Module Name: src
Committed By: jdolecek
Date: Thu Apr 9 19:26:38 UTC 2020
Modified Files:
src/sys/arch/xen/conf: std.xenversion
src/sys/arch/xen/include: xen.h
src/sys/arch/xen/x86: xen_bus_dma.c
src/sys/arch/xen/xen: balloon.c xengnt.c
Log Message:
update to __XEN_INTERFACE_VERSION__ 0x0003020a aka Xen 3.2.10
this brings grant memory v2 support:
- status separated from flags - revoking access needs just memory barrier,
no need for expensive cmpxchg16 any more
- sub-page hypervisor copy-only grants, to be used by xennet(4)
- 64-bit frame, i.e. support for DomU RAM >16TB
the grant table is now always allocated on boot to maximum size, it's now
never grown in runtime; switch back to regular kmem_alloc()/kmem_free()
code now requires v2 support, no compatibility for grant version 1 retained -
Xen v2 support predates all currently supported Xen versions
also interface for baloon changed slightly, code updated
To generate a diff of this commit:
cvs rdiff -u -r1.1 -r1.2 src/sys/arch/xen/conf/std.xenversion
cvs rdiff -u -r1.44 -r1.45 src/sys/arch/xen/include/xen.h
cvs rdiff -u -r1.28 -r1.29 src/sys/arch/xen/x86/xen_bus_dma.c
cvs rdiff -u -r1.19 -r1.20 src/sys/arch/xen/xen/balloon.c
cvs rdiff -u -r1.30 -r1.31 src/sys/arch/xen/xen/xengnt.c
Please note that diffs are not public domain; they are subject to the
copyright notices on the relevant files.
Modified files:
Index: src/sys/arch/xen/conf/std.xenversion
diff -u src/sys/arch/xen/conf/std.xenversion:1.1 src/sys/arch/xen/conf/std.xenversion:1.2
--- src/sys/arch/xen/conf/std.xenversion:1.1 Thu Apr 9 14:39:10 2020
+++ src/sys/arch/xen/conf/std.xenversion Thu Apr 9 19:26:37 2020
@@ -1,6 +1,6 @@
-# $NetBSD: std.xenversion,v 1.1 2020/04/09 14:39:10 jdolecek Exp $
+# $NetBSD: std.xenversion,v 1.2 2020/04/09 19:26:37 jdolecek Exp $
#
# Xen options shared for all archs (i386, amd64)
-options __XEN_INTERFACE_VERSION__=0x00030208 # Xen 3.1 interface
+options __XEN_INTERFACE_VERSION__=0x0003020a # Xen 3.2.10 interface
Index: src/sys/arch/xen/include/xen.h
diff -u src/sys/arch/xen/include/xen.h:1.44 src/sys/arch/xen/include/xen.h:1.45
--- src/sys/arch/xen/include/xen.h:1.44 Thu May 9 17:09:50 2019
+++ src/sys/arch/xen/include/xen.h Thu Apr 9 19:26:37 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: xen.h,v 1.44 2019/05/09 17:09:50 bouyer Exp $ */
+/* $NetBSD: xen.h,v 1.45 2020/04/09 19:26:37 jdolecek Exp $ */
/*
*
@@ -170,20 +170,6 @@ xen_atomic_xchg(volatile XATOMIC_T *ptr,
return result;
}
-static inline uint16_t
-xen_atomic_cmpxchg16(volatile uint16_t *ptr, uint16_t val, uint16_t newval)
-{
- unsigned long result;
-
- __asm volatile(__LOCK_PREFIX
- "cmpxchgw %w1,%2"
- :"=a" (result)
- :"q"(newval), "m" (*ptr), "0" (val)
- :"memory");
-
- return result;
-}
-
static __inline void
xen_atomic_setbits_l (volatile XATOMIC_T *ptr, unsigned long bits) {
#ifdef __x86_64__
Index: src/sys/arch/xen/x86/xen_bus_dma.c
diff -u src/sys/arch/xen/x86/xen_bus_dma.c:1.28 src/sys/arch/xen/x86/xen_bus_dma.c:1.29
--- src/sys/arch/xen/x86/xen_bus_dma.c:1.28 Mon Sep 3 16:29:29 2018
+++ src/sys/arch/xen/x86/xen_bus_dma.c Thu Apr 9 19:26:37 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: xen_bus_dma.c,v 1.28 2018/09/03 16:29:29 riastradh Exp $ */
+/* $NetBSD: xen_bus_dma.c,v 1.29 2020/04/09 19:26:37 jdolecek Exp $ */
/* NetBSD bus_dma.c,v 1.21 2005/04/16 07:53:35 yamt Exp */
/*-
@@ -32,7 +32,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: xen_bus_dma.c,v 1.28 2018/09/03 16:29:29 riastradh Exp $");
+__KERNEL_RCSID(0, "$NetBSD: xen_bus_dma.c,v 1.29 2020/04/09 19:26:37 jdolecek Exp $");
#include <sys/param.h>
#include <sys/systm.h>
@@ -94,7 +94,7 @@ _xen_alloc_contig(bus_size_t size, bus_s
set_xen_guest_handle(res.extent_start, &mfn);
res.nr_extents = 1;
res.extent_order = 0;
- res.address_bits = 0;
+ res.mem_flags = 0;
res.domid = DOMID_SELF;
error = HYPERVISOR_memory_op(XENMEM_decrease_reservation, &res);
if (error != 1) {
@@ -113,14 +113,14 @@ _xen_alloc_contig(bus_size_t size, bus_s
set_xen_guest_handle(res.extent_start, &mfn);
res.nr_extents = 1;
res.extent_order = order;
- res.address_bits = get_order(high) + PAGE_SHIFT;
+ res.mem_flags = XENMEMF_address_bits(get_order(high) + PAGE_SHIFT);
res.domid = DOMID_SELF;
error = HYPERVISOR_memory_op(XENMEM_increase_reservation, &res);
if (error != 1) {
#ifdef DEBUG
printf("xen_alloc_contig: XENMEM_increase_reservation "
- "failed: %d (order %d address_bits %d)\n",
- error, order, res.address_bits);
+ "failed: %d (order %d mem_flags %d)\n",
+ error, order, res.mem_flags);
#endif
error = ENOMEM;
pg = NULL;
@@ -166,7 +166,7 @@ failed:
set_xen_guest_handle(res.extent_start, &mfn);
res.nr_extents = 1;
res.extent_order = 0;
- res.address_bits = 32;
+ res.mem_flags = XENMEMF_address_bits(32);
res.domid = DOMID_SELF;
if (HYPERVISOR_memory_op(XENMEM_increase_reservation, &res)
< 0) {
Index: src/sys/arch/xen/xen/balloon.c
diff -u src/sys/arch/xen/xen/balloon.c:1.19 src/sys/arch/xen/xen/balloon.c:1.20
--- src/sys/arch/xen/xen/balloon.c:1.19 Sun Jun 24 20:28:58 2018
+++ src/sys/arch/xen/xen/balloon.c Thu Apr 9 19:26:38 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: balloon.c,v 1.19 2018/06/24 20:28:58 jdolecek Exp $ */
+/* $NetBSD: balloon.c,v 1.20 2020/04/09 19:26:38 jdolecek Exp $ */
/*-
* Copyright (c) 2010 The NetBSD Foundation, Inc.
@@ -75,7 +75,7 @@
#endif
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: balloon.c,v 1.19 2018/06/24 20:28:58 jdolecek Exp $");
+__KERNEL_RCSID(0, "$NetBSD: balloon.c,v 1.20 2020/04/09 19:26:38 jdolecek Exp $");
#include <sys/inttypes.h>
#include <sys/device.h>
@@ -378,7 +378,7 @@ balloon_inflate(struct balloon_xenbus_so
xen_pfn_t *mfn_list = sc->sc_mfn_list;
struct xen_memory_reservation reservation = {
- .address_bits = 0,
+ .mem_flags = 0,
.extent_order = 0,
.domid = DOMID_SELF
};
@@ -449,7 +449,7 @@ balloon_deflate(struct balloon_xenbus_so
xen_pfn_t *mfn_list = sc->sc_mfn_list;
struct xen_memory_reservation reservation = {
- .address_bits = 0,
+ .mem_flags = 0,
.extent_order = 0,
.domid = DOMID_SELF
};
Index: src/sys/arch/xen/xen/xengnt.c
diff -u src/sys/arch/xen/xen/xengnt.c:1.30 src/sys/arch/xen/xen/xengnt.c:1.31
--- src/sys/arch/xen/xen/xengnt.c:1.30 Thu Apr 9 15:54:41 2020
+++ src/sys/arch/xen/xen/xengnt.c Thu Apr 9 19:26:38 2020
@@ -1,4 +1,4 @@
-/* $NetBSD: xengnt.c,v 1.30 2020/04/09 15:54:41 bouyer Exp $ */
+/* $NetBSD: xengnt.c,v 1.31 2020/04/09 19:26:38 jdolecek Exp $ */
/*
* Copyright (c) 2006 Manuel Bouyer.
@@ -26,7 +26,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: xengnt.c,v 1.30 2020/04/09 15:54:41 bouyer Exp $");
+__KERNEL_RCSID(0, "$NetBSD: xengnt.c,v 1.31 2020/04/09 19:26:38 jdolecek Exp $");
#include <sys/types.h>
#include <sys/param.h>
@@ -49,7 +49,8 @@ __KERNEL_RCSID(0, "$NetBSD: xengnt.c,v 1
#define DPRINTF(x)
#endif
-#define NR_GRANT_ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(grant_entry_t))
+#define NR_GRANT_ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(grant_entry_v2_t))
+#define NR_GRANT_STATUS_PER_PAGE (PAGE_SIZE / sizeof(grant_status_t))
/* External tools reserve first few grant table entries. */
#define NR_RESERVED_ENTRIES 8
@@ -58,6 +59,8 @@ __KERNEL_RCSID(0, "$NetBSD: xengnt.c,v 1
int gnt_nr_grant_frames;
/* Maximum number of frames that can make up the grant table */
int gnt_max_grant_frames;
+/* Number of grant status frames */
+int gnt_status_frames;
/* table of free grant entries */
grant_ref_t *gnt_entries;
@@ -67,17 +70,20 @@ int last_gnt_entry;
#define XENGNT_NO_ENTRY 0xffffffff
/* VM address of the grant table */
-grant_entry_t *grant_table;
+grant_entry_v2_t *grant_table;
+grant_status_t *grant_status;
kmutex_t grant_lock;
static grant_ref_t xengnt_get_entry(void);
static void xengnt_free_entry(grant_ref_t);
static int xengnt_more_entries(void);
+static int xengnt_map_status(void);
void
xengnt_init(void)
{
struct gnttab_query_size query;
+ struct gnttab_set_version gntversion;
int rc;
int nr_grant_entries;
int i;
@@ -88,7 +94,16 @@ xengnt_init(void)
gnt_max_grant_frames = 4; /* Legacy max number of frames */
else
gnt_max_grant_frames = query.max_nr_frames;
- gnt_nr_grant_frames = 0;
+
+ /*
+ * Always allocate max number of grant frames, never expand in runtime
+ */
+ gnt_nr_grant_frames = gnt_max_grant_frames;
+
+ gntversion.version = 2;
+ rc = HYPERVISOR_grant_table_op(GNTTABOP_set_version, &gntversion, 1);
+ if (rc < 0 || gntversion.version != 2)
+ panic("GNTTABOP_set_version 2 failed %d", rc);
nr_grant_entries =
gnt_max_grant_frames * NR_GRANT_ENTRIES_PER_PAGE;
@@ -96,12 +111,20 @@ xengnt_init(void)
grant_table = (void *)uvm_km_alloc(kernel_map,
gnt_max_grant_frames * PAGE_SIZE, 0, UVM_KMF_VAONLY);
if (grant_table == NULL)
- panic("xengnt_init() no VM space");
+ panic("xengnt_init() table no VM space");
+
gnt_entries = kmem_alloc((nr_grant_entries + 1) * sizeof(grant_ref_t),
KM_SLEEP);
for (i = 0; i <= nr_grant_entries; i++)
gnt_entries[i] = XENGNT_NO_ENTRY;
+ gnt_status_frames =
+ round_page(nr_grant_entries * sizeof(grant_status_t)) / PAGE_SIZE;
+ grant_status = (void *)uvm_km_alloc(kernel_map,
+ gnt_status_frames * PAGE_SIZE, 0, UVM_KMF_VAONLY);
+ if (grant_status == NULL)
+ panic("xengnt_init() status no VM space");
+
mutex_init(&grant_lock, MUTEX_DEFAULT, IPL_VM);
xengnt_resume();
@@ -124,6 +147,7 @@ xengnt_resume(void)
if (xengnt_more_entries() != 0)
panic("xengnt_resume: can't restore grant frames");
}
+ xengnt_map_status();
mutex_exit(&grant_lock);
return true;
}
@@ -144,13 +168,67 @@ xengnt_suspend(void) {
gnt_entries[i] = XENGNT_NO_ENTRY;
}
- /* Remove virtual => machine mapping */
+ /* Remove virtual => machine mapping for grant table */
pmap_kremove((vaddr_t)grant_table, gnt_nr_grant_frames * PAGE_SIZE);
+
+ /* Remove virtual => machine mapping for status table */
+ pmap_kremove((vaddr_t)grant_status, gnt_status_frames * PAGE_SIZE);
+
pmap_update(pmap_kernel());
mutex_exit(&grant_lock);
return true;
}
+/*
+ * Get status frames and enter them into the VA space.
+ */
+static int
+xengnt_map_status(void)
+{
+ gnttab_get_status_frames_t getstatus;
+ u_long *pages;
+ size_t sz;
+
+ KASSERT(mutex_owned(&grant_lock));
+
+ sz = gnt_status_frames * sizeof(u_long);
+ pages = kmem_alloc(sz, KM_NOSLEEP);
+ if (pages == NULL)
+ return ENOMEM;
+
+ getstatus.dom = DOMID_SELF;
+ getstatus.nr_frames = gnt_status_frames;
+ set_xen_guest_handle(getstatus.frame_list, pages);
+
+ /*
+ * get the status frames, and return the list of their virtual
+ * addresses in 'pages'
+ */
+ if (HYPERVISOR_grant_table_op(GNTTABOP_get_status_frames,
+ &getstatus, 1) != 0)
+ panic("%s: get_status_frames failed", __func__);
+ if (getstatus.status != GNTST_okay) {
+ aprint_error("%s: get_status_frames returned %d\n",
+ __func__, getstatus.status);
+ kmem_free(pages, sz);
+ return ENOMEM;
+ }
+
+ /*
+ * map between status_table addresses and the machine addresses of
+ * the status table frames
+ */
+ for (int i = 0; i < gnt_status_frames; i++) {
+ pmap_kenter_ma(((vaddr_t)grant_status) + i * PAGE_SIZE,
+ ((paddr_t)pages[i]) << PAGE_SHIFT,
+ VM_PROT_WRITE, 0);
+ }
+ pmap_update(pmap_kernel());
+
+ kmem_free(pages, sz);
+
+ return 0;
+}
/*
* Add another page to the grant table
@@ -170,7 +248,7 @@ xengnt_more_entries(void)
return ENOMEM;
sz = nframes_new * sizeof(u_long);
- pages = kmem_intr_alloc(sz, KM_NOSLEEP);
+ pages = kmem_alloc(sz, KM_NOSLEEP);
if (pages == NULL)
return ENOMEM;
@@ -209,7 +287,7 @@ xengnt_more_entries(void)
if (setup.status != GNTST_okay) {
aprint_error("%s: setup table returned %d\n",
__func__, setup.status);
- kmem_intr_free(pages, sz);
+ kmem_free(pages, sz);
return ENOMEM;
}
}
@@ -244,7 +322,7 @@ xengnt_more_entries(void)
last_gnt_entry++;
}
gnt_nr_grant_frames = nframes_new;
- kmem_intr_free(pages, sz);
+ kmem_free(pages, sz);
return 0;
}
@@ -258,13 +336,11 @@ xengnt_get_entry(void)
static struct timeval xengnt_nonmemtime;
static const struct timeval xengnt_nonmemintvl = {5,0};
- if (last_gnt_entry == 0) {
- if (xengnt_more_entries()) {
- if (ratecheck(&xengnt_nonmemtime, &xengnt_nonmemintvl))
- printf("xengnt_get_entry: out of grant "
- "table entries\n");
- return XENGNT_NO_ENTRY;
- }
+ if (__predict_false(last_gnt_entry == 0)) {
+ if (ratecheck(&xengnt_nonmemtime, &xengnt_nonmemintvl))
+ printf("xengnt_get_entry: out of grant "
+ "table entries\n");
+ return XENGNT_NO_ENTRY;
}
KASSERT(gnt_entries[last_gnt_entry] == XENGNT_NO_ENTRY);
last_gnt_entry--;
@@ -303,14 +379,14 @@ xengnt_grant_access(domid_t dom, paddr_t
return ENOMEM;
}
- grant_table[*entryp].frame = ma >> PAGE_SHIFT;
- grant_table[*entryp].domid = dom;
+ grant_table[*entryp].full_page.frame = ma >> PAGE_SHIFT;
+ grant_table[*entryp].hdr.domid = dom;
/*
* ensure that the above values reach global visibility
* before permitting frame's access (done when we set flags)
*/
xen_rmb();
- grant_table[*entryp].flags =
+ grant_table[*entryp].hdr.flags =
GTF_permit_access | (ro ? GTF_readonly : 0);
mutex_exit(&grant_lock);
return 0;
@@ -319,21 +395,29 @@ xengnt_grant_access(domid_t dom, paddr_t
void
xengnt_revoke_access(grant_ref_t entry)
{
- uint16_t flags, nflags;
+ grant_table[entry].hdr.flags = 0;
+ xen_mb(); /* Concurrent access by hypervisor */
+
+ if (__predict_false((grant_status[entry] & (GTF_reading|GTF_writing))
+ != 0))
+ printf("xengnt_revoke_access(%u): still in use\n",
+ entry);
+ else {
- nflags = grant_table[entry].flags;
+ /*
+ * The read of grant_status needs to have acquire semantics.
+ * Reads already have that on x86, so need only protect
+ * against compiler reordering. May need full barrier
+ * on other architectures.
+ */
+ __insn_barrier();
+ }
- do {
- if ((flags = nflags) & (GTF_reading|GTF_writing))
- panic("xengnt_revoke_access: still in use");
- nflags = xen_atomic_cmpxchg16(&grant_table[entry].flags,
- flags, 0);
- } while (nflags != flags);
xengnt_free_entry(entry);
}
int
xengnt_status(grant_ref_t entry)
{
- return (grant_table[entry].flags & (GTF_reading|GTF_writing));
+ return grant_status[entry] & (GTF_reading|GTF_writing);
}
