System/FFI/LibC: Round-robin mem_task selection for malloc

This fixes an issue where multiple HTTP requests would arrive
simultaneously, and the subsequent calls to @rsa_import would
end up clobbering the shared mem_task's CHeapCtrl, leading to #GP.

Eventually, we may want to simplify this to only init the CHeapCtrl,
rather than creating the entire CTask, but for now it seems ok.

System/FFI/LibC.HC

@@ -118,11 +118,18 @@ U0 ntohs()
     POP_SYSV_REGS
 }
 
+U64 @malloc(I64 size)
+{
+    U64 res = MAlloc(size, malloc_mem_task[malloc_current_mem_task % MALLOC_MEM_TASK_COUNT]->code_heap);
+    malloc_current_mem_task++;
+    return res;
+}
+
 U0 malloc()
 {
     PUSH_SYSV_REGS
     GET_SYSV_ARGS
-    MAlloc(p0, slon_mem_task->code_heap);
+    @malloc(p0);
     POP_SYSV_REGS
 }
 

System/Setup/Environment.HC

@@ -51,6 +51,13 @@ I64 @t(Bool _condition, I64 _true, I64 _false)
     return _false;
 }
 
+U0 @slon_mem_task_loop()
+{
+    while (1) {
+        Sleep(1);
+    };
+}
+
 // Before doing anything else, we:
 
 // 1. Mark memory in code heap below 0x1000000 as used.
@@ -62,6 +69,26 @@ sys_code_bp->mem_free_lst = ShrinkMemBlkByPags(sys_code_bp->mem_free_lst, 131072
 // 3. Enable SSE
 @sse_enable;
 
+// 4. Init mem_tasks
+
+CTask* slon_mem_task = Spawn(&@slon_mem_task_loop, , "SlonMemTask");
+
+#define MALLOC_MEM_TASK_COUNT 4
+CTask** malloc_mem_task = CAlloc(sizeof(CTask*) * MALLOC_MEM_TASK_COUNT, slon_mem_task);
+I64 malloc_current_mem_task = 0;
+
+U0 @malloc_mem_tasks_init()
+{
+    U8* scratch_buffer[64];
+    I64 i;
+    for (i = 0; i < MALLOC_MEM_TASK_COUNT; i++) {
+        StrPrint(scratch_buffer, "SlonMallocTask%d", i);
+        malloc_mem_task[i] = Spawn(&@slon_mem_task_loop, , scratch_buffer);
+    }
+}
+
+@malloc_mem_tasks_init;
+
 U0 dd() { DocDump(adam_task->put_doc); }
 //@patch_jmp_rel32(&Dbg2, &Reboot); // Reboot instead of crashing to the debugger
 U0 NoBeep(I8, Bool) {};