xnu-2050.48.11.tar.gz

[apple/xnu.git] / bsd / dev / unix_startup.c

diff --git a/bsd/dev/unix_startup.c b/bsd/dev/unix_startup.c
index daef72f73baabf8d71b5e21ea0f9847c05eac504..a1d8f5200890c20cf367346e9e0cda4bf2c12035 100644 (file)
--- a/bsd/dev/unix_startup.c
+++ b/bsd/dev/unix_startup.c
@@ -1,5 +1,5 @@
 /*
 /*

- * Copyright (c) 2000-2004 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2010 Apple Inc. All rights reserved.

  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 


@@ -39,23 +39,34 @@
 
 #include <sys/param.h>
 #include <sys/buf_internal.h>
 
 #include <sys/param.h>
 #include <sys/buf_internal.h>

+#include <sys/file_internal.h>
+#include <sys/proc_internal.h>

 #include <sys/clist.h>
 #include <sys/clist.h>

+#include <sys/mcache.h>

 #include <sys/mbuf.h>
 #include <sys/systm.h>
 #include <sys/tty.h>
 #include <sys/vnode.h>
 #include <sys/sysctl.h>
 #include <sys/mbuf.h>
 #include <sys/systm.h>
 #include <sys/tty.h>
 #include <sys/vnode.h>
 #include <sys/sysctl.h>

-#include <dev/ppc/cons.h>
+#include <machine/cons.h>
+#include <pexpert/pexpert.h>
+#include <sys/socketvar.h>
+#include <pexpert/pexpert.h>

 
 

+extern uint32_t kern_maxvnodes;

 extern vm_map_t mb_map;
 
 extern vm_map_t mb_map;
 

-extern u_long   tcp_sendspace;
-extern u_long   tcp_recvspace;
+#if INET || INET6
+extern uint32_t   tcp_sendspace;
+extern uint32_t   tcp_recvspace;
+#endif

 
 

-void            bsd_bufferinit(void);
+void            bsd_bufferinit(void) __attribute__((section("__TEXT, initcode")));

 extern void     md_prepare_for_shutdown(int, int, char *);
 
 extern void     md_prepare_for_shutdown(int, int, char *);
 

-int            bsd_mbuf_cluster_reserve(void);
+unsigned int   bsd_mbuf_cluster_reserve(boolean_t *);
+void bsd_scale_setup(int);
+void bsd_exec_setup(int);

 
 /*
  * Declare these as initialized data so we can patch them.
 
 /*
  * Declare these as initialized data so we can patch them.


@@ -63,28 +74,29 @@ int         bsd_mbuf_cluster_reserve(void);
 
 #ifdef NBUF
 int             max_nbuf_headers = NBUF;
 
 #ifdef NBUF
 int             max_nbuf_headers = NBUF;

-int             niobuf = NBUF / 2;
+int             niobuf_headers = (NBUF / 2) + 2048;

 int            nbuf_hashelements = NBUF;
 int            nbuf_hashelements = NBUF;

-int            nbuf = NBUF;
+int            nbuf_headers = NBUF;

 #else
 int             max_nbuf_headers = 0;
 #else
 int             max_nbuf_headers = 0;

-int             niobuf = 0;
+int             niobuf_headers = 0;

 int            nbuf_hashelements = 0;
 int            nbuf_hashelements = 0;

-int            nbuf = 0;
+int            nbuf_headers = 0;

 #endif
 
 #endif
 

-SYSCTL_INT (_kern, OID_AUTO, nbuf, CTLFLAG_RD, &nbuf, 0, "");
-SYSCTL_INT (_kern, OID_AUTO, maxnbuf, CTLFLAG_RW, &max_nbuf_headers, 0, "");
+SYSCTL_INT (_kern, OID_AUTO, nbuf, CTLFLAG_RD | CTLFLAG_LOCKED, &nbuf_headers, 0, "");
+SYSCTL_INT (_kern, OID_AUTO, maxnbuf, CTLFLAG_RW | CTLFLAG_LOCKED, &max_nbuf_headers, 0, "");

 
 __private_extern__ int customnbuf = 0;
 
 __private_extern__ int customnbuf = 0;

-int             srv = 0;       /* Flag indicates a server boot when set */
+int             serverperfmode = 0;    /* Flag indicates a server boot when set */

 int             ncl = 0;
 int             ncl = 0;

+static unsigned int mbuf_poolsz;

 
 vm_map_t        buffer_map;
 vm_map_t        bufferhdr_map;
 
 vm_map_t        buffer_map;
 vm_map_t        bufferhdr_map;

+static int vnodes_sized = 0;

 
 

-
-extern void     bsd_startupearly(void);
+extern void     bsd_startupearly(void) __attribute__((section("__TEXT, initcode")));

 
 void
 bsd_startupearly(void)
 
 void
 bsd_startupearly(void)


@@ -95,28 +107,30 @@ bsd_startupearly(void)
 
        /* clip the number of buf headers upto 16k */
        if (max_nbuf_headers == 0)
 
        /* clip the number of buf headers upto 16k */
        if (max_nbuf_headers == 0)

-               max_nbuf_headers = atop(sane_size / 50);        /* Get 2% of ram, but no more than we can map */
+               max_nbuf_headers = atop_kernel(sane_size / 50); /* Get 2% of ram, but no more than we can map */

        if ((customnbuf == 0) && (max_nbuf_headers > 16384))
                max_nbuf_headers = 16384;
        if ((customnbuf == 0) && (max_nbuf_headers > 16384))
                max_nbuf_headers = 16384;

-       if (max_nbuf_headers < 256)
-               max_nbuf_headers = 256;
+       if (max_nbuf_headers < CONFIG_MIN_NBUF)
+               max_nbuf_headers = CONFIG_MIN_NBUF;

 
        /* clip the number of hash elements  to 200000 */
        if ( (customnbuf == 0 ) && nbuf_hashelements == 0) {
 
        /* clip the number of hash elements  to 200000 */
        if ( (customnbuf == 0 ) && nbuf_hashelements == 0) {

-               nbuf_hashelements = atop(sane_size / 50);
+               nbuf_hashelements = atop_kernel(sane_size / 50);

                if (nbuf_hashelements > 200000)
                        nbuf_hashelements = 200000;
        } else
                nbuf_hashelements = max_nbuf_headers;
 
                if (nbuf_hashelements > 200000)
                        nbuf_hashelements = 200000;
        } else
                nbuf_hashelements = max_nbuf_headers;
 

-       if (niobuf == 0)
-               niobuf = max_nbuf_headers;
-       if (niobuf > 4096)
-               niobuf = 4096;
-       if (niobuf < 128)
-               niobuf = 128;
+       if (niobuf_headers == 0) {
+               if (max_nbuf_headers < 4096)
+                       niobuf_headers = max_nbuf_headers;
+               else
+                       niobuf_headers = (max_nbuf_headers / 2) + 2048;
+       }
+       if (niobuf_headers < CONFIG_MIN_NIOBUF)
+               niobuf_headers = CONFIG_MIN_NIOBUF;

 
 

-       size = (max_nbuf_headers + niobuf) * sizeof(struct buf);
+       size = (max_nbuf_headers + niobuf_headers) * sizeof(struct buf);

        size = round_page(size);
 
        ret = kmem_suballoc(kernel_map,
        size = round_page(size);
 
        ret = kmem_suballoc(kernel_map,


@@ -138,35 +152,49 @@ bsd_startupearly(void)
        if (ret != KERN_SUCCESS)
                panic("Failed to allocate bufferhdr_map");
 
        if (ret != KERN_SUCCESS)
                panic("Failed to allocate bufferhdr_map");
 

-       buf = (struct buf *) firstaddr;
-       bzero(buf, size);
+       buf_headers = (struct buf *) firstaddr;
+       bzero(buf_headers, size);

 
 

+#if SOCKETS

        {
        {

+               static const unsigned int       maxspace = 128 * 1024;

                int             scale;
 
                int             scale;
 

-               nmbclusters = bsd_mbuf_cluster_reserve() / MCLBYTES;
+               nmbclusters = bsd_mbuf_cluster_reserve(NULL) / MCLBYTES;

 
 

+#if INET || INET6

                if ((scale = nmbclusters / NMBCLUSTERS) > 1) {
                        tcp_sendspace *= scale;
                        tcp_recvspace *= scale;
 
                if ((scale = nmbclusters / NMBCLUSTERS) > 1) {
                        tcp_sendspace *= scale;
                        tcp_recvspace *= scale;
 

-                       if (tcp_sendspace > (32 * 1024))
-                               tcp_sendspace = 32 * 1024;
-                       if (tcp_recvspace > (32 * 1024))
-                               tcp_recvspace = 32 * 1024;
+                       if (tcp_sendspace > maxspace)
+                               tcp_sendspace = maxspace;
+                       if (tcp_recvspace > maxspace)
+                               tcp_recvspace = maxspace;

                }
                }

+#endif /* INET || INET6 */
+       }
+#endif /* SOCKETS */
+
+       if (vnodes_sized == 0) {
+               if (!PE_get_default("kern.maxvnodes", &desiredvnodes, sizeof(desiredvnodes))) {
+                       /*
+                        * Size vnodes based on memory 
+                        * Number vnodes  is (memsize/64k) + 1024 
+                        * This is the calculation that is used by launchd in tiger
+                        * we are clipping the max based on 16G 
+                        * ie ((16*1024*1024*1024)/(64 *1024)) + 1024 = 263168;
+                        * CONFIG_VNODES is set to 263168 for "medium" configurations (the default)
+                        * but can be smaller or larger. 
+                        */
+                       desiredvnodes  = (sane_size/65536) + 1024;
+#ifdef CONFIG_VNODES
+                               if (desiredvnodes > CONFIG_VNODES)
+                                       desiredvnodes = CONFIG_VNODES;
+#endif
+               }
+               vnodes_sized = 1;

        }
        }

-
-       /*
-        * Size vnodes based on memory 
-        * Number vnodes  is (memsize/64k) + 1024 
-        * This is the calculation that is used by launchd in tiger
-        * we are clipping the max based on 16G 
-        * ie ((16*1024*1024*1024)/(64 *1024)) + 1024 = 263168;
-        */
-       desiredvnodes  = (sane_size/65536) + 1024;
-       if (desiredvnodes > 263168)
-               desiredvnodes = 263168;

 }
 
 void
 }
 
 void


@@ -174,10 +202,14 @@ bsd_bufferinit(void)
 {
        kern_return_t   ret;
 
 {
        kern_return_t   ret;
 

-       cons.t_dev = makedev(12, 0);
+       /*
+        * Note: Console device initialized in kminit() from bsd_autoconf()
+        * prior to call to us in bsd_init().
+        */

 
        bsd_startupearly();
 
 
        bsd_startupearly();
 

+#if SOCKETS

        ret = kmem_suballoc(kernel_map,
                            (vm_offset_t *) & mbutl,
                            (vm_size_t) (nmbclusters * MCLBYTES),
        ret = kmem_suballoc(kernel_map,
                            (vm_offset_t *) & mbutl,
                            (vm_size_t) (nmbclusters * MCLBYTES),


@@ -187,6 +219,7 @@ bsd_bufferinit(void)
 
        if (ret != KERN_SUCCESS)
                panic("Failed to allocate mb_map\n");
 
        if (ret != KERN_SUCCESS)
                panic("Failed to allocate mb_map\n");

+#endif /* SOCKETS */

 
        /*
         * Set up buffers, so they can be used to read disk labels.
 
        /*
         * Set up buffers, so they can be used to read disk labels.


@@ -194,6 +227,14 @@ bsd_bufferinit(void)
        bufinit();
 }
 
        bufinit();
 }
 

+/* 512 MB (K32) or 2 GB (K64) hard limit on size of the mbuf pool */
+#if !defined(__LP64__)
+#define        MAX_MBUF_POOL   (512 << MBSHIFT)
+#else
+#define        MAX_MBUF_POOL   (2ULL << GBSHIFT)
+#endif /* !__LP64__ */
+#define        MAX_NCL         (MAX_MBUF_POOL >> MCLSHIFT)
+

 /*
  * this has been broken out into a separate routine that
  * can be called from the x86 early vm initialization to
 /*
  * this has been broken out into a separate routine that
  * can be called from the x86 early vm initialization to


@@ -201,16 +242,103 @@ bsd_bufferinit(void)
  * DMA hardware that can't fully address all of the physical
  * memory that is present.
  */
  * DMA hardware that can't fully address all of the physical
  * memory that is present.
  */

-int
-bsd_mbuf_cluster_reserve(void)
+unsigned int
+bsd_mbuf_cluster_reserve(boolean_t *overridden)

 {
 {

-       if (sane_size > (64 * 1024 * 1024) || ncl) {
+       int mbuf_pool = 0;
+       static boolean_t was_overridden = FALSE;
+
+       /* If called more than once, return the previously calculated size */
+       if (mbuf_poolsz != 0)
+               goto done;
+
+       /*
+        * Some of these are parsed in parse_bsd_args(), but for x86 we get
+        * here early from i386_vm_init() and so we parse them now, in order
+        * to correctly compute the size of the low-memory VM pool.  It is
+        * redundant but rather harmless.
+        */
+       (void) PE_parse_boot_argn("ncl", &ncl, sizeof (ncl));
+       (void) PE_parse_boot_argn("mbuf_pool", &mbuf_pool, sizeof (mbuf_pool));
+
+       /*
+        * Convert "mbuf_pool" from MB to # of 2KB clusters; it is
+        * equivalent to "ncl", except that it uses different unit.
+        */
+       if (mbuf_pool != 0)
+               ncl = (mbuf_pool << MBSHIFT) >> MCLSHIFT;
+
+        if (sane_size > (64 * 1024 * 1024) || ncl != 0) {
+
+               if (ncl || serverperfmode)
+                       was_overridden = TRUE;

 
                if ((nmbclusters = ncl) == 0) {
 
                if ((nmbclusters = ncl) == 0) {

-                       if ((nmbclusters = ((sane_size / 16)/MCLBYTES)) > 32768)
-                               nmbclusters = 32768;
+                       /* Auto-configure the mbuf pool size */
+                       nmbclusters = mbuf_default_ncl(serverperfmode, sane_size);
+               } else {
+                       /* Make sure it's not odd in case ncl is manually set */
+                       if (nmbclusters & 0x1)
+                               --nmbclusters;
+
+                       /* And obey the upper limit */
+                       if (nmbclusters > MAX_NCL)
+                               nmbclusters = MAX_NCL;

                }
                }

+
+               /* Round it down to nearest multiple of 4KB clusters */
+               nmbclusters = P2ROUNDDOWN(nmbclusters, NCLPBG);

        }
        }

+       mbuf_poolsz = nmbclusters << MCLSHIFT;
+done:
+       if (overridden)
+               *overridden = was_overridden;
+
+       return (mbuf_poolsz);
+}
+#if defined(__LP64__)
+extern int tcp_tcbhashsize;
+extern int max_cached_sock_count;
+#endif 

 
 

-       return (nmbclusters * MCLBYTES);
+
+void
+bsd_scale_setup(int scale)
+{
+#if defined(__LP64__)
+       if ((scale > 0) && (serverperfmode == 0)) {
+               maxproc *= scale;
+               maxprocperuid = (maxproc * 2) / 3;
+       }
+       /* Apply server scaling rules */
+       if ((scale >  0) && (serverperfmode !=0)) {
+               maxproc = 2500 * scale;
+               hard_maxproc = maxproc;
+               /* no fp usage */
+               maxprocperuid = (maxproc*3)/4;
+               maxfiles = (150000 * scale);
+               maxfilesperproc = maxfiles/2;
+               desiredvnodes = maxfiles;
+               vnodes_sized = 1;
+               if (scale > 4) {
+                       /* clip them at 32G level */
+                       somaxconn = 2048;
+                       /* 64G or more the hash size is 32k */
+                       if (scale > 7) {
+                               /* clip at 64G level */
+                               tcp_tcbhashsize = 16 *1024;
+                               max_cached_sock_count = 165000;
+                       } else {
+                               tcp_tcbhashsize = 32 *1024;
+                               max_cached_sock_count = 60000 + ((scale-1) * 15000);
+                       }
+               } else {
+                       somaxconn = 512*scale;
+                       tcp_tcbhashsize = 4*1024*scale;
+                       max_cached_sock_count = 60000 + ((scale-1) * 15000);
+               }
+       }
+#endif
+       bsd_exec_setup(scale);

 }
 }

+