[RFC XEN PATCH v4 28/41] xen/pmem: release PMEM pages on HVM domain destruction

[Haozhong Zhang <haozhong.zhang@intel.com>]

A new step RELMEM_pmem is added and taken before RELMEM_xen to release
all PMEM pages mapped to a HVM domain.

Signed-off-by: Haozhong Zhang <haozhong.zhang@intel.com>
---
Cc: Jan Beulich <jbeulich@suse.com>
Cc: Andrew Cooper <andrew.cooper3@citrix.com>
Cc: George Dunlap <George.Dunlap@eu.citrix.com>
Cc: Ian Jackson <ian.jackson@eu.citrix.com>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: Stefano Stabellini <sstabellini@kernel.org>
Cc: Tim Deegan <tim@xen.org>
Cc: Wei Liu <wei.liu2@citrix.com>
---
 xen/arch/x86/domain.c        | 32 ++++++++++++++++++++++++++++----
 xen/arch/x86/mm.c            |  9 +++++++--
 xen/common/pmem.c            | 10 ++++++++++
 xen/include/asm-x86/domain.h |  1 +
 xen/include/xen/pmem.h       |  6 ++++++
 5 files changed, 52 insertions(+), 6 deletions(-)

diff --git a/xen/arch/x86/domain.c b/xen/arch/x86/domain.c
index e1bf2d9e9d..613a8b4250 100644
--- a/xen/arch/x86/domain.c
+++ b/xen/arch/x86/domain.c
@@ -1801,11 +1801,15 @@ static int relinquish_memory(
 {
     struct page_info  *page;
     unsigned long     x, y;
+    bool              is_pmem_list = (list == &d->pmem_page_list);
     int               ret = 0;
 
     /* Use a recursive lock, as we may enter 'free_domheap_page'. */
     spin_lock_recursive(&d->page_alloc_lock);
 
+    if ( is_pmem_list )
+        spin_lock(&d->pmem_lock);
+
     while ( (page = page_list_remove_head(list)) )
     {
         /* Grab a reference to the page so it won't disappear from under us. */
@@ -1887,8 +1891,9 @@ static int relinquish_memory(
             }
         }
 
-        /* Put the page on the list and /then/ potentially free it. */
-        page_list_add_tail(page, &d->arch.relmem_list);
+        if ( !is_pmem_list )
+            /* Put the page on the list and /then/ potentially free it. */
+            page_list_add_tail(page, &d->arch.relmem_list);
         put_page(page);
 
         if ( hypercall_preempt_check() )
@@ -1898,10 +1903,13 @@ static int relinquish_memory(
         }
     }
 
-    /* list is empty at this point. */
-    page_list_move(list, &d->arch.relmem_list);
+    if ( !is_pmem_list )
+        /* list is empty at this point. */
+        page_list_move(list, &d->arch.relmem_list);
 
  out:
+    if ( is_pmem_list )
+        spin_unlock(&d->pmem_lock);
     spin_unlock_recursive(&d->page_alloc_lock);
     return ret;
 }
@@ -1968,13 +1976,29 @@ int domain_relinquish_resources(struct domain *d)
                 return ret;
         }
 
+#ifndef CONFIG_NVDIMM_PMEM
         d->arch.relmem = RELMEM_xen;
+#else
+        d->arch.relmem = RELMEM_pmem;
+#endif
 
         spin_lock(&d->page_alloc_lock);
         page_list_splice(&d->arch.relmem_list, &d->page_list);
         INIT_PAGE_LIST_HEAD(&d->arch.relmem_list);
         spin_unlock(&d->page_alloc_lock);
 
+#ifdef CONFIG_NVDIMM_PMEM
+        /* Fallthrough. Relinquish every page of PMEM. */
+    case RELMEM_pmem:
+        if ( is_hvm_domain(d) )
+        {
+            ret = relinquish_memory(d, &d->pmem_page_list, ~0UL);
+            if ( ret )
+                return ret;
+        }
+        d->arch.relmem = RELMEM_xen;
+#endif
+
         /* Fallthrough. Relinquish every page of memory. */
     case RELMEM_xen:
         ret = relinquish_memory(d, &d->xenpage_list, ~0UL);
diff --git a/xen/arch/x86/mm.c b/xen/arch/x86/mm.c
index 9a224cf1bb..9386e88eb1 100644
--- a/xen/arch/x86/mm.c
+++ b/xen/arch/x86/mm.c
@@ -106,6 +106,7 @@
 #include <xen/efi.h>
 #include <xen/grant_table.h>
 #include <xen/hypercall.h>
+#include <xen/pmem.h>
 #include <asm/paging.h>
 #include <asm/shadow.h>
 #include <asm/page.h>
@@ -2306,8 +2307,12 @@ void put_page(struct page_info *page)
 
     if ( unlikely((nx & PGC_count_mask) == 0) )
     {
-        if ( !is_pmem_page(page) /* PMEM page is not allocated from Xen heap. */
-             && cleanup_page_cacheattr(page) == 0 )
+#ifdef CONFIG_NVDIMM_PMEM
+        if ( is_pmem_page(page) )
+            pmem_page_cleanup(page);
+        else
+#endif
+        if ( cleanup_page_cacheattr(page) == 0 )
             free_domheap_page(page);
         else
             gdprintk(XENLOG_WARNING,
diff --git a/xen/common/pmem.c b/xen/common/pmem.c
index d2c5518329..a0d23cdfbe 100644
--- a/xen/common/pmem.c
+++ b/xen/common/pmem.c
@@ -733,6 +733,16 @@ int pmem_populate(struct xen_pmem_map_args *args)
     return rc;
 }
 
+void pmem_page_cleanup(struct page_info *page)
+{
+    ASSERT(is_pmem_page(page));
+    ASSERT((page->count_info & PGC_count_mask) == 0);
+
+    page->count_info = PGC_pmem_page | PGC_state_free;
+    page_set_owner(page, NULL);
+    set_gpfn_from_mfn(page_to_mfn(page), INVALID_M2P_ENTRY);
+}
+
 int __init pmem_dom0_setup_permission(struct domain *d)
 {
     struct list_head *cur;
diff --git a/xen/include/asm-x86/domain.h b/xen/include/asm-x86/domain.h
index f69911918e..e6f575244d 100644
--- a/xen/include/asm-x86/domain.h
+++ b/xen/include/asm-x86/domain.h
@@ -305,6 +305,7 @@ struct arch_domain
     enum {
         RELMEM_not_started,
         RELMEM_shared,
+        RELMEM_pmem,
         RELMEM_xen,
         RELMEM_l4,
         RELMEM_l3,
diff --git a/xen/include/xen/pmem.h b/xen/include/xen/pmem.h
index 2dab90530b..dfbc412065 100644
--- a/xen/include/xen/pmem.h
+++ b/xen/include/xen/pmem.h
@@ -21,6 +21,7 @@
 #ifdef CONFIG_NVDIMM_PMEM
 
 #include <public/sysctl.h>
+#include <xen/mm.h>
 #include <xen/types.h>
 
 int pmem_register(unsigned long smfn, unsigned long emfn, unsigned int pxm);
@@ -46,6 +47,7 @@ struct xen_pmem_map_args {
 };
 
 int pmem_populate(struct xen_pmem_map_args *args);
+void pmem_page_cleanup(struct page_info *page);
 
 #else /* !CONFIG_X86 */
 
@@ -64,6 +66,10 @@ static inline int pmem_populate(...)
     return -ENOSYS;
 }
 
+static inline void pmem_page_cleanup(...)
+{
+}
+
 #endif /* CONFIG_X86 */
 
 #endif /* CONFIG_NVDIMM_PMEM */
-- 
2.15.1


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel