Module Name: src
Committed By: jym
Date: Sun May 31 20:15:37 UTC 2009
Modified Files:
src/sys/arch/x86/include [jym-xensuspend]: pmap.h
src/sys/arch/x86/x86 [jym-xensuspend]: pmap.c
src/sys/arch/xen/include [jym-xensuspend]: xenbus.h
src/sys/arch/xen/include/amd64 [jym-xensuspend]: hypercalls.h
src/sys/arch/xen/include/i386 [jym-xensuspend]: hypercalls.h
src/sys/arch/xen/x86 [jym-xensuspend]: x86_xpmap.c xen_bus_dma.c
src/sys/arch/xen/xen [jym-xensuspend]: evtchn.c if_xennet_xenbus.c
xbd_xenbus.c xen_machdep.c xencons.c xennetback_xenbus.c
src/sys/arch/xen/xenbus [jym-xensuspend]: xenbus_probe.c
Log Message:
Modifications for the Xen suspend/migrate/resume branch:
- introduce xenbus_device_{suspend,resume}() functions. These are routines
used to suspend/resume MI parts of the Xenbus device interfaces, like updating
frontend/backend devices' paths found in XenStore.
- introduce HYPERVISOR_sysctl(), an hypercall used only by Xentools to obtain
information from hypervisor (listing VMs, printing console, etc.). I use it
to query xenconsole from ddb(), as a last resort in case of a panic() in
dom0 (xm being not available). Currently unused in the branch; could be, if
requested.
- disable the rwlock(9) used to protect code that could use transient MFNs.
It could trigger nasty context switches in place it should not to.
- fix some bugs in the xennet/xbd suspend/resume pmf(9) handlers.
- following XenSource's design, talk_to_otherend() is now called
watch_otherend(), and free_otherend_details() is used by Xenbus device
suspend/resume routines.
- some slight modifications in pmap regarding APDP. Introduce an inline
function (pmap_unmap_apdp_pde()) that clears APDP entry for the current pmap.
- similarly, implement pmap_unmap_all_apdp_pdes() that iterates through all
pmaps and tears down APDP, as Xen does not handle them properly.
TODO/XXX:
- pmap_unmap_apdp_pde() does not handle APDP shadow entry of PAE. It will,
once I figure out how PAE uses it.
- revisit the pmap locking issue regarding transient MFNs. As NetBSD does not
use kernel preemption and MP for Xen, this could be skipped momentarily. See
http://mail-index.netbsd.org/port-xen/2009/04/27/msg004903.html for details.
- fix a bug regarding grant tables which could technically DoS a dom0 if
ridiculously high consumer/producer indexes are passed down in the ring during
a resume.
All in all, once the grant table index issue and APDP PAE are fixed, next step
is to torture test this branch.
Tested under i386 PAE and non-PAE, Xen3 dom0 and domU. amd64 is only compile
tested.
To generate a diff of this commit:
cvs rdiff -u -r1.21.2.1 -r1.21.2.2 src/sys/arch/x86/include/pmap.h
cvs rdiff -u -r1.77.2.2 -r1.77.2.3 src/sys/arch/x86/x86/pmap.c
cvs rdiff -u -r1.11 -r1.11.6.1 src/sys/arch/xen/include/xenbus.h
cvs rdiff -u -r1.5 -r1.5.4.1 src/sys/arch/xen/include/amd64/hypercalls.h
cvs rdiff -u -r1.8 -r1.8.4.1 src/sys/arch/xen/include/i386/hypercalls.h
cvs rdiff -u -r1.12.4.2 -r1.12.4.3 src/sys/arch/xen/x86/x86_xpmap.c
cvs rdiff -u -r1.14.2.1 -r1.14.2.2 src/sys/arch/xen/x86/xen_bus_dma.c
cvs rdiff -u -r1.42.2.2 -r1.42.2.3 src/sys/arch/xen/xen/evtchn.c
cvs rdiff -u -r1.33.2.2 -r1.33.2.3 src/sys/arch/xen/xen/if_xennet_xenbus.c
cvs rdiff -u -r1.38.2.2 -r1.38.2.3 src/sys/arch/xen/xen/xbd_xenbus.c
cvs rdiff -u -r1.4.12.2 -r1.4.12.3 src/sys/arch/xen/xen/xen_machdep.c
cvs rdiff -u -r1.31.2.1 -r1.31.2.2 src/sys/arch/xen/xen/xencons.c
cvs rdiff -u -r1.27.2.2 -r1.27.2.3 src/sys/arch/xen/xen/xennetback_xenbus.c
cvs rdiff -u -r1.27.2.1 -r1.27.2.2 src/sys/arch/xen/xenbus/xenbus_probe.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/x86/include/pmap.h
diff -u src/sys/arch/x86/include/pmap.h:1.21.2.1 src/sys/arch/x86/include/pmap.h:1.21.2.2
--- src/sys/arch/x86/include/pmap.h:1.21.2.1 Wed May 13 17:18:44 2009
+++ src/sys/arch/x86/include/pmap.h Sun May 31 20:15:36 2009
@@ -1,4 +1,4 @@
-/* $NetBSD: pmap.h,v 1.21.2.1 2009/05/13 17:18:44 jym Exp $ */
+/* $NetBSD: pmap.h,v 1.21.2.2 2009/05/31 20:15:36 jym Exp $ */
/*
*
@@ -400,6 +400,7 @@
int pmap_enter_ma(struct pmap *, vaddr_t, paddr_t, paddr_t,
vm_prot_t, u_int, int);
bool pmap_extract_ma(pmap_t, vaddr_t, paddr_t *);
+void pmap_unmap_all_apdp_pdes(void);
paddr_t vtomach(vaddr_t);
#define vtomfn(va) (vtomach(va) >> PAGE_SHIFT)
Index: src/sys/arch/x86/x86/pmap.c
diff -u src/sys/arch/x86/x86/pmap.c:1.77.2.2 src/sys/arch/x86/x86/pmap.c:1.77.2.3
--- src/sys/arch/x86/x86/pmap.c:1.77.2.2 Wed May 13 17:18:45 2009
+++ src/sys/arch/x86/x86/pmap.c Sun May 31 20:15:36 2009
@@ -1,4 +1,4 @@
-/* $NetBSD: pmap.c,v 1.77.2.2 2009/05/13 17:18:45 jym Exp $ */
+/* $NetBSD: pmap.c,v 1.77.2.3 2009/05/31 20:15:36 jym Exp $ */
/*
* Copyright (c) 2007 Manuel Bouyer.
@@ -154,7 +154,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.77.2.2 2009/05/13 17:18:45 jym Exp $");
+__KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.77.2.3 2009/05/31 20:15:36 jym Exp $");
#include "opt_user_ldt.h"
#include "opt_lockdebug.h"
@@ -741,6 +741,54 @@
(kernel && (pmap->pm_kernel_cpus & ci->ci_cpumask) != 0));
}
+/*
+ * Flush the content of APDP_PDE
+ */
+static inline
+void pmap_unmap_apdp_pde(void) {
+
+ int i;
+
+ for (i = 0; i < PDP_SIZE; i++) {
+ pmap_pte_set(&APDP_PDE[i], 0);
+#ifdef PAE
+ /* clear shadow entry too */
+ pmap_pte_set(&APDP_PDE_SHADOW[i], 0);
+#endif
+ }
+
+}
+
+/*
+ * Flush all APDP entries found in pmaps
+ * Required during Xen save/restore operations, as it does not
+ * handle alternative recursive mappings properly
+ */
+void
+pmap_unmap_all_apdp_pdes(void) {
+
+ // XXX JYM PAE
+
+ int s;
+ struct pmap *pm;
+
+ s = splvm();
+
+ mutex_enter(&pmaps_lock);
+ LIST_FOREACH(pm, &pmaps, pm_list) {
+ xpq_queue_pte_update(
+ xpmap_ptom(pmap_pdirpa(pm, PDIR_SLOT_APTE)),
+ 0);
+ }
+ xpq_flush_queue();
+
+ mutex_exit(&pmaps_lock);
+
+ splx(s);
+
+}
+
+
static void
pmap_apte_flush(struct pmap *pmap)
{
@@ -932,7 +980,7 @@
KASSERT(curcpu()->ci_pmap == pmap2);
#endif
#if defined(MULTIPROCESSOR)
- pmap_pte_set(APDP_PDE, 0);
+ pmap_unmap_apdp_pde();
pmap_pte_flush();
pmap_apte_flush(pmap2);
#endif
@@ -2308,13 +2356,7 @@
xen_acquire_reader_ptom_lock();
if (xpmap_ptom_masked(pmap_pdirpa(pmap, 0)) == (*APDP_PDE & PG_FRAME)) {
kpreempt_disable();
- for (i = 0; i < PDP_SIZE; i++) {
- pmap_pte_set(&APDP_PDE[i], 0);
-#ifdef PAE
- /* clear shadow entry too */
- pmap_pte_set(&APDP_PDE_SHADOW[i], 0);
-#endif
- }
+ pmap_unmap_apdp_pde();
pmap_pte_flush();
pmap_apte_flush(pmap_kernel());
kpreempt_enable();
@@ -2780,14 +2822,7 @@
* been freed
*/
if (*APDP_PDE) {
- int i;
- for (i = 0; i < PDP_SIZE; i++) {
- pmap_pte_set(&APDP_PDE[i], 0);
-#ifdef PAE
- /* clear shadow entry too */
- pmap_pte_set(&APDP_PDE_SHADOW[i], 0);
-#endif
- }
+ pmap_unmap_apdp_pde();
}
/* lldt() does pmap_pte_flush() */
#else /* XEN */
@@ -3026,11 +3061,10 @@
#ifdef XEN
/*
* pmap_extract_ma: extract a MA for the given VA
- */
-
-/*
- * XXX JYM replace functions calling pmap_extract_ma by wrappers, as they are
- * NOT safe in regards to ptom locking
+ * When used directly in a Xen domain, caller must ensure that this function
+ * should not be used in a context where a reference to a MFN could be kept
+ * between a suspend, resume, or migrate.
+ * XXX JYM revisit for locking (MFNs may be invalid after a migration)
*/
bool
pmap_extract_ma(struct pmap *pmap, vaddr_t va, paddr_t *pap)
Index: src/sys/arch/xen/include/xenbus.h
diff -u src/sys/arch/xen/include/xenbus.h:1.11 src/sys/arch/xen/include/xenbus.h:1.11.6.1
--- src/sys/arch/xen/include/xenbus.h:1.11 Wed Oct 29 13:35:35 2008
+++ src/sys/arch/xen/include/xenbus.h Sun May 31 20:15:36 2009
@@ -1,4 +1,4 @@
-/* $NetBSD: xenbus.h,v 1.11 2008/10/29 13:35:35 cegger Exp $ */
+/* $NetBSD: xenbus.h,v 1.11.6.1 2009/05/31 20:15:36 jym Exp $ */
/******************************************************************************
* xenbus.h
*
@@ -257,6 +257,8 @@
void xenbus_dev_fatal(struct xenbus_device *dev, int err, const char *fmt,
...);
+bool xenbus_device_suspend(struct xenbus_device *);
+bool xenbus_device_resume(struct xenbus_device *);
#endif /* _ASM_XEN_XENBUS_H */
Index: src/sys/arch/xen/include/amd64/hypercalls.h
diff -u src/sys/arch/xen/include/amd64/hypercalls.h:1.5 src/sys/arch/xen/include/amd64/hypercalls.h:1.5.4.1
--- src/sys/arch/xen/include/amd64/hypercalls.h:1.5 Thu Nov 13 01:45:48 2008
+++ src/sys/arch/xen/include/amd64/hypercalls.h Sun May 31 20:15:36 2009
@@ -1,4 +1,4 @@
-/* $NetBSD: hypercalls.h,v 1.5 2008/11/13 01:45:48 cegger Exp $ */
+/* $NetBSD: hypercalls.h,v 1.5.4.1 2009/05/31 20:15:36 jym Exp $ */
/******************************************************************************
* hypercall.h
*
@@ -406,4 +406,9 @@
return _hypercall1(int, mca, mc);
}
+static inline int
+HYPERVISOR_sysctl(void *sysctl)
+{
+ return _hypercall1(int, sysctl, sysctl);
+}
#endif /* __HYPERCALL_H__ */
Index: src/sys/arch/xen/include/i386/hypercalls.h
diff -u src/sys/arch/xen/include/i386/hypercalls.h:1.8 src/sys/arch/xen/include/i386/hypercalls.h:1.8.4.1
--- src/sys/arch/xen/include/i386/hypercalls.h:1.8 Thu Nov 13 18:35:20 2008
+++ src/sys/arch/xen/include/i386/hypercalls.h Sun May 31 20:15:37 2009
@@ -1,4 +1,4 @@
-/* $NetBSD: hypercalls.h,v 1.8 2008/11/13 18:35:20 cegger Exp $ */
+/* $NetBSD: hypercalls.h,v 1.8.4.1 2009/05/31 20:15:37 jym Exp $ */
/*
* Copyright (c) 2006 Manuel Bouyer.
@@ -796,4 +796,16 @@
return ret;
}
+
+static __inline int
+HYPERVISOR_sysctl(void *sysctl)
+{
+ int ret;
+ unsigned long ign1;
+
+ _hypercall(__HYPERVISOR_sysctl, _harg("1" (sysctl)),
+ _harg("=a" (ret), "=b" (ign1)));
+
+ return ret;
+}
#endif /* _XENI386_HYPERVISOR_H_ */
Index: src/sys/arch/xen/x86/x86_xpmap.c
diff -u src/sys/arch/xen/x86/x86_xpmap.c:1.12.4.2 src/sys/arch/xen/x86/x86_xpmap.c:1.12.4.3
--- src/sys/arch/xen/x86/x86_xpmap.c:1.12.4.2 Wed May 13 17:18:50 2009
+++ src/sys/arch/xen/x86/x86_xpmap.c Sun May 31 20:15:37 2009
@@ -1,4 +1,4 @@
-/* $NetBSD: x86_xpmap.c,v 1.12.4.2 2009/05/13 17:18:50 jym Exp $ */
+/* $NetBSD: x86_xpmap.c,v 1.12.4.3 2009/05/31 20:15:37 jym Exp $ */
/*
* Copyright (c) 2006 Mathieu Ropert <[email protected]>
@@ -79,7 +79,7 @@
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: x86_xpmap.c,v 1.12.4.2 2009/05/13 17:18:50 jym Exp $");
+__KERNEL_RCSID(0, "$NetBSD: x86_xpmap.c,v 1.12.4.3 2009/05/31 20:15:37 jym Exp $");
#include "opt_xen.h"
#include "opt_ddb.h"
@@ -139,6 +139,8 @@
*
* The thread responsible for the domU suspension will acquire an exclusive
* (writer) lock.
+ *
+ * XXX JYM the locking will need revisit - rwlock(9) is currently inadequate
*/
static krwlock_t xen_ptom_lock;
@@ -149,17 +151,17 @@
void
xen_release_ptom_lock(void) {
- rw_exit(&xen_ptom_lock);
+ /* rw_exit(&xen_ptom_lock); */
}
void
xen_acquire_reader_ptom_lock(void) {
- rw_enter(&xen_ptom_lock, RW_READER);
+ /* rw_enter(&xen_ptom_lock, RW_READER); */
}
void
xen_acquire_writer_ptom_lock(void) {
- rw_enter(&xen_ptom_lock, RW_WRITER);
+ /* rw_enter(&xen_ptom_lock, RW_WRITER); */
}
void xen_failsafe_handler(void);
Index: src/sys/arch/xen/x86/xen_bus_dma.c
diff -u src/sys/arch/xen/x86/xen_bus_dma.c:1.14.2.1 src/sys/arch/xen/x86/xen_bus_dma.c:1.14.2.2
--- src/sys/arch/xen/x86/xen_bus_dma.c:1.14.2.1 Mon Feb 9 00:03:55 2009
+++ src/sys/arch/xen/x86/xen_bus_dma.c Sun May 31 20:15:37 2009
@@ -1,4 +1,4 @@
-/* $NetBSD: xen_bus_dma.c,v 1.14.2.1 2009/02/09 00:03:55 jym Exp $ */
+/* $NetBSD: xen_bus_dma.c,v 1.14.2.2 2009/05/31 20:15:37 jym 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 1.14.2.1 2009/02/09 00:03:55 jym Exp $");
+__KERNEL_RCSID(0, "$NetBSD: xen_bus_dma.c,v 1.14.2.2 2009/05/31 20:15:37 jym Exp $");
#include <sys/param.h>
#include <sys/systm.h>
@@ -89,6 +89,7 @@
if (error)
return (error);
+ xen_acquire_reader_ptom_lock();
for (pg = mlistp->tqh_first; pg != NULL; pg = pg->pageq.queue.tqe_next) {
pa = VM_PAGE_TO_PHYS(pg);
mfn = xpmap_ptom(pa) >> PAGE_SHIFT;
@@ -125,6 +126,7 @@
}
#endif
}
+
/* Get the new contiguous memory extent */
#ifdef XEN3
xenguest_handle(res.extent_start) = &mfn;
@@ -169,10 +171,13 @@
uvm_pagefree(pg);
}
}
+
/* Flush updates through and flush the TLB */
xpq_queue_tlb_flush();
xpq_flush_queue();
splx(s);
+
+ xen_release_ptom_lock();
return 0;
failed:
Index: src/sys/arch/xen/xen/evtchn.c
diff -u src/sys/arch/xen/xen/evtchn.c:1.42.2.2 src/sys/arch/xen/xen/evtchn.c:1.42.2.3
--- src/sys/arch/xen/xen/evtchn.c:1.42.2.2 Wed May 13 17:18:50 2009
+++ src/sys/arch/xen/xen/evtchn.c Sun May 31 20:15:37 2009
@@ -1,4 +1,4 @@
-/* $NetBSD: evtchn.c,v 1.42.2.2 2009/05/13 17:18:50 jym Exp $ */
+/* $NetBSD: evtchn.c,v 1.42.2.3 2009/05/31 20:15:37 jym Exp $ */
/*
* Copyright (c) 2006 Manuel Bouyer.
@@ -64,7 +64,7 @@
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: evtchn.c,v 1.42.2.2 2009/05/13 17:18:50 jym Exp $");
+__KERNEL_RCSID(0, "$NetBSD: evtchn.c,v 1.42.2.3 2009/05/31 20:15:37 jym Exp $");
#include "opt_xen.h"
#include "isa.h"
@@ -192,7 +192,7 @@
ctrl_if_init();
#endif
debug_port = bind_virq_to_evtch(VIRQ_DEBUG);
- aprint_verbose("debug virtual interrupt using event channel %d\n",
+ aprint_verbose("VIRQ_DEBUG interrupt using event channel %d\n",
debug_port);
/*
* Don't call event_set_handler(), we'll use a shortcut. Just set
Index: src/sys/arch/xen/xen/if_xennet_xenbus.c
diff -u src/sys/arch/xen/xen/if_xennet_xenbus.c:1.33.2.2 src/sys/arch/xen/xen/if_xennet_xenbus.c:1.33.2.3
--- src/sys/arch/xen/xen/if_xennet_xenbus.c:1.33.2.2 Wed May 13 17:18:50 2009
+++ src/sys/arch/xen/xen/if_xennet_xenbus.c Sun May 31 20:15:37 2009
@@ -1,4 +1,4 @@
-/* $NetBSD: if_xennet_xenbus.c,v 1.33.2.2 2009/05/13 17:18:50 jym Exp $ */
+/* $NetBSD: if_xennet_xenbus.c,v 1.33.2.3 2009/05/31 20:15:37 jym Exp $ */
/*
* Copyright (c) 2006 Manuel Bouyer.
@@ -95,7 +95,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: if_xennet_xenbus.c,v 1.33.2.2 2009/05/13 17:18:50 jym Exp $");
+__KERNEL_RCSID(0, "$NetBSD: if_xennet_xenbus.c,v 1.33.2.3 2009/05/31 20:15:37 jym Exp $");
#include "opt_xen.h"
#include "opt_nfs_boot.h"
@@ -393,20 +393,21 @@
sc->sc_tx_ring.sring = tx_ring;
sc->sc_rx_ring.sring = rx_ring;
+ /* resume shared structures and tell backend that we are ready */
+ xennet_xenbus_resume(self, PMF_F_NONE);
+
#if NRND > 0
rnd_attach_source(&sc->sc_rnd_source, device_xname(sc->sc_dev),
RND_TYPE_NET, 0);
#endif
- /* initialize shared structures and tell backend that we are ready */
- xennet_xenbus_resume(self, PMF_F_NONE);
-
if (!pmf_device_register(self,
- xennet_xenbus_suspend,
+ xennet_xenbus_suspend,
xennet_xenbus_resume))
aprint_error_dev(self, "couldn't establish power handler\n");
else
pmf_class_network_register(self, ifp);
+
}
static int
@@ -477,6 +478,16 @@
paddr_t ma;
const char *errmsg;
+ /* invalidate the RX and TX rings */
+ if (sc->sc_backend_status == BEST_SUSPENDED) {
+ /*
+ * Device was suspended, so ensure that access associated to
+ * the previous RX and TX rings are revoked.
+ */
+ xengnt_revoke_access(sc->sc_tx_ring_gntref);
+ xengnt_revoke_access(sc->sc_rx_ring_gntref);
+ }
+
sc->sc_tx_ring_gntref = GRANT_INVALID_REF;
sc->sc_rx_ring_gntref = GRANT_INVALID_REF;
@@ -546,6 +557,11 @@
return false;
}
xennet_alloc_rx_buffer(sc);
+
+ if (sc->sc_backend_status == BEST_SUSPENDED) {
+ xenbus_device_resume(sc->sc_xbusd);
+ }
+
sc->sc_backend_status = BEST_CONNECTED;
return true;
@@ -562,22 +578,34 @@
int s;
struct xennet_xenbus_softc *sc = device_private(dev);
- /* xennet_stop() is called by pmf(9) before xennet_xenbus_suspend() */
+ /*
+ * xennet_stop() is called by pmf(9) before xennet_xenbus_suspend(),
+ * so we do not mask event channel here
+ */
s = splnet();
-
/* process any outstanding TX responses, then collect RX packets */
xennet_handler(sc);
while (sc->sc_tx_ring.sring->rsp_prod != sc->sc_tx_ring.rsp_cons) {
tsleep(xennet_xenbus_suspend, PRIBIO, "xnet_suspend", hz/2);
xennet_handler(sc);
}
- xennet_free_rx_buffer(sc);
+
+ /*
+ * dom0 may still use references to the grants we gave away
+ * earlier during RX buffers allocation. So we do not free RX buffers
+ * here, as dom0 does not expect the guest domain to suddenly revoke
+ * access to these grants.
+ */
sc->sc_backend_status = BEST_SUSPENDED;
+ event_remove_handler(sc->sc_evtchn, &xennet_handler, sc);
+
splx(s);
- event_remove_handler(sc->sc_evtchn, &xennet_handler, sc);
+ xenbus_device_suspend(sc->sc_xbusd);
+ aprint_verbose_dev(dev, "removed event channel %d\n", sc->sc_evtchn);
+
return true;
}
@@ -786,11 +814,6 @@
* RX buffers to catch-up with backend's consumption
*/
req->rxreq_gntref = GRANT_INVALID_REF;
-
- /*
- * ring needs more requests to be pushed in, allocate some
- * RX buffers to catch-up with backend's consumption
- */
if (sc->sc_free_rxreql >= SC_NLIVEREQ(sc) &&
__predict_true(sc->sc_backend_status == BEST_CONNECTED)) {
xennet_alloc_rx_buffer(sc);
@@ -836,6 +859,7 @@
else
ifp->if_opackets++;
xengnt_revoke_access(req->txreq_gntref);
+
m_freem(req->txreq_m);
SLIST_INSERT_HEAD(&sc->sc_txreq_head, req, txreq_next);
}
Index: src/sys/arch/xen/xen/xbd_xenbus.c
diff -u src/sys/arch/xen/xen/xbd_xenbus.c:1.38.2.2 src/sys/arch/xen/xen/xbd_xenbus.c:1.38.2.3
--- src/sys/arch/xen/xen/xbd_xenbus.c:1.38.2.2 Wed May 13 17:18:50 2009
+++ src/sys/arch/xen/xen/xbd_xenbus.c Sun May 31 20:15:37 2009
@@ -1,4 +1,4 @@
-/* $NetBSD: xbd_xenbus.c,v 1.38.2.2 2009/05/13 17:18:50 jym Exp $ */
+/* $NetBSD: xbd_xenbus.c,v 1.38.2.3 2009/05/31 20:15:37 jym Exp $ */
/*
* Copyright (c) 2006 Manuel Bouyer.
@@ -31,7 +31,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: xbd_xenbus.c,v 1.38.2.2 2009/05/13 17:18:50 jym Exp $");
+__KERNEL_RCSID(0, "$NetBSD: xbd_xenbus.c,v 1.38.2.3 2009/05/31 20:15:37 jym Exp $");
#include "opt_xen.h"
#include "rnd.h"
@@ -260,7 +260,7 @@
panic("%s: can't alloc ring", device_xname(self));
sc->sc_ring.sring = ring;
- /* initialize shared structures and tell backend that we are ready */
+ /* resume shared structures and tell backend that we are ready */
xbd_xenbus_resume(self, PMF_F_NONE);
#if NRND > 0
@@ -346,14 +346,13 @@
sc->sc_dksc.sc_dkdev.dk_stats->io_busy > 0)
tsleep(xbd_xenbus_suspend, PRIBIO, "xbdsuspend", hz/2);
+ hypervisor_mask_event(sc->sc_evtchn);
sc->sc_backend_status = BLKIF_STATE_SUSPENDED;
- splx(s);
+ event_remove_handler(sc->sc_evtchn, xbd_handler, sc);
- hypervisor_mask_event(sc->sc_evtchn);
- if (event_remove_handler(sc->sc_evtchn, xbd_handler, sc) != 0)
- aprint_error_dev(dev,
- "can't remove handler: xbd_handler\n");
+ splx(s);
+ xenbus_device_suspend(sc->sc_xbusd);
aprint_verbose_dev(dev, "removed event channel %d\n", sc->sc_evtchn);
return true;
@@ -370,6 +369,14 @@
const char *errmsg;
sc = device_private(dev);
+
+ if (sc->sc_backend_status == BLKIF_STATE_SUSPENDED) {
+ /*
+ * Device was suspended, so ensure that access associated to
+ * the block I/O ring is revoked.
+ */
+ xengnt_revoke_access(sc->sc_ring_gntref);
+ }
sc->sc_ring_gntref = GRANT_INVALID_REF;
ring = sc->sc_ring.sring;
@@ -381,7 +388,6 @@
* for the block device
*/
xen_acquire_reader_ptom_lock();
-
(void)pmap_extract_ma(pmap_kernel(), (vaddr_t)ring, &ma);
error = xenbus_grant_ring(sc->sc_xbusd, ma, &sc->sc_ring_gntref);
xen_release_ptom_lock();
@@ -437,13 +443,11 @@
* already initialized - we use a shortcut
*/
sc->sc_backend_status = BLKIF_STATE_CONNECTED;
+ xenbus_device_resume(sc->sc_xbusd);
hypervisor_enable_event(sc->sc_evtchn);
xenbus_switch_state(sc->sc_xbusd, NULL, XenbusStateConnected);
}
-// XXX JYM
- printf("read_otherend_details '%s' (%d)\n", sc->sc_xbusd->xbusd_otherend, sc->sc_xbusd->xbusd_otherend_id);
-
return true;
abort_transaction:
Index: src/sys/arch/xen/xen/xen_machdep.c
diff -u src/sys/arch/xen/xen/xen_machdep.c:1.4.12.2 src/sys/arch/xen/xen/xen_machdep.c:1.4.12.3
--- src/sys/arch/xen/xen/xen_machdep.c:1.4.12.2 Wed May 13 17:18:50 2009
+++ src/sys/arch/xen/xen/xen_machdep.c Sun May 31 20:15:37 2009
@@ -1,4 +1,4 @@
-/* $NetBSD: xen_machdep.c,v 1.4.12.2 2009/05/13 17:18:50 jym Exp $ */
+/* $NetBSD: xen_machdep.c,v 1.4.12.3 2009/05/31 20:15:37 jym Exp $ */
/*
* Copyright (c) 2006 Manuel Bouyer.
@@ -63,7 +63,7 @@
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: xen_machdep.c,v 1.4.12.2 2009/05/13 17:18:50 jym Exp $");
+__KERNEL_RCSID(0, "$NetBSD: xen_machdep.c,v 1.4.12.3 2009/05/31 20:15:37 jym Exp $");
#include "opt_xen.h"
@@ -246,7 +246,7 @@
* it should not call this function to register
* machdep.sleep_state sysctl
*/
- KASSERT( !(xen_start_info.flags & SIF_INITDOMAIN) );
+ KASSERT( !(xendomain_is_dom0()) );
ret = sysctl_createv(NULL, 0, NULL, NULL, CTLFLAG_READWRITE,
CTLTYPE_INT, "sleep_state", NULL, sysctl_xen_sleepstate, 0,
@@ -286,6 +286,13 @@
xen_suspendclocks();
xen_acquire_writer_ptom_lock();
+
+ /*
+ * Xen lazy evaluation of recursive mappings requires
+ * to flush the APDP entries
+ */
+ pmap_unmap_all_apdp_pdes();
+
/*
* save/restore code does not translate these MFNs to their
* associated PFNs, so we must do it
Index: src/sys/arch/xen/xen/xencons.c
diff -u src/sys/arch/xen/xen/xencons.c:1.31.2.1 src/sys/arch/xen/xen/xencons.c:1.31.2.2
--- src/sys/arch/xen/xen/xencons.c:1.31.2.1 Mon Feb 9 00:03:55 2009
+++ src/sys/arch/xen/xen/xencons.c Sun May 31 20:15:37 2009
@@ -1,4 +1,4 @@
-/* $NetBSD: xencons.c,v 1.31.2.1 2009/02/09 00:03:55 jym Exp $ */
+/* $NetBSD: xencons.c,v 1.31.2.2 2009/05/31 20:15:37 jym Exp $ */
/*
* Copyright (c) 2006 Manuel Bouyer.
@@ -63,7 +63,7 @@
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: xencons.c,v 1.31.2.1 2009/02/09 00:03:55 jym Exp $");
+__KERNEL_RCSID(0, "$NetBSD: xencons.c,v 1.31.2.2 2009/05/31 20:15:37 jym Exp $");
#include "opt_xen.h"
@@ -228,13 +228,13 @@
int evtch;
- if (xen_start_info.flags & SIF_INITDOMAIN) {
+ if (xendomain_is_dom0()) {
evtch = unbind_virq_from_evtch(VIRQ_CONSOLE);
hypervisor_mask_event(evtch);
if (event_remove_handler(evtch, xencons_intr,
xencons_console_device) != 0)
aprint_error_dev(dev,
- "can't remove handler: xencons_intr\n");
+ "can't remove handler: xencons_intr\n");
} else {
#ifdef XEN3
evtch = xen_start_info.console_evtchn;
@@ -242,7 +242,7 @@
if (event_remove_handler(evtch, xencons_handler,
xencons_console_device) != 0)
aprint_error_dev(dev,
- "can't remove handler: xencons_handler\n");
+ "can't remove handler: xencons_handler\n");
#endif
}
Index: src/sys/arch/xen/xen/xennetback_xenbus.c
diff -u src/sys/arch/xen/xen/xennetback_xenbus.c:1.27.2.2 src/sys/arch/xen/xen/xennetback_xenbus.c:1.27.2.3
--- src/sys/arch/xen/xen/xennetback_xenbus.c:1.27.2.2 Wed May 13 17:18:51 2009
+++ src/sys/arch/xen/xen/xennetback_xenbus.c Sun May 31 20:15:37 2009
@@ -1,4 +1,4 @@
-/* $NetBSD: xennetback_xenbus.c,v 1.27.2.2 2009/05/13 17:18:51 jym Exp $ */
+/* $NetBSD: xennetback_xenbus.c,v 1.27.2.3 2009/05/31 20:15:37 jym Exp $ */
/*
* Copyright (c) 2006 Manuel Bouyer.
@@ -214,11 +214,9 @@
0, 0, &mlist, NB_XMIT_PAGES_BATCH, 0) != 0)
panic("xennetback_init: uvm_pglistalloc");
- xen_acquire_reader_ptom_lock();
for (i = 0, pg = mlist.tqh_first; pg != NULL;
pg = pg->pageq.queue.tqe_next, i++)
mcl_pages[i] = xpmap_ptom(VM_PAGE_TO_PHYS(pg)) >> PAGE_SHIFT;
- xen_release_ptom_lock();
if (i != NB_XMIT_PAGES_BATCH)
@@ -862,7 +860,6 @@
return;
}
- xen_acquire_reader_ptom_lock();
while (!IFQ_IS_EMPTY(&ifp->if_snd)) {
XENPRINTF(("pkt\n"));
req_prod = xneti->xni_rxring.sring->req_prod;
@@ -1087,7 +1084,6 @@
break;
}
}
- xen_release_ptom_lock();
splx(s);
}
Index: src/sys/arch/xen/xenbus/xenbus_probe.c
diff -u src/sys/arch/xen/xenbus/xenbus_probe.c:1.27.2.1 src/sys/arch/xen/xenbus/xenbus_probe.c:1.27.2.2
--- src/sys/arch/xen/xenbus/xenbus_probe.c:1.27.2.1 Mon Feb 9 00:03:55 2009
+++ src/sys/arch/xen/xenbus/xenbus_probe.c Sun May 31 20:15:37 2009
@@ -1,4 +1,4 @@
-/* $NetBSD: xenbus_probe.c,v 1.27.2.1 2009/02/09 00:03:55 jym Exp $ */
+/* $NetBSD: xenbus_probe.c,v 1.27.2.2 2009/05/31 20:15:37 jym Exp $ */
/******************************************************************************
* Talks to Xen Store to figure out what devices we have.
*
@@ -29,7 +29,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: xenbus_probe.c,v 1.27.2.1 2009/02/09 00:03:55 jym Exp $");
+__KERNEL_RCSID(0, "$NetBSD: xenbus_probe.c,v 1.27.2.2 2009/05/31 20:15:37 jym Exp $");
#if 0
#define DPRINTK(fmt, args...) \
@@ -73,6 +73,13 @@
static bool xenbus_suspend(device_t PMF_FN_PROTO);
static bool xenbus_resume(device_t PMF_FN_PROTO);
+/* routines gathering device information from XenStore */
+static int read_otherend_details(struct xenbus_device *,
+ const char *, const char *);
+static int read_backend_details (struct xenbus_device *);
+static int read_frontend_details(struct xenbus_device *);
+static void free_otherend_details(struct xenbus_device *);
+
CFATTACH_DECL_NEW(xenbus, 0, xenbus_match, xenbus_attach,
NULL, NULL);
@@ -130,6 +137,29 @@
return true;
}
+/*
+ * Suspend a xenbus device
+ */
+bool
+xenbus_device_suspend(struct xenbus_device *dev) {
+ free_otherend_details(dev);
+
+ return true;
+}
+
+/*
+ * Resume a xenbus device
+ */
+bool
+xenbus_device_resume(struct xenbus_device *dev) {
+
+ if (dev->xbusd_type == XENBUS_FRONTEND_DEVICE) {
+ read_backend_details(dev);
+ }
+
+ return true;
+}
+
void
xenbus_backend_register(struct xenbus_backend_driver *xbakd)
{
@@ -180,8 +210,7 @@
printf("missing other end from %s\n", xendev->xbusd_path);
xenbus_dev_fatal(xendev, -ENOENT, "missing other end from %s",
xendev->xbusd_path);
- free(xendev->xbusd_otherend, M_DEVBUF);
- xendev->xbusd_otherend = NULL;
+ free_otherend_details(xendev);
return ENOENT;
}
@@ -201,15 +230,12 @@
return read_otherend_details(xendev, "frontend-id", "frontend");
}
-#if unused
static void
free_otherend_details(struct xenbus_device *dev)
{
free(dev->xbusd_otherend, M_DEVBUF);
dev->xbusd_otherend = NULL;
}
-#endif
-
static void
free_otherend_watch(struct xenbus_device *dev)
@@ -269,9 +295,19 @@
xdev->xbusd_u.b.b_cookie : xdev->xbusd_u.f.f_dev, state);
}
+#ifdef unused
static int
talk_to_otherend(struct xenbus_device *dev)
{
+ return xenbus_watch_path2(dev, dev->xbusd_otherend, "state",
+ &dev->xbusd_otherend_watch,
+ otherend_changed);
+}
+#endif
+
+static int
+watch_otherend(struct xenbus_device *dev)
+{
free_otherend_watch(dev);
return xenbus_watch_path2(dev, dev->xbusd_otherend, "state",
@@ -382,7 +418,7 @@
}
SLIST_INSERT_HEAD(&xenbus_device_list,
xbusd, xbusd_entries);
- talk_to_otherend(xbusd);
+ watch_otherend(xbusd);
}
free(dir, M_DEVBUF);
return err;
@@ -489,7 +525,7 @@
KASSERT(xenbus_lookup_device_path(xbusd->xbusd_path) == xbusd);
SLIST_REMOVE(&xenbus_device_list, xbusd, xenbus_device, xbusd_entries);
free_otherend_watch(xbusd);
- free(xbusd->xbusd_otherend, M_DEVBUF);
+ free_otherend_details(xbusd);
xenbus_switch_state(xbusd, NULL, XenbusStateClosed);
free(xbusd, M_DEVBUF);
return 0;
