The main goal of this is to generate timer interrupts only when there is
some work to do. When CPU is busy interrupts are generating at full rate
of hz + stathz to fullfill scheduler and timekeeping requirements. But
when CPU is idle, only minimum set of interrupts (down to 8 interrupts per
second per CPU now), needed to handle scheduled callouts is executed.
This allows significantly increase idle CPU sleep time, increasing effect
of static power-saving technologies. Also it should reduce host CPU load
on virtualized systems, when guest system is idle.
There is set of tunables, also available as writable sysctls, allowing to
control wanted event timer subsystem behavior:
kern.eventtimer.timer - allows to choose event timer hardware to use.
On x86 there is up to 4 different kinds of timers. Depending on whether
chosen timer is per-CPU, behavior of other options slightly differs.
kern.eventtimer.periodic - allows to choose periodic and one-shot
operation mode. In periodic mode, current timer hardware taken as the only
source of time for time events. This mode is quite alike to previous kernel
behavior. One-shot mode instead uses currently selected time counter
hardware to schedule all needed events one by one and program timer to
generate interrupt exactly in specified time. Default value depends of
chosen timer capabilities, but one-shot mode is preferred, until other is
forced by user or hardware.
kern.eventtimer.singlemul - in periodic mode specifies how much times
higher timer frequency should be, to not strictly alias hardclock() and
statclock() events. Default values are 2 and 4, but could be reduced to 1
if extra interrupts are unwanted.
kern.eventtimer.idletick - makes each CPU to receive every timer interrupt
independently of whether they busy or not. By default this options is
disabled. If chosen timer is per-CPU and runs in periodic mode, this option
has no effect - all interrupts are generating.
As soon as this patch modifies cpu_idle() on some platforms, I have also
refactored one on x86. Now it makes use of MONITOR/MWAIT instrunctions
(if supported) under high sleep/wakeup rate, as fast alternative to other
methods. It allows SMP scheduler to wake up sleeping CPUs much faster
without using IPI, significantly increasing performance on some highly
task-switching loads.
Tested by: many (on i386, amd64, sparc64 and powerc)
H/W donated by: Gheorghe Ardelean
Sponsored by: iXsystems, Inc.
PMAP_DIAGNOSTIC was eliminated from amd64/i386, and, in fact, the
non-MIPS parts of the kernel, several years ago. Any of the interesting
checks were turned into KASSERT()s. Basically, the motivation was that
lots of people run with INVARIANTS but no one runs with DIAGNOSTIC.
panic strings needn't and shouldn't have a terminating newline.
Finally, there is one functional change. The sched_pin() in
pmap_remove_pages() is an artifact of the way we temporarily map page
table pages on i386. (The mappings are processor private. We don't do
a system-wide shootdown.) It isn't needed by MIPS.
Tested by: jchandra
Submitted by: alc
1. On n64, use XKPHYS to map page table pages instead of KSEG0. Maintain
just one freepages list on n64.
The changes are mainly to introduce MIPS_PHYS_TO_DIRECT(pa),
MIPS_DIRECT_TO_PHYS(), which will use KSEG0 in 32 bit compilation
and XKPHYS in 64 bit compilation.
2. Change macro based PMAP_LMEM_MAP1(), PMAP_LMEM_MAP2(), PMAP_LMEM_UNMAP()
to inline functions.
3. Introduce MIPS_DIRECT_MAPPABLE(pa), which will further reduce the cases
in which we will need to have a special case for 64 bit compilation.
4. Update CP0 hazard definitions for CPU_RMI - the cpu does not need any
nops
Reviewed by: neel
In particular, provide pagesize and pagesizes array, the canary value
for SSP use, number of host CPUs and osreldate.
Tested by: marius (sparc64)
MFC after: 1 month
1. Move dirty bit emulation code that is duplicted for kernel and user
in trap.c to a function pmap_emulate_modified() in pmap.c.
2. While doing dirty bit emulation, it is not necessary to update the
TLB entry on all CPUs using smp_rendezvous(), we can just update the
TLB entry on the current CPU, and let the other CPUs update their TLB
entry lazily if they get an exception.
Reviewed by: alc, neel
r211130 in favor of this more general fix.
This fixes a compilation error for mips 64-bit little endian build.
libexec/rtld-elf/mips/reloc.c:196: warning: right shift count >= width of type
Suggested by: stefanf, jchandra, bde
IPI to a specific CPU by its cpuid. Replace calls to ipi_selected() that
constructed a mask for a single CPU with calls to ipi_cpu() instead. This
will matter more in the future when we transition from cpumask_t to
cpuset_t for CPU masks in which case building a CPU mask is more expensive.
Submitted by: peter, sbruno
Reviewed by: rookie
Obtained from: Yahoo! (x86)
MFC after: 1 month
pmap_page_wired_mappings() counts the number of pv entries for the
specified page that have the pv entry wired flag set to TRUE.
pmap_enter() correctly initializes this flag. However,
pmap_change_wiring() doesn't update the corresponding pv entry flag,
only the PTE. So, the count returned by pmap_page_wired_mappings()
will sometimes be wrong.
In the short term, the best fix would be to eliminate the pv entry
flag and use only the PTE. That flag is wasting non-trivial memory.
Remove pv_wired flag, and use PTE flag to count the wired mappings.
Reviewed by: alc
- 32 bit compilation will still use old 2 level page tables
- re-arrange pmap code so that adding another level is easier
- pmap code for 3 level page tables for n64
- update TLB handler to traverse 3 levels in n64
Reviewed by: jmallett
now it uses a very dumb first-touch allocation policy. This will change in
the future.
- Each architecture indicates the maximum number of supported memory domains
via a new VM_NDOMAIN parameter in <machine/vmparam.h>.
- Each cpu now has a PCPU_GET(domain) member to indicate the memory domain
a CPU belongs to. Domain values are dense and numbered from 0.
- When a platform supports multiple domains, the default freelist
(VM_FREELIST_DEFAULT) is split up into N freelists, one for each domain.
The MD code is required to populate an array of mem_affinity structures.
Each entry in the array defines a range of memory (start and end) and a
domain for the range. Multiple entries may be present for a single
domain. The list is terminated by an entry where all fields are zero.
This array of structures is used to split up phys_avail[] regions that
fall in VM_FREELIST_DEFAULT into per-domain freelists.
- Each memory domain has a separate lookup-array of freelists that is
used when fulfulling a physical memory allocation. Right now the
per-domain freelists are listed in a round-robin order for each domain.
In the future a table such as the ACPI SLIT table may be used to order
the per-domain lookup lists based on the penalty for each memory domain
relative to a specific domain. The lookup lists may be examined via a
new vm.phys.lookup_lists sysctl.
- The first-touch policy is implemented by using PCPU_GET(domain) to
pick a lookup list when allocating memory.
Reviewed by: alc
alc@.
The UMA zone based allocation is replaced by a scheme that creates
a new free page list for the KSEG0 region, and a new function
in sys/vm that allocates pages from a specific free page list.
This also fixes a race condition introduced by the UMA based page table
page allocation code. Dropping the page queue and pmap locks before
the call to uma_zfree, and re-acquiring them afterwards will introduce
a race condtion(noted by alc@).
The changes are :
- Revert the earlier changes in MIPS pmap.c that added UMA zone for
page table pages.
- Add a new freelist VM_FREELIST_HIGHMEM to MIPS vmparam.h for memory that
is not directly mapped (in 32bit kernel). Normal page allocations will first
try the HIGHMEM freelist and then the default(direct mapped) freelist.
- Add a new function 'vm_page_t vm_page_alloc_freelist(int flind, int
order, int req)' to vm/vm_page.c to allocate a page from a specified
freelist. The MIPS page table pages will be allocated using this function
from the freelist containing direct mapped pages.
- Move the page initialization code from vm_phys_alloc_contig() to a
new function vm_page_alloc_init(), and use this function to initialize
pages in vm_page_alloc_freelist() too.
- Split the function vm_phys_alloc_pages(int pool, int order) to create
vm_phys_alloc_freelist_pages(int flind, int pool, int order), and use
this function from both vm_page_alloc_freelist() and vm_phys_alloc_pages().
Reviewed by: alc
library:
o) Increase inline unit / large function growth limits for MIPS to accommodate
the needs of the Simple Executive, which uses a shocking amount of inlining.
o) Remove TARGET_OCTEON and use CPU_CNMIPS to do things required by cnMIPS and
the Octeon SoC.
o) Add OCTEON_VENDOR_LANNER to use Lanner's allocation of vendor-specific
board numbers, specifically to support the MR320.
o) Add OCTEON_BOARD_CAPK_0100ND to hard-wire configuration for the CAPK-0100nd,
which improperly uses an evaluation board's board number and breaks board
detection at runtime. This board is sold by Portwell as the CAM-0100.
o) Add support for the RTC available on some Octeon boards.
o) Add support for the Octeon PCI bus. Note that rman_[sg]et_virtual for IO
ports can not work unless building for n64.
o) Clean up the CompactFlash driver to use Simple Executive macros and
structures where possible (it would be advisable to use the Simple Executive
API to set the PIO mode, too, but that is not done presently.) Also use
structures from FreeBSD's ATA layer rather than structures copied from
Linux.
o) Print available Octeon SoC features on boot.
o) Add support for the Octeon timecounter.
o) Use the Simple Executive's routines rather than local copies for doing reads
and writes to 64-bit addresses and use its macros for various device
addresses rather than using local copies.
o) Rename octeon_board_real to octeon_is_simulation to reduce differences with
Cavium-provided code originally written for Linux. Also make it use the
same simplified test that the Simple Executive and Linux both use rather
than our complex one.
o) Add support for the Octeon CIU, which is the main interrupt unit, as a bus
to use normal interrupt allocation and setup routines.
o) Use the Simple Executive's bootmem facility to allocate physical memory for
the kernel, rather than assuming we know which addresses we can steal.
NB: This may reduce the amount of RAM the kernel reports you as having if
you are leaving large temporary allocations made by U-Boot allocated
when starting FreeBSD.
o) Add a port of the Cavium-provided Ethernet driver for Linux. This changes
Ethernet interface naming from rgmxN to octeN. The new driver has vast
improvements over the old one, both in performance and functionality, but
does still have some features which have not been ported entirely and there
may be unimplemented code that can be hit in everyday use. I will make
every effort to correct those as they are reported.
o) Support loading the kernel on non-contiguous cores.
o) Add very conservative support for harvesting randomness from the Octeon
random number device.
o) Turn SMP on by default.
o) Clean up the style of the Octeon kernel configurations a little and make
them compile with -march=octeon.
o) Add support for the Lanner MR320 and the CAPK-0100nd to the Simple
Executive.
o) Modify the Simple Executive to build on FreeBSD and to build without
executive-config.h or cvmx-config.h. In the future we may want to
revert part of these changes and supply executive-config.h and
cvmx-config.h and access to the options contained in those files via
kernel configuration files.
o) Modify the Simple Executive USB routines to support getting and setting
of the USB PID.
Move inappropriate stuff in cpu.h elsewhere:
{s,g}et_intr_mask -> md_var.h
num_tlbentries -> tlb.h
Remove #define clockframe trapframe and fix clock, which was the only place
this was used.
All the rest of this stuff was unused.
# we're not quite minimal yet, since we duplicate a few status register things
# here...
Inspired by: bde@
Use int32/intptr casts for exception vector names.
Define MIPS_SR_INT_MASK again
Change MIPS_XKPHYS_CCA_* to MIPS_CCA_* since we can use them in many contexts
Minor gratuitous whitespace churn
The problem with setting it there is that the last CPU to come up
wins, it seems. This also removes one more ifdef in locore.S, a noble
goal too. Since they are unused, and pollute cpu.h, remove them.
Submitted by: bde.h (cpu.h pollution)
Approved in theory by: jmallet@
Updated PTE/PDE macros from http://svn.freebsd.org/base/user/jmallett/octeon
Introduce pmap_segshift() macro, use pmap_segmap() in place of pmap_pde, and
remove pmap_pde().
Approved by: rrs (mentor)
Obtained from: jmallett@
If we save/restore the PageMask, the value set by the bootloader will
persist, and will cause problems later in TLB exception handler.
This caused a crash in AR71xx boards.
Also fixes the EntryHi mask in pte.h
Reported by: Luiz Otavio O Souza <lists.br@gmail.com>
Tested by: Luiz Otavio O Souza <lists.br@gmail.com>
Approved by: rrs (mentor)
Initial support for n32 and n64 ABIs from
http://svn.freebsd.org/base/user/jmallett/octeon
Changes are:
- syscall, exception and trap support for n32/n64 ABIs
- 64-bit address space defines
- _jmp_buf for n32/n64
- casts between registers and ptr/int updated to work on n32/n64
Approved by: rrs(mentor), jmallett
PTE flag cleanup from http://svn.freebsd.org/base/user/jmallett/octeon
- Rename PTE_xx flags to match their MIPS names
- Use the new pte_set/test/clear macros uniformly, instead of a mixture
of mips_pg_xxx(), pmap_pte_x() macros and direct access.
- Remove unused macros and defines from pte.h and pmap.c
Discussed on freebsd-mips@
Approved by: rrs(mentor), jmallett
