Skip to content

K
Kernel

openeuler / Kernel
1 年多 前同步成功

通知 8
Star 0
Fork 0
K
Kernel
切换分支/标签
  1. 07 5月, 2014 9 次提交
  3. 04 5月, 2014 1 次提交
    • M
      arm64: fixmap: fix missing sub-page offset for earlyprintk · f774b7d1
      由 Marc Zyngier 提交于 
      Commit d57c33c5 (add generic fixmap.h) added (among other
similar things) set_fixmap_io to deal with early ioremap of devices.

More recently, commit bf4b558e (arm64: add early_ioremap support)
converted the arm64 earlyprintk to use set_fixmap_io. A side effect of
this conversion is that my virtual machines have stopped booting when
I pass "earlyprintk=uart8250-8bit,0x3f8" to the guest kernel.

Turns out that the new earlyprintk code doesn't care at all about
sub-page offsets, and just assumes that the earlyprintk device will
be page-aligned. Obviously, that doesn't play well with the above example.

Further investigation shows that set_fixmap_io uses __set_fixmap instead
of __set_fixmap_offset. A fix is to introduce a set_fixmap_offset_io that
uses the latter, and to remove the superflous call to fix_to_virt
(which only returns the value that set_fixmap_io has already given us).

With this applied, my VMs are back in business. Tested on a Cortex-A57
platform with kvmtool as platform emulation.

Cc: Will Deacon <will.deacon@arm.com>
Acked-by: NMark Salter <msalter@redhat.com>
Acked-by: NArnd Bergmann <arnd@arndb.de>
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Signed-off-by: NCatalin Marinas <catalin.marinas@arm.com>
      f774b7d1
  5. 01 5月, 2014 1 次提交
  7. 28 4月, 2014 4 次提交
    • S
      ftrace/module: Hardcode ftrace_module_init() call into load_module() · a949ae56
      由 Steven Rostedt (Red Hat) 提交于 
      A race exists between module loading and enabling of function tracer.

	CPU 1				CPU 2
	-----				-----
  load_module()
   module->state = MODULE_STATE_COMING

				register_ftrace_function()
				 mutex_lock(&ftrace_lock);
				 ftrace_startup()
				  update_ftrace_function();
				   ftrace_arch_code_modify_prepare()
				    set_all_module_text_rw();
				   <enables-ftrace>
				    ftrace_arch_code_modify_post_process()
				     set_all_module_text_ro();

				[ here all module text is set to RO,
				  including the module that is
				  loading!! ]

   blocking_notifier_call_chain(MODULE_STATE_COMING);
    ftrace_init_module()

     [ tries to modify code, but it's RO, and fails!
       ftrace_bug() is called]

When this race happens, ftrace_bug() will produces a nasty warning and
all of the function tracing features will be disabled until reboot.

The simple solution is to treate module load the same way the core
kernel is treated at boot. To hardcode the ftrace function modification
of converting calls to mcount into nops. This is done in init/main.c
there's no reason it could not be done in load_module(). This gives
a better control of the changes and doesn't tie the state of the
module to its notifiers as much. Ftrace is special, it needs to be
treated as such.

The reason this would work, is that the ftrace_module_init() would be
called while the module is in MODULE_STATE_UNFORMED, which is ignored
by the set_all_module_text_ro() call.

Link: http://lkml.kernel.org/r/1395637826-3312-1-git-send-email-indou.takao@jp.fujitsu.comReported-by: NTakao Indoh <indou.takao@jp.fujitsu.com>
Acked-by: NRusty Russell <rusty@rustcorp.com.au>
Cc: stable@vger.kernel.org # 2.6.38+
Signed-off-by: NSteven Rostedt <rostedt@goodmis.org>
      a949ae56
    • T
      genirq: x86: Ensure that dynamic irq allocation does not conflict · 62a08ae2
      由 Thomas Gleixner 提交于 
      On x86 the allocation of irq descriptors may allocate interrupts which
are in the range of the GSI interrupts. That's wrong as those
interrupts are hardwired and we don't have the irq domain translation
like PPC. So one of these interrupts can be hooked up later to one of
the devices which are hard wired to it and the io_apic init code for
that particular interrupt line happily reuses that descriptor with a
completely different configuration so hell breaks lose.

Inside x86 we allocate dynamic interrupts from above nr_gsi_irqs,
except for a few usage sites which have not yet blown up in our face
for whatever reason. But for drivers which need an irq range, like the
GPIO drivers, we have no limit in place and we don't want to expose
such a detail to a driver.

To cure this introduce a function which an architecture can implement
to impose a lower bound on the dynamic interrupt allocations.

Implement it for x86 and set the lower bound to nr_gsi_irqs, which is
the end of the hardwired interrupt space, so all dynamic allocations
happen above.

That not only allows the GPIO driver to work sanely, it also protects
the bogus callsites of create_irq_nr() in hpet, uv, irq_remapping and
htirq code. They need to be cleaned up as well, but that's a separate
issue.
Reported-by: NJin Yao <yao.jin@linux.intel.com>
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Tested-by: NMika Westerberg <mika.westerberg@linux.intel.com>
Cc: Mathias Nyman <mathias.nyman@linux.intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Grant Likely <grant.likely@linaro.org>
Cc: H. Peter Anvin <hpa@linux.intel.com>
Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc: Krogerus Heikki <heikki.krogerus@intel.com>
Cc: Linus Walleij <linus.walleij@linaro.org>
Link: http://lkml.kernel.org/r/alpine.DEB.2.02.1404241617360.28206@ionos.tec.linutronix.deSigned-off-by: NThomas Gleixner <tglx@linutronix.de>
      62a08ae2
    • R
      linux/interrupt.h: fix new kernel-doc warnings · def5f127
      由 Randy Dunlap 提交于 
      Fix new kernel-doc warnings in <linux/interrupt.h>:

Warning(include/linux/interrupt.h:219): No description found for parameter 'cpumask'
Warning(include/linux/interrupt.h:219): Excess function parameter 'mask' description in 'irq_set_affinity'
Warning(include/linux/interrupt.h:236): No description found for parameter 'cpumask'
Warning(include/linux/interrupt.h:236): Excess function parameter 'mask' description in 'irq_force_affinity'
Signed-off-by: NRandy Dunlap <rdunlap@infradead.org>
Link: http://lkml.kernel.org/r/535DD2FD.7030804@infradead.orgSigned-off-by: NThomas Gleixner <tglx@linutronix.de>
      def5f127
    • W
      word-at-a-time: avoid undefined behaviour in zero_bytemask macro · ec6931b2
      由 Will Deacon 提交于 
      The asm-generic, big-endian version of zero_bytemask creates a mask of
bytes preceding the first zero-byte by left shifting ~0ul based on the
position of the first zero byte.

Unfortunately, if the first (top) byte is zero, the output of
prep_zero_mask has only the top bit set, resulting in undefined C
behaviour as we shift left by an amount equal to the width of the type.
As it happens, GCC doesn't manage to spot this through the call to fls(),
but the issue remains if architectures choose to implement their shift
instructions differently.

An example would be arch/arm/ (AArch32), where LSL Rd, Rn, #32 results
in Rd == 0x0, whilst on arch/arm64 (AArch64) LSL Xd, Xn, #64 results in
Xd == Xn.

Rather than check explicitly for the problematic shift, this patch adds
an extra shift by 1, replacing fls with __fls. Since zero_bytemask is
never called with a zero argument (has_zero() is used to check the data
first), we don't need to worry about calling __fls(0), which is
undefined.

Cc: <stable@vger.kernel.org>
Cc: Victor Kamensky <victor.kamensky@linaro.org>
Signed-off-by: NWill Deacon <will.deacon@arm.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      ec6931b2
  9. 25 4月, 2014 3 次提交
    • M
      tty: Fix race condition between __tty_buffer_request_room and flush_to_ldisc · 6a20dbd6
      由 Manfred Schlaegl 提交于 
      The race was introduced while development of linux-3.11 by
e8437d7e and
e9975fde.
Originally it was found and reproduced on linux-3.12.15 and
linux-3.12.15-rt25, by sending 500 byte blocks with 115kbaud to the
target uart in a loop with 100 milliseconds delay.

In short:
 1. The consumer flush_to_ldisc is on to remove the head tty_buffer.
 2. The producer adds a number of bytes, so that a new tty_buffer must
	be allocated and added by __tty_buffer_request_room.
 3. The consumer removes the head tty_buffer element, without handling
	newly committed data.

Detailed example:
 * Initial buffer:
   * Head, Tail -> 0: used=250; commit=250; read=240; next=NULL
 * Consumer: ''flush_to_ldisc''
   * consumed 10 Byte
   * buffer:
     * Head, Tail -> 0: used=250; commit=250; read=250; next=NULL
{{{
		count = head->commit - head->read;	// count = 0
		if (!count) {				// enter
			// INTERRUPTED BY PRODUCER ->
			if (head->next == NULL)
				break;
			buf->head = head->next;
			tty_buffer_free(port, head);
			continue;
		}
}}}
 * Producer: tty_insert_flip_... 10 bytes + tty_flip_buffer_push
   * buffer:
     * Head, Tail -> 0: used=250; commit=250; read=250; next=NULL
   * added 6 bytes: head-element filled to maximum.
     * buffer:
       * Head, Tail -> 0: used=256; commit=250; read=250; next=NULL
   * added 4 bytes: __tty_buffer_request_room is called
     * buffer:
       * Head -> 0: used=256; commit=256; read=250; next=1
       * Tail -> 1: used=4; commit=0; read=250 next=NULL
   * push (tty_flip_buffer_push)
     * buffer:
       * Head -> 0: used=256; commit=256; read=250; next=1
       * Tail -> 1: used=4; commit=4; read=250 next=NULL
 * Consumer
{{{
		count = head->commit - head->read;
		if (!count) {
			// INTERRUPTED BY PRODUCER <-
			if (head->next == NULL)		// -> no break
				break;
			buf->head = head->next;
			tty_buffer_free(port, head);
			// ERROR: tty_buffer head freed -> 6 bytes lost
			continue;
		}
}}}

This patch reintroduces a spin_lock to protect this case. Perhaps later
a lock-less solution could be found.
Signed-off-by: NManfred Schlaegl <manfred.schlaegl@gmx.at>
Cc: stable <stable@vger.kernel.org> # 3.11
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      6a20dbd6
    • R
      of/irq: do irq resolution in platform_get_irq · 9ec36caf
      由 Rob Herring 提交于 
      Currently we get the following kind of errors if we try to use interrupt
phandles to irqchips that have not yet initialized:

irq: no irq domain found for /ocp/pinmux@48002030 !
------------[ cut here ]------------
WARNING: CPU: 0 PID: 1 at drivers/of/platform.c:171 of_device_alloc+0x144/0x184()
Modules linked in:
CPU: 0 PID: 1 Comm: swapper/0 Not tainted 3.12.0-00038-g42a9708 #1012
(show_stack+0x14/0x1c)
(dump_stack+0x6c/0xa0)
(warn_slowpath_common+0x64/0x84)
(warn_slowpath_null+0x1c/0x24)
(of_device_alloc+0x144/0x184)
(of_platform_device_create_pdata+0x44/0x9c)
(of_platform_bus_create+0xd0/0x170)
(of_platform_bus_create+0x12c/0x170)
(of_platform_populate+0x60/0x98)

This is because we're wrongly trying to populate resources that are not
yet available. It's perfectly valid to create irqchips dynamically, so
let's fix up the issue by resolving the interrupt resources when
platform_get_irq is called.

And then we also need to accept the fact that some irqdomains do not
exist that early on, and only get initialized later on. So we can
make the current WARN_ON into just into a pr_debug().

We still attempt to populate irq resources when we create the devices.
This allows current drivers which don't use platform_get_irq to continue
to function. Once all drivers are fixed, this code can be removed.
Suggested-by: NRussell King <linux@arm.linux.org.uk>
Signed-off-by: NRob Herring <robh@kernel.org>
Signed-off-by: NTony Lindgren <tony@atomide.com>
Tested-by: NTony Lindgren <tony@atomide.com>
Cc: stable@vger.kernel.org # v3.10+
Signed-off-by: NGrant Likely <grant.likely@linaro.org>
      9ec36caf
    • G
      phy: core: make NULL a valid phy reference if !CONFIG_GENERIC_PHY · 2b97789f
      由 Grygorii Strashko 提交于 
      This fixes a regression on Keystone 2 platforms caused by patch
57303488
"usb: dwc3: adapt dwc3 core to use Generic PHY Framework" which adds
optional support of generic phy in DWC3 core.

On Keystone 2 platforms the USB is not working now because
CONFIG_GENERIC_PHY isn't set and, as result, Generic PHY APIs stubs
return -ENOSYS always. The log shows:
 dwc3 2690000.dwc3: failed to initialize core
 dwc3: probe of 2690000.dwc3 failed with error -38

Hence, fix it by making NULL a valid phy reference in Generic PHY
APIs stubs in the same way as it was done by the patch
04c2faca "drivers: phy: Make NULL
a valid phy reference".
Acked-by: NFelipe Balbi <balbi@ti.com>
Acked-by: NSantosh Shilimkar <santosh.shilimkar@ti.com>
Signed-off-by: NGrygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: NKishon Vijay Abraham I <kishon@ti.com>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      2b97789f
  11. 24 4月, 2014 2 次提交
  13. 23 4月, 2014 1 次提交
  15. 22 4月, 2014 1 次提交
    • J
      locks: rename file-private locks to "open file description locks" · 0d3f7a2d
      由 Jeff Layton 提交于 
      File-private locks have been merged into Linux for v3.15, and *now*
people are commenting that the name and macro definitions for the new
file-private locks suck.

...and I can't even disagree. The names and command macros do suck.

We're going to have to live with these for a long time, so it's
important that we be happy with the names before we're stuck with them.
The consensus on the lists so far is that they should be rechristened as
"open file description locks".

The name isn't a big deal for the kernel, but the command macros are not
visually distinct enough from the traditional POSIX lock macros. The
glibc and documentation folks are recommending that we change them to
look like F_OFD_{GETLK|SETLK|SETLKW}. That lessens the chance that a
programmer will typo one of the commands wrong, and also makes it easier
to spot this difference when reading code.

This patch makes the following changes that I think are necessary before
v3.15 ships:

1) rename the command macros to their new names. These end up in the uapi
   headers and so are part of the external-facing API. It turns out that
   glibc doesn't actually use the fcntl.h uapi header, but it's hard to
   be sure that something else won't. Changing it now is safest.

2) make the the /proc/locks output display these as type "OFDLCK"

Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Carlos O'Donell <carlos@redhat.com>
Cc: Stefan Metzmacher <metze@samba.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Frank Filz <ffilzlnx@mindspring.com>
Cc: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: NJeff Layton <jlayton@redhat.com>
      0d3f7a2d
  17. 21 4月, 2014 8 次提交
  19. 20 4月, 2014 2 次提交
    • H
      Input: Add INPUT_PROP_TOPBUTTONPAD device property · f37c0134
      由 Hans de Goede 提交于 
      On some newer laptops with a trackpoint the physical buttons for the
trackpoint have been removed to allow for a larger touchpad. On these
laptops the buttonpad has clearly marked areas on the top which are to be
used as trackpad buttons.

Users of the event device-node need to know about this, so that they can
properly interpret BTN_LEFT events as being a left / right / middle click
depending on where on the button pad the clicking finger is.

This commits adds a INPUT_PROP_TOPBUTTONPAD device property which drivers
for such buttonpads will use to signal to the user that this buttonpad not
only has the normal bottom button area, but also a top button area.
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NDmitry Torokhov <dmitry.torokhov@gmail.com>
      f37c0134
    • H
      Input: serio - add firmware_id sysfs attribute · 0456c66f
      由 Hans de Goede 提交于 
      serio devices exposed via platform firmware interfaces such as ACPI may
provide additional identifying information of use to userspace.

We don't associate the serio devices with the firmware device (we don't
set it as parent), so there's no way for userspace to make use of this
information.

We cannot change the parent for serio devices instantiated though a
firmware interface as that would break suspend / resume ordering.

Therefore this patch adds a new firmware_id sysfs attribute so that
userspace can get a string from there with any additional identifying
information the firmware interface may provide.
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Acked-by: NPeter Hutterer <peter.hutterer@who-t.net>
Signed-off-by: NDmitry Torokhov <dmitry.torokhov@gmail.com>
      0456c66f
  21. 19 4月, 2014 5 次提交
    • M
      mm: use paravirt friendly ops for NUMA hinting ptes · 29c77870
      由 Mel Gorman 提交于 
      David Vrabel identified a regression when using automatic NUMA balancing
under Xen whereby page table entries were getting corrupted due to the
use of native PTE operations.  Quoting him

	Xen PV guest page tables require that their entries use machine
	addresses if the preset bit (_PAGE_PRESENT) is set, and (for
	successful migration) non-present PTEs must use pseudo-physical
	addresses.  This is because on migration MFNs in present PTEs are
	translated to PFNs (canonicalised) so they may be translated back
	to the new MFN in the destination domain (uncanonicalised).

	pte_mknonnuma(), pmd_mknonnuma(), pte_mknuma() and pmd_mknuma()
	set and clear the _PAGE_PRESENT bit using pte_set_flags(),
	pte_clear_flags(), etc.

	In a Xen PV guest, these functions must translate MFNs to PFNs
	when clearing _PAGE_PRESENT and translate PFNs to MFNs when setting
	_PAGE_PRESENT.

His suggested fix converted p[te|md]_[set|clear]_flags to using
paravirt-friendly ops but this is overkill.  He suggested an alternative
of using p[te|md]_modify in the NUMA page table operations but this is
does more work than necessary and would require looking up a VMA for
protections.

This patch modifies the NUMA page table operations to use paravirt
friendly operations to set/clear the flags of interest.  Unfortunately
this will take a performance hit when updating the PTEs on
CONFIG_PARAVIRT but I do not see a way around it that does not break
Xen.
Signed-off-by: NMel Gorman <mgorman@suse.de>
Acked-by: NDavid Vrabel <david.vrabel@citrix.com>
Tested-by: NDavid Vrabel <david.vrabel@citrix.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Peter Anvin <hpa@zytor.com>
Cc: Fengguang Wu <fengguang.wu@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Steven Noonan <steven@uplinklabs.net>
Cc: Rik van Riel <riel@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Cyrill Gorcunov <gorcunov@gmail.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      29c77870
    • P
      wait: explain the shadowing and type inconsistencies · 8b32201d
      由 Peter Zijlstra 提交于 
      Stick in a comment before someone else tries to fix the sparse warning
this generates.
Suggested-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NPeter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/n/tip-o2ro6f3vkxklni0bc8f7m68s@git.kernel.orgSigned-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      8b32201d
    • V
      Shiraz has moved · 9cc23682
      由 Viresh Kumar 提交于 
      shiraz.hashim@st.com email-id doesn't exist anymore as he has left the
company.  Replace ST's id with shiraz.linux.kernel@gmail.com.

It also updates .mailmap file to fix address for 'git shortlog'.
Signed-off-by: NViresh Kumar <viresh.kumar@linaro.org>
Cc: Shiraz Hashim <shiraz.linux.kernel@gmail.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      9cc23682
    • V
      net: sctp: cache auth_enable per endpoint · b14878cc
      由 Vlad Yasevich 提交于 
      Currently, it is possible to create an SCTP socket, then switch
auth_enable via sysctl setting to 1 and crash the system on connect:

Oops[#1]:
CPU: 0 PID: 0 Comm: swapper Not tainted 3.14.1-mipsgit-20140415 #1
task: ffffffff8056ce80 ti: ffffffff8055c000 task.ti: ffffffff8055c000
[...]
Call Trace:
[<ffffffff8043c4e8>] sctp_auth_asoc_set_default_hmac+0x68/0x80
[<ffffffff8042b300>] sctp_process_init+0x5e0/0x8a4
[<ffffffff8042188c>] sctp_sf_do_5_1B_init+0x234/0x34c
[<ffffffff804228c8>] sctp_do_sm+0xb4/0x1e8
[<ffffffff80425a08>] sctp_endpoint_bh_rcv+0x1c4/0x214
[<ffffffff8043af68>] sctp_rcv+0x588/0x630
[<ffffffff8043e8e8>] sctp6_rcv+0x10/0x24
[<ffffffff803acb50>] ip6_input+0x2c0/0x440
[<ffffffff8030fc00>] __netif_receive_skb_core+0x4a8/0x564
[<ffffffff80310650>] process_backlog+0xb4/0x18c
[<ffffffff80313cbc>] net_rx_action+0x12c/0x210
[<ffffffff80034254>] __do_softirq+0x17c/0x2ac
[<ffffffff800345e0>] irq_exit+0x54/0xb0
[<ffffffff800075a4>] ret_from_irq+0x0/0x4
[<ffffffff800090ec>] rm7k_wait_irqoff+0x24/0x48
[<ffffffff8005e388>] cpu_startup_entry+0xc0/0x148
[<ffffffff805a88b0>] start_kernel+0x37c/0x398
Code: dd0900b8  000330f8  0126302d <dcc60000> 50c0fff1  0047182a  a48306a0
03e00008  00000000
---[ end trace b530b0551467f2fd ]---
Kernel panic - not syncing: Fatal exception in interrupt

What happens while auth_enable=0 in that case is, that
ep->auth_hmacs is initialized to NULL in sctp_auth_init_hmacs()
when endpoint is being created.

After that point, if an admin switches over to auth_enable=1,
the machine can crash due to NULL pointer dereference during
reception of an INIT chunk. When we enter sctp_process_init()
via sctp_sf_do_5_1B_init() in order to respond to an INIT chunk,
the INIT verification succeeds and while we walk and process
all INIT params via sctp_process_param() we find that
net->sctp.auth_enable is set, therefore do not fall through,
but invoke sctp_auth_asoc_set_default_hmac() instead, and thus,
dereference what we have set to NULL during endpoint
initialization phase.

The fix is to make auth_enable immutable by caching its value
during endpoint initialization, so that its original value is
being carried along until destruction. The bug seems to originate
from the very first days.

Fix in joint work with Daniel Borkmann.
Reported-by: NJoshua Kinard <kumba@gentoo.org>
Signed-off-by: NVlad Yasevich <vyasevic@redhat.com>
Signed-off-by: NDaniel Borkmann <dborkman@redhat.com>
Acked-by: NNeil Horman <nhorman@tuxdriver.com>
Tested-by: NJoshua Kinard <kumba@gentoo.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      b14878cc
    • D
      libata/ahci: accommodate tag ordered controllers · 8a4aeec8
      由 Dan Williams 提交于 
      The AHCI spec allows implementations to issue commands in tag order
rather than FIFO order:

	5.3.2.12 P:SelectCmd
	HBA sets pSlotLoc = (pSlotLoc + 1) mod (CAP.NCS + 1)
	or HBA selects the command to issue that has had the
	PxCI bit set to '1' longer than any other command
	pending to be issued.

The result is that commands posted sequentially (time-wise) may play out
of sequence when issued by hardware.

This behavior has likely been hidden by drives that arrange for commands
to complete in issue order.  However, it appears recent drives (two from
different vendors that we have found so far) inflict out-of-order
completions as a matter of course.  So, we need to take care to maintain
ordered submission, otherwise we risk triggering a drive to fall out of
sequential-io automation and back to random-io processing, which incurs
large latency and degrades throughput.

This issue was found in simple benchmarks where QD=2 seq-write
performance was 30-50% *greater* than QD=32 seq-write performance.

Tagging for -stable and making the change globally since it has a low
risk-to-reward ratio.  Also, word is that recent versions of an unnamed
OS also does it this way now.  So, drives in the field are already
experienced with this tag ordering scheme.

Cc: <stable@vger.kernel.org>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Ed Ciechanowski <ed.ciechanowski@intel.com>
Reviewed-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NTejun Heo <tj@kernel.org>
      8a4aeec8
  23. 18 4月, 2014 3 次提交