picolibc: Enable TLS support [v4]

Allocate and initialize a thread-local block for each thread at the
top of the stack.

Set the tls base when switching to a new thread.

Add tdata/tbss linker instructions to cortex_m and risc-v scripts.

Signed-off-by: Keith Packard <keithp@keithp.com>

---

v2:
	Squash fixes

v3:
	Replace tabs with spaces

v4:
	Add tbss to fe310 linker script

core/include/thread.h

@@ -176,6 +176,10 @@ struct _thread {
     const char *name;               /**< thread's name                  */
     int stack_size;                 /**< thread's stack size            */
 #endif
+/* enable TLS only when Picolibc is compiled with TLS enabled */
+#ifdef PICOLIBC_TLS
+    void *tls;                      /**< thread local storage ptr */
+#endif
 #ifdef HAVE_THREAD_ARCH_T
     thread_arch_t arch;             /**< architecture dependent part    */
 #endif

core/sched.c

@@ -40,6 +40,10 @@
 #include <inttypes.h>
 #endif
 
+#ifdef PICOLIBC_TLS
+#include <picotls.h>
+#endif
+
 /* Needed by OpenOCD to read sched_threads */
 #if defined(__APPLE__) && defined(__MACH__)
  #define FORCE_USED_SECTION __attribute__((used)) __attribute__((section( \
@@ -148,6 +152,9 @@ int __attribute__((used)) sched_run(void)
     next_thread->status = STATUS_RUNNING;
     sched_active_pid = next_thread->pid;
     sched_active_thread = next_thread;
+#ifdef PICOLIBC_TLS
+    _set_tls(next_thread->tls);
+#endif
 
 #ifdef MODULE_MPU_STACK_GUARD
     mpu_configure(

core/thread.c

@@ -20,6 +20,9 @@
 
 #include <errno.h>
 #include <stdio.h>
+#ifdef PICOLIBC_TLS
+#include <picotls.h>
+#endif
 
 #include "assert.h"
 #include "thread.h"
@@ -215,6 +218,13 @@ kernel_pid_t thread_create(char *stack, int stacksize, uint8_t priority,
     /* allocate our thread control block at the top of our stackspace */
     thread_t *thread = (thread_t *)(stack + stacksize);
 
+#ifdef PICOLIBC_TLS
+    stacksize -= _tls_size();
+
+    thread->tls = stack + stacksize;
+    _init_tls(thread->tls);
+#endif
+
 #if defined(DEVELHELP) || defined(SCHED_TEST_STACK)
     if (flags & THREAD_CREATE_STACKTEST) {
         /* assign each int of the stack the value of it's address */

cpu/cortexm_common/ldscripts/cortexm_base.ld

@@ -96,6 +96,42 @@ SECTIONS
         _efixed = .;            /* End of text section */
     } > rom
 
+    /*
+     * TLS relocations are offsets relative to the address
+     * of the first TLS symbol. That means we just need to
+     * allocate them all together so that the TLS region
+     * is compact when allocated for each thread.
+     */
+
+    /*
+     * TLS initialization data is loaded into ROM so that
+     * each thread can get its values initialized from there
+     * at startup
+     */
+    .tdata :
+    {
+         __tdata_start = .;
+        *(.tdata .tdata.* .gnu.linkonce.td.*)
+        __tdata_end = .;
+    } > rom
+    __tdata_source = LOADADDR(.tdata);
+    __tdata_size = SIZEOF(.tdata);
+
+    /*
+     * TLS zeroed data is relocated as if it immediately followed
+     * the tdata values. However, the linker 'magically' erases the
+     * memory allocation so that no ROM is consumed by this
+     * section
+     */
+    .tbss :
+    {
+        *(.tbss .tbss.* .gnu.linkonce.tb.*)
+        *(.tcommon)
+        __tbss_end = .;
+    } > rom
+    __tls_size = __tbss_end - __tdata_start;
+    __tbss_size = __tls_size - __tdata_size;
+
     /* .ARM.exidx is sorted, so has to go in its own output section.  */
     PROVIDE_HIDDEN (__exidx_start = .);
     .ARM.exidx :

cpu/fe310/ldscripts/fe310_base.ld

@@ -27,6 +27,7 @@ PHDRS
   flash PT_LOAD;
   ram_init PT_LOAD;
   ram PT_NULL;
+  tls PT_TLS;
 }
 
 SECTIONS
@@ -118,6 +119,42 @@ SECTIONS
     KEEP (*(.dtors))
   } >flash AT>flash :flash
 
+  /*
+   * TLS relocations are offsets relative to the address
+   * of the first TLS symbol. That means we just need to
+   * allocate them all together so that the TLS region
+   * is compact when allocated for each thread.
+   */
+
+  /*
+   * TLS initialization data is loaded into ROM so that
+   * each thread can get its values initialized from there
+   * at startup
+   */
+  .tdata :
+  {
+       __tdata_start = .;
+      *(.tdata .tdata.* .gnu.linkonce.td.*)
+      __tdata_end = .;
+  } >flash AT>flash :tls
+  __tdata_source = LOADADDR(.tdata);
+  __tdata_size = SIZEOF(.tdata);
+
+  /*
+   * TLS zeroed data is relocated as if it immediately followed
+   * the tdata values. However, the linker 'magically' erases the
+   * memory allocation so that no ROM is consumed by this
+   * section
+   */
+  .tbss :
+  {
+      *(.tbss .tbss.* .gnu.linkonce.tb.*)
+      *(.tcommon)
+      __tbss_end = .;
+  } >flash : tls
+  __tls_size = __tbss_end - __tdata_start;
+  __tbss_size = __tls_size - __tdata_size;
+
   .lalign         :
   {
     . = ALIGN(4);