URL
https://opencores.org/ocsvn/openrisc/openrisc/trunk
Subversion Repositories openrisc
[/] [openrisc/] [trunk/] [gnu-stable/] [gcc-4.5.1/] [gcc/] [ada/] [s-asthan-vms-alpha.adb] - Rev 859
Go to most recent revision | Compare with Previous | Blame | View Log
------------------------------------------------------------------------------ -- -- -- GNAT RUN-TIME COMPONENTS -- -- -- -- S Y S T E M . A S T _ H A N D L I N G -- -- -- -- B o d y -- -- -- -- Copyright (C) 1996-2009, Free Software Foundation, Inc. -- -- -- -- GNAT is free software; you can redistribute it and/or modify it under -- -- terms of the GNU General Public License as published by the Free Soft- -- -- ware Foundation; either version 3, or (at your option) any later ver- -- -- sion. GNAT is distributed in the hope that it will be useful, but WITH- -- -- OUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY -- -- or FITNESS FOR A PARTICULAR PURPOSE. -- -- -- -- As a special exception under Section 7 of GPL version 3, you are granted -- -- additional permissions described in the GCC Runtime Library Exception, -- -- version 3.1, as published by the Free Software Foundation. -- -- -- -- You should have received a copy of the GNU General Public License and -- -- a copy of the GCC Runtime Library Exception along with this program; -- -- see the files COPYING3 and COPYING.RUNTIME respectively. If not, see -- -- <http://www.gnu.org/licenses/>. -- -- -- -- GNAT was originally developed by the GNAT team at New York University. -- -- Extensive contributions were provided by Ada Core Technologies Inc. -- -- -- ------------------------------------------------------------------------------ -- This is the OpenVMS/Alpha version with System; use System; with System.IO; with System.Machine_Code; with System.Parameters; with System.Storage_Elements; with System.Tasking; with System.Tasking.Rendezvous; with System.Tasking.Initialization; with System.Tasking.Utilities; with System.Task_Primitives; with System.Task_Primitives.Operations; with System.Task_Primitives.Operations.DEC; -- with Ada.Finalization; -- removed, because of problem with controlled attribute ??? with Ada.Task_Attributes; with Ada.Exceptions; use Ada.Exceptions; with Ada.Unchecked_Conversion; package body System.AST_Handling is package ATID renames Ada.Task_Identification; package SP renames System.Parameters; package ST renames System.Tasking; package STR renames System.Tasking.Rendezvous; package STI renames System.Tasking.Initialization; package STU renames System.Tasking.Utilities; package SSE renames System.Storage_Elements; package STPO renames System.Task_Primitives.Operations; package STPOD renames System.Task_Primitives.Operations.DEC; AST_Lock : aliased System.Task_Primitives.RTS_Lock; -- This is a global lock; it is used to execute in mutual exclusion -- from all other AST tasks. It is only used by Lock_AST and -- Unlock_AST. procedure Lock_AST (Self_ID : ST.Task_Id); -- Locks out other AST tasks. Preceding a section of code by Lock_AST and -- following it by Unlock_AST creates a critical region. procedure Unlock_AST (Self_ID : ST.Task_Id); -- Releases lock previously set by call to Lock_AST. -- All nested locks must be released before other tasks competing for the -- tasking lock are released. -------------- -- Lock_AST -- -------------- procedure Lock_AST (Self_ID : ST.Task_Id) is begin STI.Defer_Abort_Nestable (Self_ID); STPO.Write_Lock (AST_Lock'Access, Global_Lock => True); end Lock_AST; ---------------- -- Unlock_AST -- ---------------- procedure Unlock_AST (Self_ID : ST.Task_Id) is begin STPO.Unlock (AST_Lock'Access, Global_Lock => True); STI.Undefer_Abort_Nestable (Self_ID); end Unlock_AST; --------------------------------- -- AST_Handler Data Structures -- --------------------------------- -- As noted in the private part of the spec of System.Aux_DEC, the -- AST_Handler type is simply a pointer to a procedure that takes -- a single 64bit parameter. The following is a local copy -- of that definition. -- We need our own copy because we need to get our hands on this -- and we cannot see the private part of System.Aux_DEC. We don't -- want to be a child of Aux_Dec because of complications resulting -- from the use of pragma Extend_System. We will use unchecked -- conversions between the two versions of the declarations. type AST_Handler is access procedure (Param : Long_Integer); -- However, this declaration is somewhat misleading, since the values -- referenced by AST_Handler values (all produced in this package by -- calls to Create_AST_Handler) are highly stylized. -- The first point is that in VMS/Alpha, procedure pointers do not in -- fact point to code, but rather to a 48-byte procedure descriptor. -- So a value of type AST_Handler is in fact a pointer to one of these -- 48-byte descriptors. type Descriptor_Type is new SSE.Storage_Array (1 .. 48); for Descriptor_Type'Alignment use Standard'Maximum_Alignment; type Descriptor_Ref is access all Descriptor_Type; -- Normally, there is only one such descriptor for a given procedure, but -- it works fine to make a copy of the single allocated descriptor, and -- use the copy itself, and we take advantage of this in the design here. -- The idea is that AST_Handler values will all point to a record with the -- following structure: -- Note: When we say it works fine, there is one delicate point, which -- is that the code for the AST procedure itself requires the original -- descriptor address. We handle this by saving the original descriptor -- address in this structure and restoring in Process_AST. type AST_Handler_Data is record Descriptor : Descriptor_Type; Original_Descriptor_Ref : Descriptor_Ref; Taskid : ATID.Task_Id; Entryno : Natural; end record; type AST_Handler_Data_Ref is access all AST_Handler_Data; function To_AST_Handler is new Ada.Unchecked_Conversion (AST_Handler_Data_Ref, System.Aux_DEC.AST_Handler); -- Each time Create_AST_Handler is called, a new value of this record -- type is created, containing a copy of the procedure descriptor for -- the routine used to handle all AST's (Process_AST), and the Task_Id -- and entry number parameters identifying the task entry involved. -- The AST_Handler value returned is a pointer to this record. Since -- the record starts with the procedure descriptor, it can be used -- by the system in the normal way to call the procedure. But now -- when the procedure gets control, it can determine the address of -- the procedure descriptor used to call it (since the ABI specifies -- that this is left sitting in register r27 on entry), and then use -- that address to retrieve the Task_Id and entry number so that it -- knows on which entry to queue the AST request. -- The next issue is where are these records placed. Since we intend -- to pass pointers to these records to asynchronous system service -- routines, they have to be on the heap, which means we have to worry -- about when to allocate them and deallocate them. -- We solve this problem by introducing a task attribute that points to -- a vector, indexed by the entry number, of AST_Handler_Data records -- for a given task. The pointer itself is a controlled object allowing -- us to write a finalization routine that frees the referenced vector. -- An entry in this vector is either initialized (Entryno non-zero) and -- can be used for any subsequent reference to the same entry, or it is -- unused, marked by the Entryno value being zero. type AST_Handler_Vector is array (Natural range <>) of AST_Handler_Data; type AST_Handler_Vector_Ref is access all AST_Handler_Vector; -- type AST_Vector_Ptr is new Ada.Finalization.Controlled with record -- removed due to problem with controlled attribute, consequence is that -- we have a memory leak if a task that has AST attribute entries is -- terminated. ??? type AST_Vector_Ptr is record Vector : AST_Handler_Vector_Ref; end record; AST_Vector_Init : AST_Vector_Ptr; -- Initial value, treated as constant, Vector will be null package AST_Attribute is new Ada.Task_Attributes (Attribute => AST_Vector_Ptr, Initial_Value => AST_Vector_Init); use AST_Attribute; ----------------------- -- AST Service Queue -- ----------------------- -- The following global data structures are used to queue pending -- AST requests. When an AST is signalled, the AST service routine -- Process_AST is called, and it makes an entry in this structure. type AST_Instance is record Taskid : ATID.Task_Id; Entryno : Natural; Param : Long_Integer; end record; -- The Taskid and Entryno indicate the entry on which this AST is to -- be queued, and Param is the parameter provided from the AST itself. AST_Service_Queue_Size : constant := 256; AST_Service_Queue_Limit : constant := 250; type AST_Service_Queue_Index is mod AST_Service_Queue_Size; -- Index used to refer to entries in the circular buffer which holds -- active AST_Instance values. The upper bound reflects the maximum -- number of AST instances that can be stored in the buffer. Since -- these entries are immediately serviced by the high priority server -- task that does the actual entry queuing, it is very unusual to have -- any significant number of entries simultaneously queued. AST_Service_Queue : array (AST_Service_Queue_Index) of AST_Instance; pragma Volatile_Components (AST_Service_Queue); -- The circular buffer used to store active AST requests AST_Service_Queue_Put : AST_Service_Queue_Index := 0; AST_Service_Queue_Get : AST_Service_Queue_Index := 0; pragma Atomic (AST_Service_Queue_Put); pragma Atomic (AST_Service_Queue_Get); -- These two variables point to the next slots in the AST_Service_Queue -- to be used for putting a new entry in and taking an entry out. This -- is a circular buffer, so these pointers wrap around. If the two values -- are equal the buffer is currently empty. The pointers are atomic to -- ensure proper synchronization between the single producer (namely the -- Process_AST procedure), and the single consumer (the AST_Service_Task). -------------------------------- -- AST Server Task Structures -- -------------------------------- -- The basic approach is that when an AST comes in, a call is made to -- the Process_AST procedure. It queues the request in the service queue -- and then wakes up an AST server task to perform the actual call to the -- required entry. We use this intermediate server task, since the AST -- procedure itself cannot wait to return, and we need some caller for -- the rendezvous so that we can use the normal rendezvous mechanism. -- It would work to have only one AST server task, but then we would lose -- all overlap in AST processing, and furthermore, we could get priority -- inversion effects resulting in starvation of AST requests. -- We therefore maintain a small pool of AST server tasks. We adjust -- the size of the pool dynamically to reflect traffic, so that we have -- a sufficient number of server tasks to avoid starvation. Max_AST_Servers : constant Natural := 16; -- Maximum number of AST server tasks that can be allocated Num_AST_Servers : Natural := 0; -- Number of AST server tasks currently active Num_Waiting_AST_Servers : Natural := 0; -- This is the number of AST server tasks that are either waiting for -- work, or just about to go to sleep and wait for work. Is_Waiting : array (1 .. Max_AST_Servers) of Boolean := (others => False); -- An array of flags showing which AST server tasks are currently waiting AST_Task_Ids : array (1 .. Max_AST_Servers) of ST.Task_Id; -- Task Id's of allocated AST server tasks task type AST_Server_Task (Num : Natural) is pragma Priority (Priority'Last); end AST_Server_Task; -- Declaration for AST server task. This task has no entries, it is -- controlled by sleep and wakeup calls at the task primitives level. type AST_Server_Task_Ptr is access all AST_Server_Task; -- Type used to allocate server tasks ----------------------- -- Local Subprograms -- ----------------------- procedure Allocate_New_AST_Server; -- Allocate an additional AST server task procedure Process_AST (Param : Long_Integer); -- This is the central routine for processing all AST's, it is referenced -- as the code address of all created AST_Handler values. See detailed -- description in body to understand how it works to have a single such -- procedure for all AST's even though it does not get any indication of -- the entry involved passed as an explicit parameter. The single explicit -- parameter Param is the parameter passed by the system with the AST. ----------------------------- -- Allocate_New_AST_Server -- ----------------------------- procedure Allocate_New_AST_Server is Dummy : AST_Server_Task_Ptr; pragma Unreferenced (Dummy); begin if Num_AST_Servers = Max_AST_Servers then return; else -- Note: it is safe to increment Num_AST_Servers immediately, since -- no one will try to activate this task until it indicates that it -- is sleeping by setting its entry in Is_Waiting to True. Num_AST_Servers := Num_AST_Servers + 1; Dummy := new AST_Server_Task (Num_AST_Servers); end if; end Allocate_New_AST_Server; --------------------- -- AST_Server_Task -- --------------------- task body AST_Server_Task is Taskid : ATID.Task_Id; Entryno : Natural; Param : aliased Long_Integer; Self_Id : constant ST.Task_Id := ST.Self; pragma Volatile (Param); begin -- By making this task independent of master, when the environment -- task is finalizing, the AST_Server_Task will be notified that it -- should terminate. STU.Make_Independent; -- Record our task Id for access by Process_AST AST_Task_Ids (Num) := Self_Id; -- Note: this entire task operates with the main task lock set, except -- when it is sleeping waiting for work, or busy doing a rendezvous -- with an AST server. This lock protects the data structures that -- are shared by multiple instances of the server task. Lock_AST (Self_Id); -- This is the main infinite loop of the task. We go to sleep and -- wait to be woken up by Process_AST when there is some work to do. loop Num_Waiting_AST_Servers := Num_Waiting_AST_Servers + 1; Unlock_AST (Self_Id); STI.Defer_Abort (Self_Id); if SP.Single_Lock then STPO.Lock_RTS; end if; STPO.Write_Lock (Self_Id); Is_Waiting (Num) := True; Self_Id.Common.State := ST.AST_Server_Sleep; STPO.Sleep (Self_Id, ST.AST_Server_Sleep); Self_Id.Common.State := ST.Runnable; STPO.Unlock (Self_Id); if SP.Single_Lock then STPO.Unlock_RTS; end if; -- If the process is finalizing, Undefer_Abort will simply end -- this task. STI.Undefer_Abort (Self_Id); -- We are awake, there is something to do! Lock_AST (Self_Id); Num_Waiting_AST_Servers := Num_Waiting_AST_Servers - 1; -- Loop here to service outstanding requests. We are always -- locked on entry to this loop. while AST_Service_Queue_Get /= AST_Service_Queue_Put loop Taskid := AST_Service_Queue (AST_Service_Queue_Get).Taskid; Entryno := AST_Service_Queue (AST_Service_Queue_Get).Entryno; Param := AST_Service_Queue (AST_Service_Queue_Get).Param; AST_Service_Queue_Get := AST_Service_Queue_Get + 1; -- This is a manual expansion of the normal call simple code declare type AA is access all Long_Integer; P : AA := Param'Unrestricted_Access; function To_ST_Task_Id is new Ada.Unchecked_Conversion (ATID.Task_Id, ST.Task_Id); begin Unlock_AST (Self_Id); STR.Call_Simple (Acceptor => To_ST_Task_Id (Taskid), E => ST.Task_Entry_Index (Entryno), Uninterpreted_Data => P'Address); exception when E : others => System.IO.Put_Line ("%Debugging event"); System.IO.Put_Line (Exception_Name (E) & " raised when trying to deliver an AST."); if Exception_Message (E)'Length /= 0 then System.IO.Put_Line (Exception_Message (E)); end if; System.IO.Put_Line ("Task type is " & "Receiver_Type"); System.IO.Put_Line ("Task id is " & ATID.Image (Taskid)); end; Lock_AST (Self_Id); end loop; end loop; end AST_Server_Task; ------------------------ -- Create_AST_Handler -- ------------------------ function Create_AST_Handler (Taskid : ATID.Task_Id; Entryno : Natural) return System.Aux_DEC.AST_Handler is Attr_Ref : Attribute_Handle; Process_AST_Ptr : constant AST_Handler := Process_AST'Access; -- Reference to standard procedure descriptor for Process_AST pragma Warnings (Off, "*alignment*"); -- Suppress harmless warnings about alignment. -- Should explain why this warning is harmless ??? function To_Descriptor_Ref is new Ada.Unchecked_Conversion (AST_Handler, Descriptor_Ref); Original_Descriptor_Ref : constant Descriptor_Ref := To_Descriptor_Ref (Process_AST_Ptr); pragma Warnings (On, "*alignment*"); begin if ATID.Is_Terminated (Taskid) then raise Program_Error; end if; Attr_Ref := Reference (Taskid); -- Allocate another server if supply is getting low if Num_Waiting_AST_Servers < 2 then Allocate_New_AST_Server; end if; -- No point in creating more if we have zillions waiting to -- be serviced. while AST_Service_Queue_Put - AST_Service_Queue_Get > AST_Service_Queue_Limit loop delay 0.01; end loop; -- If no AST vector allocated, or the one we have is too short, then -- allocate one of right size and initialize all entries except the -- one we will use to unused. Note that the assignment automatically -- frees the old allocated table if there is one. if Attr_Ref.Vector = null or else Attr_Ref.Vector'Length < Entryno then Attr_Ref.Vector := new AST_Handler_Vector (1 .. Entryno); for E in 1 .. Entryno loop Attr_Ref.Vector (E).Descriptor := Original_Descriptor_Ref.all; Attr_Ref.Vector (E).Original_Descriptor_Ref := Original_Descriptor_Ref; Attr_Ref.Vector (E).Taskid := Taskid; Attr_Ref.Vector (E).Entryno := E; end loop; end if; return To_AST_Handler (Attr_Ref.Vector (Entryno)'Unrestricted_Access); end Create_AST_Handler; ---------------------------- -- Expand_AST_Packet_Pool -- ---------------------------- procedure Expand_AST_Packet_Pool (Requested_Packets : Natural; Actual_Number : out Natural; Total_Number : out Natural) is pragma Unreferenced (Requested_Packets); begin -- The AST implementation of GNAT does not permit dynamic expansion -- of the pool, so we simply add no entries and return the total. If -- it is necessary to expand the allocation, then this package body -- must be recompiled with a larger value for AST_Service_Queue_Size. Actual_Number := 0; Total_Number := AST_Service_Queue_Size; end Expand_AST_Packet_Pool; ----------------- -- Process_AST -- ----------------- procedure Process_AST (Param : Long_Integer) is Handler_Data_Ptr : AST_Handler_Data_Ref; -- This variable is set to the address of the descriptor through -- which Process_AST is called. Since the descriptor is part of -- an AST_Handler value, this is also the address of this value, -- from which we can obtain the task and entry number information. function To_Address is new Ada.Unchecked_Conversion (ST.Task_Id, System.Task_Primitives.Task_Address); begin System.Machine_Code.Asm (Template => "addq $27,0,%0", Outputs => AST_Handler_Data_Ref'Asm_Output ("=r", Handler_Data_Ptr), Volatile => True); System.Machine_Code.Asm (Template => "ldq $27,%0", Inputs => Descriptor_Ref'Asm_Input ("m", Handler_Data_Ptr.Original_Descriptor_Ref), Volatile => True); AST_Service_Queue (AST_Service_Queue_Put) := AST_Instance' (Taskid => Handler_Data_Ptr.Taskid, Entryno => Handler_Data_Ptr.Entryno, Param => Param); -- OpenVMS Programming Concepts manual, chapter 8.2.3: -- "Implicit synchronization can be achieved for data that is shared -- for write by using only AST routines to write the data, since only -- one AST can be running at any one time." -- This subprogram runs at AST level so is guaranteed to be -- called sequentially at a given access level. AST_Service_Queue_Put := AST_Service_Queue_Put + 1; -- Need to wake up processing task. If there is no waiting server -- then we have temporarily run out, but things should still be -- OK, since one of the active ones will eventually pick up the -- service request queued in the AST_Service_Queue. for J in 1 .. Num_AST_Servers loop if Is_Waiting (J) then Is_Waiting (J) := False; -- Sleeps are handled by ASTs on VMS, so don't call Wakeup STPOD.Interrupt_AST_Handler (To_Address (AST_Task_Ids (J))); exit; end if; end loop; end Process_AST; begin STPO.Initialize_Lock (AST_Lock'Access, STPO.Global_Task_Level); end System.AST_Handling;
Go to most recent revision | Compare with Previous | Blame | View Log