Skip to content

R
raspberrypi-kernel

openeuler / raspberrypi-kernel

通知 13
Star 1
Fork 0
R
raspberrypi-kernel
切换分支/标签
  1. 15 2月, 2013 1 次提交
  3. 08 2月, 2013 1 次提交
  5. 10 1月, 2013 5 次提交
  7. 15 11月, 2012 1 次提交
  9. 09 10月, 2012 1 次提交
  11. 24 8月, 2012 1 次提交
  13. 12 4月, 2012 1 次提交
  15. 08 4月, 2012 3 次提交
  17. 16 3月, 2012 1 次提交
  19. 19 12月, 2011 1 次提交
    • A
      powerpc: POWER7 optimised copy_to_user/copy_from_user using VMX · a66086b8
      由 Anton Blanchard 提交于 
      Implement a POWER7 optimised copy_to_user/copy_from_user using VMX.
For large aligned copies this new loop is over 10% faster, and for
large unaligned copies it is over 200% faster.

If we take a fault we fall back to the old version, this keeps
things relatively simple and easy to verify.

On POWER7 unaligned stores rarely slow down - they only flush when
a store crosses a 4KB page boundary. Furthermore this flush is
handled completely in hardware and should be 20-30 cycles.

Unaligned loads on the other hand flush much more often - whenever
crossing a 128 byte cache line, or a 32 byte sector if either sector
is an L1 miss.

Considering this information we really want to get the loads aligned
and not worry about the alignment of the stores. Microbenchmarks
confirm that this approach is much faster than the current unaligned
copy loop that uses shifts and rotates to ensure both loads and
stores are aligned.

We also want to try and do the stores in cacheline aligned, cacheline
sized chunks. If the store queue is unable to merge an entire
cacheline of stores then the L2 cache will have to do a
read/modify/write. Even worse, we will serialise this with the stores
in the next iteration of the copy loop since both iterations hit
the same cacheline.

Based on this, the new loop does the following things:

1 - 127 bytes
Get the source 8 byte aligned and use 8 byte loads and stores. Pretty
boring and similar to how the current loop works.

128 - 4095 bytes
Get the source 8 byte aligned and use 8 byte loads and stores,
1 cacheline at a time. We aren't doing the stores in cacheline
aligned chunks so we will potentially serialise once per cacheline.
Even so it is much better than the loop we have today.

4096 - bytes
If both source and destination have the same alignment get them both
16 byte aligned, then get the destination cacheline aligned. Do
cacheline sized loads and stores using VMX.

If source and destination do not have the same alignment, we get the
destination cacheline aligned, and use permute to do aligned loads.

In both cases the VMX loop should be optimal - we always do aligned
loads and stores and are always doing stores in cacheline aligned,
cacheline sized chunks.

To be able to use VMX we must be careful about interrupts and
sleeping. We don't use the VMX loop when in an interrupt (which should
be rare anyway) and we wrap the VMX loop in disable/enable_pagefault
and fall back to the existing copy_tofrom_user loop if we do need to
sleep.

The VMX breakpoint of 4096 bytes was chosen using this microbenchmark:

http://ozlabs.org/~anton/junkcode/copy_to_user.c

Since we are using VMX and there is a cost to saving and restoring
the user VMX state there are two broad cases we need to benchmark:

- Best case - userspace never uses VMX

- Worst case - userspace always uses VMX

In reality a userspace process will sit somewhere between these two
extremes. Since we need to test both aligned and unaligned copies we
end up with 4 combinations. The point at which the VMX loop begins to
win is:

0% VMX
aligned		2048 bytes
unaligned	2048 bytes

100% VMX
aligned		16384 bytes
unaligned	8192 bytes

Considering this is a microbenchmark, the data is hot in cache and
the VMX loop has better store queue merging properties we set the
breakpoint to 4096 bytes, a little below the unaligned breakpoints.

Some future optimisations we can look at:

- Looking at the perf data, a significant part of the cost when a
  task is always using VMX is the extra exception we take to restore
  the VMX state. As such we should do something similar to the x86
  optimisation that restores FPU state for heavy users. ie:

        /*
         * If the task has used fpu the last 5 timeslices, just do a full
         * restore of the math state immediately to avoid the trap; the
         * chances of needing FPU soon are obviously high now
         */
        preload_fpu = tsk_used_math(next_p) && next_p->fpu_counter > 5;

  and

        /*
         * fpu_counter contains the number of consecutive context switches
         * that the FPU is used. If this is over a threshold, the lazy fpu
         * saving becomes unlazy to save the trap. This is an unsigned char
         * so that after 256 times the counter wraps and the behavior turns
         * lazy again; this to deal with bursty apps that only use FPU for
         * a short time
         */

- We could create a paca bit to mirror the VMX enabled MSR bit and check
  that first, avoiding multiple calls to calling enable_kernel_altivec.
  That should help with iovec based system calls like readv.

- We could have two VMX breakpoints, one for when we know the user VMX
  state is loaded into the registers and one when it isn't. This could
  be a second bit in the paca so we can calculate the break points quickly.

- One suggestion from Ben was to save and restore the VSX registers
  we use inline instead of using enable_kernel_altivec.

[BenH: Fixed a problem with preempt and fixed build without CONFIG_ALTIVEC]
Signed-off-by: NAnton Blanchard <anton@samba.org>
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>
      a66086b8
  21. 25 11月, 2011 1 次提交
  23. 12 7月, 2011 1 次提交
    • P
      powerpc, KVM: Split HVMODE_206 cpu feature bit into separate HV and architecture bits · 969391c5
      由 Paul Mackerras 提交于 
      This replaces the single CPU_FTR_HVMODE_206 bit with two bits, one to
indicate that we have a usable hypervisor mode, and another to indicate
that the processor conforms to PowerISA version 2.06.  We also add
another bit to indicate that the processor conforms to ISA version 2.01
and set that for PPC970 and derivatives.

Some PPC970 chips (specifically those in Apple machines) have a
hypervisor mode in that MSR[HV] is always 1, but the hypervisor mode
is not useful in the sense that there is no way to run any code in
supervisor mode (HV=0 PR=0).  On these processors, the LPES0 and LPES1
bits in HID4 are always 0, and we use that as a way of detecting that
hypervisor mode is not useful.

Where we have a feature section in assembly code around code that
only applies on POWER7 in hypervisor mode, we use a construct like

END_FTR_SECTION_IFSET(CPU_FTR_HVMODE | CPU_FTR_ARCH_206)

The definition of END_FTR_SECTION_IFSET is such that the code will
be enabled (not overwritten with nops) only if all bits in the
provided mask are set.

Note that the CPU feature check in __tlbie() only needs to check the
ARCH_206 bit, not the HVMODE bit, because __tlbie() can only get called
if we are running bare-metal, i.e. in hypervisor mode.
Signed-off-by: NPaul Mackerras <paulus@samba.org>
Signed-off-by: NAlexander Graf <agraf@suse.de>
      969391c5
  25. 19 5月, 2011 1 次提交
  27. 04 5月, 2011 2 次提交
    • P
      powerpc: Save Come-From Address Register (CFAR) in exception frame · 48404f2e
      由 Paul Mackerras 提交于 
      Recent 64-bit server processors (POWER6 and POWER7) have a "Come-From
Address Register" (CFAR), that records the address of the most recent
branch or rfid (return from interrupt) instruction for debugging purposes.

This saves the value of the CFAR in the exception entry code and stores
it in the exception frame.  We also make xmon print the CFAR value in
its register dump code.

Rather than extend the pt_regs struct at this time, we steal the orig_gpr3
field, which is only used for system calls, and use it for the CFAR value
for all exceptions/interrupts other than system calls.  This means we
don't save the CFAR on system calls, which is not a great problem since
system calls tend not to happen unexpectedly, and also avoids adding the
overhead of reading the CFAR to the system call entry path.
Signed-off-by: NPaul Mackerras <paulus@samba.org>
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>
      48404f2e
    • T
      powerpc: Add Initiate Coprocessor Store Word (icswx) support · 851d2e2f
      由 Tseng-Hui (Frank) Lin 提交于 
      Icswx is a PowerPC instruction to send data to a co-processor. On Book-S
processors the LPAR_ID and process ID (PID) of the owning process are
registered in the window context of the co-processor at initialization
time. When the icswx instruction is executed the L2 generates a cop-reg
transaction on PowerBus. The transaction has no address and the
processor does not perform an MMU access to authenticate the transaction.
The co-processor compares the LPAR_ID and the PID included in the
transaction and the LPAR_ID and PID held in the window context to
determine if the process is authorized to generate the transaction.

The OS needs to assign a 16-bit PID for the process. This cop-PID needs
to be updated during context switch. The cop-PID needs to be destroyed
when the context is destroyed.
Signed-off-by: NSonny Rao <sonnyrao@linux.vnet.ibm.com>
Signed-off-by: NTseng-Hui (Frank) Lin <thlin@linux.vnet.ibm.com>
Signed-off-by: NAnton Blanchard <anton@samba.org>
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>
      851d2e2f
  29. 27 4月, 2011 2 次提交
  31. 20 4月, 2011 1 次提交
  33. 12 4月, 2011 2 次提交
  35. 02 2月, 2011 1 次提交
  37. 29 11月, 2010 1 次提交
  39. 02 9月, 2010 1 次提交
    • A
      powerpc: Feature nop out reservation clear when stcx checks address · f89451fb
      由 Anton Blanchard 提交于 
      The POWER architecture does not require stcx to check that it is operating
on the same address as the larx. This means it is possible for an
an exception handler to execute a larx, get a reservation, decide
not to do the stcx and then return back with an active reservation. If the
interrupted code was in the middle of a larx/stcx sequence the stcx could
incorrectly succeed.

All recent POWER CPUs check the address before letting the stcx succeed
so we can create a CPU feature and nop it out. As Ben suggested, we can
only do this in our syscall path because there is a remote possibility
some kernel code gets interrupted by an exception that ends up operating
on the same cacheline.

Thanks to Paul Mackerras and Derek Williams for the idea.

To test this I used a very simple null syscall (actually getppid) testcase
at http://ozlabs.org/~anton/junkcode/null_syscall.c

I tested against 2.6.35-git10 with the following changes against the
pseries_defconfig:

CONFIG_VIRT_CPU_ACCOUNTING=n
CONFIG_AUDIT=n
CONFIG_PPC_4K_PAGES=n
CONFIG_PPC_64K_PAGES=y
CONFIG_FORCE_MAX_ZONEORDER=9
CONFIG_PPC_SUBPAGE_PROT=n
CONFIG_FUNCTION_TRACER=n
CONFIG_FUNCTION_GRAPH_TRACER=n
CONFIG_IRQSOFF_TRACER=n
CONFIG_STACK_TRACER=n

to remove the overhead of virtual CPU accounting, syscall auditing and
the ftrace mcount tracers. 64kB pages were enabled to minimise TLB misses.

POWER6: +8.2%
POWER7: +7.0%

Another suggestion was to use a larx to something in the L1 instead of a stcx.
This was almost as fast as removing the larx on POWER6, but only 3.5% faster
on POWER7. We can use this to speed up the reservation clear in our
exception exit code.
Signed-off-by: NAnton Blanchard <anton@samba.org>
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>
      f89451fb
  41. 22 6月, 2010 1 次提交
    • K
      powerpc, hw_breakpoints: Implement hw_breakpoints for 64-bit server processors · 5aae8a53
      由 K.Prasad 提交于 
      Implement perf-events based hw-breakpoint interfaces for PowerPC
64-bit server (Book III S) processors.  This allows access to a
given location to be used as an event that can be counted or
profiled by the perf_events subsystem.

This is done using the DABR (data breakpoint register), which can
also be used for process debugging via ptrace.  When perf_event
hw_breakpoint support is configured in, the perf_event subsystem
manages the DABR and arbitrates access to it, and ptrace then
creates a perf_event when it is requested to set a data breakpoint.

[Adopted suggestions from Paul Mackerras <paulus@samba.org> to
- emulate_step() all system-wide breakpoints and single-step only the
  per-task breakpoints
- perform arch-specific cleanup before unregistration through
  arch_unregister_hw_breakpoint()
]
Signed-off-by: NK.Prasad <prasad@linux.vnet.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>
      5aae8a53
  43. 09 6月, 2010 1 次提交
    • M
      powerpc: Enable asymmetric SMT scheduling on POWER7 · 76cbd8a8
      由 Michael Neuling 提交于 
      The POWER7 core has dynamic SMT mode switching which is controlled by
the hypervisor.  There are 3 SMT modes:
	SMT1 uses thread  0
	SMT2 uses threads 0 & 1
	SMT4 uses threads 0, 1, 2 & 3
When in any particular SMT mode, all threads have the same performance
as each other (ie. at any moment in time, all threads perform the same).

The SMT mode switching works such that when linux has threads 2 & 3 idle
and 0 & 1 active, it will cede (H_CEDE hypercall) threads 2 and 3 in the
idle loop and the hypervisor will automatically switch to SMT2 for that
core (independent of other cores).  The opposite is not true, so if
threads 0 & 1 are idle and 2 & 3 are active, we will stay in SMT4 mode.

Similarly if thread 0 is active and threads 1, 2 & 3 are idle, we'll go
into SMT1 mode.

If we can get the core into a lower SMT mode (SMT1 is best), the threads
will perform better (since they share less core resources).  Hence when
we have idle threads, we want them to be the higher ones.

This adds a feature bit for asymmetric packing to powerpc and then
enables it on POWER7.
Signed-off-by: NMichael Neuling <mikey@neuling.org>
Signed-off-by: NPeter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: linuxppc-dev@ozlabs.org
LKML-Reference: <20100608045702.31FB5CC8C7@localhost.localdomain>
Signed-off-by: NIngo Molnar <mingo@elte.hu>
      76cbd8a8
  45. 21 5月, 2010 1 次提交
  47. 05 5月, 2010 2 次提交
  49. 17 2月, 2010 1 次提交
    • A
      powerpc: Use lwsync for acquire barrier if CPU supports it · 5a0e9b57
      由 Anton Blanchard 提交于 
      Nick Piggin discovered that lwsync barriers around locks were faster than isync
on 970. That was a long time ago and I completely dropped the ball in testing
his patches across other ppc64 processors.

Turns out the idea helps on other chips. Using a microbenchmark that
uses a lot of threads to contend on a global pthread mutex (and therefore a
global futex), POWER6 improves 8% and POWER7 improves 2%. I checked POWER5
and while I couldn't measure an improvement, there was no regression.

This patch uses the lwsync patching code to replace the isyncs with lwsyncs
on CPUs that support the instruction. We were marking POWER3 and RS64 as lwsync
capable but in reality they treat it as a full sync (ie slow). Remove the
CPU_FTR_LWSYNC bit from these CPUs so they continue to use the faster isync
method.
Signed-off-by: NAnton Blanchard <anton@samba.org>
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>
      5a0e9b57
  51. 17 3月, 2009 1 次提交
  53. 23 2月, 2009 1 次提交
  55. 21 12月, 2008 2 次提交
  57. 16 12月, 2008 1 次提交