[PATCH v11 0/6] populate/unpopulate memory when domain on static allocation

[PATCH v11 0/6] populate/unpopulate memory when domain on static allocation

[Penny Zheng]

Today when a domain unpopulates the memory on runtime, they will always
hand the memory over to the heap allocator. And it will be a problem if it
is a static domain.
Pages used as guest RAM for static domain shall always be reserved to this
domain only, and not be used for any other purposes, so they shall never go
back to heap allocator.

This patch serie intends to fix this issue, by adding pages on the new list
resv_page_list after having taken them off the "normal" list, when unpopulating
memory, and retrieving pages from resv page list(resv_page_list) when
populating memory.

---
v11 changes:
- printing message ahead of the assertion, which should also be
XENLOG_G_* kind of log level
- commit message tweak
- move "xen: rename free_staticmem_pages to unprepare_staticmem_pages" ahead
of "xen: retrieve reserved pages on populate_physmap"
- with assignment having failed and the page not exposed to the guest at any
point, there is no need for scrubbing
---
v10 changes:
- let Arm keep #define PGC_static 0 private, with the generic fallback
remaining in page_alloc.c
- change ASSERT(d) to ASSERT_UNREACHABLE() to be more robust looking
forward, and also add a printk() to log the problem
- mention the the removal of #ifdef CONFIG_STATIC_MEMORY in commit
message
- commit message typo fix
- Do not skip the list addition in that one special case
- add lock on the fail path
- new commit "xen: rename free_staticmem_pages to unprepare_staticmem_pages"
---
v9 changes:
- move free_domheap_page into else-condition
- considering scrubbing static pages, domain dying case and opt_scrub_domheap
both do not apply to static pages.
- as unowned static pages don't make themselves to free_domstatic_page
at the moment, remove else-condition and add ASSERT(d) at the top of the
function
- remove macro helper put_static_page, and just expand its code inside
free_domstatic_page
- Use ASSERT_ALLOC_CONTEXT() in acquire_reserved_page
- Add free_staticmem_pages to undo prepare_staticmem_pages when
assign_domstatic_pages fails
- Remove redundant static in error message
---
v8 changes:
- introduce new helper free_domstatic_page
- let put_page call free_domstatic_page for static page, when last ref
drops
- #define PGC_static zero when !CONFIG_STATIC_MEMORY, as it is used
outside page_alloc.c
- #ifdef-ary around is_domain_using_staticmem() is not needed anymore
- order as a parameter is not needed here, as all staticmem operations are
limited to order-0 regions
- move d->page_alloc_lock after operation on d->resv_page_list
- As concurrent free/allocate could modify the resv_page_list, we still
need the lock
---
v7 changes:
- protect free_staticmem_pages with heap_lock to match its reverse function
acquire_staticmem_pages
- IS_ENABLED(CONFIG_STATIC_MEMORY) would not be needed anymore
- add page on the rsv_page_list *after* it has been freed
- remove the lock, since we add the page to rsv_page_list after it has
been totally freed.
---
v6 changes:
- rename PGC_staticmem to PGC_static
- remove #ifdef aroud function declaration
- use domain instead of sub-systems
- move non-zero is_domain_using_staticmem() from ARM header to common
header
- move PGC_static !CONFIG_STATIC_MEMORY definition to common header
- drop the lock before returning
---
v5 changes:
- introduce three new commits
- In order to avoid stub functions, we #define PGC_staticmem to non-zero only
when CONFIG_STATIC_MEMORY
- use "unlikely()" around pg->count_info & PGC_staticmem
- remove pointless "if", since mark_page_free() is going to set count_info
to PGC_state_free and by consequence clear PGC_staticmem
- move #define PGC_staticmem 0 to mm.h
- guard "is_domain_using_staticmem" under CONFIG_STATIC_MEMORY
- #define is_domain_using_staticmem zero if undefined
- extract common codes for assigning pages into a helper assign_domstatic_pages
- refine commit message
- remove stub function acquire_reserved_page
- Alloc/free of memory can happen concurrently. So access to rsv_page_list
needs to be protected with a spinlock
---
v4 changes:
- commit message refinement
- miss dropping __init in acquire_domstatic_pages
- add the page back to the reserved list in case of error
- remove redundant printk
- refine log message and make it warn level
- guard "is_domain_using_staticmem" under CONFIG_STATIC_MEMORY
- #define is_domain_using_staticmem zero if undefined
---
v3 changes:
- fix possible racy issue in free_staticmem_pages()
- introduce a stub free_staticmem_pages() for the !CONFIG_STATIC_MEMORY case
- move the change to free_heap_pages() to cover other potential call sites
- change fixed width type uint32_t to unsigned int
- change "flags" to a more descriptive name "cdf"
- change name from "is_domain_static()" to "is_domain_using_staticmem"
- have page_list_del() just once out of the if()
- remove resv_pages counter
- make arch_free_heap_page be an expression, not a compound statement.
- move #ifndef is_domain_using_staticmem to the common header file
- remove #ifdef CONFIG_STATIC_MEMORY-ary
- remove meaningless page_to_mfn(page) in error log
---
v2 changes:
- let "flags" live in the struct domain. So other arch can take
advantage of it in the future
- change name from "is_domain_on_static_allocation" to "is_domain_static()"
- put reserved pages on resv_page_list after having taken them off
the "normal" list
- introduce acquire_reserved_page to retrieve reserved pages from
resv_page_list
- forbid non-zero-order requests in populate_physmap
- let is_domain_static return ((void)(d), false) on x86
- fix coding style
Penny Zheng (6):
  xen: do not free reserved memory into heap
  xen/arm: introduce CDF_staticmem
  xen: unpopulate memory when domain is static
  xen: introduce prepare_staticmem_pages
  xen: rename free_staticmem_pages to unprepare_staticmem_pages
  xen: retrieve reserved pages on populate_physmap

 xen/arch/arm/domain_build.c   |   5 +-
 xen/arch/arm/include/asm/mm.h |   6 +-
 xen/arch/arm/mm.c             |   5 +-
 xen/arch/arm/setup.c          |   3 +-
 xen/common/domain.c           |   4 +
 xen/common/memory.c           |  23 +++++
 xen/common/page_alloc.c       | 171 ++++++++++++++++++++++++++--------
 xen/include/xen/domain.h      |   8 ++
 xen/include/xen/mm.h          |   8 +-
 xen/include/xen/sched.h       |   3 +
 10 files changed, 189 insertions(+), 47 deletions(-)

-- 
2.25.1

[PATCH v11 1/6] xen: do not free reserved memory into heap

[Penny Zheng]

Pages used as guest RAM for static domain, shall be reserved to this
domain only.
So in case reserved pages being used for other purpose, users
shall not free them back to heap, even when last ref gets dropped.

This commit introduces a new helper free_domstatic_page to free
static page in runtime, and free_staticmem_pages will be called by it
in runtime, so let's drop the __init flag.

Wrapper #ifdef CONFIG_STATIC_MEMORY around function declaration(
free_staticmem_pages, free_domstatic_page, etc) is kinds of redundant,
so we decide to remove it here.

Signed-off-by: Penny Zheng <penny.zheng@arm.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Julien Grall <jgrall@amazon.com>
---
v11 changes:
- printing message ahead of the assertion, which should also be
XENLOG_G_* kind of log level
---
v10 changes:
- let Arm keep #define PGC_static 0 private, with the generic fallback
remaining in page_alloc.c
- change ASSERT(d) to ASSERT_UNREACHABLE() to be more robust looking
forward, and also add a printk() to log the problem
- mention the the removal of #ifdef CONFIG_STATIC_MEMORY in commit
message
---
v9 changes:
- move free_domheap_page into else-condition
- considering scrubbing static pages, domain dying case and opt_scrub_domheap
both donot apply to static pages.
- as unowned static pages don't make themselves to free_domstatic_page
at the moment, remove else-condition and add ASSERT(d) at the top of the
function
---
v8 changes:
- introduce new helper free_domstatic_page
- let put_page call free_domstatic_page for static page, when last ref
drops
- #define PGC_static zero when !CONFIG_STATIC_MEMORY, as it is used
outside page_alloc.c
---
v7 changes:
- protect free_staticmem_pages with heap_lock to match its reverse function
acquire_staticmem_pages
---
v6 changes:
- adapt to PGC_static
- remove #ifdef aroud function declaration
---
v5 changes:
- In order to avoid stub functions, we #define PGC_staticmem to non-zero only
when CONFIG_STATIC_MEMORY
- use "unlikely()" around pg->count_info & PGC_staticmem
- remove pointless "if", since mark_page_free() is going to set count_info
to PGC_state_free and by consequence clear PGC_staticmem
- move #define PGC_staticmem 0 to mm.h
---
v4 changes:
- no changes
---
v3 changes:
- fix possible racy issue in free_staticmem_pages()
- introduce a stub free_staticmem_pages() for the !CONFIG_STATIC_MEMORY case
- move the change to free_heap_pages() to cover other potential call sites
- fix the indentation
---
v2 changes:
- new commit
---
 xen/arch/arm/include/asm/mm.h |  6 +++++-
 xen/arch/arm/mm.c             |  5 ++++-
 xen/common/page_alloc.c       | 40 ++++++++++++++++++++++++++++++++---
 xen/include/xen/mm.h          |  3 +--
 4 files changed, 47 insertions(+), 7 deletions(-)

diff --git a/xen/arch/arm/include/asm/mm.h b/xen/arch/arm/include/asm/mm.h
index da25251cda..749fbefa0c 100644
--- a/xen/arch/arm/include/asm/mm.h
+++ b/xen/arch/arm/include/asm/mm.h
@@ -121,9 +121,13 @@ struct page_info
   /* Page is Xen heap? */
 #define _PGC_xen_heap     PG_shift(2)
 #define PGC_xen_heap      PG_mask(1, 2)
-  /* Page is static memory */
+#ifdef CONFIG_STATIC_MEMORY
+/* Page is static memory */
 #define _PGC_static    PG_shift(3)
 #define PGC_static     PG_mask(1, 3)
+#else
+#define PGC_static     0
+#endif
 /* ... */
 /* Page is broken? */
 #define _PGC_broken       PG_shift(7)
diff --git a/xen/arch/arm/mm.c b/xen/arch/arm/mm.c
index b42cddb1b4..fbdab5598c 100644
--- a/xen/arch/arm/mm.c
+++ b/xen/arch/arm/mm.c
@@ -1496,7 +1496,10 @@ void put_page(struct page_info *page)
 
     if ( unlikely((nx & PGC_count_mask) == 0) )
     {
-        free_domheap_page(page);
+        if ( unlikely(nx & PGC_static) )
+            free_domstatic_page(page);
+        else
+            free_domheap_page(page);
     }
 }
 
diff --git a/xen/common/page_alloc.c b/xen/common/page_alloc.c
index bfd4150be7..0c50dee4c5 100644
--- a/xen/common/page_alloc.c
+++ b/xen/common/page_alloc.c
@@ -2694,12 +2694,14 @@ struct domain *get_pg_owner(domid_t domid)
 
 #ifdef CONFIG_STATIC_MEMORY
 /* Equivalent of free_heap_pages to free nr_mfns pages of static memory. */
-void __init free_staticmem_pages(struct page_info *pg, unsigned long nr_mfns,
-                                 bool need_scrub)
+void free_staticmem_pages(struct page_info *pg, unsigned long nr_mfns,
+                          bool need_scrub)
 {
     mfn_t mfn = page_to_mfn(pg);
     unsigned long i;
 
+    spin_lock(&heap_lock);
+
     for ( i = 0; i < nr_mfns; i++ )
     {
         mark_page_free(&pg[i], mfn_add(mfn, i));
@@ -2710,9 +2712,41 @@ void __init free_staticmem_pages(struct page_info *pg, unsigned long nr_mfns,
             scrub_one_page(pg);
         }
 
-        /* In case initializing page of static memory, mark it PGC_static. */
         pg[i].count_info |= PGC_static;
     }
+
+    spin_unlock(&heap_lock);
+}
+
+void free_domstatic_page(struct page_info *page)
+{
+    struct domain *d = page_get_owner(page);
+    bool drop_dom_ref;
+
+    if ( unlikely(!d) )
+    {
+        printk(XENLOG_G_ERR
+               "The about-to-free static page %"PRI_mfn" must be owned by a domain\n",
+               mfn_x(page_to_mfn(page)));
+        ASSERT_UNREACHABLE();
+        return;
+    }
+
+    ASSERT_ALLOC_CONTEXT();
+
+    /* NB. May recursively lock from relinquish_memory(). */
+    spin_lock_recursive(&d->page_alloc_lock);
+
+    arch_free_heap_page(d, page);
+
+    drop_dom_ref = !domain_adjust_tot_pages(d, -1);
+
+    spin_unlock_recursive(&d->page_alloc_lock);
+
+    free_staticmem_pages(page, 1, scrub_debug);
+
+    if ( drop_dom_ref )
+        put_domain(d);
 }
 
 /*
diff --git a/xen/include/xen/mm.h b/xen/include/xen/mm.h
index 35b065146f..deadf4b2a1 100644
--- a/xen/include/xen/mm.h
+++ b/xen/include/xen/mm.h
@@ -85,13 +85,12 @@ bool scrub_free_pages(void);
 } while ( false )
 #define FREE_XENHEAP_PAGE(p) FREE_XENHEAP_PAGES(p, 0)
 
-#ifdef CONFIG_STATIC_MEMORY
 /* These functions are for static memory */
 void free_staticmem_pages(struct page_info *pg, unsigned long nr_mfns,
                           bool need_scrub);
+void free_domstatic_page(struct page_info *page);
 int acquire_domstatic_pages(struct domain *d, mfn_t smfn, unsigned int nr_mfns,
                             unsigned int memflags);
-#endif
 
 /* Map machine page range in Xen virtual address space. */
 int map_pages_to_xen(
-- 
2.25.1

[PATCH v11 2/6] xen/arm: introduce CDF_staticmem

[Penny Zheng]

In order to have an easy and quick way to find out whether this domain memory
is statically configured, this commit introduces a new flag CDF_staticmem and a
new helper is_domain_using_staticmem() to tell.

Signed-off-by: Penny Zheng <penny.zheng@arm.com>
Acked-by: Julien Grall <jgrall@amazon.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
---
v11 changes:
- no change
---
v10 changes:
- no change
---
v9 changes:
- no change
---
v8 changes:
- #ifdef-ary around is_domain_using_staticmem() is not needed anymore
---
v7 changes:
- IS_ENABLED(CONFIG_STATIC_MEMORY) would not be needed anymore
---
v6 changes:
- move non-zero is_domain_using_staticmem() from ARM header to common
header
---
v5 changes:
- guard "is_domain_using_staticmem" under CONFIG_STATIC_MEMORY
- #define is_domain_using_staticmem zero if undefined
---
v4 changes:
- no changes
---
v3 changes:
- change name from "is_domain_static()" to "is_domain_using_staticmem"
---
v2 changes:
- change name from "is_domain_on_static_allocation" to "is_domain_static()
---
 xen/arch/arm/domain_build.c | 5 ++++-
 xen/include/xen/domain.h    | 8 ++++++++
 2 files changed, 12 insertions(+), 1 deletion(-)

diff --git a/xen/arch/arm/domain_build.c b/xen/arch/arm/domain_build.c
index 3fd1186b53..b76a84e8f5 100644
--- a/xen/arch/arm/domain_build.c
+++ b/xen/arch/arm/domain_build.c
@@ -3287,9 +3287,12 @@ void __init create_domUs(void)
         if ( !dt_device_is_compatible(node, "xen,domain") )
             continue;
 
+        if ( dt_find_property(node, "xen,static-mem", NULL) )
+            flags |= CDF_staticmem;
+
         if ( dt_property_read_bool(node, "direct-map") )
         {
-            if ( !IS_ENABLED(CONFIG_STATIC_MEMORY) || !dt_find_property(node, "xen,static-mem", NULL) )
+            if ( !(flags & CDF_staticmem) )
                 panic("direct-map is not valid for domain %s without static allocation.\n",
                       dt_node_name(node));
 
diff --git a/xen/include/xen/domain.h b/xen/include/xen/domain.h
index 628b14b086..2c8116afba 100644
--- a/xen/include/xen/domain.h
+++ b/xen/include/xen/domain.h
@@ -35,6 +35,14 @@ void arch_get_domain_info(const struct domain *d,
 /* Should domain memory be directly mapped? */
 #define CDF_directmap            (1U << 1)
 #endif
+/* Is domain memory on static allocation? */
+#ifdef CONFIG_STATIC_MEMORY
+#define CDF_staticmem            (1U << 2)
+#else
+#define CDF_staticmem            0
+#endif
+
+#define is_domain_using_staticmem(d) ((d)->cdf & CDF_staticmem)
 
 /*
  * Arch-specifics.
-- 
2.25.1

[PATCH v11 3/6] xen: unpopulate memory when domain is static

[Penny Zheng]

Today when a domain unpopulates the memory on runtime, they will always
hand the memory back to the heap allocator. And it will be a problem if domain
is static.

Pages as guest RAM for static domain shall be reserved to only this domain
and not be used for any other purposes, so they shall never go back to heap
allocator.

This commit puts reserved page on the new list resv_page_list after
it has been freed.

Signed-off-by: Penny Zheng <penny.zheng@arm.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Julien Grall <jgrall@amazon.com>
---
v11 change:
- commit message tweak
---
v10 change:
- Do not skip the list addition in that one special case
---
v9 change:
- remove macro helper put_static_page, and just expand its code inside
free_domstatic_page
---
v8 changes:
- adapt this patch for newly introduced free_domstatic_page
- order as a parameter is not needed here, as all staticmem operations are
limited to order-0 regions
- move d->page_alloc_lock after operation on d->resv_page_list
---
v7 changes:
- Add page on the rsv_page_list *after* it has been freed
---
v6 changes:
- refine in-code comment
- move PGC_static !CONFIG_STATIC_MEMORY definition to common header
---
v5 changes:
- adapt this patch for PGC_staticmem
---
v4 changes:
- no changes
---
v3 changes:
- have page_list_del() just once out of the if()
- remove resv_pages counter
- make arch_free_heap_page be an expression, not a compound statement.
---
v2 changes:
- put reserved pages on resv_page_list after having taken them off
the "normal" list
---
 xen/common/domain.c     | 4 ++++
 xen/common/page_alloc.c | 7 +++++--
 xen/include/xen/sched.h | 3 +++
 3 files changed, 12 insertions(+), 2 deletions(-)

diff --git a/xen/common/domain.c b/xen/common/domain.c
index 7062393e37..c23f449451 100644
--- a/xen/common/domain.c
+++ b/xen/common/domain.c
@@ -604,6 +604,10 @@ struct domain *domain_create(domid_t domid,
     INIT_PAGE_LIST_HEAD(&d->page_list);
     INIT_PAGE_LIST_HEAD(&d->extra_page_list);
     INIT_PAGE_LIST_HEAD(&d->xenpage_list);
+#ifdef CONFIG_STATIC_MEMORY
+    INIT_PAGE_LIST_HEAD(&d->resv_page_list);
+#endif
+
 
     spin_lock_init(&d->node_affinity_lock);
     d->node_affinity = NODE_MASK_ALL;
diff --git a/xen/common/page_alloc.c b/xen/common/page_alloc.c
index 0c50dee4c5..26a2fad4e3 100644
--- a/xen/common/page_alloc.c
+++ b/xen/common/page_alloc.c
@@ -2741,10 +2741,13 @@ void free_domstatic_page(struct page_info *page)
 
     drop_dom_ref = !domain_adjust_tot_pages(d, -1);
 
-    spin_unlock_recursive(&d->page_alloc_lock);
-
     free_staticmem_pages(page, 1, scrub_debug);
 
+    /* Add page on the resv_page_list *after* it has been freed. */
+    page_list_add_tail(page, &d->resv_page_list);
+
+    spin_unlock_recursive(&d->page_alloc_lock);
+
     if ( drop_dom_ref )
         put_domain(d);
 }
diff --git a/xen/include/xen/sched.h b/xen/include/xen/sched.h
index 1cf629e7ec..956e0f9dca 100644
--- a/xen/include/xen/sched.h
+++ b/xen/include/xen/sched.h
@@ -381,6 +381,9 @@ struct domain
     struct page_list_head page_list;  /* linked list */
     struct page_list_head extra_page_list; /* linked list (size extra_pages) */
     struct page_list_head xenpage_list; /* linked list (size xenheap_pages) */
+#ifdef CONFIG_STATIC_MEMORY
+    struct page_list_head resv_page_list; /* linked list */
+#endif
 
     /*
      * This field should only be directly accessed by domain_adjust_tot_pages()
-- 
2.25.1

[PATCH v11 4/6] xen: introduce prepare_staticmem_pages

[Penny Zheng]

Later, we want to use acquire_domstatic_pages() for populating memory
for static domain on runtime, however, there are a lot of pointless work
(checking mfn_valid(), scrubbing the free part, cleaning the cache...)
considering we know the page is valid and belong to the guest.

This commit splits acquire_staticmem_pages() in two parts, and
introduces prepare_staticmem_pages to bypass all "pointless work".

Signed-off-by: Penny Zheng <penny.zheng@arm.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Julien Grall <jgrall@amazon.com>
---
v11 changes:
- no change
---
v10 changes:
- no change
---
v9 changes:
- no change
---
v8 changes:
- no change
---
v7 changes:
- no change
---
v6 changes:
- adapt to PGC_static
---
v5 changes:
- new commit
---
 xen/common/page_alloc.c | 61 ++++++++++++++++++++++++-----------------
 1 file changed, 36 insertions(+), 25 deletions(-)

diff --git a/xen/common/page_alloc.c b/xen/common/page_alloc.c
index 26a2fad4e3..adcc16e4f6 100644
--- a/xen/common/page_alloc.c
+++ b/xen/common/page_alloc.c
@@ -2752,26 +2752,13 @@ void free_domstatic_page(struct page_info *page)
         put_domain(d);
 }
 
-/*
- * Acquire nr_mfns contiguous reserved pages, starting at #smfn, of
- * static memory.
- * This function needs to be reworked if used outside of boot.
- */
-static struct page_info * __init acquire_staticmem_pages(mfn_t smfn,
-                                                         unsigned long nr_mfns,
-                                                         unsigned int memflags)
+static bool __init prepare_staticmem_pages(struct page_info *pg,
+                                           unsigned long nr_mfns,
+                                           unsigned int memflags)
 {
     bool need_tlbflush = false;
     uint32_t tlbflush_timestamp = 0;
     unsigned long i;
-    struct page_info *pg;
-
-    ASSERT(nr_mfns);
-    for ( i = 0; i < nr_mfns; i++ )
-        if ( !mfn_valid(mfn_add(smfn, i)) )
-            return NULL;
-
-    pg = mfn_to_page(smfn);
 
     spin_lock(&heap_lock);
 
@@ -2782,7 +2769,7 @@ static struct page_info * __init acquire_staticmem_pages(mfn_t smfn,
         {
             printk(XENLOG_ERR
                    "pg[%lu] Static MFN %"PRI_mfn" c=%#lx t=%#x\n",
-                   i, mfn_x(smfn) + i,
+                   i, mfn_x(page_to_mfn(pg)) + i,
                    pg[i].count_info, pg[i].tlbflush_timestamp);
             goto out_err;
         }
@@ -2806,6 +2793,38 @@ static struct page_info * __init acquire_staticmem_pages(mfn_t smfn,
     if ( need_tlbflush )
         filtered_flush_tlb_mask(tlbflush_timestamp);
 
+    return true;
+
+ out_err:
+    while ( i-- )
+        pg[i].count_info = PGC_static | PGC_state_free;
+
+    spin_unlock(&heap_lock);
+
+    return false;
+}
+
+/*
+ * Acquire nr_mfns contiguous reserved pages, starting at #smfn, of
+ * static memory.
+ * This function needs to be reworked if used outside of boot.
+ */
+static struct page_info * __init acquire_staticmem_pages(mfn_t smfn,
+                                                         unsigned long nr_mfns,
+                                                         unsigned int memflags)
+{
+    unsigned long i;
+    struct page_info *pg;
+
+    ASSERT(nr_mfns);
+    for ( i = 0; i < nr_mfns; i++ )
+        if ( !mfn_valid(mfn_add(smfn, i)) )
+            return NULL;
+
+    pg = mfn_to_page(smfn);
+    if ( !prepare_staticmem_pages(pg, nr_mfns, memflags) )
+        return NULL;
+
     /*
      * Ensure cache and RAM are consistent for platforms where the guest
      * can control its own visibility of/through the cache.
@@ -2814,14 +2833,6 @@ static struct page_info * __init acquire_staticmem_pages(mfn_t smfn,
         flush_page_to_ram(mfn_x(smfn) + i, !(memflags & MEMF_no_icache_flush));
 
     return pg;
-
- out_err:
-    while ( i-- )
-        pg[i].count_info = PGC_static | PGC_state_free;
-
-    spin_unlock(&heap_lock);
-
-    return NULL;
 }
 
 /*
-- 
2.25.1

[PATCH v11 5/6] xen: rename free_staticmem_pages to unprepare_staticmem_pages

[Penny Zheng]

The name of free_staticmem_pages is inappropriate, considering it is
the opposite of function prepare_staticmem_pages.

Rename free_staticmem_pages to unprepare_staticmem_pages.

Signed-off-by: Penny Zheng <penny.zheng@arm.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
---
v11 changes:
- moved ahead of "xen: retrieve reserved pages on populate_physmap"
---
v10 changes:
- new commit
---
 xen/arch/arm/setup.c    |  3 ++-
 xen/common/page_alloc.c | 13 ++++++++-----
 xen/include/xen/mm.h    |  4 ++--
 3 files changed, 12 insertions(+), 8 deletions(-)

diff --git a/xen/arch/arm/setup.c b/xen/arch/arm/setup.c
index 500307edc0..4662997c7e 100644
--- a/xen/arch/arm/setup.c
+++ b/xen/arch/arm/setup.c
@@ -639,7 +639,8 @@ static void __init init_staticmem_pages(void)
             if ( mfn_x(bank_end) <= mfn_x(bank_start) )
                 return;
 
-            free_staticmem_pages(mfn_to_page(bank_start), bank_pages, false);
+            unprepare_staticmem_pages(mfn_to_page(bank_start),
+                                      bank_pages, false);
         }
     }
 #endif
diff --git a/xen/common/page_alloc.c b/xen/common/page_alloc.c
index adcc16e4f6..18d34d1b69 100644
--- a/xen/common/page_alloc.c
+++ b/xen/common/page_alloc.c
@@ -2693,9 +2693,12 @@ struct domain *get_pg_owner(domid_t domid)
 }
 
 #ifdef CONFIG_STATIC_MEMORY
-/* Equivalent of free_heap_pages to free nr_mfns pages of static memory. */
-void free_staticmem_pages(struct page_info *pg, unsigned long nr_mfns,
-                          bool need_scrub)
+/*
+ * It is the opposite of prepare_staticmem_pages, and it aims to unprepare
+ * nr_mfns pages of static memory.
+ */
+void unprepare_staticmem_pages(struct page_info *pg, unsigned long nr_mfns,
+                               bool need_scrub)
 {
     mfn_t mfn = page_to_mfn(pg);
     unsigned long i;
@@ -2741,7 +2744,7 @@ void free_domstatic_page(struct page_info *page)
 
     drop_dom_ref = !domain_adjust_tot_pages(d, -1);
 
-    free_staticmem_pages(page, 1, scrub_debug);
+    unprepare_staticmem_pages(page, 1, scrub_debug);
 
     /* Add page on the resv_page_list *after* it has been freed. */
     page_list_add_tail(page, &d->resv_page_list);
@@ -2862,7 +2865,7 @@ int __init acquire_domstatic_pages(struct domain *d, mfn_t smfn,
 
     if ( assign_pages(pg, nr_mfns, d, memflags) )
     {
-        free_staticmem_pages(pg, nr_mfns, memflags & MEMF_no_scrub);
+        unprepare_staticmem_pages(pg, nr_mfns, memflags & MEMF_no_scrub);
         return -EINVAL;
     }
 
diff --git a/xen/include/xen/mm.h b/xen/include/xen/mm.h
index deadf4b2a1..93db3c4418 100644
--- a/xen/include/xen/mm.h
+++ b/xen/include/xen/mm.h
@@ -86,8 +86,8 @@ bool scrub_free_pages(void);
 #define FREE_XENHEAP_PAGE(p) FREE_XENHEAP_PAGES(p, 0)
 
 /* These functions are for static memory */
-void free_staticmem_pages(struct page_info *pg, unsigned long nr_mfns,
-                          bool need_scrub);
+void unprepare_staticmem_pages(struct page_info *pg, unsigned long nr_mfns,
+                               bool need_scrub);
 void free_domstatic_page(struct page_info *page);
 int acquire_domstatic_pages(struct domain *d, mfn_t smfn, unsigned int nr_mfns,
                             unsigned int memflags);
-- 
2.25.1

[PATCH v11 6/6] xen: retrieve reserved pages on populate_physmap

[Penny Zheng]

When a static domain populates memory through populate_physmap at runtime,
it shall retrieve reserved pages from resv_page_list to make sure that
guest RAM is still restricted in statically configured memory regions.
This commit also introduces a new helper acquire_reserved_page to make it work.

Signed-off-by: Penny Zheng <penny.zheng@arm.com>
---
v11 change:
- with assignment having failed and the page not exposed to the guest at any
point, there is no need for scrubbing
---
v10 changes:
- add lock on the fail path
---
v9 changes:
- Use ASSERT_ALLOC_CONTEXT() in acquire_reserved_page
- Add free_staticmem_pages to undo prepare_staticmem_pages when
assign_domstatic_pages
- Remove redundant static in error message
---
v8 changes:
- As concurrent free/allocate could modify the resv_page_list, we still
need the lock
---
v7 changes:
- remove the lock, since we add the page to rsv_page_list after it has
been totally freed.
---
v6 changes:
- drop the lock before returning
---
v5 changes:
- extract common codes for assigning pages into a helper assign_domstatic_pages
- refine commit message
- remove stub function acquire_reserved_page
- Alloc/free of memory can happen concurrently. So access to rsv_page_list
needs to be protected with a spinlock
---
v4 changes：
- miss dropping __init in acquire_domstatic_pages
- add the page back to the reserved list in case of error
- remove redundant printk
- refine log message and make it warn level
---
v3 changes:
- move is_domain_using_staticmem to the common header file
- remove #ifdef CONFIG_STATIC_MEMORY-ary
- remove meaningless page_to_mfn(page) in error log
---
v2 changes:
- introduce acquire_reserved_page to retrieve reserved pages from
resv_page_list
- forbid non-zero-order requests in populate_physmap
- let is_domain_static return ((void)(d), false) on x86
---
 xen/common/memory.c     | 23 +++++++++++++
 xen/common/page_alloc.c | 76 ++++++++++++++++++++++++++++++++---------
 xen/include/xen/mm.h    |  1 +
 3 files changed, 83 insertions(+), 17 deletions(-)

diff --git a/xen/common/memory.c b/xen/common/memory.c
index bc89442ba5..ae8163a738 100644
--- a/xen/common/memory.c
+++ b/xen/common/memory.c
@@ -245,6 +245,29 @@ static void populate_physmap(struct memop_args *a)
 
                 mfn = _mfn(gpfn);
             }
+            else if ( is_domain_using_staticmem(d) )
+            {
+                /*
+                 * No easy way to guarantee the retrieved pages are contiguous,
+                 * so forbid non-zero-order requests here.
+                 */
+                if ( a->extent_order != 0 )
+                {
+                    gdprintk(XENLOG_WARNING,
+                             "Cannot allocate static order-%u pages for %pd\n",
+                             a->extent_order, d);
+                    goto out;
+                }
+
+                mfn = acquire_reserved_page(d, a->memflags);
+                if ( mfn_eq(mfn, INVALID_MFN) )
+                {
+                    gdprintk(XENLOG_WARNING,
+                             "%pd: failed to retrieve a reserved page\n",
+                             d);
+                    goto out;
+                }
+            }
             else
             {
                 page = alloc_domheap_pages(d, a->extent_order, a->memflags);
diff --git a/xen/common/page_alloc.c b/xen/common/page_alloc.c
index 18d34d1b69..93d504c3c4 100644
--- a/xen/common/page_alloc.c
+++ b/xen/common/page_alloc.c
@@ -2755,9 +2755,8 @@ void free_domstatic_page(struct page_info *page)
         put_domain(d);
 }
 
-static bool __init prepare_staticmem_pages(struct page_info *pg,
-                                           unsigned long nr_mfns,
-                                           unsigned int memflags)
+static bool prepare_staticmem_pages(struct page_info *pg, unsigned long nr_mfns,
+                                    unsigned int memflags)
 {
     bool need_tlbflush = false;
     uint32_t tlbflush_timestamp = 0;
@@ -2838,21 +2837,9 @@ static struct page_info * __init acquire_staticmem_pages(mfn_t smfn,
     return pg;
 }
 
-/*
- * Acquire nr_mfns contiguous pages, starting at #smfn, of static memory,
- * then assign them to one specific domain #d.
- */
-int __init acquire_domstatic_pages(struct domain *d, mfn_t smfn,
-                                   unsigned int nr_mfns, unsigned int memflags)
+static int assign_domstatic_pages(struct domain *d, struct page_info *pg,
+                                  unsigned int nr_mfns, unsigned int memflags)
 {
-    struct page_info *pg;
-
-    ASSERT_ALLOC_CONTEXT();
-
-    pg = acquire_staticmem_pages(smfn, nr_mfns, memflags);
-    if ( !pg )
-        return -ENOENT;
-
     if ( !d || (memflags & (MEMF_no_owner | MEMF_no_refcount)) )
     {
         /*
@@ -2871,6 +2858,61 @@ int __init acquire_domstatic_pages(struct domain *d, mfn_t smfn,
 
     return 0;
 }
+
+/*
+ * Acquire nr_mfns contiguous pages, starting at #smfn, of static memory,
+ * then assign them to one specific domain #d.
+ */
+int __init acquire_domstatic_pages(struct domain *d, mfn_t smfn,
+                                   unsigned int nr_mfns, unsigned int memflags)
+{
+    struct page_info *pg;
+
+    ASSERT_ALLOC_CONTEXT();
+
+    pg = acquire_staticmem_pages(smfn, nr_mfns, memflags);
+    if ( !pg )
+        return -ENOENT;
+
+    if ( assign_domstatic_pages(d, pg, nr_mfns, memflags) )
+        return -EINVAL;
+
+    return 0;
+}
+
+/*
+ * Acquire a page from reserved page list(resv_page_list), when populating
+ * memory for static domain on runtime.
+ */
+mfn_t acquire_reserved_page(struct domain *d, unsigned int memflags)
+{
+    struct page_info *page;
+
+    ASSERT_ALLOC_CONTEXT();
+
+    /* Acquire a page from reserved page list(resv_page_list). */
+    spin_lock(&d->page_alloc_lock);
+    page = page_list_remove_head(&d->resv_page_list);
+    spin_unlock(&d->page_alloc_lock);
+    if ( unlikely(!page) )
+        return INVALID_MFN;
+
+    if ( !prepare_staticmem_pages(page, 1, memflags) )
+        goto fail;
+
+    if ( assign_domstatic_pages(d, page, 1, memflags) )
+        goto fail_assign;
+
+    return page_to_mfn(page);
+
+ fail_assign:
+    unprepare_staticmem_pages(page, 1, false);
+ fail:
+    spin_lock(&d->page_alloc_lock);
+    page_list_add_tail(page, &d->resv_page_list);
+    spin_unlock(&d->page_alloc_lock);
+    return INVALID_MFN;
+}
 #endif
 
 /*
diff --git a/xen/include/xen/mm.h b/xen/include/xen/mm.h
index 93db3c4418..a925028ab3 100644
--- a/xen/include/xen/mm.h
+++ b/xen/include/xen/mm.h
@@ -198,6 +198,7 @@ struct npfec {
 #else
 #define MAX_ORDER 20 /* 2^20 contiguous pages */
 #endif
+mfn_t acquire_reserved_page(struct domain *d, unsigned int memflags);
 
 /* Private domain structs for DOMID_XEN, DOMID_IO, etc. */
 extern struct domain *dom_xen, *dom_io;
-- 
2.25.1

Re: [PATCH v11 0/6] populate/unpopulate memory when domain on static allocation

[Stefano Stabellini]

I looked at the series and patches 1-5 are ready to be committed.

Patch 6 needs an Ack, preferably from Jan as Jan had questions on this
patch in v9.

I plan to commit patches 1-5 in the next couple of days (and ideally
patch 6 too if an Ack comes from Jan).

Cheers,

Stefano


On Wed, 31 Aug 2022, Penny Zheng wrote:
> Today when a domain unpopulates the memory on runtime, they will always
> hand the memory over to the heap allocator. And it will be a problem if it
> is a static domain.
> Pages used as guest RAM for static domain shall always be reserved to this
> domain only, and not be used for any other purposes, so they shall never go
> back to heap allocator.
> 
> This patch serie intends to fix this issue, by adding pages on the new list
> resv_page_list after having taken them off the "normal" list, when unpopulating
> memory, and retrieving pages from resv page list(resv_page_list) when
> populating memory.
> 
> ---
> v11 changes:
> - printing message ahead of the assertion, which should also be
> XENLOG_G_* kind of log level
> - commit message tweak
> - move "xen: rename free_staticmem_pages to unprepare_staticmem_pages" ahead
> of "xen: retrieve reserved pages on populate_physmap"
> - with assignment having failed and the page not exposed to the guest at any
> point, there is no need for scrubbing
> ---
> v10 changes:
> - let Arm keep #define PGC_static 0 private, with the generic fallback
> remaining in page_alloc.c
> - change ASSERT(d) to ASSERT_UNREACHABLE() to be more robust looking
> forward, and also add a printk() to log the problem
> - mention the the removal of #ifdef CONFIG_STATIC_MEMORY in commit
> message
> - commit message typo fix
> - Do not skip the list addition in that one special case
> - add lock on the fail path
> - new commit "xen: rename free_staticmem_pages to unprepare_staticmem_pages"
> ---
> v9 changes:
> - move free_domheap_page into else-condition
> - considering scrubbing static pages, domain dying case and opt_scrub_domheap
> both do not apply to static pages.
> - as unowned static pages don't make themselves to free_domstatic_page
> at the moment, remove else-condition and add ASSERT(d) at the top of the
> function
> - remove macro helper put_static_page, and just expand its code inside
> free_domstatic_page
> - Use ASSERT_ALLOC_CONTEXT() in acquire_reserved_page
> - Add free_staticmem_pages to undo prepare_staticmem_pages when
> assign_domstatic_pages fails
> - Remove redundant static in error message
> ---
> v8 changes:
> - introduce new helper free_domstatic_page
> - let put_page call free_domstatic_page for static page, when last ref
> drops
> - #define PGC_static zero when !CONFIG_STATIC_MEMORY, as it is used
> outside page_alloc.c
> - #ifdef-ary around is_domain_using_staticmem() is not needed anymore
> - order as a parameter is not needed here, as all staticmem operations are
> limited to order-0 regions
> - move d->page_alloc_lock after operation on d->resv_page_list
> - As concurrent free/allocate could modify the resv_page_list, we still
> need the lock
> ---
> v7 changes:
> - protect free_staticmem_pages with heap_lock to match its reverse function
> acquire_staticmem_pages
> - IS_ENABLED(CONFIG_STATIC_MEMORY) would not be needed anymore
> - add page on the rsv_page_list *after* it has been freed
> - remove the lock, since we add the page to rsv_page_list after it has
> been totally freed.
> ---
> v6 changes:
> - rename PGC_staticmem to PGC_static
> - remove #ifdef aroud function declaration
> - use domain instead of sub-systems
> - move non-zero is_domain_using_staticmem() from ARM header to common
> header
> - move PGC_static !CONFIG_STATIC_MEMORY definition to common header
> - drop the lock before returning
> ---
> v5 changes:
> - introduce three new commits
> - In order to avoid stub functions, we #define PGC_staticmem to non-zero only
> when CONFIG_STATIC_MEMORY
> - use "unlikely()" around pg->count_info & PGC_staticmem
> - remove pointless "if", since mark_page_free() is going to set count_info
> to PGC_state_free and by consequence clear PGC_staticmem
> - move #define PGC_staticmem 0 to mm.h
> - guard "is_domain_using_staticmem" under CONFIG_STATIC_MEMORY
> - #define is_domain_using_staticmem zero if undefined
> - extract common codes for assigning pages into a helper assign_domstatic_pages
> - refine commit message
> - remove stub function acquire_reserved_page
> - Alloc/free of memory can happen concurrently. So access to rsv_page_list
> needs to be protected with a spinlock
> ---
> v4 changes:
> - commit message refinement
> - miss dropping __init in acquire_domstatic_pages
> - add the page back to the reserved list in case of error
> - remove redundant printk
> - refine log message and make it warn level
> - guard "is_domain_using_staticmem" under CONFIG_STATIC_MEMORY
> - #define is_domain_using_staticmem zero if undefined
> ---
> v3 changes:
> - fix possible racy issue in free_staticmem_pages()
> - introduce a stub free_staticmem_pages() for the !CONFIG_STATIC_MEMORY case
> - move the change to free_heap_pages() to cover other potential call sites
> - change fixed width type uint32_t to unsigned int
> - change "flags" to a more descriptive name "cdf"
> - change name from "is_domain_static()" to "is_domain_using_staticmem"
> - have page_list_del() just once out of the if()
> - remove resv_pages counter
> - make arch_free_heap_page be an expression, not a compound statement.
> - move #ifndef is_domain_using_staticmem to the common header file
> - remove #ifdef CONFIG_STATIC_MEMORY-ary
> - remove meaningless page_to_mfn(page) in error log
> ---
> v2 changes:
> - let "flags" live in the struct domain. So other arch can take
> advantage of it in the future
> - change name from "is_domain_on_static_allocation" to "is_domain_static()"
> - put reserved pages on resv_page_list after having taken them off
> the "normal" list
> - introduce acquire_reserved_page to retrieve reserved pages from
> resv_page_list
> - forbid non-zero-order requests in populate_physmap
> - let is_domain_static return ((void)(d), false) on x86
> - fix coding style
> Penny Zheng (6):
>   xen: do not free reserved memory into heap
>   xen/arm: introduce CDF_staticmem
>   xen: unpopulate memory when domain is static
>   xen: introduce prepare_staticmem_pages
>   xen: rename free_staticmem_pages to unprepare_staticmem_pages
>   xen: retrieve reserved pages on populate_physmap
> 
>  xen/arch/arm/domain_build.c   |   5 +-
>  xen/arch/arm/include/asm/mm.h |   6 +-
>  xen/arch/arm/mm.c             |   5 +-
>  xen/arch/arm/setup.c          |   3 +-
>  xen/common/domain.c           |   4 +
>  xen/common/memory.c           |  23 +++++
>  xen/common/page_alloc.c       | 171 ++++++++++++++++++++++++++--------
>  xen/include/xen/domain.h      |   8 ++
>  xen/include/xen/mm.h          |   8 +-
>  xen/include/xen/sched.h       |   3 +
>  10 files changed, 189 insertions(+), 47 deletions(-)
> 
> -- 
> 2.25.1
> 

Re: [PATCH v11 0/6] populate/unpopulate memory when domain on static allocation

[Julien Grall]

Hi Stefano,

On 31/08/2022 23:44, Stefano Stabellini wrote:
> I looked at the series and patches 1-5 are ready to be committed.
> 
> Patch 6 needs an Ack, preferably from Jan as Jan had questions on this
> patch in v9.
> 
> I plan to commit patches 1-5 in the next couple of days (and ideally
> patch 6 too if an Ack comes from Jan).

IIRC Jan is away this week. So I will have a look at the patch later today.

Please don't commit patch #1-5 until patch #6 is ready as they don't 
make sense to be merged alone.

Cheers,

-- 
Julien Grall

Re: [PATCH v11 5/6] xen: rename free_staticmem_pages to unprepare_staticmem_pages

[Julien Grall]

Hi Penny,

On 31/08/2022 03:40, Penny Zheng wrote:
> The name of free_staticmem_pages is inappropriate, considering it is
> the opposite of function prepare_staticmem_pages.
> 
> Rename free_staticmem_pages to unprepare_staticmem_pages.
> 
> Signed-off-by: Penny Zheng <penny.zheng@arm.com>
> Acked-by: Jan Beulich <jbeulich@suse.com>

Reviewed-by: Julien Grall <jgrall@amazon.com>

Cheers,

-- 
Julien Grall

Re: [PATCH v11 6/6] xen: retrieve reserved pages on populate_physmap

[Julien Grall]

Hi Penny,

On 31/08/2022 03:40, Penny Zheng wrote:
> +/*
> + * Acquire a page from reserved page list(resv_page_list), when populating
> + * memory for static domain on runtime.
> + */
> +mfn_t acquire_reserved_page(struct domain *d, unsigned int memflags)
> +{
> +    struct page_info *page;
> +
> +    ASSERT_ALLOC_CONTEXT();
> +
> +    /* Acquire a page from reserved page list(resv_page_list). */
> +    spin_lock(&d->page_alloc_lock);
> +    page = page_list_remove_head(&d->resv_page_list);
> +    spin_unlock(&d->page_alloc_lock);
> +    if ( unlikely(!page) )
> +        return INVALID_MFN;
> +
> +    if ( !prepare_staticmem_pages(page, 1, memflags) )
> +        goto fail;
> +
> +    if ( assign_domstatic_pages(d, page, 1, memflags) )
> +        goto fail_assign;
> +
> +    return page_to_mfn(page);
> +
> + fail_assign:
> +    unprepare_staticmem_pages(page, 1, false);

Looking at assign_domstatic_pages(). It will already call 
unprepare_staticmem_pages() in one of the error path. It doesn't look 
like the latter can be called twice on a page.

To be honest, I find a bit odd that assign_domstatic_pages() is calling 
unprepare_staticmem_pages() because the former doesn't call the 
"prepare" function.

AFAICT, this is an issue introduced in this patch. So I would remove the 
call from assign_domstatic_pages() and then let the caller calls 
unprepare_staticmem_pages() (this would need to be added in 
acquire_domstatic_pages()).

Also, I think it would be good to explain why we don't need to scrub. 
Something like:

"The page was never accessible by the domain. So scrubbing can be skipped".

Cheers,

-- 
Julien Grall

RE: [PATCH v11 6/6] xen: retrieve reserved pages on populate_physmap

[Penny Zheng]

Hi Julien

> -----Original Message-----
> From: Julien Grall <julien@xen.org>
> Sent: Thursday, September 1, 2022 9:53 PM
> To: Penny Zheng <Penny.Zheng@arm.com>; xen-devel@lists.xenproject.org
> Cc: Wei Chen <Wei.Chen@arm.com>; Andrew Cooper
> <andrew.cooper3@citrix.com>; George Dunlap <george.dunlap@citrix.com>;
> Jan Beulich <jbeulich@suse.com>; Stefano Stabellini <sstabellini@kernel.org>;
> Wei Liu <wl@xen.org>
> Subject: Re: [PATCH v11 6/6] xen: retrieve reserved pages on
> populate_physmap
> 
> Hi Penny,
> 
> On 31/08/2022 03:40, Penny Zheng wrote:
> > +/*
> > + * Acquire a page from reserved page list(resv_page_list), when
> > +populating
> > + * memory for static domain on runtime.
> > + */
> > +mfn_t acquire_reserved_page(struct domain *d, unsigned int memflags)
> > +{
> > +    struct page_info *page;
> > +
> > +    ASSERT_ALLOC_CONTEXT();
> > +
> > +    /* Acquire a page from reserved page list(resv_page_list). */
> > +    spin_lock(&d->page_alloc_lock);
> > +    page = page_list_remove_head(&d->resv_page_list);
> > +    spin_unlock(&d->page_alloc_lock);
> > +    if ( unlikely(!page) )
> > +        return INVALID_MFN;
> > +
> > +    if ( !prepare_staticmem_pages(page, 1, memflags) )
> > +        goto fail;
> > +
> > +    if ( assign_domstatic_pages(d, page, 1, memflags) )
> > +        goto fail_assign;
> > +
> > +    return page_to_mfn(page);
> > +
> > + fail_assign:
> > +    unprepare_staticmem_pages(page, 1, false);
> 
> Looking at assign_domstatic_pages(). It will already call
> unprepare_staticmem_pages() in one of the error path. It doesn't look like
> the latter can be called twice on a page.
> 
> To be honest, I find a bit odd that assign_domstatic_pages() is calling
> unprepare_staticmem_pages() because the former doesn't call the "prepare"
> function.
> 
> AFAICT, this is an issue introduced in this patch. So I would remove the call
> from assign_domstatic_pages() and then let the caller calls
> unprepare_staticmem_pages() (this would need to be added in
> acquire_domstatic_pages()).
> 

True, true, thanks for pointing out!!

> Also, I think it would be good to explain why we don't need to scrub.
> Something like:
> 
> "The page was never accessible by the domain. So scrubbing can be skipped".
> 

Ok,  I'll add in-code comment

> Cheers,
> 
> --
> Julien Grall