|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [Xen-devel] [PATCH v2 4/6] ioreq-server: on-demand creation of ioreq server
This patch only creates the ioreq server when the legacy HVM parameters
are touched by an emulator. It also lays some groundwork for supporting
multiple IOREQ servers. For instance, it introduces ioreq server reference
counting which is not strictly necessary at this stage but will become so
when ioreq servers can be destroyed prior the domain dying.
There is a significant change in the layout of the special pages reserved
in xc_hvm_build_x86.c. This is so that we can 'grow' them downwards without
moving pages such as the xenstore page when building a domain that can
support more than one emulator.
Signed-off-by: Paul Durrant <paul.durrant@xxxxxxxxxx>
---
tools/libxc/xc_hvm_build_x86.c | 40 +++--
xen/arch/x86/hvm/hvm.c | 355 ++++++++++++++++++++++++++--------------
2 files changed, 258 insertions(+), 137 deletions(-)
diff --git a/tools/libxc/xc_hvm_build_x86.c b/tools/libxc/xc_hvm_build_x86.c
index dd3b522..b65e702 100644
--- a/tools/libxc/xc_hvm_build_x86.c
+++ b/tools/libxc/xc_hvm_build_x86.c
@@ -41,13 +41,12 @@
#define SPECIALPAGE_PAGING 0
#define SPECIALPAGE_ACCESS 1
#define SPECIALPAGE_SHARING 2
-#define SPECIALPAGE_BUFIOREQ 3
-#define SPECIALPAGE_XENSTORE 4
-#define SPECIALPAGE_IOREQ 5
-#define SPECIALPAGE_IDENT_PT 6
-#define SPECIALPAGE_CONSOLE 7
-#define NR_SPECIAL_PAGES 8
-#define special_pfn(x) (0xff000u - NR_SPECIAL_PAGES + (x))
+#define SPECIALPAGE_XENSTORE 3
+#define SPECIALPAGE_IDENT_PT 4
+#define SPECIALPAGE_CONSOLE 5
+#define SPECIALPAGE_IOREQ 6
+#define NR_SPECIAL_PAGES SPECIALPAGE_IOREQ + 2 /* ioreq server needs 2
pages */
+#define special_pfn(x) (0xff000u - 1 - (x))
#define VGA_HOLE_SIZE (0x20)
@@ -114,7 +113,7 @@ static void build_hvm_info(void *hvm_info_page, uint64_t
mem_size,
/* Memory parameters. */
hvm_info->low_mem_pgend = lowmem_end >> PAGE_SHIFT;
hvm_info->high_mem_pgend = highmem_end >> PAGE_SHIFT;
- hvm_info->reserved_mem_pgstart = special_pfn(0);
+ hvm_info->reserved_mem_pgstart = special_pfn(0) - NR_SPECIAL_PAGES;
/* Finish with the checksum. */
for ( i = 0, sum = 0; i < hvm_info->length; i++ )
@@ -473,6 +472,23 @@ static int setup_guest(xc_interface *xch,
munmap(hvm_info_page, PAGE_SIZE);
/* Allocate and clear special pages. */
+
+ DPRINTF("%d SPECIAL PAGES:\n", NR_SPECIAL_PAGES);
+ DPRINTF(" PAGING: %"PRI_xen_pfn"\n",
+ (xen_pfn_t)special_pfn(SPECIALPAGE_PAGING));
+ DPRINTF(" ACCESS: %"PRI_xen_pfn"\n",
+ (xen_pfn_t)special_pfn(SPECIALPAGE_ACCESS));
+ DPRINTF(" SHARING: %"PRI_xen_pfn"\n",
+ (xen_pfn_t)special_pfn(SPECIALPAGE_SHARING));
+ DPRINTF(" STORE: %"PRI_xen_pfn"\n",
+ (xen_pfn_t)special_pfn(SPECIALPAGE_XENSTORE));
+ DPRINTF(" IDENT_PT: %"PRI_xen_pfn"\n",
+ (xen_pfn_t)special_pfn(SPECIALPAGE_IDENT_PT));
+ DPRINTF(" CONSOLE: %"PRI_xen_pfn"\n",
+ (xen_pfn_t)special_pfn(SPECIALPAGE_CONSOLE));
+ DPRINTF(" IOREQ: %"PRI_xen_pfn"\n",
+ (xen_pfn_t)special_pfn(SPECIALPAGE_IOREQ));
+
for ( i = 0; i < NR_SPECIAL_PAGES; i++ )
{
xen_pfn_t pfn = special_pfn(i);
@@ -488,10 +504,6 @@ static int setup_guest(xc_interface *xch,
xc_set_hvm_param(xch, dom, HVM_PARAM_STORE_PFN,
special_pfn(SPECIALPAGE_XENSTORE));
- xc_set_hvm_param(xch, dom, HVM_PARAM_BUFIOREQ_PFN,
- special_pfn(SPECIALPAGE_BUFIOREQ));
- xc_set_hvm_param(xch, dom, HVM_PARAM_IOREQ_PFN,
- special_pfn(SPECIALPAGE_IOREQ));
xc_set_hvm_param(xch, dom, HVM_PARAM_CONSOLE_PFN,
special_pfn(SPECIALPAGE_CONSOLE));
xc_set_hvm_param(xch, dom, HVM_PARAM_PAGING_RING_PFN,
@@ -500,6 +512,10 @@ static int setup_guest(xc_interface *xch,
special_pfn(SPECIALPAGE_ACCESS));
xc_set_hvm_param(xch, dom, HVM_PARAM_SHARING_RING_PFN,
special_pfn(SPECIALPAGE_SHARING));
+ xc_set_hvm_param(xch, dom, HVM_PARAM_IOREQ_PFN,
+ special_pfn(SPECIALPAGE_IOREQ));
+ xc_set_hvm_param(xch, dom, HVM_PARAM_BUFIOREQ_PFN,
+ special_pfn(SPECIALPAGE_IOREQ) - 1);
/*
* Identity-map page table is required for running with CR0.PG=0 when
diff --git a/xen/arch/x86/hvm/hvm.c b/xen/arch/x86/hvm/hvm.c
index d9586b2..fb2dd73 100644
--- a/xen/arch/x86/hvm/hvm.c
+++ b/xen/arch/x86/hvm/hvm.c
@@ -367,22 +367,9 @@ bool_t hvm_io_pending(struct vcpu *v)
return ( p->state != STATE_IOREQ_NONE );
}
-void hvm_do_resume(struct vcpu *v)
+static void hvm_wait_on_io(struct domain *d, ioreq_t *p)
{
- struct domain *d = v->domain;
- struct hvm_ioreq_server *s = d->arch.hvm_domain.ioreq_server;
- ioreq_t *p;
-
- check_wakeup_from_wait();
-
- if ( is_hvm_vcpu(v) )
- pt_restore_timer(v);
-
- if ( !s )
- goto check_inject_trap;
-
/* NB. Optimised for common case (p->state == STATE_IOREQ_NONE). */
- p = get_ioreq(s, v->vcpu_id);
while ( p->state != STATE_IOREQ_NONE )
{
switch ( p->state )
@@ -398,12 +385,29 @@ void hvm_do_resume(struct vcpu *v)
break;
default:
gdprintk(XENLOG_ERR, "Weird HVM iorequest state %d.\n", p->state);
- domain_crash(v->domain);
+ domain_crash(d);
return; /* bail */
}
}
+}
+
+void hvm_do_resume(struct vcpu *v)
+{
+ struct domain *d = v->domain;
+ struct hvm_ioreq_server *s = d->arch.hvm_domain.ioreq_server;
+
+ check_wakeup_from_wait();
+
+ if ( is_hvm_vcpu(v) )
+ pt_restore_timer(v);
+
+ if ( s )
+ {
+ ioreq_t *p = get_ioreq(s, v->vcpu_id);
+
+ hvm_wait_on_io(d, p);
+ }
- check_inject_trap:
/* Inject pending hw/sw trap */
if ( v->arch.hvm_vcpu.inject_trap.vector != -1 )
{
@@ -412,11 +416,13 @@ void hvm_do_resume(struct vcpu *v)
}
}
-static void hvm_init_ioreq_page(
- struct domain *d, struct hvm_ioreq_page *iorp)
+static void hvm_init_ioreq_page(struct hvm_ioreq_server *s, int buf)
{
+ struct hvm_ioreq_page *iorp;
+
+ iorp = ( buf ) ? &s->buf_ioreq : &s->ioreq;
+
spin_lock_init(&iorp->lock);
- domain_pause(d);
}
void destroy_ring_for_helper(
@@ -432,16 +438,13 @@ void destroy_ring_for_helper(
}
}
-static void hvm_destroy_ioreq_page(
- struct domain *d, struct hvm_ioreq_page *iorp)
+static void hvm_destroy_ioreq_page(struct hvm_ioreq_server *s, int buf)
{
- spin_lock(&iorp->lock);
+ struct hvm_ioreq_page *iorp;
- ASSERT(d->is_dying);
+ iorp = ( buf ) ? &s->buf_ioreq : &s->ioreq;
destroy_ring_for_helper(&iorp->va, iorp->page);
-
- spin_unlock(&iorp->lock);
}
int prepare_ring_for_helper(
@@ -489,8 +492,10 @@ int prepare_ring_for_helper(
}
static int hvm_set_ioreq_page(
- struct domain *d, struct hvm_ioreq_page *iorp, unsigned long gmfn)
+ struct hvm_ioreq_server *s, int buf, unsigned long gmfn)
{
+ struct domain *d = s->domain;
+ struct hvm_ioreq_page *iorp;
struct page_info *page;
void *va;
int rc;
@@ -498,22 +503,17 @@ static int hvm_set_ioreq_page(
if ( (rc = prepare_ring_for_helper(d, gmfn, &page, &va)) )
return rc;
- spin_lock(&iorp->lock);
+ iorp = ( buf ) ? &s->buf_ioreq : &s->ioreq;
if ( (iorp->va != NULL) || d->is_dying )
{
- destroy_ring_for_helper(&iorp->va, iorp->page);
- spin_unlock(&iorp->lock);
+ destroy_ring_for_helper(&va, page);
return -EINVAL;
}
iorp->va = va;
iorp->page = page;
- spin_unlock(&iorp->lock);
-
- domain_unpause(d);
-
return 0;
}
@@ -557,38 +557,6 @@ static int handle_pvh_io(
return X86EMUL_OKAY;
}
-static int hvm_init_ioreq_server(struct domain *d)
-{
- struct hvm_ioreq_server *s;
- int i;
-
- s = xzalloc(struct hvm_ioreq_server);
- if ( !s )
- return -ENOMEM;
-
- s->domain = d;
-
- for ( i = 0; i < MAX_HVM_VCPUS; i++ )
- s->ioreq_evtchn[i] = -1;
- s->buf_ioreq_evtchn = -1;
-
- hvm_init_ioreq_page(d, &s->ioreq);
- hvm_init_ioreq_page(d, &s->buf_ioreq);
-
- d->arch.hvm_domain.ioreq_server = s;
- return 0;
-}
-
-static void hvm_deinit_ioreq_server(struct domain *d)
-{
- struct hvm_ioreq_server *s = d->arch.hvm_domain.ioreq_server;
-
- hvm_destroy_ioreq_page(d, &s->ioreq);
- hvm_destroy_ioreq_page(d, &s->buf_ioreq);
-
- xfree(s);
-}
-
static void hvm_update_ioreq_server_evtchn(struct hvm_ioreq_server *s)
{
struct domain *d = s->domain;
@@ -650,6 +618,123 @@ static void hvm_ioreq_server_remove_vcpu(struct
hvm_ioreq_server *s, struct vcpu
}
}
+static int hvm_create_ioreq_server(struct domain *d, domid_t domid)
+{
+ struct hvm_ioreq_server *s;
+ unsigned long pfn;
+ struct vcpu *v;
+ int i, rc;
+
+ rc = -EEXIST;
+ if ( d->arch.hvm_domain.ioreq_server != NULL )
+ goto fail_exist;
+
+ gdprintk(XENLOG_INFO, "%s: %d\n", __func__, d->domain_id);
+
+ rc = -ENOMEM;
+ s = xzalloc(struct hvm_ioreq_server);
+ if ( !s )
+ goto fail_alloc;
+
+ s->domain = d;
+ s->domid = domid;
+
+ for ( i = 0; i < MAX_HVM_VCPUS; i++ )
+ s->ioreq_evtchn[i] = -1;
+ s->buf_ioreq_evtchn = -1;
+
+ /* Initialize shared pages */
+ pfn = d->arch.hvm_domain.params[HVM_PARAM_IOREQ_PFN];
+
+ hvm_init_ioreq_page(s, 0);
+ if ( (rc = hvm_set_ioreq_page(s, 0, pfn)) < 0 )
+ goto fail_set_ioreq;
+
+ pfn = d->arch.hvm_domain.params[HVM_PARAM_BUFIOREQ_PFN];
+
+ hvm_init_ioreq_page(s, 1);
+ if ( (rc = hvm_set_ioreq_page(s, 1, pfn)) < 0 )
+ goto fail_set_buf_ioreq;
+
+ domain_pause(d);
+
+ for_each_vcpu ( d, v )
+ {
+ if ( (rc = hvm_ioreq_server_add_vcpu(s, v)) < 0 )
+ goto fail_add_vcpu;
+ }
+
+ d->arch.hvm_domain.ioreq_server = s;
+
+ domain_unpause(d);
+
+ return 0;
+
+fail_add_vcpu:
+ for_each_vcpu ( d, v )
+ hvm_ioreq_server_remove_vcpu(s, v);
+ domain_unpause(d);
+ hvm_destroy_ioreq_page(s, 1);
+fail_set_buf_ioreq:
+ hvm_destroy_ioreq_page(s, 0);
+fail_set_ioreq:
+ xfree(s);
+fail_alloc:
+fail_exist:
+ return rc;
+}
+
+static void hvm_destroy_ioreq_server(struct domain *d)
+{
+ struct hvm_ioreq_server *s;
+ struct vcpu *v;
+
+ gdprintk(XENLOG_INFO, "%s: %d\n", __func__, d->domain_id);
+
+ s = d->arch.hvm_domain.ioreq_server;
+ if ( !s )
+ return;
+
+ domain_pause(d);
+
+ d->arch.hvm_domain.ioreq_server = NULL;
+
+ for_each_vcpu ( d, v )
+ hvm_ioreq_server_remove_vcpu(s, v);
+
+ domain_unpause(d);
+
+ hvm_destroy_ioreq_page(s, 1);
+ hvm_destroy_ioreq_page(s, 0);
+
+ xfree(s);
+}
+
+static int hvm_get_ioreq_server_buf_port(struct domain *d, evtchn_port_t *port)
+{
+ struct hvm_ioreq_server *s = d->arch.hvm_domain.ioreq_server;
+
+ if ( !s )
+ return -ENOENT;
+
+ *port = s->buf_ioreq_evtchn;
+ return 0;
+}
+
+static int hvm_get_ioreq_server_pfn(struct domain *d, int buf, xen_pfn_t *pfn)
+{
+ struct hvm_ioreq_server *s = d->arch.hvm_domain.ioreq_server;
+ int i;
+
+ if ( !s )
+ return -ENOENT;
+
+ i = ( buf ) ? HVM_PARAM_BUFIOREQ_PFN : HVM_PARAM_IOREQ_PFN;
+ *pfn = d->arch.hvm_domain.params[i];
+
+ return 0;
+}
+
static int hvm_replace_event_channel(struct vcpu *v, domid_t remote_domid,
int *p_port)
{
@@ -665,14 +750,22 @@ static int hvm_replace_event_channel(struct vcpu *v,
domid_t remote_domid,
return 0;
}
-static int hvm_set_ioreq_server_domid(struct hvm_ioreq_server *s, domid_t
domid)
+static int hvm_set_ioreq_server_domid(struct domain *d, domid_t domid)
{
- struct domain *d = s->domain;
+ struct hvm_ioreq_server *s = d->arch.hvm_domain.ioreq_server;
struct vcpu *v;
int rc = 0;
domain_pause(d);
+ rc = -ENOENT;
+ if ( !s )
+ goto done;
+
+ rc = 0;
+ if ( s->domid == domid )
+ goto done;
+
if ( d->vcpu[0] )
{
rc = hvm_replace_event_channel(d->vcpu[0], domid,
&s->buf_ioreq_evtchn);
@@ -697,27 +790,6 @@ done:
return rc;
}
-static int hvm_set_ioreq_server_pfn(struct hvm_ioreq_server *s, unsigned long
pfn)
-{
- struct domain *d = s->domain;
- int rc;
-
- rc = hvm_set_ioreq_page(d, &s->ioreq, pfn);
- if ( rc < 0 )
- return rc;
-
- hvm_update_ioreq_server_evtchn(s);
-
- return 0;
-}
-
-static int hvm_set_ioreq_server_buf_pfn(struct hvm_ioreq_server *s, unsigned
long pfn)
-{
- struct domain *d = s->domain;
-
- return hvm_set_ioreq_page(d, &s->buf_ioreq, pfn);
-}
-
int hvm_domain_initialise(struct domain *d)
{
int rc;
@@ -785,20 +857,14 @@ int hvm_domain_initialise(struct domain *d)
rtc_init(d);
- rc = hvm_init_ioreq_server(d);
- if ( rc != 0 )
- goto fail2;
-
register_portio_handler(d, 0xe9, 1, hvm_print_line);
rc = hvm_funcs.domain_initialise(d);
if ( rc != 0 )
- goto fail3;
+ goto fail2;
return 0;
- fail3:
- hvm_deinit_ioreq_server(d);
fail2:
rtc_deinit(d);
stdvga_deinit(d);
@@ -822,7 +888,7 @@ void hvm_domain_relinquish_resources(struct domain *d)
if ( hvm_funcs.nhvm_domain_relinquish_resources )
hvm_funcs.nhvm_domain_relinquish_resources(d);
- hvm_deinit_ioreq_server(d);
+ hvm_destroy_ioreq_server(d);
msixtbl_pt_cleanup(d);
@@ -1497,9 +1563,12 @@ int hvm_vcpu_initialise(struct vcpu *v)
&& (rc = nestedhvm_vcpu_initialise(v)) < 0 ) /* teardown:
nestedhvm_vcpu_destroy */
goto fail5;
- rc = hvm_ioreq_server_add_vcpu(s, v);
- if ( rc < 0 )
- goto fail6;
+ if ( s )
+ {
+ rc = hvm_ioreq_server_add_vcpu(s, v);
+ if ( rc < 0 )
+ goto fail6;
+ }
if ( v->vcpu_id == 0 )
{
@@ -1537,7 +1606,8 @@ void hvm_vcpu_destroy(struct vcpu *v)
struct domain *d = v->domain;
struct hvm_ioreq_server *s = d->arch.hvm_domain.ioreq_server;
- hvm_ioreq_server_remove_vcpu(s, v);
+ if ( s )
+ hvm_ioreq_server_remove_vcpu(s, v);
nestedhvm_vcpu_destroy(v);
@@ -1662,19 +1732,12 @@ int hvm_buffered_io_send(ioreq_t *p)
return 1;
}
-bool_t hvm_send_assist_req(struct vcpu *v, ioreq_t *proto_p)
+static bool_t hvm_send_assist_req_to_server(struct hvm_ioreq_server *s,
+ struct vcpu *v,
+ ioreq_t *proto_p)
{
struct domain *d = v->domain;
- struct hvm_ioreq_server *s = d->arch.hvm_domain.ioreq_server;
- ioreq_t *p;
-
- if ( unlikely(!vcpu_start_shutdown_deferral(v)) )
- return 0; /* implicitly bins the i/o operation */
-
- if ( !s )
- return 0;
-
- p = get_ioreq(s, v->vcpu_id);
+ ioreq_t *p = get_ioreq(s, v->vcpu_id);
if ( unlikely(p->state != STATE_IOREQ_NONE) )
{
@@ -1705,6 +1768,20 @@ bool_t hvm_send_assist_req(struct vcpu *v, ioreq_t
*proto_p)
return 1;
}
+bool_t hvm_send_assist_req(struct vcpu *v, ioreq_t *p)
+{
+ struct domain *d = v->domain;
+ struct hvm_ioreq_server *s = d->arch.hvm_domain.ioreq_server;
+
+ if ( unlikely(!vcpu_start_shutdown_deferral(v)) )
+ return 0;
+
+ if ( !s )
+ return 0;
+
+ return hvm_send_assist_req_to_server(s, v, p);
+}
+
void hvm_hlt(unsigned long rflags)
{
struct vcpu *curr = current;
@@ -4336,14 +4413,6 @@ long do_hvm_op(unsigned long op,
XEN_GUEST_HANDLE_PARAM(void) arg)
switch ( a.index )
{
- case HVM_PARAM_IOREQ_PFN:
- rc = hvm_set_ioreq_server_pfn(d->arch.hvm_domain.ioreq_server,
- a.value);
- break;
- case HVM_PARAM_BUFIOREQ_PFN:
- rc =
hvm_set_ioreq_server_buf_pfn(d->arch.hvm_domain.ioreq_server,
- a.value);
- break;
case HVM_PARAM_CALLBACK_IRQ:
hvm_set_callback_via(d, a.value);
hvm_latch_shinfo_size(d);
@@ -4397,8 +4466,9 @@ long do_hvm_op(unsigned long op,
XEN_GUEST_HANDLE_PARAM(void) arg)
if ( a.value == DOMID_SELF )
a.value = curr_d->domain_id;
- rc =
hvm_set_ioreq_server_domid(d->arch.hvm_domain.ioreq_server,
- a.value);
+ rc = hvm_create_ioreq_server(d, a.value);
+ if ( rc == -EEXIST )
+ rc = hvm_set_ioreq_server_domid(d, a.value);
break;
case HVM_PARAM_ACPI_S_STATE:
/* Not reflexive, as we must domain_pause(). */
@@ -4493,12 +4563,47 @@ long do_hvm_op(unsigned long op,
XEN_GUEST_HANDLE_PARAM(void) arg)
{
switch ( a.index )
{
- case HVM_PARAM_BUFIOREQ_EVTCHN: {
- struct hvm_ioreq_server *s = d->arch.hvm_domain.ioreq_server;
+ case HVM_PARAM_IOREQ_PFN:
+ case HVM_PARAM_BUFIOREQ_PFN:
+ case HVM_PARAM_BUFIOREQ_EVTCHN:
+ /* May need to create server */
+ rc = hvm_create_ioreq_server(d, curr_d->domain_id);
+ if ( rc != 0 && rc != -EEXIST )
+ goto param_fail;
+
+ switch ( a.index )
+ {
+ case HVM_PARAM_IOREQ_PFN: {
+ xen_pfn_t pfn;
+
+ if ( (rc = hvm_get_ioreq_server_pfn(d, 0, &pfn)) < 0 )
+ goto param_fail;
+
+ a.value = pfn;
+ break;
+ }
+ case HVM_PARAM_BUFIOREQ_PFN: {
+ xen_pfn_t pfn;
+
+ if ( (rc = hvm_get_ioreq_server_pfn(d, 1, &pfn)) < 0 )
+ goto param_fail;
+
+ a.value = pfn;
+ break;
+ }
+ case HVM_PARAM_BUFIOREQ_EVTCHN: {
+ evtchn_port_t port;
- a.value = s->buf_ioreq_evtchn;
+ if ( (rc = hvm_get_ioreq_server_buf_port(d, &port)) < 0 )
+ goto param_fail;
+
+ a.value = port;
+ break;
+ }
+ default:
+ BUG();
+ }
break;
- }
case HVM_PARAM_ACPI_S_STATE:
a.value = d->arch.hvm_domain.is_s3_suspended ? 3 : 0;
break;
--
1.7.10.4
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel
|
 |
Lists.xenproject.org is hosted with RackSpace, monitoring our |