David Brownell <david-b@pacbell.net> Clean up some Cortex-M3 reset handling.
authoroharboe <oharboe@b42882b7-edfa-0310-969c-e2dbd0fdcd60>
Wed, 19 Aug 2009 06:30:08 +0000 (06:30 +0000)
committeroharboe <oharboe@b42882b7-edfa-0310-969c-e2dbd0fdcd60>
Wed, 19 Aug 2009 06:30:08 +0000 (06:30 +0000)
 - AIRCR_SYSRESETREQ is generic; use it on any system where
  SRST won't fly, not just on Stellaris-based ones.

 - Reformat and improve comments about the Stellaris quirk; and
  xref the only public docs (an email) about the issue.

It seems that *most* Stellaris chips have this problem.  Tempest
parts aren't yet in general sampling; and if rev B silicon for
earlier chips exists, it's not very visible yet.

git-svn-id: svn://svn.berlios.de/openocd/trunk@2595 b42882b7-edfa-0310-969c-e2dbd0fdcd60

src/target/cortex_m3.c

index 626157431f995a838b1291af900edb87d24a80e6..ea20dfd14fd47b32a1e2935226b4713ded3ec34b 100644 (file)
@@ -760,11 +760,14 @@ int cortex_m3_assert_reset(target_t *target)
                target_state_name(target));
 
        enum reset_types jtag_reset_config = jtag_get_reset_config();
+
+       /*
+        * We can reset Cortex-M3 targets using just the NVIC without
+        * requiring SRST, getting a SoC reset (or a core-only reset)
+        * instead of a system reset.
+        */
        if (!(jtag_reset_config & RESET_HAS_SRST))
-       {
-               LOG_ERROR("Can't assert SRST");
-               return ERROR_FAIL;
-       }
+               assert_srst = 0;
 
        /* Enable debug requests */
        mem_ap_read_atomic_u32(swjdp, DCB_DHCSR, &cortex_m3->dcb_dhcsr);
@@ -794,15 +797,21 @@ int cortex_m3_assert_reset(target_t *target)
                mem_ap_write_atomic_u32(swjdp, DCB_DEMCR, TRCENA | VC_HARDERR | VC_BUSERR | VC_CORERESET);
        }
 
-       /* following hack is to handle luminary reset
-        * when srst is asserted the luminary device seesm to also clear the debug registers
-        * which does not match the armv7 debug TRM */
-
+       /*
+        * When nRST is asserted on most Stellaris devices, it clears some of
+        * the debug state.  The ARMv7M and Cortex-M3 TRMs say that's wrong;
+        * and OpenOCD depends on those TRMs.  So we won't use SRST on those
+        * chips.  (Only power-on reset should affect debug state, beyond a
+        * few specified bits; not the chip's nRST input, wired to SRST.)
+        *
+        * REVISIT current errata specs don't seem to cover this issue.
+        * Do we have more details than this email?
+        *   https://lists.berlios.de/pipermail
+        *      /openocd-development/2008-August/003065.html
+        */
        if (strcmp(target->variant, "lm3s") == 0)
        {
-               /* get revision of lm3s target, only early silicon has this issue
-                * Fury Rev B, DustDevil Rev B, Tempest all ok */
-
+               /* Check for silicon revisions with the issue. */
                uint32_t did0;
 
                if (target_read_u32(target, 0x400fe000, &did0) == ERROR_OK)
@@ -816,10 +825,16 @@ int cortex_m3_assert_reset(target_t *target)
 
                                case 1:
                                case 3:
-                                       /* only Fury/DustDevil rev A suffer reset problems */
+                                       /* Fury and DustDevil rev A have
+                                        * this nRST problem.  It should
+                                        * be fixed in rev B silicon.
+                                        */
                                        if (((did0 >> 8) & 0xff) == 0)
                                                assert_srst = 0;
                                        break;
+                               case 4:
+                                       /* Tempest should be fine. */
+                                       break;
                        }
                }
        }
@@ -838,13 +853,20 @@ int cortex_m3_assert_reset(target_t *target)
        }
        else
        {
-               /* this causes the luminary device to reset using the watchdog */
-               mem_ap_write_atomic_u32(swjdp, NVIC_AIRCR, AIRCR_VECTKEY | AIRCR_SYSRESETREQ);
-               LOG_DEBUG("Using Luminary Reset: SYSRESETREQ");
+               /* Use a standard Cortex-M3 software reset mechanism.
+                * SYSRESETREQ will reset SoC peripherals outside the
+                * core, like watchdog timers, if the SoC wires it up
+                * correctly.  Else VECRESET can reset just the core.
+                */
+               mem_ap_write_atomic_u32(swjdp, NVIC_AIRCR,
+                               AIRCR_VECTKEY | AIRCR_SYSRESETREQ);
+               LOG_DEBUG("Using Cortex-M3 SYSRESETREQ");
 
                {
-                       /* I do not know why this is necessary, but it fixes strange effects
-                        * (step/resume cause a NMI after reset) on LM3S6918 -- Michael Schwingen */
+                       /* I do not know why this is necessary, but it
+                        * fixes strange effects (step/resume cause NMI
+                        * after reset) on LM3S6918 -- Michael Schwingen
+                        */
                        uint32_t tmp;
                        mem_ap_read_atomic_u32(swjdp, NVIC_AIRCR, &tmp);
                }