The patch titled
Subject: mm: frontswap: lazy initialization to allow tmem backends to build/run as modules
has been added to the -mm tree. Its filename is
mm-frontswap-lazy-initialization-to-allow-tmem-backends-to-build-run-as-modules.patch
Before you just go and hit "reply", please:
a) Consider who else should be cc'ed
b) Prefer to cc a suitable mailing list as well
c) Ideally: find the original patch on the mailing list and do a
reply-to-all to that, adding suitable additional cc's
*** Remember to use Documentation/SubmitChecklist when testing your code ***
The -mm tree is included into linux-next and is updated
there every 3-4 working days
------------------------------------------------------
From: Dan Magenheimer <dan.magenheimer@xxxxxxxxxx>
Subject: mm: frontswap: lazy initialization to allow tmem backends to build/run as modules
With the goal of allowing tmem backends (zcache, ramster, Xen tmem) to be
built/loaded as modules rather than built-in and enabled by a boot
parameter, this patch provides "lazy initialization", allowing backends to
register to frontswap even after swapon was run. Before a backend
registers all calls to init are recorded and the creation of tmem_pools
delayed until a backend registers or until a frontswap store is attempted.
Signed-off-by: Stefan Hengelein <ilendir@xxxxxxxxxxxxxx>
Signed-off-by: Florian Schmaus <fschmaus@xxxxxxxxx>
Signed-off-by: Andor Daam <andor.daam@xxxxxxxxxxxxxx>
Signed-off-by: Dan Magenheimer <dan.magenheimer@xxxxxxxxxx>
[v1: Fixes per Seth Jennings suggestions]
[v2: Removed FRONTSWAP_HAS_.. ]
[v3: Fix up per Bob Liu <lliubbo@xxxxxxxxx> recommendations]
[v4: Fix up per Andrew's comments]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@xxxxxxxxxx>
Signed-off-by: Bob Liu <lliubbo@xxxxxxxxx>
Cc: Wanpeng Li <liwanp@xxxxxxxxxxxxxxxxxx>
Cc: Dan Magenheimer <dan.magenheimer@xxxxxxxxxx>
Cc: Minchan Kim <minchan@xxxxxxxxxx>
Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
---
mm/frontswap.c | 94 ++++++++++++++++++++++++++++++++++++++++++-----
1 file changed, 84 insertions(+), 10 deletions(-)
diff -puN mm/frontswap.c~mm-frontswap-lazy-initialization-to-allow-tmem-backends-to-build-run-as-modules mm/frontswap.c
--- a/mm/frontswap.c~mm-frontswap-lazy-initialization-to-allow-tmem-backends-to-build-run-as-modules
+++ a/mm/frontswap.c
@@ -80,6 +80,46 @@ static inline void inc_frontswap_succ_st
static inline void inc_frontswap_failed_stores(void) { }
static inline void inc_frontswap_invalidates(void) { }
#endif
+
+/*
+ * Due to the asynchronous nature of the backends loading potentially
+ * _after_ the swap system has been activated, we have chokepoints
+ * on all frontswap functions to not call the backend until the backend
+ * has registered.
+ *
+ * Specifically when no backend is registered (nobody called
+ * frontswap_register_ops) all calls to frontswap_init (which is done via
+ * swapon -> enable_swap_info -> frontswap_init) are registered and remembered
+ * (via the setting of need_init bitmap) but fail to create tmem_pools. When a
+ * backend registers with frontswap at some later point the previous
+ * calls to frontswap_init are executed (by iterating over the need_init
+ * bitmap) to create tmem_pools and set the respective poolids. All of that is
+ * guarded by us using atomic bit operations on the 'need_init' bitmap.
+ *
+ * This would not guards us against the user deciding to call swapoff right as
+ * we are calling the backend to initialize (so swapon is in action).
+ * Fortunatly for us, the swapon_mutex has been taked by the callee so we are
+ * OK. The other scenario where calls to frontswap_store (called via
+ * swap_writepage) is racing with frontswap_invalidate_area (called via
+ * swapoff) is again guarded by the swap subsystem.
+ *
+ * While no backend is registered all calls to frontswap_[store|load|
+ * invalidate_area|invalidate_page] are ignored or fail.
+ *
+ * The time between the backend being registered and the swap file system
+ * calling the backend (via the frontswap_* functions) is indeterminate as
+ * backend_registered is not atomic_t (or a value guarded by a spinlock).
+ * That is OK as we are comfortable missing some of these calls to the newly
+ * registered backend.
+ *
+ * Obviously the opposite (unloading the backend) must be done after all
+ * the frontswap_[store|load|invalidate_area|invalidate_page] start
+ * ignorning or failing the requests - at which point backend_registered
+ * would have to be made in some fashion atomic.
+ */
+static DECLARE_BITMAP(need_init, MAX_SWAPFILES);
+static bool backend_registered __read_mostly;
+
/*
* Register operations for frontswap, returning previous thus allowing
* detection of multiple backends and possible nesting.
@@ -87,9 +127,22 @@ static inline void inc_frontswap_invalid
struct frontswap_ops frontswap_register_ops(struct frontswap_ops *ops)
{
struct frontswap_ops old = frontswap_ops;
+ int i;
frontswap_ops = *ops;
frontswap_enabled = true;
+
+ for (i = 0; i < MAX_SWAPFILES; i++) {
+ if (test_and_clear_bit(i, need_init))
+ (*frontswap_ops.init)(i);
+ }
+ /*
+ * We MUST have backend_registered set _after_ the frontswap_init's
+ * have been called. Otherwise __frontswap_store might fail. Hence
+ * the barrier to make sure compiler does not re-order us.
+ */
+ barrier();
+ backend_registered = true;
return old;
}
EXPORT_SYMBOL(frontswap_register_ops);
@@ -119,10 +172,16 @@ void __frontswap_init(unsigned type)
{
struct swap_info_struct *sis = swap_info[type];
- BUG_ON(sis == NULL);
- if (sis->frontswap_map == NULL)
- return;
- frontswap_ops.init(type);
+ if (backend_registered) {
+ BUG_ON(sis == NULL);
+ if (sis->frontswap_map == NULL)
+ return;
+ (*frontswap_ops.init)(type);
+ } else {
+ BUG_ON(type > MAX_SWAPFILES);
+ set_bit(type, need_init);
+ }
+
}
EXPORT_SYMBOL(__frontswap_init);
@@ -147,6 +206,11 @@ int __frontswap_store(struct page *page)
struct swap_info_struct *sis = swap_info[type];
pgoff_t offset = swp_offset(entry);
+ if (!backend_registered) {
+ inc_frontswap_failed_stores();
+ return ret;
+ }
+
BUG_ON(!PageLocked(page));
BUG_ON(sis == NULL);
if (frontswap_test(sis, offset))
@@ -186,6 +250,9 @@ int __frontswap_load(struct page *page)
struct swap_info_struct *sis = swap_info[type];
pgoff_t offset = swp_offset(entry);
+ if (!backend_registered)
+ return ret;
+
BUG_ON(!PageLocked(page));
BUG_ON(sis == NULL);
if (frontswap_test(sis, offset))
@@ -209,6 +276,9 @@ void __frontswap_invalidate_page(unsigne
{
struct swap_info_struct *sis = swap_info[type];
+ if (!backend_registered)
+ return;
+
BUG_ON(sis == NULL);
if (frontswap_test(sis, offset)) {
frontswap_ops.invalidate_page(type, offset);
@@ -226,12 +296,15 @@ void __frontswap_invalidate_area(unsigne
{
struct swap_info_struct *sis = swap_info[type];
- BUG_ON(sis == NULL);
- if (sis->frontswap_map == NULL)
- return;
- frontswap_ops.invalidate_area(type);
- atomic_set(&sis->frontswap_pages, 0);
- memset(sis->frontswap_map, 0, sis->max / sizeof(long));
+ if (backend_registered) {
+ BUG_ON(sis == NULL);
+ if (sis->frontswap_map == NULL)
+ return;
+ (*frontswap_ops.invalidate_area)(type);
+ atomic_set(&sis->frontswap_pages, 0);
+ memset(sis->frontswap_map, 0, sis->max / sizeof(long));
+ }
+ clear_bit(type, need_init);
}
EXPORT_SYMBOL(__frontswap_invalidate_area);
@@ -364,6 +437,7 @@ static int __init init_frontswap(void)
debugfs_create_u64("invalidates", S_IRUGO,
root, &frontswap_invalidates);
#endif
+ frontswap_enabled = 1;
return 0;
}
_
Patches currently in -mm which might be from dan.magenheimer@xxxxxxxxxx are
linux-next.patch
mm-frontswap-lazy-initialization-to-allow-tmem-backends-to-build-run-as-modules.patch
frontswap-make-frontswap_init-use-a-pointer-for-the-ops.patch
mm-frontswap-cleanup-code.patch
frontswap-get-rid-of-swap_lock-dependency.patch
mm-cleancache-lazy-initialization-to-allow-tmem-backends-to-build-run-as-modules.patch
cleancache-make-cleancache_init-use-a-pointer-for-the-ops.patch
mm-cleancache-clean-up-cleancache_enabled.patch
xen-tmem-enable-xen-tmem-shim-to-be-built-loaded-as-a-module.patch
zcache-tmem-better-error-checking-on-frontswap_register_ops-return-value.patch
staging-zcache-enable-ramster-to-be-built-loaded-as-a-module.patch
staging-zcache-enable-zcache-to-be-built-loaded-as-a-module.patch
zcache-convert-to-idr_alloc.patch
--
To unsubscribe from this list: send the line "unsubscribe mm-commits" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
