]> git.ipfire.org Git - thirdparty/kernel/stable-queue.git/commitdiff
3.9-stable patches
authorGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 18 Jul 2013 05:25:08 +0000 (22:25 -0700)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 18 Jul 2013 05:25:08 +0000 (22:25 -0700)
added patches:
acpica-do-not-use-extended-sleep-registers-unless-hw-reduced-bit-is-set.patch
acpi-ec-add-hp-folio-13-to-ec_dmi_table-in-order-to-skip-dsdt-scan.patch
acpi-pm-fix-corner-case-in-acpi_bus_update_power.patch
ahci-remove-pmp-link-online-check-in-fbs-eh.patch
btrfs-fix-estale-with-btrfs-send.patch
btrfs-only-do-the-tree_mod_log_free_eb-if-this-is-our-last.patch
drivers-rtc-rtc-rv3029c2.c-fix-disabling-aie-irq.patch
genirq-fix-can_request_irq-for-irqs-without-an-action.patch
nbd-correct-disconnect-behavior.patch
ocfs2-xattr-fix-inlined-xattr-reflink.patch
pci-finish-sr-iov-vf-setup-before-adding-the-device.patch
pci-fix-refcount-issue-in-pci_create_root_bus-error-recovery-path.patch
timer-fix-jiffies-wrap-behavior-of-round_jiffies_common.patch
xen-pcifront-deal-with-toolstack-missing-xenbusstateclosing-state.patch
xen-time-remove-blocked-time-accounting-from-xen-clockchip.patch

17 files changed:
queue-3.9/acpi-ec-add-hp-folio-13-to-ec_dmi_table-in-order-to-skip-dsdt-scan.patch [new file with mode: 0644]
queue-3.9/acpi-pm-fix-corner-case-in-acpi_bus_update_power.patch [new file with mode: 0644]
queue-3.9/acpica-do-not-use-extended-sleep-registers-unless-hw-reduced-bit-is-set.patch [new file with mode: 0644]
queue-3.9/ahci-remove-pmp-link-online-check-in-fbs-eh.patch [new file with mode: 0644]
queue-3.9/btrfs-fix-estale-with-btrfs-send.patch [new file with mode: 0644]
queue-3.9/btrfs-only-do-the-tree_mod_log_free_eb-if-this-is-our-last.patch [new file with mode: 0644]
queue-3.9/drivers-rtc-rtc-rv3029c2.c-fix-disabling-aie-irq.patch [new file with mode: 0644]
queue-3.9/ext4-check-error-return-from-ext4_write_inline_data_end.patch [deleted file]
queue-3.9/genirq-fix-can_request_irq-for-irqs-without-an-action.patch [new file with mode: 0644]
queue-3.9/nbd-correct-disconnect-behavior.patch [new file with mode: 0644]
queue-3.9/ocfs2-xattr-fix-inlined-xattr-reflink.patch [new file with mode: 0644]
queue-3.9/pci-finish-sr-iov-vf-setup-before-adding-the-device.patch [new file with mode: 0644]
queue-3.9/pci-fix-refcount-issue-in-pci_create_root_bus-error-recovery-path.patch [new file with mode: 0644]
queue-3.9/series
queue-3.9/timer-fix-jiffies-wrap-behavior-of-round_jiffies_common.patch [new file with mode: 0644]
queue-3.9/xen-pcifront-deal-with-toolstack-missing-xenbusstateclosing-state.patch [new file with mode: 0644]
queue-3.9/xen-time-remove-blocked-time-accounting-from-xen-clockchip.patch [new file with mode: 0644]

diff --git a/queue-3.9/acpi-ec-add-hp-folio-13-to-ec_dmi_table-in-order-to-skip-dsdt-scan.patch b/queue-3.9/acpi-ec-add-hp-folio-13-to-ec_dmi_table-in-order-to-skip-dsdt-scan.patch
new file mode 100644 (file)
index 0000000..2629ccd
--- /dev/null
@@ -0,0 +1,37 @@
+From eff9a4b62b14cf0d9913e3caf1f26f8b7a6105c9 Mon Sep 17 00:00:00 2001
+From: Lan Tianyu <tianyu.lan@intel.com>
+Date: Wed, 5 Jun 2013 02:27:51 +0000
+Subject: ACPI / EC: Add HP Folio 13 to ec_dmi_table in order to skip DSDT scan
+
+From: Lan Tianyu <tianyu.lan@intel.com>
+
+commit eff9a4b62b14cf0d9913e3caf1f26f8b7a6105c9 upstream.
+
+HP Folio 13's BIOS defines CMOS RTC Operation Region and the EC's
+_REG method will access that region.  To allow the CMOS RTC region
+handler to be installed before the EC _REG method is first invoked,
+add ec_skip_dsdt_scan() as HP Folio 13's callback to ec_dmi_table.
+
+References: https://bugzilla.kernel.org/show_bug.cgi?id=54621
+Reported-and-tested-by: Stefan Nagy <public@stefan-nagy.at>
+Signed-off-by: Lan Tianyu <tianyu.lan@intel.com>
+Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
+Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+
+---
+ drivers/acpi/ec.c |    4 ++++
+ 1 file changed, 4 insertions(+)
+
+--- a/drivers/acpi/ec.c
++++ b/drivers/acpi/ec.c
+@@ -983,6 +983,10 @@ static struct dmi_system_id __initdata e
+       ec_enlarge_storm_threshold, "CLEVO hardware", {
+       DMI_MATCH(DMI_SYS_VENDOR, "CLEVO Co."),
+       DMI_MATCH(DMI_PRODUCT_NAME, "M720T/M730T"),}, NULL},
++      {
++      ec_skip_dsdt_scan, "HP Folio 13", {
++      DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
++      DMI_MATCH(DMI_PRODUCT_NAME, "HP Folio 13"),}, NULL},
+       {},
+ };
diff --git a/queue-3.9/acpi-pm-fix-corner-case-in-acpi_bus_update_power.patch b/queue-3.9/acpi-pm-fix-corner-case-in-acpi_bus_update_power.patch
new file mode 100644 (file)
index 0000000..fa5907d
--- /dev/null
@@ -0,0 +1,68 @@
+From 91bdad0b6237c25a7bf8fd4604d0cc64a2005a23 Mon Sep 17 00:00:00 2001
+From: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com>
+Date: Thu, 4 Jul 2013 13:22:11 +0200
+Subject: ACPI / PM: Fix corner case in acpi_bus_update_power()
+
+From: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com>
+
+commit 91bdad0b6237c25a7bf8fd4604d0cc64a2005a23 upstream.
+
+The role of acpi_bus_update_power() is to update the given ACPI
+device object's power.state field to reflect the current physical
+state of the device (as inferred from the configuration of power
+resources and _PSC, if available).  For this purpose it calls
+acpi_device_set_power() that should update the power resources'
+reference counters and set power.state as appropriate.  However,
+that doesn't work if the "new" state is D1, D2 or D3hot and the
+the current value of power.state means D3cold, because in that
+case acpi_device_set_power() will refuse to transition the device
+from D3cold to non-D0.
+
+To address this problem, make acpi_bus_update_power() call
+acpi_power_transition() directly to update the power resources'
+reference counters and only use acpi_device_set_power() to put
+the device into D0 if the current physical state of it cannot
+be determined.
+
+Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
+Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+
+---
+ drivers/acpi/device_pm.c |   23 ++++++++++++++++++-----
+ 1 file changed, 18 insertions(+), 5 deletions(-)
+
+--- a/drivers/acpi/device_pm.c
++++ b/drivers/acpi/device_pm.c
+@@ -295,14 +295,27 @@ int acpi_bus_update_power(acpi_handle ha
+       if (result)
+               return result;
+-      if (state == ACPI_STATE_UNKNOWN)
++      if (state == ACPI_STATE_UNKNOWN) {
+               state = ACPI_STATE_D0;
+-
+-      result = acpi_device_set_power(device, state);
+-      if (!result && state_p)
++              result = acpi_device_set_power(device, state);
++              if (result)
++                      return result;
++      } else {
++              if (device->power.flags.power_resources) {
++                      /*
++                       * We don't need to really switch the state, bu we need
++                       * to update the power resources' reference counters.
++                       */
++                      result = acpi_power_transition(device, state);
++                      if (result)
++                              return result;
++              }
++              device->power.state = state;
++      }
++      if (state_p)
+               *state_p = state;
+-      return result;
++      return 0;
+ }
+ EXPORT_SYMBOL_GPL(acpi_bus_update_power);
diff --git a/queue-3.9/acpica-do-not-use-extended-sleep-registers-unless-hw-reduced-bit-is-set.patch b/queue-3.9/acpica-do-not-use-extended-sleep-registers-unless-hw-reduced-bit-is-set.patch
new file mode 100644 (file)
index 0000000..c431bcc
--- /dev/null
@@ -0,0 +1,48 @@
+From 7cec7048fe22e3e92389da2cd67098f6c4284e7f Mon Sep 17 00:00:00 2001
+From: Lv Zheng <lv.zheng@intel.com>
+Date: Sat, 8 Jun 2013 00:59:18 +0000
+Subject: ACPICA: Do not use extended sleep registers unless HW-reduced bit is set
+
+From: Lv Zheng <lv.zheng@intel.com>
+
+commit 7cec7048fe22e3e92389da2cd67098f6c4284e7f upstream.
+
+Previous implementation incorrectly used the ACPI 5.0 extended
+sleep registers if they were simply populated. This caused
+problems on some non-HW-reduced machines. As per the ACPI spec,
+they should only be used if the HW-reduced bit is set.  Lv Zheng,
+ACPICA BZ 1020.
+
+References: https://bugzilla.kernel.org/show_bug.cgi?id=54181
+References: https://bugs.acpica.org/show_bug.cgi?id=1020
+Reported-by: Daniel Rowe <bart@fathom13.com>
+Bisected-by: Brint E. Kriebel <kernel@bekit.net>
+Signed-off-by: Lv Zheng <lv.zheng@intel.com>
+Signed-off-by: Bob Moore <robert.moore@intel.com>
+Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
+Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+
+---
+ drivers/acpi/acpica/hwxfsleep.c |    8 +++++---
+ 1 file changed, 5 insertions(+), 3 deletions(-)
+
+--- a/drivers/acpi/acpica/hwxfsleep.c
++++ b/drivers/acpi/acpica/hwxfsleep.c
+@@ -240,12 +240,14 @@ static acpi_status acpi_hw_sleep_dispatc
+           &acpi_sleep_dispatch[function_id];
+ #if (!ACPI_REDUCED_HARDWARE)
+-
+       /*
+        * If the Hardware Reduced flag is set (from the FADT), we must
+-       * use the extended sleep registers
++       * use the extended sleep registers (FADT). Note: As per the ACPI
++       * specification, these extended registers are to be used for HW-reduced
++       * platforms only. They are not general-purpose replacements for the
++       * legacy PM register sleep support.
+        */
+-      if (acpi_gbl_reduced_hardware || acpi_gbl_FADT.sleep_control.address) {
++      if (acpi_gbl_reduced_hardware) {
+               status = sleep_functions->extended_function(sleep_state);
+       } else {
+               /* Legacy sleep */
diff --git a/queue-3.9/ahci-remove-pmp-link-online-check-in-fbs-eh.patch b/queue-3.9/ahci-remove-pmp-link-online-check-in-fbs-eh.patch
new file mode 100644 (file)
index 0000000..40651f4
--- /dev/null
@@ -0,0 +1,34 @@
+From 912b9ac683b112615d5605686f1dc086402ce9f7 Mon Sep 17 00:00:00 2001
+From: Shane Huang <shane.huang@amd.com>
+Date: Sat, 8 Jun 2013 16:00:16 +0800
+Subject: ahci: remove pmp link online check in FBS EH
+
+From: Shane Huang <shane.huang@amd.com>
+
+commit 912b9ac683b112615d5605686f1dc086402ce9f7 upstream.
+
+ata_link_online() check in ahci_error_intr() is unnecessary, it should
+be removed otherwise may lead to lockup with FBS enabled PMP.
+http://marc.info/?l=linux-ide&m=137050421603272&w=2
+
+Reported-by: Yu Liu <liuyu.ac@gmail.com>
+Signed-off-by: Shane Huang <shane.huang@amd.com>
+Signed-off-by: Tejun Heo <tj@kernel.org>
+Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+
+---
+ drivers/ata/libahci.c |    3 +--
+ 1 file changed, 1 insertion(+), 2 deletions(-)
+
+--- a/drivers/ata/libahci.c
++++ b/drivers/ata/libahci.c
+@@ -1560,8 +1560,7 @@ static void ahci_error_intr(struct ata_p
+               u32 fbs = readl(port_mmio + PORT_FBS);
+               int pmp = fbs >> PORT_FBS_DWE_OFFSET;
+-              if ((fbs & PORT_FBS_SDE) && (pmp < ap->nr_pmp_links) &&
+-                  ata_link_online(&ap->pmp_link[pmp])) {
++              if ((fbs & PORT_FBS_SDE) && (pmp < ap->nr_pmp_links)) {
+                       link = &ap->pmp_link[pmp];
+                       fbs_need_dec = true;
+               }
diff --git a/queue-3.9/btrfs-fix-estale-with-btrfs-send.patch b/queue-3.9/btrfs-fix-estale-with-btrfs-send.patch
new file mode 100644 (file)
index 0000000..ded9cfa
--- /dev/null
@@ -0,0 +1,73 @@
+From 139f807a1eba1e484941a98fb93ee32ad859a6a1 Mon Sep 17 00:00:00 2001
+From: Josef Bacik <jbacik@fusionio.com>
+Date: Mon, 20 May 2013 11:26:50 -0400
+Subject: Btrfs: fix estale with btrfs send
+
+From: Josef Bacik <jbacik@fusionio.com>
+
+commit 139f807a1eba1e484941a98fb93ee32ad859a6a1 upstream.
+
+This fixes bugzilla 57491.  If we take a snapshot of a fs with a unlink ongoing
+and then try to send that root we will run into problems.  When comparing with a
+parent root we will search the parents and the send roots commit_root, which if
+we've just created the snapshot will include the file that needs to be evicted
+by the orphan cleanup.  So when we find a changed extent we will try and copy
+that info into the send stream, but when we lookup the inode we use the normal
+root, which no longer has the inode because the orphan cleanup deleted it.  The
+best solution I have for this is to check our otransid with the generation of
+the commit root and if they match just commit the transaction again, that way we
+get the changes from the orphan cleanup.  With this patch the reproducer I made
+for this bugzilla no longer returns ESTALE when trying to do the send.  Thanks,
+
+Reported-by: Chris Wilson <jakdaw@gmail.com>
+Signed-off-by: Josef Bacik <jbacik@fusionio.com>
+Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+
+---
+ fs/btrfs/send.c |   35 +++++++++++++++++++++++++++++++++++
+ 1 file changed, 35 insertions(+)
+
+--- a/fs/btrfs/send.c
++++ b/fs/btrfs/send.c
+@@ -4579,6 +4579,41 @@ long btrfs_ioctl_send(struct file *mnt_f
+       send_root = BTRFS_I(file_inode(mnt_file))->root;
+       fs_info = send_root->fs_info;
++      /*
++       * This is done when we lookup the root, it should already be complete
++       * by the time we get here.
++       */
++      WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE);
++
++      /*
++       * If we just created this root we need to make sure that the orphan
++       * cleanup has been done and committed since we search the commit root,
++       * so check its commit root transid with our otransid and if they match
++       * commit the transaction to make sure everything is updated.
++       */
++      down_read(&send_root->fs_info->extent_commit_sem);
++      if (btrfs_header_generation(send_root->commit_root) ==
++          btrfs_root_otransid(&send_root->root_item)) {
++              struct btrfs_trans_handle *trans;
++
++              up_read(&send_root->fs_info->extent_commit_sem);
++
++              trans = btrfs_attach_transaction_barrier(send_root);
++              if (IS_ERR(trans)) {
++                      if (PTR_ERR(trans) != -ENOENT) {
++                              ret = PTR_ERR(trans);
++                              goto out;
++                      }
++                      /* ENOENT means theres no transaction */
++              } else {
++                      ret = btrfs_commit_transaction(trans, send_root);
++                      if (ret)
++                              goto out;
++              }
++      } else {
++              up_read(&send_root->fs_info->extent_commit_sem);
++      }
++
+       arg = memdup_user(arg_, sizeof(*arg));
+       if (IS_ERR(arg)) {
+               ret = PTR_ERR(arg);
diff --git a/queue-3.9/btrfs-only-do-the-tree_mod_log_free_eb-if-this-is-our-last.patch b/queue-3.9/btrfs-only-do-the-tree_mod_log_free_eb-if-this-is-our-last.patch
new file mode 100644 (file)
index 0000000..e98ce09
--- /dev/null
@@ -0,0 +1,38 @@
+From 7fb7d76f96bfcbea25007d190ba828b18e13d29d Mon Sep 17 00:00:00 2001
+From: Josef Bacik <jbacik@fusionio.com>
+Date: Mon, 1 Jul 2013 16:10:16 -0400
+Subject: Btrfs: only do the tree_mod_log_free_eb if this is our last
+ ref
+
+From: Josef Bacik <jbacik@fusionio.com>
+
+commit 7fb7d76f96bfcbea25007d190ba828b18e13d29d upstream.
+
+There is another bug in the tree mod log stuff in that we're calling
+tree_mod_log_free_eb every single time a block is cow'ed.  The problem with this
+is that if this block is shared by multiple snapshots we will call this multiple
+times per block, so if we go to rewind the mod log for this block we'll BUG_ON()
+in __tree_mod_log_rewind because we try to rewind a free twice.  We only want to
+call tree_mod_log_free_eb if we are actually freeing the block.  With this patch
+I no longer hit the panic in __tree_mod_log_rewind.  Thanks,
+
+Reviewed-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
+Signed-off-by: Josef Bacik <jbacik@fusionio.com>
+Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+
+---
+ fs/btrfs/ctree.c |    3 ++-
+ 1 file changed, 2 insertions(+), 1 deletion(-)
+
+--- a/fs/btrfs/ctree.c
++++ b/fs/btrfs/ctree.c
+@@ -1049,7 +1049,8 @@ static noinline int __btrfs_cow_block(st
+               btrfs_set_node_ptr_generation(parent, parent_slot,
+                                             trans->transid);
+               btrfs_mark_buffer_dirty(parent);
+-              tree_mod_log_free_eb(root->fs_info, buf);
++              if (last_ref)
++                      tree_mod_log_free_eb(root->fs_info, buf);
+               btrfs_free_tree_block(trans, root, buf, parent_start,
+                                     last_ref);
+       }
diff --git a/queue-3.9/drivers-rtc-rtc-rv3029c2.c-fix-disabling-aie-irq.patch b/queue-3.9/drivers-rtc-rtc-rv3029c2.c-fix-disabling-aie-irq.patch
new file mode 100644 (file)
index 0000000..45fdf9a
--- /dev/null
@@ -0,0 +1,34 @@
+From 29ecd78c0fd6ee05f2c6b07b23823a6ae43c13ff Mon Sep 17 00:00:00 2001
+From: Axel Lin <axel.lin@ingics.com>
+Date: Wed, 3 Jul 2013 15:06:45 -0700
+Subject: drivers/rtc/rtc-rv3029c2.c: fix disabling AIE irq
+
+From: Axel Lin <axel.lin@ingics.com>
+
+commit 29ecd78c0fd6ee05f2c6b07b23823a6ae43c13ff upstream.
+
+In the disable AIE irq code path, current code passes "1" to enable
+parameter of rv3029c2_rtc_i2c_alarm_set_irq().  Thus it does not disable
+AIE irq.
+
+Signed-off-by: Axel Lin <axel.lin@ingics.com>
+Acked-by: Heiko Schocher <hs@denx.de>
+Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
+Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
+Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+
+---
+ drivers/rtc/rtc-rv3029c2.c |    2 +-
+ 1 file changed, 1 insertion(+), 1 deletion(-)
+
+--- a/drivers/rtc/rtc-rv3029c2.c
++++ b/drivers/rtc/rtc-rv3029c2.c
+@@ -310,7 +310,7 @@ static int rv3029c2_rtc_i2c_set_alarm(st
+               dev_dbg(&client->dev, "alarm IRQ armed\n");
+       } else {
+               /* disable AIE irq */
+-              ret = rv3029c2_rtc_i2c_alarm_set_irq(client, 1);
++              ret = rv3029c2_rtc_i2c_alarm_set_irq(client, 0);
+               if (ret)
+                       return ret;
diff --git a/queue-3.9/ext4-check-error-return-from-ext4_write_inline_data_end.patch b/queue-3.9/ext4-check-error-return-from-ext4_write_inline_data_end.patch
deleted file mode 100644 (file)
index 8622963..0000000
+++ /dev/null
@@ -1,41 +0,0 @@
-From 42c832debbbf819f6c4ad8601baa559c44105ba4 Mon Sep 17 00:00:00 2001
-From: Theodore Ts'o <tytso@mit.edu>
-Date: Mon, 1 Jul 2013 08:12:39 -0400
-Subject: ext4: check error return from ext4_write_inline_data_end()
-
-From: Theodore Ts'o <tytso@mit.edu>
-
-commit 42c832debbbf819f6c4ad8601baa559c44105ba4 upstream.
-
-The function ext4_write_inline_data_end() can return an error.  So we
-need to assign it to a signed integer variable to check for an error
-return (since copied is an unsigned int).
-
-Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
-Cc: Zheng Liu <wenqing.lz@taobao.com>
-Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
-
----
- fs/ext4/inode.c |   11 +++++++----
- 1 file changed, 7 insertions(+), 4 deletions(-)
-
---- a/fs/ext4/inode.c
-+++ b/fs/ext4/inode.c
-@@ -1095,10 +1095,13 @@ static int ext4_generic_write_end(struct
-       struct inode *inode = mapping->host;
-       handle_t *handle = ext4_journal_current_handle();
--      if (ext4_has_inline_data(inode))
--              copied = ext4_write_inline_data_end(inode, pos, len,
--                                                  copied, page);
--      else
-+      if (ext4_has_inline_data(inode)) {
-+              ret = ext4_write_inline_data_end(inode, pos, len,
-+                                               copied, page);
-+              if (ret < 0)
-+                      goto errout;
-+              copied = ret;
-+      } else
-               copied = block_write_end(file, mapping, pos,
-                                        len, copied, page, fsdata);
diff --git a/queue-3.9/genirq-fix-can_request_irq-for-irqs-without-an-action.patch b/queue-3.9/genirq-fix-can_request_irq-for-irqs-without-an-action.patch
new file mode 100644 (file)
index 0000000..3eaca83
--- /dev/null
@@ -0,0 +1,44 @@
+From 2779db8d37d4b542d9ca2575f5f178dbeaca6c86 Mon Sep 17 00:00:00 2001
+From: Ben Hutchings <ben@decadent.org.uk>
+Date: Fri, 28 Jun 2013 02:40:30 +0100
+Subject: genirq: Fix can_request_irq() for IRQs without an action
+
+From: Ben Hutchings <ben@decadent.org.uk>
+
+commit 2779db8d37d4b542d9ca2575f5f178dbeaca6c86 upstream.
+
+Commit 02725e7471b8 ('genirq: Use irq_get/put functions'),
+inadvertently changed can_request_irq() to return 0 for IRQs that have
+no action.  This causes pcibios_lookup_irq() to select only IRQs that
+already have an action with IRQF_SHARED set, or to fail if there are
+none.  Change can_request_irq() to return 1 for IRQs that have no
+action (if the first two conditions are met).
+
+Reported-by: Bjarni Ingi Gislason <bjarniig@rhi.hi.is>
+Tested-by: Bjarni Ingi Gislason <bjarniig@rhi.hi.is> (against 3.2)
+Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
+Cc: 709647@bugs.debian.org
+Link: http://bugs.debian.org/709647
+Link: http://lkml.kernel.org/r/1372383630.23847.40.camel@deadeye.wl.decadent.org.uk
+Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
+Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+
+---
+ kernel/irq/manage.c |    6 +++---
+ 1 file changed, 3 insertions(+), 3 deletions(-)
+
+--- a/kernel/irq/manage.c
++++ b/kernel/irq/manage.c
+@@ -555,9 +555,9 @@ int can_request_irq(unsigned int irq, un
+               return 0;
+       if (irq_settings_can_request(desc)) {
+-              if (desc->action)
+-                      if (irqflags & desc->action->flags & IRQF_SHARED)
+-                              canrequest =1;
++              if (!desc->action ||
++                  irqflags & desc->action->flags & IRQF_SHARED)
++                      canrequest = 1;
+       }
+       irq_put_desc_unlock(desc, flags);
+       return canrequest;
diff --git a/queue-3.9/nbd-correct-disconnect-behavior.patch b/queue-3.9/nbd-correct-disconnect-behavior.patch
new file mode 100644 (file)
index 0000000..08fb212
--- /dev/null
@@ -0,0 +1,73 @@
+From c378f70adbc1bbecd9e6db145019f14b2f688c7c Mon Sep 17 00:00:00 2001
+From: Paul Clements <paul.clements@steeleye.com>
+Date: Wed, 3 Jul 2013 15:09:04 -0700
+Subject: nbd: correct disconnect behavior
+
+From: Paul Clements <paul.clements@steeleye.com>
+
+commit c378f70adbc1bbecd9e6db145019f14b2f688c7c upstream.
+
+Currently, when a disconnect is requested by the user (via NBD_DISCONNECT
+ioctl) the return from NBD_DO_IT is undefined (it is usually one of
+several error codes).  This means that nbd-client does not know if a
+manual disconnect was performed or whether a network error occurred.
+Because of this, nbd-client's persist mode (which tries to reconnect after
+error, but not after manual disconnect) does not always work correctly.
+
+This change fixes this by causing NBD_DO_IT to always return 0 if a user
+requests a disconnect.  This means that nbd-client can correctly either
+persist the connection (if an error occurred) or disconnect (if the user
+requested it).
+
+Signed-off-by: Paul Clements <paul.clements@steeleye.com>
+Acked-by: Rob Landley <rob@landley.net>
+Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
+Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
+Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+
+---
+ drivers/block/nbd.c |    7 ++++++-
+ include/linux/nbd.h |    1 +
+ 2 files changed, 7 insertions(+), 1 deletion(-)
+
+--- a/drivers/block/nbd.c
++++ b/drivers/block/nbd.c
+@@ -623,8 +623,10 @@ static int __nbd_ioctl(struct block_devi
+               if (!nbd->sock)
+                       return -EINVAL;
++              nbd->disconnect = 1;
++
+               nbd_send_req(nbd, &sreq);
+-                return 0;
++              return 0;
+       }
+  
+       case NBD_CLEAR_SOCK: {
+@@ -654,6 +656,7 @@ static int __nbd_ioctl(struct block_devi
+                               nbd->sock = SOCKET_I(inode);
+                               if (max_part > 0)
+                                       bdev->bd_invalidated = 1;
++                              nbd->disconnect = 0; /* we're connected now */
+                               return 0;
+                       } else {
+                               fput(file);
+@@ -743,6 +746,8 @@ static int __nbd_ioctl(struct block_devi
+               set_capacity(nbd->disk, 0);
+               if (max_part > 0)
+                       ioctl_by_bdev(bdev, BLKRRPART, 0);
++              if (nbd->disconnect) /* user requested, ignore socket errors */
++                      return 0;
+               return nbd->harderror;
+       }
+--- a/include/linux/nbd.h
++++ b/include/linux/nbd.h
+@@ -41,6 +41,7 @@ struct nbd_device {
+       u64 bytesize;
+       pid_t pid; /* pid of nbd-client, if attached */
+       int xmit_timeout;
++      int disconnect; /* a disconnect has been requested by user */
+ };
+ #endif
diff --git a/queue-3.9/ocfs2-xattr-fix-inlined-xattr-reflink.patch b/queue-3.9/ocfs2-xattr-fix-inlined-xattr-reflink.patch
new file mode 100644 (file)
index 0000000..6e110d8
--- /dev/null
@@ -0,0 +1,83 @@
+From ef962df057aaafd714f5c22ba3de1be459571fdf Mon Sep 17 00:00:00 2001
+From: Junxiao Bi <junxiao.bi@oracle.com>
+Date: Wed, 3 Jul 2013 15:01:03 -0700
+Subject: ocfs2: xattr: fix inlined xattr reflink
+
+From: Junxiao Bi <junxiao.bi@oracle.com>
+
+commit ef962df057aaafd714f5c22ba3de1be459571fdf upstream.
+
+Inlined xattr shared free space of inode block with inlined data or data
+extent record, so the size of the later two should be adjusted when
+inlined xattr is enabled.  See ocfs2_xattr_ibody_init().  But this isn't
+done well when reflink.  For inode with inlined data, its max inlined
+data size is adjusted in ocfs2_duplicate_inline_data(), no problem.  But
+for inode with data extent record, its record count isn't adjusted.  Fix
+it, or data extent record and inlined xattr may overwrite each other,
+then cause data corruption or xattr failure.
+
+One panic caused by this bug in our test environment is the following:
+
+  kernel BUG at fs/ocfs2/xattr.c:1435!
+  invalid opcode: 0000 [#1] SMP
+  Pid: 10871, comm: multi_reflink_t Not tainted 2.6.39-300.17.1.el5uek #1
+  RIP: ocfs2_xa_offset_pointer+0x17/0x20 [ocfs2]
+  RSP: e02b:ffff88007a587948  EFLAGS: 00010283
+  RAX: 0000000000000000 RBX: 0000000000000010 RCX: 00000000000051e4
+  RDX: ffff880057092060 RSI: 0000000000000f80 RDI: ffff88007a587a68
+  RBP: ffff88007a587948 R08: 00000000000062f4 R09: 0000000000000000
+  R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000000010
+  R13: ffff88007a587a68 R14: 0000000000000001 R15: ffff88007a587c68
+  FS:  00007fccff7f06e0(0000) GS:ffff88007fc00000(0000) knlGS:0000000000000000
+  CS:  e033 DS: 0000 ES: 0000 CR0: 000000008005003b
+  CR2: 00000000015cf000 CR3: 000000007aa76000 CR4: 0000000000000660
+  DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
+  DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
+  Process multi_reflink_t
+  Call Trace:
+    ocfs2_xa_reuse_entry+0x60/0x280 [ocfs2]
+    ocfs2_xa_prepare_entry+0x17e/0x2a0 [ocfs2]
+    ocfs2_xa_set+0xcc/0x250 [ocfs2]
+    ocfs2_xattr_ibody_set+0x98/0x230 [ocfs2]
+    __ocfs2_xattr_set_handle+0x4f/0x700 [ocfs2]
+    ocfs2_xattr_set+0x6c6/0x890 [ocfs2]
+    ocfs2_xattr_user_set+0x46/0x50 [ocfs2]
+    generic_setxattr+0x70/0x90
+    __vfs_setxattr_noperm+0x80/0x1a0
+    vfs_setxattr+0xa9/0xb0
+    setxattr+0xc3/0x120
+    sys_fsetxattr+0xa8/0xd0
+    system_call_fastpath+0x16/0x1b
+
+Signed-off-by: Junxiao Bi <junxiao.bi@oracle.com>
+Reviewed-by: Jie Liu <jeff.liu@oracle.com>
+Acked-by: Joel Becker <jlbec@evilplan.org>
+Cc: Mark Fasheh <mfasheh@suse.com>
+Cc: Sunil Mushran <sunil.mushran@gmail.com>
+Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
+Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
+Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+
+---
+ fs/ocfs2/xattr.c |   10 ++++++++++
+ 1 file changed, 10 insertions(+)
+
+--- a/fs/ocfs2/xattr.c
++++ b/fs/ocfs2/xattr.c
+@@ -6499,6 +6499,16 @@ static int ocfs2_reflink_xattr_inline(st
+       }
+       new_oi = OCFS2_I(args->new_inode);
++      /*
++       * Adjust extent record count to reserve space for extended attribute.
++       * Inline data count had been adjusted in ocfs2_duplicate_inline_data().
++       */
++      if (!(new_oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) &&
++          !(ocfs2_inode_is_fast_symlink(args->new_inode))) {
++              struct ocfs2_extent_list *el = &new_di->id2.i_list;
++              le16_add_cpu(&el->l_count, -(inline_size /
++                                      sizeof(struct ocfs2_extent_rec)));
++      }
+       spin_lock(&new_oi->ip_lock);
+       new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL | OCFS2_INLINE_XATTR_FL;
+       new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
diff --git a/queue-3.9/pci-finish-sr-iov-vf-setup-before-adding-the-device.patch b/queue-3.9/pci-finish-sr-iov-vf-setup-before-adding-the-device.patch
new file mode 100644 (file)
index 0000000..5badedf
--- /dev/null
@@ -0,0 +1,50 @@
+From fbf33f516bdbcc2ab1ba1e54dfb720b0cfaa6874 Mon Sep 17 00:00:00 2001
+From: Xudong Hao <xudong.hao@intel.com>
+Date: Fri, 31 May 2013 12:21:29 +0800
+Subject: PCI: Finish SR-IOV VF setup before adding the device
+
+From: Xudong Hao <xudong.hao@intel.com>
+
+commit fbf33f516bdbcc2ab1ba1e54dfb720b0cfaa6874 upstream.
+
+Commit 4f535093cf "PCI: Put pci_dev in device tree as early as possible"
+moves device registering from pci_bus_add_devices() to pci_device_add().
+That causes problems for virtual functions because device_add(&virtfn->dev)
+is called before setting the virtfn->is_virtfn flag, which then causes Xen
+to report PCI virtual functions as PCI physical functions.
+
+Fix it by setting virtfn->is_virtfn before calling pci_device_add().
+
+[Jiang Liu]: Move the setting of virtfn->is_virtfn ahead further for better
+readability and modify changelog.
+
+Signed-off-by: Xudong Hao <xudong.hao@intel.com>
+Signed-off-by: Jiang Liu <jiang.liu@huawei.com>
+Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
+Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+
+---
+ drivers/pci/iov.c |    5 ++---
+ 1 file changed, 2 insertions(+), 3 deletions(-)
+
+--- a/drivers/pci/iov.c
++++ b/drivers/pci/iov.c
+@@ -92,6 +92,8 @@ static int virtfn_add(struct pci_dev *de
+       pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_DID, &virtfn->device);
+       pci_setup_device(virtfn);
+       virtfn->dev.parent = dev->dev.parent;
++      virtfn->physfn = pci_dev_get(dev);
++      virtfn->is_virtfn = 1;
+       for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
+               res = dev->resource + PCI_IOV_RESOURCES + i;
+@@ -113,9 +115,6 @@ static int virtfn_add(struct pci_dev *de
+       pci_device_add(virtfn, virtfn->bus);
+       mutex_unlock(&iov->dev->sriov->lock);
+-      virtfn->physfn = pci_dev_get(dev);
+-      virtfn->is_virtfn = 1;
+-
+       rc = pci_bus_add_device(virtfn);
+       sprintf(buf, "virtfn%u", id);
+       rc = sysfs_create_link(&dev->dev.kobj, &virtfn->dev.kobj, buf);
diff --git a/queue-3.9/pci-fix-refcount-issue-in-pci_create_root_bus-error-recovery-path.patch b/queue-3.9/pci-fix-refcount-issue-in-pci_create_root_bus-error-recovery-path.patch
new file mode 100644 (file)
index 0000000..f7a05cd
--- /dev/null
@@ -0,0 +1,54 @@
+From 343df771e671d821478dd3ef525a0610b808dbf8 Mon Sep 17 00:00:00 2001
+From: Jiang Liu <liuj97@gmail.com>
+Date: Fri, 7 Jun 2013 01:10:08 +0800
+Subject: PCI: Fix refcount issue in pci_create_root_bus() error recovery path
+
+From: Jiang Liu <liuj97@gmail.com>
+
+commit 343df771e671d821478dd3ef525a0610b808dbf8 upstream.
+
+After calling device_register(&bridge->dev), the bridge is reference-
+counted, and it is illegal to call kfree() on it except in the release
+function.
+
+[bhelgaas: changelog, use put_device() after device_register() failure]
+Signed-off-by: Jiang Liu <jiang.liu@huawei.com>
+Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
+Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+
+---
+ drivers/pci/probe.c |   14 ++++++++------
+ 1 file changed, 8 insertions(+), 6 deletions(-)
+
+--- a/drivers/pci/probe.c
++++ b/drivers/pci/probe.c
+@@ -1694,12 +1694,16 @@ struct pci_bus *pci_create_root_bus(stru
+       bridge->dev.release = pci_release_bus_bridge_dev;
+       dev_set_name(&bridge->dev, "pci%04x:%02x", pci_domain_nr(b), bus);
+       error = pcibios_root_bridge_prepare(bridge);
+-      if (error)
+-              goto bridge_dev_reg_err;
++      if (error) {
++              kfree(bridge);
++              goto err_out;
++      }
+       error = device_register(&bridge->dev);
+-      if (error)
+-              goto bridge_dev_reg_err;
++      if (error) {
++              put_device(&bridge->dev);
++              goto err_out;
++      }
+       b->bridge = get_device(&bridge->dev);
+       device_enable_async_suspend(b->bridge);
+       pci_set_bus_of_node(b);
+@@ -1753,8 +1757,6 @@ struct pci_bus *pci_create_root_bus(stru
+ class_dev_reg_err:
+       put_device(&bridge->dev);
+       device_unregister(&bridge->dev);
+-bridge_dev_reg_err:
+-      kfree(bridge);
+ err_out:
+       kfree(b);
+       return NULL;
index 5f40419a78c7693c8b6a2bee4abc6bace5da64f4..f3f8d9ca1723c86f79d04c6e1d04af0df0afc13d 100644 (file)
@@ -6,10 +6,24 @@ jbd2-move-superblock-checksum-calculation-to-jbd2_write_superblock.patch
 jbd2-fix-theoretical-race-in-jbd2__journal_restart.patch
 ext4-fix-corruption-when-online-resizing-a-fs-with-1k-block-size.patch
 ext3-ext4-don-t-mess-with-dir_file-f_pos-in-htree_dirblock_to_tree.patch
-ext4-check-error-return-from-ext4_write_inline_data_end.patch
 usb-gadget-f_mass_storage-add-missing-memory-barrier-for-thread_wakeup_needed.patch
 xhci-check-for-failed-dma-pool-allocation.patch
 usb-host-xhci-plat-release-mem-region-while-removing-module.patch
 drivers-hv-switch-to-use-mb-instead-of-smp_mb.patch
 pcmcia-at91_cf-fix-gpio_get_value-in-at91_cf_get_status.patch
 cgroup-fix-umount-vs-cgroup_event_remove-race.patch
+xen-time-remove-blocked-time-accounting-from-xen-clockchip.patch
+xen-pcifront-deal-with-toolstack-missing-xenbusstateclosing-state.patch
+genirq-fix-can_request_irq-for-irqs-without-an-action.patch
+drivers-rtc-rtc-rv3029c2.c-fix-disabling-aie-irq.patch
+acpi-ec-add-hp-folio-13-to-ec_dmi_table-in-order-to-skip-dsdt-scan.patch
+acpica-do-not-use-extended-sleep-registers-unless-hw-reduced-bit-is-set.patch
+acpi-pm-fix-corner-case-in-acpi_bus_update_power.patch
+ocfs2-xattr-fix-inlined-xattr-reflink.patch
+nbd-correct-disconnect-behavior.patch
+pci-finish-sr-iov-vf-setup-before-adding-the-device.patch
+pci-fix-refcount-issue-in-pci_create_root_bus-error-recovery-path.patch
+ahci-remove-pmp-link-online-check-in-fbs-eh.patch
+timer-fix-jiffies-wrap-behavior-of-round_jiffies_common.patch
+btrfs-fix-estale-with-btrfs-send.patch
+btrfs-only-do-the-tree_mod_log_free_eb-if-this-is-our-last.patch
diff --git a/queue-3.9/timer-fix-jiffies-wrap-behavior-of-round_jiffies_common.patch b/queue-3.9/timer-fix-jiffies-wrap-behavior-of-round_jiffies_common.patch
new file mode 100644 (file)
index 0000000..d69cf10
--- /dev/null
@@ -0,0 +1,40 @@
+From 9e04d3804d3ac97d8c03a41d78d0f0674b5d01e1 Mon Sep 17 00:00:00 2001
+From: Bart Van Assche <bart.vanassche@gmail.com>
+Date: Tue, 21 May 2013 20:43:50 +0200
+Subject: timer: Fix jiffies wrap behavior of round_jiffies_common()
+
+From: Bart Van Assche <bart.vanassche@gmail.com>
+
+commit 9e04d3804d3ac97d8c03a41d78d0f0674b5d01e1 upstream.
+
+Direct compare of jiffies related values does not work in the wrap
+around case. Replace it with time_is_after_jiffies().
+
+Signed-off-by: Bart Van Assche <bvanassche@acm.org>
+Cc: Arjan van de Ven <arjan@infradead.org>
+Cc: Stephen Rothwell <sfr@canb.auug.org.au>
+Link: http://lkml.kernel.org/r/519BC066.5080600@acm.org
+Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
+Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+
+---
+ kernel/timer.c |    8 +++++---
+ 1 file changed, 5 insertions(+), 3 deletions(-)
+
+--- a/kernel/timer.c
++++ b/kernel/timer.c
+@@ -148,9 +148,11 @@ static unsigned long round_jiffies_commo
+       /* now that we have rounded, subtract the extra skew again */
+       j -= cpu * 3;
+-      if (j <= jiffies) /* rounding ate our timeout entirely; */
+-              return original;
+-      return j;
++      /*
++       * Make sure j is still in the future. Otherwise return the
++       * unmodified value.
++       */
++      return time_is_after_jiffies(j) ? j : original;
+ }
+ /**
diff --git a/queue-3.9/xen-pcifront-deal-with-toolstack-missing-xenbusstateclosing-state.patch b/queue-3.9/xen-pcifront-deal-with-toolstack-missing-xenbusstateclosing-state.patch
new file mode 100644 (file)
index 0000000..40bf0cd
--- /dev/null
@@ -0,0 +1,91 @@
+From 098b1aeaf4d6149953b8f1f8d55c21d85536fbff Mon Sep 17 00:00:00 2001
+From: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+Date: Mon, 10 Jun 2013 16:48:09 -0400
+Subject: xen/pcifront: Deal with toolstack missing 'XenbusStateClosing' state.
+
+From: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+
+commit 098b1aeaf4d6149953b8f1f8d55c21d85536fbff upstream.
+
+There are two tool-stack that can instruct the Xen PCI frontend
+and backend to change states: 'xm' (Python code with a daemon),
+and 'xl' (C library - does not keep state changes).
+
+With the 'xm', the path to disconnect a single PCI device (xm pci-detach
+<guest> <BDF>) is:
+
+4(Connected)->7(Reconfiguring*)-> 8(Reconfigured)-> 4(Connected)->5(Closing*).
+
+The * is for states that the tool-stack sets. For 'xl', it is similar:
+
+4(Connected)->7(Reconfiguring*)-> 8(Reconfigured)-> 4(Connected)
+
+Both of them also tear down the XenBus structure, so the backend
+state ends up going in the 3(Initialised) and calls pcifront_xenbus_remove.
+
+When a PCI device is plugged back in (xm pci-attach <guest> <BDF>)
+both of them follow the same pattern:
+
+2(InitWait*), 3(Initialized*), 4(Connected*)->4(Connected).
+
+[xen-pcifront ignores the 2,3 state changes and only acts when
+4 (Connected) has been reached]
+
+Note that this is for a _single_ PCI device. If there were two
+PCI devices and only one was disconnected 'xm' would show the same
+state changes.
+
+The problem is that git commit 3d925320e9e2de162bd138bf97816bda8c3f71be
+("xen/pcifront: Use Xen-SWIOTLB when initting if required") introduced
+a mechanism to initialize the SWIOTLB when the Xen PCI front moves to
+Connected state. It also had some aggressive seatbelt code check that
+would warn the user if one tried to change to Connected state without
+hitting first the Closing state:
+
+ pcifront pci-0: PCI frontend already installed!
+
+However, that code can be relaxed and we can continue on working
+even if the frontend is instructed to be the 'Connected' state with
+no devices and then gets tickled to be in 'Connected' state again.
+
+In other words, this 4(Connected)->5(Closing)->4(Connected) state
+was expected, while 4(Connected)->.... anything but 5(Closing)->4(Connected)
+was not. This patch removes that aggressive check and allows
+Xen pcifront to work with the 'xl' toolstack (for one or more
+PCI devices) and with 'xm' toolstack (for more than two PCI
+devices).
+
+Acked-by: Bjorn Helgaas <bhelgaas@google.com>
+Cc: linux-pci@vger.kernel.org
+[v2: Added in the description about two PCI devices]
+Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+
+---
+ drivers/pci/xen-pcifront.c |    7 +++----
+ 1 file changed, 3 insertions(+), 4 deletions(-)
+
+--- a/drivers/pci/xen-pcifront.c
++++ b/drivers/pci/xen-pcifront.c
+@@ -678,10 +678,9 @@ static int pcifront_connect_and_init_dma
+       if (!pcifront_dev) {
+               dev_info(&pdev->xdev->dev, "Installing PCI frontend\n");
+               pcifront_dev = pdev;
+-      } else {
+-              dev_err(&pdev->xdev->dev, "PCI frontend already installed!\n");
++      } else
+               err = -EEXIST;
+-      }
++
+       spin_unlock(&pcifront_dev_lock);
+       if (!err && !swiotlb_nr_tbl()) {
+@@ -848,7 +847,7 @@ static int pcifront_try_connect(struct p
+               goto out;
+       err = pcifront_connect_and_init_dma(pdev);
+-      if (err) {
++      if (err && err != -EEXIST) {
+               xenbus_dev_fatal(pdev->xdev, err,
+                                "Error setting up PCI Frontend");
+               goto out;
diff --git a/queue-3.9/xen-time-remove-blocked-time-accounting-from-xen-clockchip.patch b/queue-3.9/xen-time-remove-blocked-time-accounting-from-xen-clockchip.patch
new file mode 100644 (file)
index 0000000..9adc37e
--- /dev/null
@@ -0,0 +1,83 @@
+From 0b0c002c340e78173789f8afaa508070d838cf3d Mon Sep 17 00:00:00 2001
+From: Laszlo Ersek <lersek@redhat.com>
+Date: Tue, 18 Oct 2011 22:42:59 +0200
+Subject: xen/time: remove blocked time accounting from xen "clockchip"
+
+From: Laszlo Ersek <lersek@redhat.com>
+
+commit 0b0c002c340e78173789f8afaa508070d838cf3d upstream.
+
+... because the "clock_event_device framework" already accounts for idle
+time through the "event_handler" function pointer in
+xen_timer_interrupt().
+
+The patch is intended as the completion of [1]. It should fix the double
+idle times seen in PV guests' /proc/stat [2]. It should be orthogonal to
+stolen time accounting (the removed code seems to be isolated).
+
+The approach may be completely misguided.
+
+[1] https://lkml.org/lkml/2011/10/6/10
+[2] http://lists.xensource.com/archives/html/xen-devel/2010-08/msg01068.html
+
+John took the time to retest this patch on top of v3.10 and reported:
+"idle time is correctly incremented for pv and hvm for the normal
+case, nohz=off and nohz=idle." so lets put this patch in.
+
+Signed-off-by: Laszlo Ersek <lersek@redhat.com>
+Signed-off-by: John Haxby <john.haxby@oracle.com>
+Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+
+---
+ arch/x86/xen/time.c |   17 ++---------------
+ 1 file changed, 2 insertions(+), 15 deletions(-)
+
+--- a/arch/x86/xen/time.c
++++ b/arch/x86/xen/time.c
+@@ -36,9 +36,8 @@ static DEFINE_PER_CPU(struct vcpu_runsta
+ /* snapshots of runstate info */
+ static DEFINE_PER_CPU(struct vcpu_runstate_info, xen_runstate_snapshot);
+-/* unused ns of stolen and blocked time */
++/* unused ns of stolen time */
+ static DEFINE_PER_CPU(u64, xen_residual_stolen);
+-static DEFINE_PER_CPU(u64, xen_residual_blocked);
+ /* return an consistent snapshot of 64-bit time/counter value */
+ static u64 get64(const u64 *p)
+@@ -115,7 +114,7 @@ static void do_stolen_accounting(void)
+ {
+       struct vcpu_runstate_info state;
+       struct vcpu_runstate_info *snap;
+-      s64 blocked, runnable, offline, stolen;
++      s64 runnable, offline, stolen;
+       cputime_t ticks;
+       get_runstate_snapshot(&state);
+@@ -125,7 +124,6 @@ static void do_stolen_accounting(void)
+       snap = &__get_cpu_var(xen_runstate_snapshot);
+       /* work out how much time the VCPU has not been runn*ing*  */
+-      blocked = state.time[RUNSTATE_blocked] - snap->time[RUNSTATE_blocked];
+       runnable = state.time[RUNSTATE_runnable] - snap->time[RUNSTATE_runnable];
+       offline = state.time[RUNSTATE_offline] - snap->time[RUNSTATE_offline];
+@@ -141,17 +139,6 @@ static void do_stolen_accounting(void)
+       ticks = iter_div_u64_rem(stolen, NS_PER_TICK, &stolen);
+       __this_cpu_write(xen_residual_stolen, stolen);
+       account_steal_ticks(ticks);
+-
+-      /* Add the appropriate number of ticks of blocked time,
+-         including any left-overs from last time. */
+-      blocked += __this_cpu_read(xen_residual_blocked);
+-
+-      if (blocked < 0)
+-              blocked = 0;
+-
+-      ticks = iter_div_u64_rem(blocked, NS_PER_TICK, &blocked);
+-      __this_cpu_write(xen_residual_blocked, blocked);
+-      account_idle_ticks(ticks);
+ }
+ /* Get the TSC speed from Xen */