Blackfin: make EVT3->EVT5 lowering more robust wrt IPEND[4]

We handle many exceptions at EVT5 (hardware error level) so that we can
catch exceptions in our exception handling code.  Today - if the global
interrupt enable bit (IPEND[4]) is set (interrupts disabled) our trap
handling code goes into a infinite loop, since we need interrupts to be
on to defer things to EVT5.

Normal kernel code should not trigger this for any reason as IPEND[4] gets
cleared early (when doing an interrupt context save) and the kernel stack
there should be sane (or something much worse is happening in the system).
But there have been a few times where this has happened, so this change
makes sure we dump a proper crash message even when things have gone south.

Signed-off-by: Robin Getz <robin.getz@analog.com>
Signed-off-by: Mike Frysinger <vapier@gentoo.org>

4 files changed, 66 insertions, 41 deletions

diff --git a/arch/blackfin/include/asm/pda.h b/arch/blackfin/include/asm/pda.h
index b42555c1431c..69b96b40c188 100644
--- a/arch/blackfin/include/asm/pda.h
+++ b/arch/blackfin/include/asm/pda.h
@@ -50,6 +50,7 @@ struct blackfin_pda {			/* Per-processor Data Area */
         unsigned long ex_optr;
         unsigned long ex_buf[4];
         unsigned long ex_imask;         /* Saved imask from exception */
+        unsigned long ex_ipend;         /* Saved IPEND from exception */
         unsigned long *ex_stack;        /* Exception stack space */
 
 #ifdef ANOMALY_05000261
diff --git a/arch/blackfin/kernel/asm-offsets.c b/arch/blackfin/kernel/asm-offsets.c
index b5df9459d6d5..8ad4f2c69961 100644
--- a/arch/blackfin/kernel/asm-offsets.c
+++ b/arch/blackfin/kernel/asm-offsets.c
@@ -145,6 +145,7 @@ int main(void)
         DEFINE(PDA_EXBUF, offsetof(struct blackfin_pda, ex_buf));
         DEFINE(PDA_EXIMASK, offsetof(struct blackfin_pda, ex_imask));
         DEFINE(PDA_EXSTACK, offsetof(struct blackfin_pda, ex_stack));
+        DEFINE(PDA_EXIPEND, offsetof(struct blackfin_pda, ex_ipend));
 #ifdef ANOMALY_05000261
         DEFINE(PDA_LFRETX, offsetof(struct blackfin_pda, last_cplb_fault_retx));
 #endif
diff --git a/arch/blackfin/kernel/traps.c b/arch/blackfin/kernel/traps.c
index bf2b2d1f8ae5..fccf741ed3b5 100644
--- a/arch/blackfin/kernel/traps.c
+++ b/arch/blackfin/kernel/traps.c
@@ -267,11 +267,6 @@ asmlinkage void trap_c(struct pt_regs *fp)
          * double faults if the stack has become corrupt
          */
 
-#ifndef CONFIG_KGDB
-        /* IPEND is skipped if KGDB isn't enabled (see entry code) */
-        fp->ipend = bfin_read_IPEND();
-#endif
-
         /* trap_c() will be called for exceptions. During exceptions
          * processing, the pc value should be set with retx value.
          * With this change we can cleanup some code in signal.c- TODO
@@ -1116,10 +1111,16 @@ void show_regs(struct pt_regs *fp)
 
         verbose_printk(KERN_NOTICE "%s", linux_banner);
 
-        verbose_printk(KERN_NOTICE "\nSEQUENCER STATUS:\t\t%s\n",
-                       print_tainted());
-        verbose_printk(KERN_NOTICE " SEQSTAT: %08lx  IPEND: %04lx  SYSCFG: %04lx\n",
-                       (long)fp->seqstat, fp->ipend, fp->syscfg);
+        verbose_printk(KERN_NOTICE "\nSEQUENCER STATUS:\t\t%s\n", print_tainted());
+        verbose_printk(KERN_NOTICE " SEQSTAT: %08lx  IPEND: %04lx  IMASK: %04lx  SYSCFG: %04lx\n",
+                (long)fp->seqstat, fp->ipend, cpu_pda[smp_processor_id()].ex_imask, fp->syscfg);
+        if (fp->ipend & EVT_IRPTEN)
+                verbose_printk(KERN_NOTICE "  Global Interrupts Disabled (IPEND[4])\n");
+        if (!(cpu_pda[smp_processor_id()].ex_imask & (EVT_IVG13 | EVT_IVG12 | EVT_IVG11 |
+                        EVT_IVG10 | EVT_IVG9 | EVT_IVG8 | EVT_IVG7 | EVT_IVTMR)))
+                verbose_printk(KERN_NOTICE "  Peripheral interrupts masked off\n");
+        if (!(cpu_pda[smp_processor_id()].ex_imask & (EVT_IVG15 | EVT_IVG14)))
+                verbose_printk(KERN_NOTICE "  Kernel interrupts masked off\n");
         if ((fp->seqstat & SEQSTAT_EXCAUSE) == VEC_HWERR) {
                 verbose_printk(KERN_NOTICE "  HWERRCAUSE: 0x%lx\n",
                         (fp->seqstat & SEQSTAT_HWERRCAUSE) >> 14);
diff --git a/arch/blackfin/mach-common/entry.S b/arch/blackfin/mach-common/entry.S
index fb1795d5be2a..4c07fcb356a2 100644
--- a/arch/blackfin/mach-common/entry.S
+++ b/arch/blackfin/mach-common/entry.S
@@ -301,25 +301,31 @@ ENTRY(_ex_replaceable)
         nop;
 
 ENTRY(_ex_trap_c)
+        /* The only thing that has been saved in this context is
+         * (R7:6,P5:4), ASTAT & SP - don't use anything else
+         */
+
+        GET_PDA(p5, r6);
+
         /* Make sure we are not in a double fault */
         p4.l = lo(IPEND);
         p4.h = hi(IPEND);
         r7 = [p4];
         CC = BITTST (r7, 5);
         if CC jump _double_fault;
+        [p5 + PDA_EXIPEND] = r7;
 
         /* Call C code (trap_c) to handle the exception, which most
          * likely involves sending a signal to the current process.
          * To avoid double faults, lower our priority to IRQ5 first.
          */
-        P5.h = _exception_to_level5;
-        P5.l = _exception_to_level5;
+        r7.h = _exception_to_level5;
+        r7.l = _exception_to_level5;
         p4.l = lo(EVT5);
         p4.h = hi(EVT5);
-        [p4] = p5;
+        [p4] = r7;
         csync;
 
-        GET_PDA(p5, r6);
 #ifndef CONFIG_DEBUG_DOUBLEFAULT
 
         /*
@@ -349,8 +355,7 @@ ENTRY(_ex_trap_c)
         BITCLR(r6, SYSCFG_SSSTEP_P);
         SYSCFG = r6;
 
-        /* Disable all interrupts, but make sure level 5 is enabled so
-         * we can switch to that level.  Save the old mask.  */
+        /* Save the current IMASK, since we change in order to jump to level 5 */
         cli r6;
         [p5 + PDA_EXIMASK] = r6;
 
@@ -358,9 +363,21 @@ ENTRY(_ex_trap_c)
         p4.h = hi(SAFE_USER_INSTRUCTION);
         retx = p4;
 
+        /* Disable all interrupts, but make sure level 5 is enabled so
+         * we can switch to that level.
+         */
         r6 = 0x3f;
         sti r6;
 
+        /* In case interrupts are disabled IPEND[4] (global interrupt disable bit)
+         * clear it (re-enabling interrupts again) by the special sequence of pushing
+         * RETI onto the stack.  This way we can lower ourselves to IVG5 even if the
+         * exception was taken after the interrupt handler was called but before it
+         * got a chance to enable global interrupts itself.
+         */
+        [--sp] = reti;
+        sp += 4;
+
         raise 5;
         jump.s _bfin_return_from_exception;
 ENDPROC(_ex_trap_c)
@@ -420,47 +437,52 @@ ENDPROC(_double_fault)
 ENTRY(_exception_to_level5)
         SAVE_ALL_SYS
 
-        GET_PDA(p4, r7);        /* Fetch current PDA */
-        r6 = [p4 + PDA_RETX];
+        GET_PDA(p5, r7);        /* Fetch current PDA */
+        r6 = [p5 + PDA_RETX];
         [sp + PT_PC] = r6;
 
-        r6 = [p4 + PDA_SYSCFG];
+        r6 = [p5 + PDA_SYSCFG];
         [sp + PT_SYSCFG] = r6;
 
-        /* Restore interrupt mask.  We haven't pushed RETI, so this
-         * doesn't enable interrupts until we return from this handler.  */
-        r6 = [p4 + PDA_EXIMASK];
-        sti r6;
-
         /* Restore the hardware error vector.  */
-        P5.h = _evt_ivhw;
-        P5.l = _evt_ivhw;
+        r7.h = _evt_ivhw;
+        r7.l = _evt_ivhw;
         p4.l = lo(EVT5);
         p4.h = hi(EVT5);
-        [p4] = p5;
+        [p4] = r7;
         csync;
 
-        p2.l = lo(IPEND);
-        p2.h = hi(IPEND);
-        csync;
-        r0 = [p2];              /* Read current IPEND */
-        [sp + PT_IPEND] = r0;   /* Store IPEND */
+#ifdef CONFIG_DEBUG_DOUBLEFAULT
+        /* Now that we have the hardware error vector programmed properly
+         * we can re-enable interrupts (IPEND[4]), so if the _trap_c causes
+         * another hardware error, we can catch it (self-nesting).
+         */
+        [--sp] = reti;
+        sp += 4;
+#endif
+
+        r7 = [p5 + PDA_EXIPEND] /* Read the IPEND from the Exception state */
+        [sp + PT_IPEND] = r7;   /* Store IPEND onto the stack */
 
         r0 = sp;        /* stack frame pt_regs pointer argument ==> r0 */
         SP += -12;
         call _trap_c;
         SP += 12;
 
-#ifdef CONFIG_DEBUG_DOUBLEFAULT
-        /* Grab ILAT */
-        p2.l = lo(ILAT);
-        p2.h = hi(ILAT);
-        r0 = [p2];
-        r1 = 0x20;  /* Did I just cause anther HW error? */
-        r0 = r0 & r1;
-        CC = R0 == R1;
-        if CC JUMP _double_fault;
-#endif
+        /* If interrupts were off during the exception (IPEND[4] = 1), turn them off
+         * before we return.
+         */
+        CC = BITTST(r7, EVT_IRPTEN_P)
+        if !CC jump 1f;
+        /* this will load a random value into the reti register - but that is OK,
+         * since we do restore it to the correct value in the 'RESTORE_ALL_SYS' macro
+         */
+        sp += -4;
+        reti = [sp++];
+1:
+        /* restore the interrupt mask (IMASK) */
+        r6 = [p5 + PDA_EXIMASK];
+        sti r6;
 
         call _ret_from_exception;
         RESTORE_ALL_SYS