[v2,1/2] x86/sgx: Use sgx_nr_used_pages for EPC page count instead of sgx_nr_free_pages

Commit Message

Reshetova, Elena March 28, 2025, 12:57 p.m. UTC

sgx_nr_free_pages is an atomic that is used to keep track of
free EPC pages and detect whenever page reclaiming should start.
Since successful execution of ENCLS[EUPDATESVN] requires empty
EPC and preferably a fast lockless way of checking for this
condition in all code paths where EPC is already used, change the
reclaiming code to track the number of used pages via
sgx_nr_used_pages instead of sgx_nr_free_pages.
For this change to work in the page reclamation code, add a new
variable, sgx_nr_total_pages, that will keep track of total
number of EPC pages.

It would have been possible to implement ENCLS[EUPDATESVN] using
existing sgx_nr_free_pages counter and a new sgx_nr_total_pages
counter, but it won't be possible to avoid taking a lock *every time*
a new EPC page is being allocated. The conversion of sgx_nr_free_pages
into sgx_nr_used_pages allows avoiding the lock in all cases except
when it is the first EPC page being allocated via a quick
atomic_long_inc_not_zero check.

Note: The serialization for sgx_nr_total_pages is not needed because
the variable is only updated during the initialization and there's no
concurrent access.

Signed-off-by: Elena Reshetova <[e-mail]>
---
 arch/x86/kernel/cpu/sgx/main.c | 15 ++++++++++-----
 1 file changed, 10 insertions(+), 5 deletions(-)

Patch

diff --git a/arch/x86/kernel/cpu/sgx/main.c b/arch/x86/kernel/cpu/sgx/main.c
index 8ce352fc72ac..b61d3bad0446 100644
--- a/arch/x86/kernel/cpu/sgx/main.c
+++ b/arch/x86/kernel/cpu/sgx/main.c
@@ -32,7 +32,8 @@  static DEFINE_XARRAY(sgx_epc_address_space);
 static LIST_HEAD(sgx_active_page_list);
 static DEFINE_SPINLOCK(sgx_reclaimer_lock);
 
-static atomic_long_t sgx_nr_free_pages = ATOMIC_LONG_INIT(0);
+static atomic_long_t sgx_nr_used_pages = ATOMIC_LONG_INIT(0);
+static unsigned long sgx_nr_total_pages;
 
 /* Nodes with one or more EPC sections. */
 static nodemask_t sgx_numa_mask;
@@ -378,8 +379,8 @@  static void sgx_reclaim_pages(void)
 
 static bool sgx_should_reclaim(unsigned long watermark)
 {
-	return atomic_long_read(&sgx_nr_free_pages) < watermark &&
-	       !list_empty(&sgx_active_page_list);
+	return (sgx_nr_total_pages - atomic_long_read(&sgx_nr_used_pages))
+	       < watermark && !list_empty(&sgx_active_page_list);
 }
 
 /*
@@ -456,7 +457,7 @@  static struct sgx_epc_page *__sgx_alloc_epc_page_from_node(int nid)
 	page->flags = 0;
 
 	spin_unlock(&node->lock);
-	atomic_long_dec(&sgx_nr_free_pages);
+	atomic_long_inc(&sgx_nr_used_pages);
 
 	return page;
 }
@@ -616,7 +617,7 @@  void sgx_free_epc_page(struct sgx_epc_page *page)
 	page->flags = SGX_EPC_PAGE_IS_FREE;
 
 	spin_unlock(&node->lock);
-	atomic_long_inc(&sgx_nr_free_pages);
+	atomic_long_dec(&sgx_nr_used_pages);
 }
 
 static bool __init sgx_setup_epc_section(u64 phys_addr, u64 size,
@@ -648,6 +649,8 @@  static bool __init sgx_setup_epc_section(u64 phys_addr, u64 size,
 		list_add_tail(&section->pages[i].list, &sgx_dirty_page_list);
 	}
 
+	sgx_nr_total_pages += nr_pages;
+
 	return true;
 }
 
@@ -848,6 +851,8 @@  static bool __init sgx_page_cache_init(void)
 		return false;
 	}
 
+	atomic_long_set(&sgx_nr_used_pages, sgx_nr_total_pages);
+
 	for_each_online_node(nid) {
 		if (!node_isset(nid, sgx_numa_mask) &&
 		    node_state(nid, N_MEMORY) && node_state(nid, N_CPU))