GitHub/mt8127/android_kernel_alcatel_ttab.git
18 years ago[NETFILTER] NAT sequence adjustment: Save eight bytes per conntrack
Harald Welte [Tue, 21 Mar 2006 01:14:12 +0000 (17:14 -0800)]
[NETFILTER] NAT sequence adjustment: Save eight bytes per conntrack

This patch reduces the size of 'struct ip_conntrack' on systems with NAT
by eight bytes.  The sequence number delta values can be int16_t, since
we only support one sequence number modification per window anyway, and
one such modification is not going to exceed 32kB ;)

Signed-off-by: Harald Welte <laforge@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[NET]: Do not lose accepted socket when -ENFILE/-EMFILE.
David S. Miller [Tue, 21 Mar 2006 01:13:49 +0000 (17:13 -0800)]
[NET]: Do not lose accepted socket when -ENFILE/-EMFILE.

Try to allocate the struct file and an unused file
descriptor before we try to pull a newly accepted
socket out of the protocol layer.

Based upon a patch by Prassana Meda.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[NET]: Reduce size of struct sk_buff on 64 bit architectures
Patrick McHardy [Tue, 21 Mar 2006 01:12:12 +0000 (17:12 -0800)]
[NET]: Reduce size of struct sk_buff on 64 bit architectures

Move skb->nf_mark next to skb->tc_index to remove a 4 byte hole between
skb->nfmark and skb->nfct and another one between skb->users and skb->head
when CONFIG_NETFILTER, CONFIG_NET_SCHED and CONFIG_NET_CLS_ACT are enabled.
For all other combinations the size stays the same.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[VLAN]: translate IF_OPER_DORMANT to netif_dormant_on()
Stefan Rompf [Tue, 21 Mar 2006 01:11:41 +0000 (17:11 -0800)]
[VLAN]: translate IF_OPER_DORMANT to netif_dormant_on()

this patch adds support to the VLAN driver to translate IF_OPER_DORMANT of the
underlying device to netif_dormant_on(). Beside clean state forwarding, this
allows running independant userspace supplicants on both the real device and
the stacked VLAN. It depends on my RFC2863 patch.

Signed-off-by: Stefan Rompf <stefan@loplof.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[NET] core: add RFC2863 operstate
Stefan Rompf [Tue, 21 Mar 2006 01:09:11 +0000 (17:09 -0800)]
[NET] core: add RFC2863 operstate

this patch adds a dormant flag to network devices, RFC2863 operstate derived
from these flags and possibility for userspace interaction. It allows drivers
to signal that a device is unusable for user traffic without disabling
queueing (and therefore the possibility for protocol establishment traffic to
flow) and a userspace supplicant (WPA, 802.1X) to mark a device unusable
without changes to the driver.

It is the result of our long discussion. However I must admit that it
represents what Jamal and I agreed on with compromises towards Krzysztof, but
Thomas and Krzysztof still disagree with some parts. Anyway I think it should
be applied.

Signed-off-by: Stefan Rompf <stefan@loplof.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ROUTE: Ensure to accept redirects from nexthop for the target.
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 01:07:49 +0000 (17:07 -0800)]
[IPV6]: ROUTE: Ensure to accept redirects from nexthop for the target.

It is possible to get redirects from nexthop of "more-specific"
routes.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ROUTE: Add accept_ra_rt_info_max_plen sysctl.
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 01:07:03 +0000 (17:07 -0800)]
[IPV6]: ROUTE: Add accept_ra_rt_info_max_plen sysctl.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ROUTE: Flag RTF_DEFAULT for Route Infomation for ::/0.
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 01:06:42 +0000 (17:06 -0800)]
[IPV6]: ROUTE: Flag RTF_DEFAULT for Route Infomation for ::/0.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ROUTE: Add experimental support for Route Information Option in RA (RFC4191).
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 01:06:24 +0000 (17:06 -0800)]
[IPV6]: ROUTE: Add experimental support for Route Information Option in RA (RFC4191).

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ROUTE: Add router_probe_interval sysctl.
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 01:05:47 +0000 (17:05 -0800)]
[IPV6]: ROUTE: Add router_probe_interval sysctl.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ROUTE: Add accept_ra_rtr_pref sysctl.
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 01:05:30 +0000 (17:05 -0800)]
[IPV6]: ROUTE: Add accept_ra_rtr_pref sysctl.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ROUTE: Add Router Reachability Probing (RFC4191).
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 01:05:13 +0000 (17:05 -0800)]
[IPV6]: ROUTE: Add Router Reachability Probing (RFC4191).

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ROUTE: Add support for Router Preference (RFC4191).
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 01:04:53 +0000 (17:04 -0800)]
[IPV6]: ROUTE: Add support for Router Preference (RFC4191).

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ROUTE: Handle finding the next best route in reachability in BACKTRACK().
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 01:04:35 +0000 (17:04 -0800)]
[IPV6]: ROUTE: Handle finding the next best route in reachability in BACKTRACK().

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ROUTE: Try finding the next best route.
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 01:01:43 +0000 (17:01 -0800)]
[IPV6]: ROUTE: Try finding the next best route.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ROUTE: Clean up rt6_select() code path in ip6_route_{intput,output}().
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 01:01:24 +0000 (17:01 -0800)]
[IPV6]: ROUTE: Clean up rt6_select() code path in ip6_route_{intput,output}().

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ROUTE: Try selecting better route for non-default routes as well.
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 01:01:06 +0000 (17:01 -0800)]
[IPV6]: ROUTE: Try selecting better route for non-default routes as well.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ROUTE: More strict check for default routers in rt6_get_dflt_router().
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 01:00:48 +0000 (17:00 -0800)]
[IPV6]: ROUTE: More strict check for default routers in rt6_get_dflt_router().

Check RTF_ADDRCONF|RTF_DEFAULT in rt6_get_dflt_router().

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ROUTE: Eliminate lock for default route pointer.
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 01:00:26 +0000 (17:00 -0800)]
[IPV6]: ROUTE: Eliminate lock for default route pointer.

And prepare for more advanced router selection.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ROUTE: Clean-up cow'ing in ip6_route_{intput,output}().
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 01:00:05 +0000 (17:00 -0800)]
[IPV6]: ROUTE: Clean-up cow'ing in ip6_route_{intput,output}().

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ROUTE: Convert rt6_cow() to rt6_alloc_cow().
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 00:59:27 +0000 (16:59 -0800)]
[IPV6]: ROUTE: Convert rt6_cow() to rt6_alloc_cow().

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ROUTE: Clean up reference counting / unlocking for returning object.
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 00:59:08 +0000 (16:59 -0800)]
[IPV6]: ROUTE: Clean up reference counting / unlocking for returning object.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ROUTE: Unify two code paths for pmtu disc.
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 00:58:48 +0000 (16:58 -0800)]
[IPV6]: ROUTE: Unify two code paths for pmtu disc.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ROUTE: Add rt6_alloc_clone() for cloning route allocation.
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 00:58:32 +0000 (16:58 -0800)]
[IPV6]: ROUTE: Add rt6_alloc_clone() for cloning route allocation.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ROUTE: Copy u.dst.error for RTF_REJECT routes when cloning.
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 00:56:50 +0000 (16:56 -0800)]
[IPV6]: ROUTE: Copy u.dst.error for RTF_REJECT routes when cloning.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ROUTE: Set appropriate information before inserting a route.
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 00:56:32 +0000 (16:56 -0800)]
[IPV6]: ROUTE: Set appropriate information before inserting a route.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ROUTE: Split up rt6_cow() for future changes.
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 00:55:51 +0000 (16:55 -0800)]
[IPV6]: ROUTE: Split up rt6_cow() for future changes.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ADDRCONF: Add accept_ra_pinfo sysctl.
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 00:55:26 +0000 (16:55 -0800)]
[IPV6]: ADDRCONF: Add accept_ra_pinfo sysctl.

This controls whether we accept Prefix Information in RAs.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ROUTE: Add accept_ra_defrtr sysctl.
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 00:55:08 +0000 (16:55 -0800)]
[IPV6]: ROUTE: Add accept_ra_defrtr sysctl.

This controls whether we accept default router information
in RAs.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ADDRCONF: Split up ipv6_generate_eui64() by device type.
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 00:54:49 +0000 (16:54 -0800)]
[IPV6]: ADDRCONF: Split up ipv6_generate_eui64() by device type.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: ADDRCONF: Use our standard algorithm for randomized ifid.
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 00:54:09 +0000 (16:54 -0800)]
[IPV6]: ADDRCONF: Use our standard algorithm for randomized ifid.

RFC 3041 describes an algorithm to generate random interface
identifier.  In RFC 3041bis, it is allowed to use different
algorithm than one described in RFC 3041.

So, let's use our standard pseudo random algorithm to simplify
our implementation.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[NET]: NEIGHBOUR: Ensure to record time to neigh->updated when neighbour's state...
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 00:52:52 +0000 (16:52 -0800)]
[NET]: NEIGHBOUR: Ensure to record time to neigh->updated when neighbour's state changed.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV6]: TUNNEL6: Don't try to add multicast route twice.
YOSHIFUJI Hideaki [Tue, 21 Mar 2006 00:51:48 +0000 (16:51 -0800)]
[IPV6]: TUNNEL6: Don't try to add multicast route twice.

Since addrconf_add_dev() has already called addrconf_add_mroute()
to added route for multicast prefix, there's no point to call it
again in addrconf_ip6_tnl_config().

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6
Linus Torvalds [Mon, 20 Mar 2006 19:57:50 +0000 (11:57 -0800)]
Merge /pub/scm/linux/kernel/git/davem/sparc-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6: (230 commits)
  [SPARC64]: Update defconfig.
  [SPARC64]: Fix 2 bugs in huge page support.
  [SPARC64]: CONFIG_BLK_DEV_RAM fix
  [SPARC64]: Optimized TSB table initialization.
  [SPARC64]: Allow CONFIG_MEMORY_HOTPLUG to build.
  [SPARC64]: Use SLAB caches for TSB tables.
  [SPARC64]: Don't kill the page allocator when growing a TSB.
  [SPARC64]: Randomize mm->mmap_base when PF_RANDOMIZE is set.
  [SPARC64]: Increase top of 32-bit process stack.
  [SPARC64]: Top-down address space allocation for 32-bit tasks.
  [SPARC64] bbc_i2c: Fix cpu check and add missing module license.
  [SPARC64]: Fix and re-enable dynamic TSB sizing.
  [SUNSU]: Fix missing spinlock initialization.
  [TG3]: Do not try to access NIC_SRAM_DATA_SIG on Sun parts.
  [SPARC64]: First cut at VIS simulator for Niagara.
  [SPARC64]: Fix system type in /proc/cpuinfo and remove bogus OBP check.
  [SPARC64]: Add SMT scheduling support for Niagara.
  [SPARC64]: Fix 32-bit truncation which broke sparsemem.
  [SPARC64]: Move over to sparsemem.
  [SPARC64]: Fix new context version SMP handling.
  ...

18 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/shaggy...
Linus Torvalds [Mon, 20 Mar 2006 18:32:33 +0000 (10:32 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/shaggy/jfs-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/shaggy/jfs-2.6:
  JFS: add uid, gid, and umask mount options
  JFS: Take logsync lock before testing mp->lsn
  JFS: kzalloc conversion
  JFS: Add missing file from fa3241d24cf1182b0ffb6e4d412c3bc2a2ab7bf6
  JFS: Use the kthread_ API
  JFS: Fix regression.  fsck complains if symlinks do not have INLINEEA attribute
  JFS: ext2 inode attributes for jfs
  JFS: semaphore to mutex conversion.
  JFS: make buddy table static
  JFS: Add back directory i_size calculations for legacy partitions

18 years agoMerge branch 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik...
Linus Torvalds [Mon, 20 Mar 2006 18:30:31 +0000 (10:30 -0800)]
Merge branch 'upstream-linus' of /linux/kernel/git/jgarzik/netdev-2.6

* 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik/netdev-2.6: (150 commits)
  [PATCH] ipw2100: Update version ipw2100 stamp to 1.2.2
  [PATCH] ipw2100: move mutex.h include from ipw2100.c to ipw2100.h
  [PATCH] ipw2100: semaphore to mutexes conversion
  [PATCH] ipw2100: Fix radiotap code gcc warning
  [PATCH] ipw2100: add radiotap headers to packtes captured in monitor mode
  [PATCH] ipw2x00: expend Copyright to 2006
  [PATCH] drivers/net/wireless/ipw2200.c: fix an array overun
  [PATCH] ieee80211: Don't update network statistics from off-channel packets.
  [PATCH] ipw2200: Update ipw2200 version stamp to 1.1.1
  [PATCH] ipw2200: switch to the new ipw2200-fw-3.0 image format
  [PATCH] ipw2200: wireless extension sensitivity threshold support
  [PATCH] ipw2200: Enables the "slow diversity" algorithm
  [PATCH] ipw2200: Set a meaningful silence threshold value
  [PATCH] ipw2200: export `debug' module param only if CONFIG_IPW2200_DEBUG
  [PATCH] ipw2200: Change debug level for firmware error logging
  [PATCH] ipw2200: Filter unsupported channels out in ad-hoc mode
  [PATCH] ipw2200: Fix ipw_sw_reset() implementation inconsistent with comment
  [PATCH] ipw2200: Fix rf_kill is activated after mode change with 'disable=1'
  [PATCH] ipw2200: remove the WPA card associates to non-WPA AP checking
  [PATCH] ipw2200: Add signal level to iwlist scan output
  ...

18 years agoMerge branch 'block-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/viro...
Linus Torvalds [Mon, 20 Mar 2006 18:28:51 +0000 (10:28 -0800)]
Merge branch 'block-fixes' of git://git./linux/kernel/git/viro/block

* 'block-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/block:
  [PATCH] fix rmmod problems with elevator attributes, clean them up
  [PATCH] elevator_t lifetime rules and sysfs fixes
  [PATCH] noise removal: cfq-iosched.c
  [PATCH] don't bother with refcounting for cfq_data
  [PATCH] fix sysfs interaction and lifetime rules handling for queues
  [PATCH] regularize blk_cleanup_queue() use
  [PATCH] fix cfq_get_queue()/ioprio_set(2) races
  [PATCH] deal with rmmod/put_io_context() races
  [PATCH] stop elv_unregister() from rogering other iosched's data, fix locking
  [PATCH] stop cfq from pinning queue down
  [PATCH] make cfq_exit_queue() prune the cfq_io_context for that queue
  [PATCH] fix the exclusion for ioprio_set()
  [PATCH] keep sync and async cfq_queue separate
  [PATCH] switch to use of ->key to get cfq_data by cfq_io_context
  [PATCH] stop leaking cfq_data in cfq_set_request()
  [PATCH] fix cfq hash lookups
  [PATCH] fix locking in queue_requests_store()
  [PATCH] fix double-free in blk_init_queue_node()
  [PATCH] don't do exit_io_context() until we know we won't be doing any IO

18 years agoMerge branch 'upstream' of git://git.kernel.org/pub/scm/linux/kernel/git/linville...
Jeff Garzik [Mon, 20 Mar 2006 09:38:50 +0000 (04:38 -0500)]
Merge branch 'upstream' of git://git./linux/kernel/git/linville/wireless-2.6

18 years agoMerge branch 'master'
Jeff Garzik [Mon, 20 Mar 2006 09:38:03 +0000 (04:38 -0500)]
Merge branch 'master'

18 years ago[SPARC64]: Update defconfig.
David S. Miller [Mon, 20 Mar 2006 09:23:43 +0000 (01:23 -0800)]
[SPARC64]: Update defconfig.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Fix 2 bugs in huge page support.
David S. Miller [Mon, 20 Mar 2006 09:17:17 +0000 (01:17 -0800)]
[SPARC64]: Fix 2 bugs in huge page support.

1) huge_pte_offset() did not check the page table hierarchy
   elements as being empty correctly, resulting in an OOPS

2) Need platform specific hugetlb_get_unmapped_area() to handle
   the top-down vs. bottom-up address space allocation strategies.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: CONFIG_BLK_DEV_RAM fix
Andrew Morton [Sun, 19 Mar 2006 20:46:55 +0000 (12:46 -0800)]
[SPARC64]: CONFIG_BLK_DEV_RAM fix

init/do_mounts_rd.c depends upon CONFIG_BLK_DEV_RAM, not CONFIG_BLK_DEV_INITRD.

Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Optimized TSB table initialization.
David S. Miller [Sun, 19 Mar 2006 07:55:11 +0000 (23:55 -0800)]
[SPARC64]: Optimized TSB table initialization.

We only need to write an invalid tag every 16 bytes,
so taking advantage of this can save many instructions
compared to the simple memset() call we make now.

A prefetching implementation is implemented for sun4u
and a block-init store version if implemented for Niagara.

The next trick is to be able to perform an init and
a copy_tsb() in parallel when growing a TSB table.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Allow CONFIG_MEMORY_HOTPLUG to build.
David S. Miller [Sun, 19 Mar 2006 03:16:23 +0000 (19:16 -0800)]
[SPARC64]: Allow CONFIG_MEMORY_HOTPLUG to build.

online_page() is straightforward, and then add a dummy
remove_memory() that returns -EINVAL just like i386.

There is no point in implementing remove_memory() since
__remove_pages() has no implementation either.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Use SLAB caches for TSB tables.
David S. Miller [Sun, 19 Mar 2006 02:12:42 +0000 (18:12 -0800)]
[SPARC64]: Use SLAB caches for TSB tables.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Don't kill the page allocator when growing a TSB.
David S. Miller [Sat, 18 Mar 2006 07:40:47 +0000 (23:40 -0800)]
[SPARC64]: Don't kill the page allocator when growing a TSB.

Try only lightly on > 1 order allocations.

If a grow fails, we are under memory pressure, so do not try
to grow the TSB for this address space any more.

If a > 0 order TSB allocation fails on a new fork, retry using
a 0 order allocation.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Randomize mm->mmap_base when PF_RANDOMIZE is set.
David S. Miller [Sat, 18 Mar 2006 01:42:57 +0000 (17:42 -0800)]
[SPARC64]: Randomize mm->mmap_base when PF_RANDOMIZE is set.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Increase top of 32-bit process stack.
David S. Miller [Sat, 18 Mar 2006 01:33:56 +0000 (17:33 -0800)]
[SPARC64]: Increase top of 32-bit process stack.

Put it one page below the top of the 32-bit address space.
This gives us ~16MB more address space to work with.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Top-down address space allocation for 32-bit tasks.
David S. Miller [Fri, 17 Mar 2006 22:41:03 +0000 (14:41 -0800)]
[SPARC64]: Top-down address space allocation for 32-bit tasks.

Currently allocations are very constrained for 32-bit processes.
It grows down-up from 0x70000000 to 0xf0000000 which gives about
2GB of stack + dynamic mmap() space.

So support the top-down method, and we need to override the
generic helper function in order to deal with D-cache coloring.

With these changes I was able to squeeze out a mmap() just over
3.6GB in size in a 32-bit process.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64] bbc_i2c: Fix cpu check and add missing module license.
David S. Miller [Fri, 17 Mar 2006 21:23:56 +0000 (13:23 -0800)]
[SPARC64] bbc_i2c: Fix cpu check and add missing module license.

Should allow cheetah_plus cpu types and don't taint
the kernel.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Fix and re-enable dynamic TSB sizing.
David S. Miller [Thu, 16 Mar 2006 10:02:32 +0000 (02:02 -0800)]
[SPARC64]: Fix and re-enable dynamic TSB sizing.

This is good for up to %50 performance improvement of some test cases.
The problem has been the race conditions, and hopefully I've plugged
them all up here.

1) There was a serious race in switch_mm() wrt. lazy TLB
   switching to and from kernel threads.

   We could erroneously skip a tsb_context_switch() and thus
   use a stale TSB across a TSB grow event.

   There is a big comment now in that function describing
   exactly how it can happen.

2) All code paths that do something with the TSB need to be
   guarded with the mm->context.lock spinlock.  This makes
   page table flushing paths properly synchronize with both
   TSB growing and TLB context changes.

3) TSB growing events are moved to the end of successful fault
   processing.  Previously it was in update_mmu_cache() but
   that is deadlock prone.  At the end of do_sparc64_fault()
   we hold no spinlocks that could deadlock the TSB grow
   sequence.  We also have dropped the address space semaphore.

While we're here, add prefetching to the copy_tsb() routine
and put it in assembler into the tsb.S file.  This piece of
code is quite time critical.

There are some small negative side effects to this code which
can be improved upon.  In particular we grab the mm->context.lock
even for the tsb insert done by update_mmu_cache() now and that's
a bit excessive.  We can get rid of that locking, and the same
lock taking in flush_tsb_user(), by disabling PSTATE_IE around
the whole operation including the capturing of the tsb pointer
and tsb_nentries value.  That would work because anyone growing
the TSB won't free up the old TSB until all cpus respond to the
TSB change cross call.

I'm not quite so confident in that optimization to put it in
right now, but eventually we might be able to and the description
is here for reference.

This code seems very solid now.  It passes several parallel GCC
bootstrap builds, and our favorite "nut cruncher" stress test which is
a full "make -j8192" build of a "make allmodconfig" kernel.  That puts
about 256 processes on each cpu's run queue, makes lots of process cpu
migrations occur, causes lots of page table and TLB flushing activity,
incurs many context version number changes, and it swaps the machine
real far out to disk even though there is 16GB of ram on this test
system. :-)

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SUNSU]: Fix missing spinlock initialization.
David S. Miller [Thu, 16 Mar 2006 08:55:30 +0000 (00:55 -0800)]
[SUNSU]: Fix missing spinlock initialization.

Caught by CONFIG_DEBUG_SPINLOCK.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[TG3]: Do not try to access NIC_SRAM_DATA_SIG on Sun parts.
David S. Miller [Tue, 14 Mar 2006 22:11:48 +0000 (14:11 -0800)]
[TG3]: Do not try to access NIC_SRAM_DATA_SIG on Sun parts.

Sun does't put an SEEPROM behind the tigon3 chip, among other things,
so accesses to these areas just give bus timeouts.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: First cut at VIS simulator for Niagara.
David S. Miller [Mon, 13 Mar 2006 09:27:34 +0000 (01:27 -0800)]
[SPARC64]: First cut at VIS simulator for Niagara.

Niagara does not implement some of the VIS instructions in
hardware, so we have to emulate them.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Fix system type in /proc/cpuinfo and remove bogus OBP check.
David S. Miller [Thu, 9 Mar 2006 01:18:19 +0000 (17:18 -0800)]
[SPARC64]: Fix system type in /proc/cpuinfo and remove bogus OBP check.

Report 'sun4v' when appropriate in /proc/cpuinfo

Remove all the verifications of the OBP version string.  Just
make sure it's there, and report it raw in the bootup logs and
via /proc/cpuinfo.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Add SMT scheduling support for Niagara.
David S. Miller [Thu, 9 Mar 2006 00:09:19 +0000 (16:09 -0800)]
[SPARC64]: Add SMT scheduling support for Niagara.

The mapping is a simple "(cpuid >> 2) == core" for now.
Later we'll add more sophisticated code that will walk
the sun4v machine description and figure this out from
there.

We should also add core mappings for jaguar and panther
processors.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Fix 32-bit truncation which broke sparsemem.
David S. Miller [Wed, 8 Mar 2006 23:57:03 +0000 (15:57 -0800)]
[SPARC64]: Fix 32-bit truncation which broke sparsemem.

The page->flags manipulations done by the D-cache dirty
state tracking was broken because the constants were not
marked with "UL" to make them 64-bit, which means we were
clobbering the upper 32-bits of page->flags all the time.

This doesn't jive well with sparsemem which stores the
section and indexing information in the top 32-bits of
page->flags.

This is yet another sparc64 bug which has been with us
forever.

While we're here, tidy up some things in bootmem_init()
and paginig_init():

1) Pass min_low_pfn to init_bootmem_node(), it's identical
   to (phys_base >> PAGE_SHIFT) but we should use consistent
   with the variable names we print in CONFIG_BOOTMEM_DEBUG

2) max_mapnr, although no longer used, was being set
   inaccurately, we shouldn't subtract pfn_base any more.

3) All the games with phys_base in the zones_*[] arrays
   we pass to free_area_init_node() are no longer necessary.

Thanks to Josh Grebe and Fabbione for the bug reports
and testing.  Fix also verified locally on an SB2500
which had a memory layout that triggered the same problem.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Move over to sparsemem.
David S. Miller [Wed, 8 Mar 2006 10:16:07 +0000 (02:16 -0800)]
[SPARC64]: Move over to sparsemem.

This has been pending for a long time, and the fact
that we waste a ton of ram on some configurations
kind of pushed things over the edge.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Fix new context version SMP handling.
David S. Miller [Tue, 7 Mar 2006 06:50:44 +0000 (22:50 -0800)]
[SPARC64]: Fix new context version SMP handling.

Don't piggy back the SMP receive signal code to do the
context version change handling.

Instead allocate another fixed PIL number for this
asynchronous cross-call.  We can't use smp_call_function()
because this thing is invoked with interrupts disabled
and a few spinlocks held.

Also, fix smp_call_function_mask() to count "cpus" correctly.
There is no guarentee that the local cpu is in the mask
yet that is exactly what this code was assuming.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Bulletproof MMU context locking.
David S. Miller [Tue, 7 Mar 2006 03:59:50 +0000 (19:59 -0800)]
[SPARC64]: Bulletproof MMU context locking.

1) Always spin_lock_init() in init_context().  The caller essentially
   clears it out, or copies the mm info from the parent.  In both
   cases we need to explicitly initialize the spinlock.

2) Always do explicit IRQ disabling while taking mm->context.lock
   and ctx_alloc_lock.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: kzalloc() conversion
Eric Sesterhenn [Mon, 6 Mar 2006 21:48:40 +0000 (13:48 -0800)]
[SPARC64]: kzalloc() conversion

this patch converts arch/sparc64 to kzalloc usage.
Crosscompile tested with allyesconfig.

Signed-off-by: Eric Sesterhenn <snakebyte@gmx.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Fix loop termination in mark_kpte_bitmap()
David S. Miller [Mon, 6 Mar 2006 06:18:50 +0000 (22:18 -0800)]
[SPARC64]: Fix loop termination in mark_kpte_bitmap()

If we were aligned, but didn't have at least 256MB left
to process, we would loop forever.

Thanks to fabbione for the report and testing the fix.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Simplify TSB insert checks.
David S. Miller [Mon, 6 Mar 2006 02:26:24 +0000 (18:26 -0800)]
[SPARC64]: Simplify TSB insert checks.

Don't try to avoid putting non-base page sized entries
into the user TSB.  It actually costs us more to check
this than it helps.

Eventually we'll have a multiple TSB scheme for user
processes.  Once a process starts using larger pages,
we'll allocate and use such a TSB.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: More SUN4V cpu mondo bug fixing.
David S. Miller [Fri, 3 Mar 2006 05:50:47 +0000 (21:50 -0800)]
[SPARC64]: More SUN4V cpu mondo bug fixing.

This cpu mondo sending interface isn't all that easy to
use correctly...

We were clearing out the wrong bits from the "mask" after getting
something other than EOK from the hypervisor.

It turns out the hypervisor can just be resent the same cpu_list[]
array, with the 0xffff "done" entries still in there, and it will do
the right thing.

So don't update or try to rebuild the cpu_list[] array to condense it.

This requires the "forward_progress" check to be done slightly
differently, but this new scheme is less bug prone than what we were
doing before.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Fix sun4v mna winfixup handling.
David S. Miller [Fri, 3 Mar 2006 04:42:53 +0000 (20:42 -0800)]
[SPARC64]: Fix sun4v mna winfixup handling.

We were clobbering a base register before we were done
using it.  Fix a comment typo while we're here.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Fix mini RTC driver reading.
David S. Miller [Fri, 3 Mar 2006 04:28:34 +0000 (20:28 -0800)]
[SPARC64]: Fix mini RTC driver reading.

Need to subtract 1900 from year and 1 from month before
giving it back to userspace.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Do not allow mapping pages within 4GB of 64-bit VA hole.
David S. Miller [Fri, 3 Mar 2006 02:12:27 +0000 (18:12 -0800)]
[SPARC64]: Do not allow mapping pages within 4GB of 64-bit VA hole.

The UltraSPARC T1 manual recommends this because the chip
could instruction prefetch into the VA hole, and this would
also make decoding  certain kinds of memory access traps
more difficult (because the chip sign extends certain pieces
of trap state).

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Fix _PAGE_EXEC handling.
David S. Miller [Thu, 2 Mar 2006 06:42:18 +0000 (22:42 -0800)]
[SPARC64]: Fix _PAGE_EXEC handling.

First of all, use the known _PAGE_EXEC_{4U,4V} value instead
of loading _PAGE_EXEC from memory.  We either know which one
to use by context, or we can code patch the test.

Next, we need to check executability of a PTE in the generic
TSB miss handler.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Fix typo in SUN4V D-TLB miss handler.
David S. Miller [Thu, 2 Mar 2006 06:27:09 +0000 (22:27 -0800)]
[SPARC64]: Fix typo in SUN4V D-TLB miss handler.

Should put FAULT_CODE_DTLB into %g3 not FAULT_CODE_ITLB.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Kill bogus function externs in asm/pgtable.h
David S. Miller [Thu, 2 Mar 2006 06:25:43 +0000 (22:25 -0800)]
[SPARC64]: Kill bogus function externs in asm/pgtable.h

These are all implemented inline earlier in the file.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Add mini-RTC driver for Starfire and SUN4V.
David S. Miller [Thu, 2 Mar 2006 01:32:46 +0000 (17:32 -0800)]
[SPARC64]: Add mini-RTC driver for Starfire and SUN4V.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Fix bugs in SUN4V cpu mondo dispatch.
David S. Miller [Tue, 28 Feb 2006 23:10:26 +0000 (15:10 -0800)]
[SPARC64]: Fix bugs in SUN4V cpu mondo dispatch.

There were several bugs in the SUN4V cpu mondo dispatch code.

In fact, if we ever got a EWOULDBLOCK or other error from
the hypervisor call, we'd potentially send a cpu mondo multiple
times to the same cpu and even worse we could loop until the
timeout resending the same mondo over and over to such cpus.

So let's bulletproof this thing as follows:

1) Implement cpu_mondo_send() and cpu_state() hypervisor calls
   in arch/sparc64/kernel/entry.S, add prototypes to asm/hypervisor.h

2) Don't build and update the cpulist using inline functions, this
   was causing the cpu mask to not get updated in the caller.

3) Disable interrupts during the entire mondo send, otherwise our
   cpu list and/or mondo block could get overwritten if we take
   an interrupt and do a cpu mondo send on the current cpu.

4) Check for all possible error return types from the cpu_mondo_send()
   hypervisor call.  In particular:

   HV_EOK) Our work is done, all cpus have received the mondo.
   HV_CPUERROR) One or more of the cpus in the cpu list we passed
                to the hypervisor are in error state.  Use cpu_state()
                calls over the entries in the cpu list to see which
ones.  Record them in "error_mask" and report this
after we are done sending the mondo to cpus which are
not in error state.
   HV_EWOULDBLOCK) We need to keep trying.

   Any other error we consider fatal, we report the event and exit
   immediately.

5) We only timeout if forward progress is not made.  Forward progress
   is defined as having at least one cpu get the mondo successfully
   in a given cpu_mondo_send() call.  Otherwise we bump a counter
   and delay a little.  If the counter hits a limit, we signal an
   error and report the event.

Also, smp_call_function_mask() error handling reports the number
of cpus incorrectly.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Fix bugs in SMP TLB context version expiration handling.
David S. Miller [Tue, 28 Feb 2006 01:56:51 +0000 (17:56 -0800)]
[SPARC64]: Fix bugs in SMP TLB context version expiration handling.

1) We must flush the TLB, duh.

2) Even if the sw context was seen to be valid, the local cpu's
   hw context can be out of date, so reload it unconditionally.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Fix indexing into kpte_linear_bitmap.
David S. Miller [Mon, 27 Feb 2006 07:09:37 +0000 (23:09 -0800)]
[SPARC64]: Fix indexing into kpte_linear_bitmap.

Need to shift back up by 3 bits to get 8-byte entry
index.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Use 13-bit context size always.
David S. Miller [Mon, 27 Feb 2006 04:37:41 +0000 (20:37 -0800)]
[SPARC64]: Use 13-bit context size always.

We no longer have the problems that require using the smaller
sizes.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Avoid dcache-dirty page state management on sun4v.
David S. Miller [Mon, 27 Feb 2006 03:44:50 +0000 (19:44 -0800)]
[SPARC64]: Avoid dcache-dirty page state management on sun4v.

It is totally wasted work, since we have no D-cache aliasing
issues on sun4v.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Bulletproof hypervisor TLB flushing.
David S. Miller [Mon, 27 Feb 2006 03:31:49 +0000 (19:31 -0800)]
[SPARC64]: Bulletproof hypervisor TLB flushing.

Check TLB flush hypervisor calls for errors and report them.

Pass HV_MMU_ALL always for now, we can add back the optimization
to avoid the I-TLB flush later.

Always explicitly page align the virtual address arguments.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Report mondo error correctly in hypervisor_xcall_deliver().
David S. Miller [Sun, 26 Feb 2006 22:51:29 +0000 (14:51 -0800)]
[SPARC64]: Report mondo error correctly in hypervisor_xcall_deliver().

It's in "arg0" not "func".

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Niagara optimized XOR functions for RAID.
David S. Miller [Sun, 26 Feb 2006 01:16:29 +0000 (17:16 -0800)]
[SPARC64]: Niagara optimized XOR functions for RAID.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Fix binfmt_aout32.c build.
Andrew Morton [Fri, 24 Feb 2006 21:21:18 +0000 (13:21 -0800)]
[SPARC64]: Fix binfmt_aout32.c build.

Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: destroy_context() needs to disable interrupts.
David S. Miller [Fri, 24 Feb 2006 05:40:15 +0000 (21:40 -0800)]
[SPARC64]: destroy_context() needs to disable interrupts.

get_new_mmu_context() can be invoked from interrupt context
now for the new SMP version wrap handling.

So disable interrupt while taking ctx_alloc_lock in destroy_context()
so we don't deadlock.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Fix TLB context allocation with SMT style shared TLBs.
David S. Miller [Thu, 23 Feb 2006 22:19:28 +0000 (14:19 -0800)]
[SPARC64]: Fix TLB context allocation with SMT style shared TLBs.

The context allocation scheme we use depends upon there being a 1<-->1
mapping from cpu to physical TLB for correctness.  Chips like Niagara
break this assumption.

So what we do is notify all cpus with a cross call when the context
version number changes, and if necessary this makes them allocate
a valid context for the address space they are running at the time.

Stress tested with make -j1024, make -j2048, and make -j4096 kernel
builds on a 32-strand, 8 core, T2000 with 16GB of ram.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Put syscall tables after trap table.
David S. Miller [Thu, 23 Feb 2006 10:28:25 +0000 (02:28 -0800)]
[SPARC64]: Put syscall tables after trap table.

Otherwise with too much stuff enabled in the kernel config
we can end up with an unaligned trap table.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Export _PAGE_E and _PAGE_CACHE to modules.
David S. Miller [Thu, 23 Feb 2006 09:55:55 +0000 (01:55 -0800)]
[SPARC64]: Export _PAGE_E and _PAGE_CACHE to modules.

SBUS flash driver needs it.

Noticed by Fabbione.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Fix %tstate ASI handling in start_thread{,32}()
David S. Miller [Thu, 23 Feb 2006 00:20:11 +0000 (16:20 -0800)]
[SPARC64]: Fix %tstate ASI handling in start_thread{,32}()

Niagara helps us find a ancient bug in the sparc64 port :-)

The ASI_* values are plain constant defines, thus signed 32-bit
on sparc64.  To put shift this into the regs->tstate value we were
doing or'ing "(ASI_PNF << 24)" into there.

ASI_PNF is 0x82 and shifted left by 24 makes that topmost bit the
sign bit in a 32-bit value.  This would get sign extended to 64-bits
and thus corrupt the top-half of the reg->tstate value.

This never caused problems in pre-Niagara cpus because the only thing
up there were the condition code values.  But Niagara has the global
register level field, and this all 1's value is illegal there so
Niagara gives an illegal instruction trap due to this bug.

I'm pretty sure this bug is about as old as the sparc64 port itself.

This also points out that we weren't setting ASI_PNF for 32-bit tasks.
We should, so fix that while we're here.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Drop %gl to 0 before re-enabling PSTATE_IE in rtrap
David S. Miller [Thu, 23 Feb 2006 00:15:45 +0000 (16:15 -0800)]
[SPARC64]: Drop %gl to 0 before re-enabling PSTATE_IE in rtrap

If we take a window fault, on SUN4V set %gl to zero before we
turn PSTATE_IE back on in %pstate.  Otherwise if we take an
interrupt we'll end up with corrupt register state.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Create a seperate kernel TSB for 4MB/256MB mappings.
David S. Miller [Wed, 22 Feb 2006 06:31:11 +0000 (22:31 -0800)]
[SPARC64]: Create a seperate kernel TSB for 4MB/256MB mappings.

It can map all of the linear kernel mappings with zero TSB hash
conflicts for systems with 16GB or less ram.  In such cases, on
SUN4V, once we load up this TSB the first time with all the
mappings, we never take a linear kernel mapping TLB miss ever
again, the hypervisor handles them all.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Make use of Niagara 256MB PTEs for kernel mappings.
David S. Miller [Wed, 22 Feb 2006 04:51:13 +0000 (20:51 -0800)]
[SPARC64]: Make use of Niagara 256MB PTEs for kernel mappings.

We use a bitmap, one bit for every 256MB of memory.  If the
bit is set we can use a 256MB PTE for linear mappings, else
we have to use a 4MB PTE.

SUN4V support is there, and we can very easily add support
for Panther cpu 256MB PTEs in the future.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Use sun4v_cpu_idle() in cpu_idle() on SUN4V.
David S. Miller [Wed, 22 Feb 2006 00:55:23 +0000 (16:55 -0800)]
[SPARC64]: Use sun4v_cpu_idle() in cpu_idle() on SUN4V.

We have to turn off the "polling nrflag" bit when we sleep
the cpu like this, so that we'll get a cross-cpu interrupt
to wake the processor up from the yield.

We also have to disable PSTATE_IE in %pstate around the yield
call and recheck need_resched() in order to avoid any races.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64] math-emu: Delete debugging printk left by previous commit.
David S. Miller [Wed, 22 Feb 2006 00:53:26 +0000 (16:53 -0800)]
[SPARC64] math-emu: Delete debugging printk left by previous commit.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Add sun4v_cpu_yield().
David S. Miller [Tue, 21 Feb 2006 23:42:09 +0000 (15:42 -0800)]
[SPARC64]: Add sun4v_cpu_yield().

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Kill cpudata->idle_volume.
David S. Miller [Tue, 21 Feb 2006 23:41:01 +0000 (15:41 -0800)]
[SPARC64]: Kill cpudata->idle_volume.

Set, but never used.

We used to use this for dynamic IRQ retargetting, but that
code died a long time ago.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Niagara optimized memset/bzero/clear_user.
David S. Miller [Tue, 21 Feb 2006 22:29:42 +0000 (14:29 -0800)]
[SPARC64]: Niagara optimized memset/bzero/clear_user.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Pass multiple CPUs at once to hypervisor cross-call API.
David S. Miller [Tue, 21 Feb 2006 21:59:47 +0000 (13:59 -0800)]
[SPARC64]: Pass multiple CPUs at once to hypervisor cross-call API.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Args to SUNW,set-trap-table are 64-bit.
David S. Miller [Tue, 21 Feb 2006 06:56:01 +0000 (22:56 -0800)]
[SPARC64]: Args to SUNW,set-trap-table are 64-bit.

They were getting truncated to 32-bit and this is very bad
when your MMU fault status area is in physical memory above
4GB on SUN4V.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Handle unimplemented FPU square-root on Niagara.
David S. Miller [Tue, 21 Feb 2006 00:02:24 +0000 (16:02 -0800)]
[SPARC64]: Handle unimplemented FPU square-root on Niagara.

The math-emu code only expects unfinished fpop traps when
emulating FPU sqrt instructions on pre-Niagara chips.
On Niagara we can get unimplemented fpop, so handle that.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC] serial: Make sure sysfs nodes get named correctly.
David S. Miller [Mon, 20 Feb 2006 22:39:16 +0000 (14:39 -0800)]
[SPARC] serial: Make sure sysfs nodes get named correctly.

Because we play this trick where we use ttyS? in increasing minor
numbers for different sunfoo.c drivers, we have to inform the TTY
layer of this.

Do so by setting the tty->name_base appropriately.

Probably there should be a generic way to do this in the serial core,
but for now...

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Typo in sun4v_data_access_exception log message.
David S. Miller [Mon, 20 Feb 2006 09:50:09 +0000 (01:50 -0800)]
[SPARC64]: Typo in sun4v_data_access_exception log message.

Should be "Dax" not "Iax".

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Handle zero-length map requests in pci_sun4v.c
David S. Miller [Mon, 20 Feb 2006 09:42:51 +0000 (01:42 -0800)]
[SPARC64]: Handle zero-length map requests in pci_sun4v.c

By simply changing the do-while loop into a plain
while loop.

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[SPARC64]: Kill stray PGLIST_NENTS check in pci_sun4v.c
David S. Miller [Mon, 20 Feb 2006 09:09:01 +0000 (01:09 -0800)]
[SPARC64]: Kill stray PGLIST_NENTS check in pci_sun4v.c

I forgot to remove the one in pci_4v_map_sg() during the
iommu batching commit.

Signed-off-by: David S. Miller <davem@davemloft.net>