arch/tile: use interrupt critical sections less

In general we want to avoid ever touching memory while within an
interrupt critical section, since the page fault path goes through
a different path from the hypervisor when in an interrupt critical
section, and we carefully decided with tilegx that we didn't need
to support this path in the kernel.  (On tilepro we did implement
that path as part of supporting atomic instructions in software.)

In practice we always need to touch the kernel stack, since that's
where we store the interrupt state before releasing the critical
section, but this change cleans up a few things.  The IRQ_ENABLE
macro is split up so that when we want to enable interrupts in a
deferred way (e.g. for cpu_idle or for interrupt return) we can
read the per-cpu enable mask before entering the critical section.
The cache-migration code is changed to use interrupt masking instead
of interrupt critical sections.  And, the interrupt-entry code is
changed so that we defer loading "tp" from per-cpu data until after
we have released the interrupt critical section.

Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>

2 files changed, 45 insertions, 36 deletions

diff --git a/arch/tile/kernel/entry.S b/arch/tile/kernel/entry.S
index ec91568df880..133c4b56a99e 100644
--- a/arch/tile/kernel/entry.S
+++ b/arch/tile/kernel/entry.S
@@ -100,8 +100,9 @@ STD_ENTRY(smp_nap)
  */
 STD_ENTRY(_cpu_idle)
         movei r1, 1
+        IRQ_ENABLE_LOAD(r2, r3)
         mtspr INTERRUPT_CRITICAL_SECTION, r1
-        IRQ_ENABLE(r2, r3)             /* unmask, but still with ICS set */
+        IRQ_ENABLE_APPLY(r2, r3)       /* unmask, but still with ICS set */
         mtspr INTERRUPT_CRITICAL_SECTION, zero
         .global _cpu_idle_nap
 _cpu_idle_nap:
diff --git a/arch/tile/kernel/intvec_64.S b/arch/tile/kernel/intvec_64.S
index 30ae76e50c44..0ae8723ea578 100644
--- a/arch/tile/kernel/intvec_64.S
+++ b/arch/tile/kernel/intvec_64.S
@@ -220,7 +220,9 @@ intvec_\vecname:
          * This routine saves just the first four registers, plus the
          * stack context so we can do proper backtracing right away,
          * and defers to handle_interrupt to save the rest.
-         * The backtracer needs pc, ex1, lr, sp, r52, and faultnum.
+         * The backtracer needs pc, ex1, lr, sp, r52, and faultnum,
+         * and needs sp set to its final location at the bottom of
+         * the stack frame.
          */
         addli   r0, r0, PTREGS_OFFSET_LR - (PTREGS_SIZE + KSTK_PTREGS_GAP)
         wh64    r0   /* cache line 7 */
@@ -450,23 +452,6 @@ intvec_\vecname:
         push_reg r5, r52
         st      r52, r4
 
-        /* Load tp with our per-cpu offset. */
-#ifdef CONFIG_SMP
-        {
-         mfspr  r20, SPR_SYSTEM_SAVE_K_0
-         moveli r21, hw2_last(__per_cpu_offset)
-        }
-        {
-         shl16insli r21, r21, hw1(__per_cpu_offset)
-         bfextu r20, r20, 0, LOG2_THREAD_SIZE-1
-        }
-        shl16insli r21, r21, hw0(__per_cpu_offset)
-        shl3add r20, r20, r21
-        ld      tp, r20
-#else
-        move    tp, zero
-#endif
-
         /*
          * If we will be returning to the kernel, we will need to
          * reset the interrupt masks to the state they had before.
@@ -489,6 +474,44 @@ intvec_\vecname:
         .endif
         st      r21, r32
 
+        /*
+         * we've captured enough state to the stack (including in
+         * particular our EX_CONTEXT state) that we can now release
+         * the interrupt critical section and replace it with our
+         * standard "interrupts disabled" mask value.  This allows
+         * synchronous interrupts (and profile interrupts) to punch
+         * through from this point onwards.
+         *
+         * It's important that no code before this point touch memory
+         * other than our own stack (to keep the invariant that this
+         * is all that gets touched under ICS), and that no code after
+         * this point reference any interrupt-specific SPR, in particular
+         * the EX_CONTEXT_K_ values.
+         */
+        .ifc \function,handle_nmi
+        IRQ_DISABLE_ALL(r20)
+        .else
+        IRQ_DISABLE(r20, r21)
+        .endif
+        mtspr   INTERRUPT_CRITICAL_SECTION, zero
+
+        /* Load tp with our per-cpu offset. */
+#ifdef CONFIG_SMP
+        {
+         mfspr  r20, SPR_SYSTEM_SAVE_K_0
+         moveli r21, hw2_last(__per_cpu_offset)
+        }
+        {
+         shl16insli r21, r21, hw1(__per_cpu_offset)
+         bfextu r20, r20, 0, LOG2_THREAD_SIZE-1
+        }
+        shl16insli r21, r21, hw0(__per_cpu_offset)
+        shl3add r20, r20, r21
+        ld      tp, r20
+#else
+        move    tp, zero
+#endif
+
 #ifdef __COLLECT_LINKER_FEEDBACK__
         /*
          * Notify the feedback routines that we were in the
@@ -513,21 +536,6 @@ intvec_\vecname:
 #endif
 
         /*
-         * we've captured enough state to the stack (including in
-         * particular our EX_CONTEXT state) that we can now release
-         * the interrupt critical section and replace it with our
-         * standard "interrupts disabled" mask value.  This allows
-         * synchronous interrupts (and profile interrupts) to punch
-         * through from this point onwards.
-         */
-        .ifc \function,handle_nmi
-        IRQ_DISABLE_ALL(r20)
-        .else
-        IRQ_DISABLE(r20, r21)
-        .endif
-        mtspr   INTERRUPT_CRITICAL_SECTION, zero
-
-        /*
          * Prepare the first 256 stack bytes to be rapidly accessible
          * without having to fetch the background data.
          */
@@ -736,9 +744,10 @@ STD_ENTRY(interrupt_return)
         beqzt   r30, .Lrestore_regs
         j       3f
 2:      TRACE_IRQS_ON
+        IRQ_ENABLE_LOAD(r20, r21)
         movei   r0, 1
         mtspr   INTERRUPT_CRITICAL_SECTION, r0
-        IRQ_ENABLE(r20, r21)
+        IRQ_ENABLE_APPLY(r20, r21)
         beqzt   r30, .Lrestore_regs
 3:
 
@@ -755,7 +764,6 @@ STD_ENTRY(interrupt_return)
          * that will save some cycles if this turns out to be a syscall.
          */
 .Lrestore_regs:
-        FEEDBACK_REENTER(interrupt_return)   /* called from elsewhere */
 
         /*
          * Rotate so we have one high bit and one low bit to test.