Merge #19341 #19345

19341: Picolibc updates r=benpicco a=keith-packard

### Contribution description

Minor updates to picolibc support for newer versions of picolibc, including 1.8.


### Testing procedure

Here's the current build result for a board I happen to have on my bench right now:

```
$ make -C examples/blinky BOARD=nucleo-f103rb FEATURES_REQUIRED=picolibc
...
/home/keithp/src/RIOT/sys/picolibc_syscalls_default/syscalls.c:319:1: error: unknown type name '_READ_WRITE_RETURN_TYPE'
  319 | _READ_WRITE_RETURN_TYPE read(int fd, void *dest, size_t count)
      | ^~~~~~~~~~~~~~~~~~~~~~~
/home/keithp/src/RIOT/sys/picolibc_syscalls_default/syscalls.c:351:1: error: unknown type name '_READ_WRITE_RETURN_TYPE'
  351 | _READ_WRITE_RETURN_TYPE write(int fd, const void *src, size_t count)
      | ^~~~~~~~~~~~~~~~~~~~~~~
...
```

### Description of changes

 1. Fix the read/write return values (Picolibc 1.8 uses POSIX standard types now)
 2. Add new symbols to linker scripts (Picolibc needs help dealing with TLS alignment)
 3. Align stack and thread local storage block during thread setup.


19345: cpu/riscv_common: convert to uword_t usage r=benpicco a=dylad

### Contribution description

This PR makes use of `uword_t` type in a few places in `cpu/riscv_common`.
This should not have any effect on produced binaries.

### Testing procedure

CI should be enough. 

### Issues/PRs references
Quickly adapt from #16994. It will help for 64 bits support one day ;)



Co-authored-by: Keith Packard <keithp@keithp.com>
Co-authored-by: Dylan Laduranty <dylan.laduranty@mesotic.com>

core/thread.c

@@ -226,9 +226,19 @@ kernel_pid_t thread_create(char *stack, int stacksize, uint8_t priority,
     thread_t *thread = (thread_t *)(uintptr_t)(stack + stacksize);
 
 #ifdef PICOLIBC_TLS
-    stacksize -= _tls_size();
+#if __PICOLIBC_MAJOR__ > 1 || __PICOLIBC_MINOR__ >= 8
+#define TLS_ALIGN       (alignof(thread_t) > _tls_align() ? alignof(thread_t) : _tls_align())
+#else
+#define TLS_ALIGN       alignof(thread_t)
+#endif
+    char *tls = stack + stacksize - _tls_size();
+    /*
+     * Make sure the TLS area is aligned as required and that the
+     * resulting stack will also be aligned as required
+     */
+    thread->tls = (void *) ((uintptr_t) tls & ~ (TLS_ALIGN - 1));
+    stacksize = (char *) thread->tls - stack;
 
-    thread->tls = stack + stacksize;
     _init_tls(thread->tls);
 #endif
 

cpu/cortexm_common/ldscripts/cortexm_base.ld

@@ -132,6 +132,10 @@ SECTIONS
     } > rom
     __tls_size = __tbss_end - __tdata_start;
     __tbss_size = __tls_size - __tdata_size;
+    PROVIDE( __tbss_offset = ADDR(.tbss) - ADDR(.tdata) );
+    PROVIDE( __tls_align = MAX(ALIGNOF(.tdata), ALIGNOF(.tbss)) );
+    PROVIDE( __arm32_tls_tcb_offset = MAX(8, __tls_align) );
+    PROVIDE( __arm64_tls_tcb_offset = MAX(16, __tls_align) );
 
     /* .ARM.exidx is sorted, so has to go in its own output section.  */
     PROVIDE_HIDDEN (__exidx_start = .);

cpu/lpc23xx/ldscripts/lpc23xx.ld

@@ -106,6 +106,9 @@ SECTIONS
     } > rom
     __tls_size = __tbss_end - __tdata_start;
     __tbss_size = __tls_size - __tdata_size;
+    PROVIDE( __tls_align = MAX(ALIGNOF(.tdata), ALIGNOF(.tbss)) );
+    PROVIDE( __arm32_tls_tcb_offset = MAX(8, __tls_align) );
+    PROVIDE( __arm64_tls_tcb_offset = MAX(16, __tls_align) );
 
     /* .ARM.exidx is sorted, so has to go in its own output section.  */
     PROVIDE_HIDDEN (__exidx_start = .);

cpu/riscv_common/irq_arch.c

@@ -30,6 +30,7 @@
 #include "sched.h"
 #include "plic.h"
 #include "clic.h"
+#include "architecture.h"
 
 #include "vendor/riscv_csr.h"
 
@@ -83,13 +84,13 @@ void riscv_irq_init(void)
  * @brief Global trap and interrupt handler
  */
 __attribute((used))
-static void handle_trap(uint32_t mcause)
+static void handle_trap(uword_t mcause)
 {
     /*  Tell RIOT to set sched_context_switch_request instead of
      *  calling thread_yield(). */
     riscv_in_isr = 1;
 
-    uint32_t trap = mcause & CPU_CSR_MCAUSE_CAUSE_MSK;
+    uword_t trap = mcause & CPU_CSR_MCAUSE_CAUSE_MSK;
 
     /* Check for INT or TRAP */
     if ((mcause & MCAUSE_INT) == MCAUSE_INT) {
@@ -129,16 +130,16 @@ static void handle_trap(uint32_t mcause)
             sched_context_switch_request = 1;
             /* Increment the return program counter past the ecall
              * instruction */
-            uint32_t return_pc = read_csr(mepc);
+            uword_t return_pc = read_csr(mepc);
             write_csr(mepc, return_pc + 4);
             break;
         }
         default:
 #ifdef DEVELHELP
             printf("Unhandled trap:\n");
             printf("  mcause: 0x%" PRIx32 "\n", trap);
-            printf("  mepc:   0x%lx\n", read_csr(mepc));
-            printf("  mtval:  0x%lx\n", read_csr(mtval));
+            printf("  mepc:   0x%" PRIx32 "\n", read_csr(mepc));
+            printf("  mtval:  0x%" PRIx32 "\n", read_csr(mtval));
 #endif
             /* Unknown trap */
             core_panic(PANIC_GENERAL_ERROR, "Unhandled trap");

cpu/riscv_common/ldscripts/riscv_base.ld

@@ -159,6 +159,8 @@ SECTIONS
   } >flash : tls
   __tls_size = __tbss_end - __tdata_start;
   __tbss_size = __tls_size - __tdata_size;
+  PROVIDE( __tbss_offset = ADDR(.tbss) - ADDR(.tdata) );
+  PROVIDE( __tls_align = MAX(ALIGNOF(.tdata), ALIGNOF(.tbss)) );
 
   .lalign         :
   {

cpu/riscv_common/periph/plic.c

@@ -28,6 +28,7 @@
 #include "assert.h"
 #include "cpu.h"
 #include "plic.h"
+#include "architecture.h"
 
 /* Local macros to calculate register offsets */
 #ifndef _REG32
@@ -42,7 +43,7 @@ static plic_isr_cb_t _ext_isrs[PLIC_NUM_INTERRUPTS];
 
 static inline volatile uint32_t *_get_claim_complete_addr(void)
 {
-    uint32_t hart_id = read_csr(mhartid);
+    uword_t hart_id = read_csr(mhartid);
 
     /* Construct the claim address */
     return &PLIC_REG(PLIC_CLAIM_OFFSET +
@@ -51,7 +52,7 @@ static inline volatile uint32_t *_get_claim_complete_addr(void)
 
 static inline volatile uint32_t *_get_threshold_addr(void)
 {
-    uint32_t hart_id = read_csr(mhartid);
+    uword_t hart_id = read_csr(mhartid);
 
     /* Construct the claim address */
     return &PLIC_REG(PLIC_THRESHOLD_OFFSET +
@@ -60,7 +61,7 @@ static inline volatile uint32_t *_get_threshold_addr(void)
 
 static inline volatile uint32_t *_get_irq_reg(unsigned irq)
 {
-    uint32_t hart_id = read_csr(mhartid);
+    uword_t hart_id = read_csr(mhartid);
 
     return &PLIC_REG(PLIC_ENABLE_OFFSET +
                      (hart_id << PLIC_ENABLE_SHIFT_PER_TARGET)) +

cpu/riscv_common/thread_arch.c

@@ -25,6 +25,7 @@
 #include "thread.h"
 #include "sched.h"
 #include "context_frame.h"
+#include "architecture.h"
 
 /**
  * @brief   Noticeable marker marking the beginning of a stack segment
@@ -101,11 +102,11 @@ char *thread_stack_init(thread_task_func_t task_func,
     memset(sf, 0, sizeof(*sf));
 
     /* set initial reg values */
-    sf->pc = (uint32_t)task_func;
-    sf->a0 = (uint32_t)arg;
+    sf->pc = (uword_t)task_func;
+    sf->a0 = (uword_t)arg;
 
     /* if the thread exits go to sched_task_exit() */
-    sf->ra = (uint32_t)sched_task_exit;
+    sf->ra = (uword_t)sched_task_exit;
 
     return (char *)stk_top;
 }

sys/picolibc_syscalls_default/syscalls.c

@@ -118,7 +118,8 @@ _exit(int n)
 {
     LOG_INFO("#! exit %i: powering off\n", n);
     pm_off();
-    while(1);
+    for (;;) {
+    }
 }
 
 /**
@@ -304,6 +305,13 @@ int open(const char *name, int flags, int mode)
 #endif
 }
 
+/*
+ * Picolibc newer than 1.8 uses standard posix types for read/write
+ * return values
+ */
+#if __PICOLIBC_MAJOR__ > 1 || __PICOLIBC_MINOR__ >= 8
+#define _READ_WRITE_RETURN_TYPE ssize_t
+#endif
 /**
  * @brief Read bytes from an open file
  *