Merge patch series "riscv: asid: switch to alternative way to fix stale TLB entries"

Sergey Matyukevich <geomatsi@gmail.com> says:

Some time ago two different patches have been posted to fix stale TLB
entries that caused applications crashes.

The patch [0] suggested 'aggregating' mm_cpumask, i.e. current cpu is not
cleared for the switched-out task in switch_mm function. For additional
explanations see the commit message by Guo Ren. The same approach is
used by arc architecture, so another good comment is for switch_mm
in arch/arc/include/asm/mmu_context.h.

The patch [1] attempted to reduce the number of TLB flushes by deferring
(and possibly avoiding) them for CPUs not running the task.

Patch [1] has been merged. However we already have two bug reports from
different vendors. So apparently something is missing in the approach
suggested in [1]. In both cases the patch [0] fixed the issue.

This patch series reverts [1] and replaces it by [0].

[0] https://lore.kernel.org/linux-riscv/20221111075902.798571-1-guoren@kernel.org/
[1] https://lore.kernel.org/linux-riscv/20220829205219.283543-1-geomatsi@gmail.com/

* b4-shazam-merge:
  riscv: asid: Fixup stale TLB entry cause application crash
  Revert "riscv: mm: notify remote harts about mmu cache updates"

Link: https://lore.kernel.org/r/20230226150137.1919750-1-geomatsi@gmail.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>

1 files changed, 20 insertions, 20 deletions

diff --git a/arch/riscv/mm/context.c b/arch/riscv/mm/context.c
index 80ce9caba8d2..0f784e3d307b 100644
--- a/arch/riscv/mm/context.c
+++ b/arch/riscv/mm/context.c
@@ -196,16 +196,6 @@ switch_mm_fast:
 
 	if (need_flush_tlb)
 		local_flush_tlb_all();
-#ifdef CONFIG_SMP
-	else {
-		cpumask_t *mask = &mm->context.tlb_stale_mask;
-
-		if (cpumask_test_cpu(cpu, mask)) {
-			cpumask_clear_cpu(cpu, mask);
-			local_flush_tlb_all_asid(cntx & asid_mask);
-		}
-	}
-#endif
 }
 
 static void set_mm_noasid(struct mm_struct *mm)
@@ -215,12 +205,24 @@ static void set_mm_noasid(struct mm_struct *mm)
 	local_flush_tlb_all();
 }
 
-static inline void set_mm(struct mm_struct *mm, unsigned int cpu)
+static inline void set_mm(struct mm_struct *prev,
+			  struct mm_struct *next, unsigned int cpu)
 {
-	if (static_branch_unlikely(&use_asid_allocator))
-		set_mm_asid(mm, cpu);
-	else
-		set_mm_noasid(mm);
+	/*
+	 * The mm_cpumask indicates which harts' TLBs contain the virtual
+	 * address mapping of the mm. Compared to noasid, using asid
+	 * can't guarantee that stale TLB entries are invalidated because
+	 * the asid mechanism wouldn't flush TLB for every switch_mm for
+	 * performance. So when using asid, keep all CPUs footmarks in
+	 * cpumask() until mm reset.
+	 */
+	cpumask_set_cpu(cpu, mm_cpumask(next));
+	if (static_branch_unlikely(&use_asid_allocator)) {
+		set_mm_asid(next, cpu);
+	} else {
+		cpumask_clear_cpu(cpu, mm_cpumask(prev));
+		set_mm_noasid(next);
+	}
 }
 
 static int __init asids_init(void)
@@ -274,7 +276,8 @@ static int __init asids_init(void)
 }
 early_initcall(asids_init);
 #else
-static inline void set_mm(struct mm_struct *mm, unsigned int cpu)
+static inline void set_mm(struct mm_struct *prev,
+			  struct mm_struct *next, unsigned int cpu)
 {
 	/* Nothing to do here when there is no MMU */
 }
@@ -327,10 +330,7 @@ void switch_mm(struct mm_struct *prev, struct mm_struct *next,
 	 */
 	cpu = smp_processor_id();
 
-	cpumask_clear_cpu(cpu, mm_cpumask(prev));
-	cpumask_set_cpu(cpu, mm_cpumask(next));
-
-	set_mm(next, cpu);
+	set_mm(prev, next, cpu);
 
 	flush_icache_deferred(next, cpu);
 }