Skip to content

cloud-kernel
cloud-kernel

openanolis / cloud-kernel
大约 1 年 前同步成功

通知 158
Star 36
Fork 7
cloud-kernel
cloud-kernel

体验新版 GitCode，发现更多精彩内容 >>

切换分支/标签
  1. 13 11月, 2008 4 次提交
  3. 11 11月, 2008 4 次提交
  5. 09 11月, 2008 1 次提交
    • I
      sched: optimize sched_clock() a bit · 7cbaef9c
      由 Ingo Molnar 提交于 
      sched_clock() uses cycles_2_ns() needlessly - which is an irq-disabling
variant of __cycles_2_ns().

Most of the time sched_clock() is called with irqs disabled already.
The few places that call it with irqs enabled need to be updated.
Signed-off-by: NIngo Molnar <mingo@elte.hu>
      7cbaef9c
  7. 08 11月, 2008 2 次提交
    • I
      sched: improve sched_clock() performance · 0d12cdd5
      由 Ingo Molnar 提交于 
      in scheduler-intense workloads native_read_tsc() overhead accounts for
20% of the system overhead:

 659567 system_call                              41222.9375
 686796 schedule                                 435.7843
 718382 __switch_to                              665.1685
 823875 switch_mm                                4526.7857
 1883122 native_read_tsc                          55385.9412
 9761990 total                                      2.8468

this is large part due to the rdtsc_barrier() that is done before
and after reading the TSC.

But sched_clock() is not a precise clock in the GTOD sense, using such
barriers is completely pointless. So remove the barriers and only use
them in vget_cycles().

This improves lat_ctx performance by about 5%.
Signed-off-by: NIngo Molnar <mingo@elte.hu>
      0d12cdd5
    • A
      oprofile: Fix p6 counter overflow check · 7c64ade5
      由 Andi Kleen 提交于 
      Fix the counter overflow check for CPUs with counter width > 32

I had a similar change in a different patch that I didn't submit
and I didn't notice the problem earlier because it was always
tested together.
Signed-off-by: NAndi Kleen <ak@linux.intel.com>
Signed-off-by: NRobert Richter <robert.richter@amd.com>
      7c64ade5
  9. 07 11月, 2008 2 次提交
    • J
      xen: make sure stray alias mappings are gone before pinning · d05fdf31
      由 Jeremy Fitzhardinge 提交于 
      Xen requires that all mappings of pagetable pages are read-only, so
that they can't be updated illegally.  As a result, if a page is being
turned into a pagetable page, we need to make sure all its mappings
are RO.

If the page had been used for ioremap or vmalloc, it may still have
left over mappings as a result of not having been lazily unmapped.
This change makes sure we explicitly mop them all up before pinning
the page.

Unlike aliases created by kmap, the there can be vmalloc aliases even
for non-high pages, so we must do the flush unconditionally.
Signed-off-by: NJeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Linux Memory Management List <linux-mm@kvack.org>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: NIngo Molnar <mingo@elte.hu>
      d05fdf31
    • J
      x86, xen: fix use of pgd_page now that it really does return a page · 47cb2ed9
      由 Jeremy Fitzhardinge 提交于 
      Impact: fix 32-bit Xen guest boot crash

On 32-bit PAE, pud_page, for no good reason, didn't really return a
struct page *.  Since Jan Beulich's fix "i386/PAE: fix pud_page()",
pud_page does return a struct page *.

Because PAE has 3 pagetable levels, the pud level is folded into the
pgd level, so pgd_page() is the same as pud_page(), and now returns
a struct page *.  Update the xen/mmu.c code which uses pgd_page()
accordingly.
Signed-off-by: NJeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>
      47cb2ed9
  11. 06 11月, 2008 9 次提交
    • E
      Revert "x86: default to reboot via ACPI" · 8d00450d
      由 Eduardo Habkost 提交于 
      This reverts commit c7ffa6c2.

the assumptio of this change was that this would not break
any existing machine. Andrey Borzenkov reported troubles with
the ACPI reboot method: the system would hang on reboot, necessiating
a power cycle. Probably more systems are affected as well.

Also, there are patches queued up for v2.6.29 to disable virtualization
on emergency_restart() - which was the original motivation of
this change.
Reported-by: NAndrey Borzenkov <arvidjaar@mail.ru>
Bisected-by: NAndrey Borzenkov <arvidjaar@mail.ru>
Signed-off-by: NEduardo Habkost <ehabkost@redhat.com>
Acked-by: NAvi Kivity <avi@redhat.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>
      8d00450d
    • H
      x86: align DirectMap in /proc/meminfo · b9c3bfc2
      由 Hugh Dickins 提交于 
      Impact: right-align /proc/meminfo consistent with other fields

When the split-LRU patches added Inactive(anon) and Inactive(file) lines
to /proc/meminfo, all counts were moved two columns rightwards to fit in.
Now move x86's DirectMap lines two columns rightwards to line up.
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>
      b9c3bfc2
    • J
      AMD IOMMU: fix lazy IO/TLB flushing in unmap path · 80be308d
      由 Joerg Roedel 提交于 
      Lazy flushing needs to take care of the unmap path too which is not yet
implemented and leads to stale IO/TLB entries. This is fixed by this
patch.
Signed-off-by: NJoerg Roedel <joerg.roedel@amd.com>
      80be308d
    • S
      x86: add smp_mb() before sending INVALIDATE_TLB_VECTOR · d6f0f39b
      由 Suresh Siddha 提交于 
      Impact: fix rare x2apic hang

On x86, x2apic mode accesses for sending IPI's don't have serializing
semantics. If the IPI receivner refers(in lock-free fashion) to some
memory setup by the sender, the need for smp_mb() before sending the
IPI becomes critical in x2apic mode.

Add the smp_mb() in native_flush_tlb_others() before sending the IPI.
Signed-off-by: NSuresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>
      d6f0f39b
    • Y
      x86: remove VISWS and PARAVIRT around NR_IRQS puzzle · 7db282fa
      由 Yinghai Lu 提交于 
      Impact: fix warning message when PARAVIRT is set in config

Remove stale #ifdef components from our IRQ sizing logic.
x86/Voyager is the only holdout.
Signed-off-by: NYinghai Lu <yinghai@kernel.org>
Signed-off-by: NIngo Molnar <mingo@elte.hu>
      7db282fa
    • B
      x86: mention ACPI in top-level Kconfig menu · da85f865
      由 Bjorn Helgaas 提交于 
      Impact: clarify menuconfig text

Mention ACPI in the top-level menu to give a clue as to where
it lives. This matches what ia64 does.
Signed-off-by: NBjorn Helgaas <bjorn.helgaas@hp.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>
      da85f865
    • Y
      x86: size NR_IRQS on 32-bit systems the same way as 64-bit · 1b489768
      由 Yinghai Lu 提交于 
      Impact: make NR_IRQS big enough for system with lots of apic/pins

If lots of IO_APIC's are there (or can be there), size the same way
as 64-bit, depending on MAX_IO_APICS and NR_CPUS.

This fixes the boot problem reported by Ben Hutchings on a 32-bit
server with 5 IO-APICs and 240 IO-APIC pins.
Signed-off-by: NYinghai <yinghai@kernel.org>
Tested-by: NBen Hutchings <bhutchings@solarflare.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>
      1b489768
    • B
      x86: don't allow nr_irqs > NR_IRQS · c78d0cf2
      由 Ben Hutchings 提交于 
      Impact: fix boot hang on 32-bit systems with more than 224 IO-APIC pins

On some 32-bit systems with a lot of IO-APICs probe_nr_irqs() can
return a value larger than NR_IRQS. This will lead to probe_irq_on()
overrunning the irq_desc array.

I hit this when running net-next-2.6 (close to 2.6.28-rc3) on a
Supermicro dual Xeon system.  NR_IRQS is 224 but probe_nr_irqs() detects
5 IOAPICs and returns 240.  Here are the log messages:

Tue Nov  4 16:53:47 2008 ACPI: IOAPIC (id[0x01] address[0xfec00000] gsi_base[0])
Tue Nov  4 16:53:47 2008 IOAPIC[0]: apic_id 1, version 32, address 0xfec00000, GSI 0-23
Tue Nov  4 16:53:47 2008 ACPI: IOAPIC (id[0x02] address[0xfec81000] gsi_base[24])
Tue Nov  4 16:53:47 2008 IOAPIC[1]: apic_id 2, version 32, address 0xfec81000, GSI 24-47
Tue Nov  4 16:53:47 2008 ACPI: IOAPIC (id[0x03] address[0xfec81400] gsi_base[48])
Tue Nov  4 16:53:47 2008 IOAPIC[2]: apic_id 3, version 32, address 0xfec81400, GSI 48-71
Tue Nov  4 16:53:47 2008 ACPI: IOAPIC (id[0x04] address[0xfec82000] gsi_base[72])
Tue Nov  4 16:53:47 2008 IOAPIC[3]: apic_id 4, version 32, address 0xfec82000, GSI 72-95
Tue Nov  4 16:53:47 2008 ACPI: IOAPIC (id[0x05] address[0xfec82400] gsi_base[96])
Tue Nov  4 16:53:47 2008 IOAPIC[4]: apic_id 5, version 32, address 0xfec82400, GSI 96-119
Tue Nov  4 16:53:47 2008 ACPI: INT_SRC_OVR (bus 0 bus_irq 0 global_irq 2 high edge)
Tue Nov  4 16:53:47 2008 ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 9 high level)
Tue Nov  4 16:53:47 2008 Enabling APIC mode:  Flat.  Using 5 I/O APICs
Signed-off-by: NBen Hutchings <bhutchings@solarflare.com>
Acked-by: NYinghai Lu <yhlu.kernel@gmail.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>
      c78d0cf2
    • I
      sched: re-tune balancing · 9fcd18c9
      由 Ingo Molnar 提交于 
      Impact: improve wakeup affinity on NUMA systems, tweak SMP systems

Given the fixes+tweaks to the wakeup-buddy code, re-tweak the domain
balancing defaults on NUMA and SMP systems.

Turn on SD_WAKE_AFFINE which was off on x86 NUMA - there's no reason
why we would not want to have wakeup affinity across nodes as well.
(we already do this in the standard NUMA template.)

lat_ctx on a NUMA box is particularly happy about this change:

before:

 |   phoenix:~/l> ./lat_ctx -s 0 2
 |   "size=0k ovr=2.60
 |   2 5.70

after:

 |   phoenix:~/l> ./lat_ctx -s 0 2
 |   "size=0k ovr=2.65
 |   2 2.07

a 2.75x speedup.

pipe-test is similarly happy about it too:

 |  phoenix:~/sched-tests> ./pipe-test
 |   18.26 usecs/loop.
 |   14.70 usecs/loop.
 |   14.38 usecs/loop.
 |   10.55 usecs/loop.              # +WAKE_AFFINE on domain0+domain1
 |   8.63 usecs/loop.
 |   8.59 usecs/loop.
 |   9.03 usecs/loop.
 |   8.94 usecs/loop.
 |   8.96 usecs/loop.
 |   8.63 usecs/loop.

Also:

 - disable SD_BALANCE_NEWIDLE on NUMA and SMP domains (keep it for siblings)
 - enable SD_WAKE_BALANCE on SMP domains

Sysbench+postgresql improves all around the board, quite significantly:

           .28-rc3-11474e2c  .28-rc3-11474e2c-tune
-------------------------------------------------
    1:             571              688    +17.08%
    2:            1236             1206    -2.55%
    4:            2381             2642    +9.89%
    8:            4958             5164    +3.99%
   16:            9580             9574    -0.07%
   32:            7128             8118    +12.20%
   64:            7342             8266    +11.18%
  128:            7342             8064    +8.95%
  256:            7519             7884    +4.62%
  512:            7350             7731    +4.93%
-------------------------------------------------
  SUM:           55412            59341    +6.62%

So it's a win both for the runup portion, the peak area and the tail.
Signed-off-by: NIngo Molnar <mingo@elte.hu>
      9fcd18c9
  13. 04 11月, 2008 2 次提交
  15. 03 11月, 2008 1 次提交
  17. 02 11月, 2008 1 次提交
    • L
      x86: Clean up late e820 resource allocation · 1f987577
      由 Linus Torvalds 提交于 
      This makes the late e820 resources use 'insert_resource_expand_to_fit()'
instead of doing a 'reserve_region_with_split()', and also avoids
marking them as IORESOURCE_BUSY.

This results in us being perfectly happy to use pre-existing PCI
resources even if they were marked as being in a reserved region, while
still avoiding any _new_ allocations in the reserved regions.  It also
makes for a simpler and more accurate resource tree.

Example resource allocation from Jonathan Corbet, who has firmware that
has an e820 reserved entry that covered a big range (e0000000-fed003ff),
and that had various PCI resources in it set up by firmware.

With old kernels, the reserved range would force us to re-allocate all
pre-existing PCI resources, and his reserved range would end up looking
like this:

	e0000000-fed003ff : reserved
	  fec00000-fec00fff : IOAPIC 0
	  fed00000-fed003ff : HPET 0

where only the pre-allocated special regions (IOAPIC and HPET) were kept
around.

With 2.6.28-rc2, which uses 'reserve_region_with_split()', Jonathan's
resource tree looked like this:

	e0000000-fe7fffff : reserved
	fe800000-fe8fffff : PCI Bus 0000:01
	 fe800000-fe8fffff : reserved
	fe900000-fe9d9aff : reserved
	fe9d9b00-fe9d9bff : 0000:00:1f.3
	 fe9d9b00-fe9d9bff : reserved
	fe9d9c00-fe9d9fff : 0000:00:1a.7
	 fe9d9c00-fe9d9fff : reserved
	fe9da000-fe9dafff : 0000:00:03.3
	 fe9da000-fe9dafff : reserved
	fe9db000-fe9dbfff : 0000:00:19.0
	 fe9db000-fe9dbfff : reserved
	fe9dc000-fe9dffff : 0000:00:1b.0
	 fe9dc000-fe9dffff : reserved
	fe9e0000-fe9fffff : 0000:00:19.0
	 fe9e0000-fe9fffff : reserved
	fea00000-fea7ffff : 0000:00:02.0
	 fea00000-fea7ffff : reserved
	fea80000-feafffff : 0000:00:02.1
	 fea80000-feafffff : reserved
	feb00000-febfffff : 0000:00:02.0
	 feb00000-febfffff : reserved
	fec00000-fed003ff : reserved
	 fec00000-fec00fff : IOAPIC 0
	 fed00000-fed003ff : HPET 0

and because the reserved entry had been split and moved into the
individual resources, and because it used the IORESOURCE_BUSY flag, the
drivers that actually wanted to _use_ those resources couldn't actually
attach to them:

	e1000e 0000:00:19.0: BAR 0: can't reserve mem region [0xfe9e0000-0xfe9fffff]
	HDA Intel 0000:00:1b.0: BAR 0: can't reserve mem region [0xfe9dc000-0xfe9dffff]

with this patch, the resource tree instead becomes

	e0000000-fed003ff : reserved
	  fe800000-fe8fffff : PCI Bus 0000:01
	  fe9d9b00-fe9d9bff : 0000:00:1f.3
	  fe9d9c00-fe9d9fff : 0000:00:1a.7
	    fe9d9c00-fe9d9fff : ehci_hcd
	  fe9da000-fe9dafff : 0000:00:03.3
	  fe9db000-fe9dbfff : 0000:00:19.0
	    fe9db000-fe9dbfff : e1000e
	  fe9dc000-fe9dffff : 0000:00:1b.0
	    fe9dc000-fe9dffff : ICH HD audio
	  fe9e0000-fe9fffff : 0000:00:19.0
	    fe9e0000-fe9fffff : e1000e
	  fea00000-fea7ffff : 0000:00:02.0
	  fea80000-feafffff : 0000:00:02.1
	  feb00000-febfffff : 0000:00:02.0
	  fec00000-fec00fff : IOAPIC 0
	  fed00000-fed003ff : HPET 0

ie the one reserved region now ends up surrounding all the PCI resources
that were allocated inside of it by firmware, and because it is not
marked BUSY, drivers have no problem attaching to the pre-allocated
resources.
Reported-and-tested-by: NJonathan Corbet <corbet@lwn.net>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Robert Hancock <hancockr@shaw.ca>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      1f987577
  19. 31 10月, 2008 14 次提交