uml: redo host capability detection and disabling
[safe/jmp/linux-2.6] / arch / um / os-Linux / aio.c
index f897140..57e3d46 100644 (file)
@@ -1,19 +1,20 @@
 /*
- * Copyright (C) 2004 Jeff Dike (jdike@addtoit.com)
+ * Copyright (C) 2004 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
  * Licensed under the GPL
  */
 
-#include <stdlib.h>
 #include <unistd.h>
+#include <sched.h>
 #include <signal.h>
 #include <errno.h>
-#include <sched.h>
-#include <sys/syscall.h>
-#include "os.h"
+#include <sys/time.h>
+#include <asm/unistd.h>
 #include "aio.h"
 #include "init.h"
+#include "kern_constants.h"
+#include "kern_util.h"
+#include "os.h"
 #include "user.h"
-#include "mode.h"
 
 struct aio_thread_req {
        enum aio_type type;
@@ -24,13 +25,11 @@ struct aio_thread_req {
        struct aio_context *aio;
 };
 
-static int aio_req_fd_r = -1;
-static int aio_req_fd_w = -1;
-
 #if defined(HAVE_AIO_ABI)
 #include <linux/aio_abi.h>
 
-/* If we have the headers, we are going to build with AIO enabled.
+/*
+ * If we have the headers, we are going to build with AIO enabled.
  * If we don't have aio in libc, we define the necessary stubs here.
  */
 
@@ -54,7 +53,8 @@ static long io_getevents(aio_context_t ctx_id, long min_nr, long nr,
 
 #endif
 
-/* The AIO_MMAP cases force the mmapped page into memory here
+/*
+ * The AIO_MMAP cases force the mmapped page into memory here
  * rather than in whatever place first touches the data.  I used
  * to do this by touching the page, but that's delicate because
  * gcc is prone to optimizing that away.  So, what's done here
@@ -68,49 +68,36 @@ static long io_getevents(aio_context_t ctx_id, long min_nr, long nr,
 static int do_aio(aio_context_t ctx, enum aio_type type, int fd, char *buf,
                  int len, unsigned long long offset, struct aio_context *aio)
 {
-       struct iocb iocb, *iocbp = &iocb;
+       struct iocb *iocbp = & ((struct iocb) {
+                                   .aio_data       = (unsigned long) aio,
+                                   .aio_fildes     = fd,
+                                   .aio_buf        = (unsigned long) buf,
+                                   .aio_nbytes     = len,
+                                   .aio_offset     = offset
+                            });
        char c;
-       int err;
 
-       iocb = ((struct iocb) { .aio_data       = (unsigned long) aio,
-                               .aio_reqprio    = 0,
-                               .aio_fildes     = fd,
-                               .aio_buf        = (unsigned long) buf,
-                               .aio_nbytes     = len,
-                               .aio_offset     = offset,
-                               .aio_reserved1  = 0,
-                               .aio_reserved2  = 0,
-                               .aio_reserved3  = 0 });
-
-       switch(type){
+       switch (type) {
        case AIO_READ:
-               iocb.aio_lio_opcode = IOCB_CMD_PREAD;
-               err = io_submit(ctx, 1, &iocbp);
+               iocbp->aio_lio_opcode = IOCB_CMD_PREAD;
                break;
        case AIO_WRITE:
-               iocb.aio_lio_opcode = IOCB_CMD_PWRITE;
-               err = io_submit(ctx, 1, &iocbp);
+               iocbp->aio_lio_opcode = IOCB_CMD_PWRITE;
                break;
        case AIO_MMAP:
-               iocb.aio_lio_opcode = IOCB_CMD_PREAD;
-               iocb.aio_buf = (unsigned long) &c;
-               iocb.aio_nbytes = sizeof(c);
-               err = io_submit(ctx, 1, &iocbp);
+               iocbp->aio_lio_opcode = IOCB_CMD_PREAD;
+               iocbp->aio_buf = (unsigned long) &c;
+               iocbp->aio_nbytes = sizeof(c);
                break;
        default:
-               printk("Bogus op in do_aio - %d\n", type);
-               err = -EINVAL;
-               break;
+               printk(UM_KERN_ERR "Bogus op in do_aio - %d\n", type);
+               return -EINVAL;
        }
 
-       if(err > 0)
-               err = 0;
-       else
-               err = -errno;
-
-       return err;
+       return (io_submit(ctx, 1, &iocbp) > 0) ? 0 : -errno;
 }
 
+/* Initialized in an initcall and unchanged thereafter */
 static aio_context_t ctx = 0;
 
 static int aio_thread(void *arg)
@@ -121,12 +108,12 @@ static int aio_thread(void *arg)
 
        signal(SIGWINCH, SIG_IGN);
 
-       while(1){
+       while (1) {
                n = io_getevents(ctx, 1, 1, &event, NULL);
-               if(n < 0){
-                       if(errno == EINTR)
+               if (n < 0) {
+                       if (errno == EINTR)
                                continue;
-                       printk("aio_thread - io_getevents failed, "
+                       printk(UM_KERN_ERR "aio_thread - io_getevents failed, "
                               "errno = %d\n", errno);
                }
                else {
@@ -134,10 +121,10 @@ static int aio_thread(void *arg)
                                { .data = (void *) (long) event.data,
                                                .err    = event.res });
                        reply_fd = ((struct aio_context *) reply.data)->reply_fd;
-                       err = os_write_file(reply_fd, &reply, sizeof(reply));
-                       if(err != sizeof(reply))
-                               printk("aio_thread - write failed, fd = %d, "
-                                      "err = %d\n", aio_req_fd_r, -err);
+                       err = write(reply_fd, &reply, sizeof(reply));
+                       if (err != sizeof(reply))
+                               printk(UM_KERN_ERR "aio_thread - write failed, "
+                                      "fd = %d, err = %d\n", reply_fd, errno);
                }
        }
        return 0;
@@ -148,40 +135,40 @@ static int aio_thread(void *arg)
 static int do_not_aio(struct aio_thread_req *req)
 {
        char c;
-       int err;
+       unsigned long long actual;
+       int n;
 
-       switch(req->type){
-       case AIO_READ:
-               err = os_seek_file(req->io_fd, req->offset);
-               if(err)
-                       goto out;
+       actual = lseek64(req->io_fd, req->offset, SEEK_SET);
+       if (actual != req->offset)
+               return -errno;
 
-               err = os_read_file(req->io_fd, req->buf, req->len);
+       switch (req->type) {
+       case AIO_READ:
+               n = read(req->io_fd, req->buf, req->len);
                break;
        case AIO_WRITE:
-               err = os_seek_file(req->io_fd, req->offset);
-               if(err)
-                       goto out;
-
-               err = os_write_file(req->io_fd, req->buf, req->len);
+               n = write(req->io_fd, req->buf, req->len);
                break;
        case AIO_MMAP:
-               err = os_seek_file(req->io_fd, req->offset);
-               if(err)
-                       goto out;
-
-               err = os_read_file(req->io_fd, &c, sizeof(c));
+               n = read(req->io_fd, &c, sizeof(c));
                break;
        default:
-               printk("do_not_aio - bad request type : %d\n", req->type);
-               err = -EINVAL;
-               break;
+               printk(UM_KERN_ERR "do_not_aio - bad request type : %d\n",
+                      req->type);
+               return -EINVAL;
        }
 
-out:
-       return err;
+       if (n < 0)
+               return -errno;
+       return 0;
 }
 
+/* These are initialized in initcalls and not changed */
+static int aio_req_fd_r = -1;
+static int aio_req_fd_w = -1;
+static int aio_pid = -1;
+static unsigned long aio_stack;
+
 static int not_aio_thread(void *arg)
 {
        struct aio_thread_req req;
@@ -189,63 +176,68 @@ static int not_aio_thread(void *arg)
        int err;
 
        signal(SIGWINCH, SIG_IGN);
-       while(1){
-               err = os_read_file(aio_req_fd_r, &req, sizeof(req));
-               if(err != sizeof(req)){
-                       if(err < 0)
-                               printk("not_aio_thread - read failed, "
-                                      "fd = %d, err = %d\n", aio_req_fd_r,
-                                      -err);
+       while (1) {
+               err = read(aio_req_fd_r, &req, sizeof(req));
+               if (err != sizeof(req)) {
+                       if (err < 0)
+                               printk(UM_KERN_ERR "not_aio_thread - "
+                                      "read failed, fd = %d, err = %d\n",
+                                      aio_req_fd_r,
+                                      errno);
                        else {
-                               printk("not_aio_thread - short read, fd = %d, "
-                                      "length = %d\n", aio_req_fd_r, err);
+                               printk(UM_KERN_ERR "not_aio_thread - short "
+                                      "read, fd = %d, length = %d\n",
+                                      aio_req_fd_r, err);
                        }
                        continue;
                }
                err = do_not_aio(&req);
                reply = ((struct aio_thread_reply) { .data      = req.aio,
-                                        .err   = err });
-               err = os_write_file(req.aio->reply_fd, &reply, sizeof(reply));
-               if(err != sizeof(reply))
-                       printk("not_aio_thread - write failed, fd = %d, "
-                              "err = %d\n", aio_req_fd_r, -err);
+                                                    .err       = err });
+               err = write(req.aio->reply_fd, &reply, sizeof(reply));
+               if (err != sizeof(reply))
+                       printk(UM_KERN_ERR "not_aio_thread - write failed, "
+                              "fd = %d, err = %d\n", req.aio->reply_fd, errno);
        }
 
        return 0;
 }
 
-static int aio_pid = -1;
-
 static int init_aio_24(void)
 {
-       unsigned long stack;
        int fds[2], err;
 
        err = os_pipe(fds, 1, 1);
-       if(err)
+       if (err)
                goto out;
 
        aio_req_fd_w = fds[0];
        aio_req_fd_r = fds[1];
+
+       err = os_set_fd_block(aio_req_fd_w, 0);
+       if (err)
+               goto out_close_pipe;
+
        err = run_helper_thread(not_aio_thread, NULL,
-                               CLONE_FILES | CLONE_VM | SIGCHLD, &stack, 0);
-       if(err < 0)
+                               CLONE_FILES | CLONE_VM, &aio_stack);
+       if (err < 0)
                goto out_close_pipe;
 
        aio_pid = err;
        goto out;
 
 out_close_pipe:
-       os_close_file(fds[0]);
-       os_close_file(fds[1]);
+       close(fds[0]);
+       close(fds[1]);
        aio_req_fd_w = -1;
        aio_req_fd_r = -1;
 out:
 #ifndef HAVE_AIO_ABI
-       printk("/usr/include/linux/aio_abi.h not present during build\n");
+       printk(UM_KERN_INFO "/usr/include/linux/aio_abi.h not present during "
+              "build\n");
 #endif
-       printk("2.6 host AIO support not used - falling back to I/O "
-              "thread\n");
+       printk(UM_KERN_INFO "2.6 host AIO support not used - falling back to "
+              "I/O thread\n");
        return 0;
 }
 
@@ -253,24 +245,23 @@ out:
 #define DEFAULT_24_AIO 0
 static int init_aio_26(void)
 {
-       unsigned long stack;
        int err;
 
-       if(io_setup(256, &ctx)){
+       if (io_setup(256, &ctx)) {
                err = -errno;
-               printk("aio_thread failed to initialize context, err = %d\n",
-                      errno);
+               printk(UM_KERN_ERR "aio_thread failed to initialize context, "
+                      "err = %d\n", errno);
                return err;
        }
 
        err = run_helper_thread(aio_thread, NULL,
-                               CLONE_FILES | CLONE_VM | SIGCHLD, &stack, 0);
-       if(err < 0)
+                               CLONE_FILES | CLONE_VM, &aio_stack);
+       if (err < 0)
                return err;
 
        aio_pid = err;
 
-       printk("Using 2.6 host AIO\n");
+       printk(UM_KERN_INFO "Using 2.6 host AIO\n");
        return 0;
 }
 
@@ -281,13 +272,15 @@ static int submit_aio_26(enum aio_type type, int io_fd, char *buf, int len,
        int err;
 
        err = do_aio(ctx, type, io_fd, buf, len, offset, aio);
-       if(err){
+       if (err) {
                reply = ((struct aio_thread_reply) { .data = aio,
                                         .err  = err });
-               err = os_write_file(aio->reply_fd, &reply, sizeof(reply));
-               if(err != sizeof(reply))
-                       printk("submit_aio_26 - write failed, "
+               err = write(aio->reply_fd, &reply, sizeof(reply));
+               if (err != sizeof(reply)) {
+                       err = -errno;
+                       printk(UM_KERN_ERR "submit_aio_26 - write failed, "
                               "fd = %d, err = %d\n", aio->reply_fd, -err);
+               }
                else err = 0;
        }
 
@@ -308,6 +301,7 @@ static int submit_aio_26(enum aio_type type, int io_fd, char *buf, int len,
 }
 #endif
 
+/* Initialized in an initcall and unchanged thereafter */
 static int aio_24 = DEFAULT_24_AIO;
 
 static int __init set_aio_24(char *name, int *add)
@@ -332,28 +326,24 @@ static int init_aio(void)
 {
        int err;
 
-       CHOOSE_MODE(({ if(!aio_24){
-                           printk("Disabling 2.6 AIO in tt mode\n");
-                           aio_24 = 1;
-                   } }), (void) 0);
-
-       if(!aio_24){
+       if (!aio_24) {
                err = init_aio_26();
-               if(err && (errno == ENOSYS)){
-                       printk("2.6 AIO not supported on the host - "
-                              "reverting to 2.4 AIO\n");
+               if (err && (errno == ENOSYS)) {
+                       printk(UM_KERN_INFO "2.6 AIO not supported on the "
+                              "host - reverting to 2.4 AIO\n");
                        aio_24 = 1;
                }
                else return err;
        }
 
-       if(aio_24)
+       if (aio_24)
                return init_aio_24();
 
        return 0;
 }
 
-/* The reason for the __initcall/__uml_exitcall asymmetry is that init_aio
+/*
+ * The reason for the __initcall/__uml_exitcall asymmetry is that init_aio
  * needs to be called when the kernel is running because it calls run_helper,
  * which needs get_free_page.  exit_aio is a __uml_exitcall because the generic
  * kernel does not run __exitcalls on shutdown, and can't because many of them
@@ -363,8 +353,10 @@ __initcall(init_aio);
 
 static void exit_aio(void)
 {
-       if(aio_pid != -1)
+       if (aio_pid != -1) {
                os_kill_process(aio_pid, 1);
+               free_stack(aio_stack, 0);
+       }
 }
 
 __uml_exitcall(exit_aio);
@@ -381,9 +373,10 @@ static int submit_aio_24(enum aio_type type, int io_fd, char *buf, int len,
        };
        int err;
 
-       err = os_write_file(aio_req_fd_w, &req, sizeof(req));
-       if(err == sizeof(req))
+       err = write(aio_req_fd_w, &req, sizeof(req));
+       if (err == sizeof(req))
                err = 0;
+       else err = -errno;
 
        return err;
 }
@@ -393,9 +386,8 @@ int submit_aio(enum aio_type type, int io_fd, char *buf, int len,
               struct aio_context *aio)
 {
        aio->reply_fd = reply_fd;
-       if(aio_24)
+       if (aio_24)
                return submit_aio_24(type, io_fd, buf, len, offset, aio);
-       else {
+       else
                return submit_aio_26(type, io_fd, buf, len, offset, aio);
-       }
 }