mmio_len;
140
-
141
-
prepare_shared_info();
142
153
143
154
if (!xen_have_vector_callback) {
144
155
ret = xen_allocate_irq(pdev);
-2
include/xen/events.h
-2
include/xen/events.h