Changeset 0366d09d in mainline for kernel/generic

Timestamp:

2023-02-07T16:49:43Z (3 years ago)

Author:

Jiří Zárevúcky <zarevucky.jiri@…>

Branches:

master, ticket/834-toolchain-update, topic/msim-upgrade, topic/simplify-dev-export

Children:

78acbc72

Parents:

1eaead4

git-author:

Jiří Zárevúcky <zarevucky.jiri@…> (2022-08-14 14:08:42)

git-committer:

Jiří Zárevúcky <zarevucky.jiri@…> (2023-02-07 16:49:43)

Message:

Avoid separate allocation for FPU context

With _Alignas, we can encode alignment in the type itself and
include it directly as a member of thread_t, thus removing
the need for a separate allocation.

The alignment requirement is never more than 16 bytes,
so this adds only minimal extra padding to the structure.

Location:

kernel/generic

Files:

• include/proc/thread.h (modified) (1 diff)
• src/proc/scheduler.c (modified) (4 diffs)
• src/proc/thread.c (modified) (5 diffs)

kernel/generic/include/proc/thread.h

@@ -151 +151 @@
         link_t joiner_link;
 
-        fpu_context_t *saved_fpu_context;
+#ifdef CONFIG_FPU
+        fpu_context_t fpu_context;
+#endif
         bool fpu_context_exists;
 

kernel/generic/src/proc/scheduler.c

@@ -97 +97 @@
         fpu_enable();
         if (THREAD->fpu_context_exists)
-                fpu_context_restore(THREAD->saved_fpu_context);
+                fpu_context_restore(&THREAD->fpu_context);
         else {
                 fpu_init();
@@ -140 +140 @@
         if (CPU->fpu_owner != NULL) {
                 irq_spinlock_lock(&CPU->fpu_owner->lock, false);
-                fpu_context_save(CPU->fpu_owner->saved_fpu_context);
+                fpu_context_save(&CPU->fpu_owner->fpu_context);
 
                 /* Don't prevent migration */
@@ -150 +150 @@
         irq_spinlock_lock(&THREAD->lock, false);
         if (THREAD->fpu_context_exists) {
-                fpu_context_restore(THREAD->saved_fpu_context);
+                fpu_context_restore(&THREAD->fpu_context);
         } else {
                 fpu_init();
@@ -325 +325 @@
 
 #if (defined CONFIG_FPU) && (!defined CONFIG_FPU_LAZY)
-                fpu_context_save(THREAD->saved_fpu_context);
+                fpu_context_save(&THREAD->fpu_context);
 #endif
                 if (!context_save(&THREAD->saved_context)) {

kernel/generic/src/proc/thread.c

@@ -102 +102 @@
 static slab_cache_t *thread_cache;
 
-#ifdef CONFIG_FPU
-slab_cache_t *fpu_context_cache;
-#endif
-
 static void *threads_getkey(odlink_t *);
 static int threads_cmp(void *, void *);
@@ -164 +160 @@
         /* call the architecture-specific part of the constructor */
         thr_constructor_arch(thread);
-
-#ifdef CONFIG_FPU
-        thread->saved_fpu_context = slab_alloc(fpu_context_cache,
-            FRAME_ATOMIC | kmflags);
-        if (!thread->saved_fpu_context)
-                return ENOMEM;
-#endif /* CONFIG_FPU */
 
         /*
@@ -198 +187 @@
         uintptr_t stack_phys =
             frame_alloc(STACK_FRAMES, kmflags, STACK_SIZE - 1);
-        if (!stack_phys) {
-#ifdef CONFIG_FPU
-                assert(thread->saved_fpu_context);
-                slab_free(fpu_context_cache, thread->saved_fpu_context);
-#endif
+        if (!stack_phys)
                 return ENOMEM;
-        }
 
         thread->kstack = (uint8_t *) PA2KA(stack_phys);
@@ -225 +209 @@
         frame_free(KA2PA(thread->kstack), STACK_FRAMES);
 
-#ifdef CONFIG_FPU
-        assert(thread->saved_fpu_context);
-        slab_free(fpu_context_cache, thread->saved_fpu_context);
-#endif
-
         return STACK_FRAMES;  /* number of frames freed */
 }
@@ -243 +222 @@
 
         atomic_store(&nrdy, 0);
-        thread_cache = slab_cache_create("thread_t", sizeof(thread_t), 0,
+        thread_cache = slab_cache_create("thread_t", sizeof(thread_t), _Alignof(thread_t),
             thr_constructor, thr_destructor, 0);
-
-#ifdef CONFIG_FPU
-        fpu_context_cache = slab_cache_create("fpu_context_t",
-            sizeof(fpu_context_t), FPU_CONTEXT_ALIGN, NULL, NULL, 0);
-#endif
 
         odict_initialize(&threads, threads_getkey, threads_cmp);