Fixed bug 2971878 .

git-svn-id: svn://svn.code.sf.net/p/chibios/svn/trunk@1747 35acf78f-673a-0410-8e92-d51de3d6d3f4

3 files changed, 34 insertions, 20 deletions

diff --git a/docs/reports/SPC563M64-80.txt b/docs/reports/SPC563M64-80.txt
index a11669746..35bbe17a6 100644
--- a/docs/reports/SPC563M64-80.txt
+++ b/docs/reports/SPC563M64-80.txt
@@ -92,7 +92,7 @@ Settings: SYSCLK=80, optimal wait states, prefetching enabled
 --- Result: SUCCESS

 ----------------------------------------------------------------------------

 --- Test Case 11.1 (Benchmark, messages #1)

---- Score : 280180 msgs/S, 560360 ctxswc/S

+--- Score : 280181 msgs/S, 560362 ctxswc/S

 --- Result: SUCCESS

 ----------------------------------------------------------------------------

 --- Test Case 11.2 (Benchmark, messages #2)

@@ -128,7 +128,7 @@ Settings: SYSCLK=80, optimal wait states, prefetching enabled
 --- Result: SUCCESS

 ----------------------------------------------------------------------------

 --- Test Case 11.10 (Benchmark, virtual timers set/reset)

---- Score : 1093664 timers/S

+--- Score : 1093672 timers/S

 --- Result: SUCCESS

 ----------------------------------------------------------------------------

 --- Test Case 11.11 (Benchmark, semaphores wait/signal)

diff --git a/os/kernel/src/chregistry.c b/os/kernel/src/chregistry.c
index 4dca52534..5ed8ace2e 100644
--- a/os/kernel/src/chregistry.c
+++ b/os/kernel/src/chregistry.c
@@ -23,6 +23,16 @@
  *

  * @addtogroup registry

  * @details Threads Registry related APIs and services.<br>

+ *          The threads Threads Registry is a double linked list that holds

+ *          all the active threads in the system.<br>

+ *          The registry is meant to be mainly a debug feature, as example

+ *          through the registry a debugger can enumerate the active threads

+ *          in any given moment or the shell can print the active threads and

+ *          their state.<br>

+ *          Another possible use is for centralized threads memory management,

+ *          terminating threads can pulse an event source and an event handler

+ *          can perform a scansion of the registry in order to recover the

+ *          memory.<br>

  *          In order to use the threads registry the @p CH_USE_REGISTRY option

  *          must be enabled in @p chconf.h.

  * @{

@@ -64,25 +74,24 @@ Thread *chRegFirstThread(void) {
  * @retval NULL         if there is no next thread.

  */

 Thread *chRegNextThread(Thread *tp) {

+  Thread *ntp;

 

   chSysLock();

+  ntp = tp->p_newer;

+  if (ntp == (Thread *)&rlist)

+    ntp = NULL;

 #if CH_USE_DYNAMIC

-  chDbgAssert(tp->p_refs > 0, "chRegNextThread(), #1",

-              "not referenced");

-  tp->p_refs--;

-#endif

-  if (tp->p_newer != (Thread *)&rlist) {

-    tp = tp->p_newer;

-#if CH_USE_DYNAMIC

-    chDbgAssert(tp->p_refs < 255, "chRegNextThread(), #2",

+  else {

+    chDbgAssert(ntp->p_refs < 255, "chRegNextThread(), #1",

                 "too many references");

-    tp->p_refs++;

-#endif

+    ntp->p_refs++;

   }

-  else

-    tp = NULL;

+#endif

   chSysUnlock();

-  return tp;

+#if CH_USE_DYNAMIC

+  chThdRelease(tp);

+#endif

+  return ntp;

 }

 

 #endif /* CH_USE_REGISTRY */

diff --git a/readme.txt b/readme.txt
index 8d5f68b48..bb47971e9 100644
--- a/readme.txt
+++ b/readme.txt
@@ -57,7 +57,14 @@
 *****************************************************************************

 

 *** 1.5.4 ***

+- FIX: Fixed missing memory recovery on reference release in chRegNextThread()

+  (bug 2971878).

 - FIX: Fixed wrong thread state macro in STM32/spi_lld.c (bug 2968142).

+- NEW: The port layer now can "capture" the implementation of individual

+  scheduler API functions in order to provide architecture-optimized

+  versions. This is done because further scheduler optimizations are

+  becoming increasingly pointless without considering architecture and

+  compiler related constraints.

 - NEW: Added support for the STM8 large memory model to the STM8 port. Now

   the assembler port code is totally inlined and the chcoreasm.asm file has

   been removed.

@@ -65,11 +72,6 @@
   subdirectory, this should make things easier for RIDE7 users. The normal

   makefile is still available of course.

 - NEW: New article in the documentation. Fixed an orphaned page (STM8 port).

-- NEW: The port layer now can "capture" the implementation of individual

-  scheduler API functions in order to provide architecture-optimized

-  versions. This is done because further scheduler optimizations are

-  becoming increasingly pointless without considering architecture and

-  compiler related constraints.

 - NEW: Documentation improvements, now the description goes on top of each

   page, doxygen defaulted it in the middle, not exactly the best for

   readability. Improved many descriptions of the various subsystems.

@@ -80,6 +82,9 @@
   The previous implementation was probably overkill and took too much space.

 - CHANGE: Exiting from a chCondWaitTimeout() because a timeout now does not

   re-acquire the mutex, ownership is lost.

+- CHANGE: The module documentation has been moved from the kernel.dox file

+  to the various source code files in order to make it easier to maintain

+  and double as source comments.

 

 *** 1.5.3 ***

 - FIX: Removed C99-style variables declarations (bug 2964418)(backported