[PATCH v2 13/35] x86/mm: attempt speculative mm faults first

From: Michel Lespinasse <michel@lespinasse.org>
To: Linux-MM <linux-mm@kvack.org>,
	linux-kernel@vger.kernel.org,
	Andrew Morton <akpm@linux-foundation.org>
Cc: kernel-team@fb.com, Laurent Dufour <ldufour@linux.ibm.com>,
	Jerome Glisse <jglisse@google.com>,
	Peter Zijlstra <peterz@infradead.org>,
	Michal Hocko <mhocko@suse.com>, Vlastimil Babka <vbabka@suse.cz>,
	Davidlohr Bueso <dave@stgolabs.net>,
	Matthew Wilcox <willy@infradead.org>,
	Liam Howlett <liam.howlett@oracle.com>,
	Rik van Riel <riel@surriel.com>,
	Paul McKenney <paulmck@kernel.org>,
	Song Liu <songliubraving@fb.com>,
	Suren Baghdasaryan <surenb@google.com>,
	Minchan Kim <minchan@google.com>,
	Joel Fernandes <joelaf@google.com>,
	David Rientjes <rientjes@google.com>,
	Axel Rasmussen <axelrasmussen@google.com>,
	Andy Lutomirski <luto@kernel.org>,
	Michel Lespinasse <michel@lespinasse.org>
Subject: [PATCH v2 13/35] x86/mm: attempt speculative mm faults first
Date: Fri, 28 Jan 2022 05:09:44 -0800	[thread overview]
Message-ID: <20220128131006.67712-14-michel@lespinasse.org> (raw)
In-Reply-To: <20220128131006.67712-1-michel@lespinasse.org>

Attempt speculative mm fault handling first, and fall back to the
existing (non-speculative) code if that fails.

The speculative handling closely mirrors the non-speculative logic.
This includes some x86 specific bits such as the access_error() call.
This is why we chose to implement the speculative handling in arch/x86
rather than in common code.

The vma is first looked up and copied, under protection of the rcu
read lock. The mmap lock sequence count is used to verify the
integrity of the copied vma, and passed to do_handle_mm_fault() to
allow checking against races with mmap writers when finalizing the fault.

Signed-off-by: Michel Lespinasse <michel@lespinasse.org>
---
 arch/x86/mm/fault.c           | 44 +++++++++++++++++++++++++++++++++++
 include/linux/mm_types.h      |  5 ++++
 include/linux/vm_event_item.h |  4 ++++
 mm/vmstat.c                   |  4 ++++
 4 files changed, 57 insertions(+)

diff --git a/arch/x86/mm/fault.c b/arch/x86/mm/fault.c
index d0074c6ed31a..99b0a358154e 100644
--- a/arch/x86/mm/fault.c
+++ b/arch/x86/mm/fault.c
@@ -1226,6 +1226,10 @@ void do_user_addr_fault(struct pt_regs *regs,
 	struct mm_struct *mm;
 	vm_fault_t fault;
 	unsigned int flags = FAULT_FLAG_DEFAULT;
+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+	struct vm_area_struct pvma;
+	unsigned long seq;
+#endif
 
 	tsk = current;
 	mm = tsk->mm;
@@ -1323,6 +1327,43 @@ void do_user_addr_fault(struct pt_regs *regs,
 	}
 #endif
 
+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+	count_vm_event(SPF_ATTEMPT);
+	seq = mmap_seq_read_start(mm);
+	if (seq & 1)
+		goto spf_abort;
+	rcu_read_lock();
+	vma = __find_vma(mm, address);
+	if (!vma || vma->vm_start > address) {
+		rcu_read_unlock();
+		goto spf_abort;
+	}
+	pvma = *vma;
+	rcu_read_unlock();
+	if (!mmap_seq_read_check(mm, seq))
+		goto spf_abort;
+	vma = &pvma;
+	if (unlikely(access_error(error_code, vma)))
+		goto spf_abort;
+	fault = do_handle_mm_fault(vma, address,
+				   flags | FAULT_FLAG_SPECULATIVE, seq, regs);
+
+	if (!(fault & VM_FAULT_RETRY))
+		goto done;
+
+	/* Quick path to respond to signals */
+	if (fault_signal_pending(fault, regs)) {
+		if (!user_mode(regs))
+			kernelmode_fixup_or_oops(regs, error_code, address,
+						 SIGBUS, BUS_ADRERR,
+						 ARCH_DEFAULT_PKEY);
+		return;
+	}
+
+spf_abort:
+	count_vm_event(SPF_ABORT);
+#endif
+
 	/*
 	 * Kernel-mode access to the user address space should only occur
 	 * on well-defined single instructions listed in the exception
@@ -1419,6 +1460,9 @@ void do_user_addr_fault(struct pt_regs *regs,
 	}
 
 	mmap_read_unlock(mm);
+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+done:
+#endif
 	if (likely(!(fault & VM_FAULT_ERROR)))
 		return;
 
diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h
index b6678578a729..305f05d2a4bc 100644
--- a/include/linux/mm_types.h
+++ b/include/linux/mm_types.h
@@ -370,6 +370,11 @@ struct anon_vma_name {
  * per VM-area/task. A VM area is any part of the process virtual memory
  * space that has a special rule for the page-fault handlers (ie a shared
  * library, the executable area etc).
+ *
+ * Note that speculative page faults make an on-stack copy of the VMA,
+ * so the structure size matters.
+ * (TODO - it would be preferable to copy only the required vma attributes
+ *  rather than the entire vma).
  */
 struct vm_area_struct {
 	/* The first cache line has the info for VMA tree walking. */
diff --git a/include/linux/vm_event_item.h b/include/linux/vm_event_item.h
index 7b2363388bfa..f00b3e36ff39 100644
--- a/include/linux/vm_event_item.h
+++ b/include/linux/vm_event_item.h
@@ -133,6 +133,10 @@ enum vm_event_item { PGPGIN, PGPGOUT, PSWPIN, PSWPOUT,
 #ifdef CONFIG_X86
 		DIRECT_MAP_LEVEL2_SPLIT,
 		DIRECT_MAP_LEVEL3_SPLIT,
+#endif
+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+		SPF_ATTEMPT,
+		SPF_ABORT,
 #endif
 		NR_VM_EVENT_ITEMS
 };
diff --git a/mm/vmstat.c b/mm/vmstat.c
index 4057372745d0..dbb0160e5558 100644
--- a/mm/vmstat.c
+++ b/mm/vmstat.c
@@ -1390,6 +1390,10 @@ const char * const vmstat_text[] = {
 	"direct_map_level2_splits",
 	"direct_map_level3_splits",
 #endif
+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+	"spf_attempt",
+	"spf_abort",
+#endif
 #endif /* CONFIG_VM_EVENT_COUNTERS || CONFIG_MEMCG */
 };
 #endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA || CONFIG_MEMCG */
-- 
2.20.1

