Age | Commit message (Collapse) | Author |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
The main impact of the direct physical mapping on the kmem module is the
slab size computation. The page allocator requires the allocation size
to be a power-of-two above the page size since it uses the buddy memory
allocation algorithm.
Custom slab allocation functions are no longer needed since the only
user was the kentry area, which has been removed recently.
The KMEM_CACHE_NOCPUPOOL flag is also no longer needed since CPU pools,
which are allocated from a kmem cache, can now always be allocated out
of the direct physical mapping.
|
|
Don't encourage anyone to use non reclaimable pools of resources, it's
a Bad Thing To Do.
|
|
|
|
|
|
Conflicts:
vm/vm_page.h
|
|
|
|
* vm/vm_object.h (struct vm_object): Use a general lock, adapt macros.
|
|
* vm/vm_object.c: Use a general lock to protect the object cache.
|
|
* vm/memory_object.c: Use a general lock to protect the default memory
manager.
|
|
Avoid accessing fields of `vm_object' objects without having it
locked. These problems have been found using a code transformation
done by Coccinelle that instrumented all accesses with a runtime
check, and manual inspection.
* vm/memory_object.c (memory_object_data_supply): Avoid accessing
fields without the lock.
* vm/vm_fault.c (vm_fault_page): Likewise.
* vm/vm_map.c (vm_map_submap): Properly lock `object'.
(vm_map_copy_overwrite): Avoid accessing fields without the lock.
(vm_map_copyin): Lock `src_object'.
* vm/vm_object.c (_vm_object_setup): Likewise.
(vm_object_allocate): Likewise.
(vm_object_terminate): Avoid accessing fields without the lock.
(vm_object_copy_slowly): Lock `new_object'.
(vm_object_copy_delayed): Lock `src_object' earlier, lock `new_copy'.
(vm_object_shadow): Lock `result'.
(vm_object_enter): Properly lock `object'. Avoid accessing fields
without the lock.
* vm/vm_pageout.c (vm_pageout_setup): Properly lock `old_object'.
|
|
* vm/vm_object.c (vm_object_collapse): Collapse unreachable branch
into assertion.
|
|
* vm/vm_user.c (vm_wire): Drop unused but set variable `host'.
|
|
* vm/vm_fault.c (vm_fault_page): Enable extra assertions.
|
|
Previously, the pageout code traversed the list of pages in an object
instead of the list of inactive pages.
* vm/vm_pageout.c (vm_pageout_scan): Fix traversing the list of
inactive pages.
|
|
Previously, the pageout code traversed the hash table chain instead of
the list of inactive pages. The code merely compiled by accident,
because the `struct page' also has a field called `next' for the hash
table chain.
* vm/vm_pageout.c (vm_pageout_scan): Fix traversing the list of
inactive pages.
|
|
* vm/vm_object.c (vm_object_terminate): Drop debugging remnants.
|
|
* vm/vm_kern.c (kmem_init): Fix panic message.
|
|
* doc/mach.texi (vm_wire): Document that the host port does not have to be
privileged.
* include/mach/mach_hosts.defs (vm_wire): Use mach_port_t instead of
host_priv_t.
* vm/vm_map.h (vm_map): Add user_wired field.
* vm/vm_map.c (vm_map_setup): Initialize user_wired field to 0.
(vm_map_pageable_common, vm_map_entry_delete, vm_map_copy_overwrite,
vm_map_copyout_page_list, vm_map_copyin_page_list): When switching
user_wired_count field of entry between 0 and non-0, accumulate the
corresponding size into the user_wired field of map.
* vm/vm_user.c (vm_wire): Turn host parameter into port parameter, and
inline a version of convert_port_to_host_priv which records whether the host
port is privileged or not. When it is not privileged, check whether
the additional amount to user_wired will overcome 64KiB.
|
|
* vm/vm_kern.c (kmem_alloc_aligned): Fix typo.
|
|
* vm/vm_kern.c (kmem_realloc): Remove function.
(kmem_alloc_wired): Adopt comment.
* vm/vm_kern.h (kmem_realloc): Remove declaration.
|
|
* vm/vm_object.c (vm_object_copy_call): Gracefully handle resource
shortage by doing the allocation earlier and aborting the function if
unsuccessful.
|
|
Import the macro definitions from the x15 kernel project, and replace
all similar definitions littered all over the place with it.
Importing this file will make importing code from the x15 kernel
easier. We are already using the red-black tree implementation and
the slab allocator from it, and we will import even more code in the
near future.
* kern/list.h: Do not define `structof', include `macros.h' instead.
* kern/rbtree.h: Likewise.
* kern/slab.c: Do not define `ARRAY_SIZE', include `macros.h' instead.
* i386/grub/misc.h: Likewise.
* i386/i386/xen.h: Do not define `barrier', include `macros.h' instead.
* kern/macro_help.h: Delete file. Replaced by `macros.h'.
* kern/macros.h: New file.
* Makefrag.am (libkernel_a_SOURCES): Add new file, remove old file.
* device/dev_master.h: Adopt accordingly.
* device/io_req.h: Likewise.
* device/net_io.h: Likewise.
* i386/intel/read_fault.c: Likewise.
* ipc/ipc_kmsg.h: Likewise.
* ipc/ipc_mqueue.h: Likewise.
* ipc/ipc_object.h: Likewise.
* ipc/ipc_port.h: Likewise.
* ipc/ipc_space.h: Likewise.
* ipc/ipc_splay.c: Likewise.
* ipc/ipc_splay.h: Likewise.
* kern/assert.h: Likewise.
* kern/ast.h: Likewise.
* kern/pc_sample.h: Likewise.
* kern/refcount.h: Likewise.
* kern/sched.h: Likewise.
* kern/sched_prim.c: Likewise.
* kern/timer.c: Likewise.
* kern/timer.h: Likewise.
* vm/vm_fault.c: Likewise.
* vm/vm_map.h: Likewise.
* vm/vm_object.h: Likewise.
* vm/vm_page.h: Likewise.
|
|
Currently, `thread_collect_scan' does nothing because `pcb_collect' is
a nop. Its body is exempt from compilation by means of the
preprocessor.
This is unfortunate as it increases the risk of bitrot, and we still
need to pay the price of rate-limiting thread_collect_scan.
* kern/thread.c (thread_collect_scan): Drop #if 0 around the body.
* vm/vm_pageout.c (vm_pageout_scan): Do not call
`consider_thread_collect' and document why.
|
|
* vm/vm_resident.c: Fix typo.
|
|
Signed-off-by: Stefan Weil <sw@weilnetz.de>
|
|
This reverts commit 5ae510e35c54009626999a88f0f1cb34d6dfc94f.
|
|
As vm_allocate does.
* vm/vm_user.c (vm_map): When `anywhere' is true, set `address' to the
minimum address of the `target_map'.
|
|
This targets having always at least 8% free memory instead of just 1%.
This has shown improving buildd stability a lot. Also increase the
reserved amount to nowadays standards.
* vm/vm_pageout.c (VM_PAGE_FREE_TARGET): Increase to 10%.
(VM_PAGE_FREE_MIN): Increase to 8%.
(VM_PAGE_FREE_RESERVED): Increase to 500 pages.
(VM_PAGEOUT_RESERVED_INTERNAL): Increase to 150 pages.
(VM_PAGEOUT_RESERVED_REALLY): Increase to 100 pages.
|
|
* vm/vm_pageout.c (vm_pageout): Set the priority to 0.
|
|
Currently, the size of struct vm_map is 68 bytes. By using a bit
field for the boolean flags, it can be made fit into a cache line.
* vm/vm_map.h (struct vm_map): Use a bit field for the boolean flags
wait_for_space and wiring_required.
|
|
Convert from K&R style function definitions to ANSI style
function definitions.
* ddb/db_access.c: Convert function prototypes from K&R to ANSI.
* ddb/db_aout.c: Likewise.
* ddb/db_break.c: Likewise.
* ddb/db_command.c: Likewise.
* ddb/db_cond.c: Likewise.
* ddb/db_examine.c: Likewise.
* ddb/db_expr.c: Likewise.
* ddb/db_ext_symtab.c: Likewise.
* ddb/db_input.c: Likewise.
* ddb/db_lex.c: Likewise.
* ddb/db_macro.c: Likewise.
* ddb/db_mp.c: Likewise.
* ddb/db_output.c: Likewise.
* ddb/db_print.c: Likewise.
* ddb/db_run.c: Likewise.
* ddb/db_sym.c: Likewise.
* ddb/db_task_thread.c: Likewise.
* ddb/db_trap.c: Likewise.
* ddb/db_variables.c: Likewise.
* ddb/db_watch.c: Likewise.
* device/blkio.c: Likewise.
* device/chario.c: Likewise.
* device/dev_lookup.c: Likewise.
* device/dev_name.c: Likewise.
* device/dev_pager.c: Likewise.
* device/ds_routines.c: Likewise.
* device/net_io.c: Likewise.
* device/subrs.c: Likewise.
* i386/i386/db_interface.c: Likewise.
* i386/i386/fpu.c: Likewise.
* i386/i386/io_map.c: Likewise.
* i386/i386/loose_ends.c: Likewise.
* i386/i386/mp_desc.c: Likewise.
* i386/i386/pcb.c: Likewise.
* i386/i386/phys.c: Likewise.
* i386/i386/trap.c: Likewise.
* i386/i386/user_ldt.c: Likewise.
* i386/i386at/com.c: Likewise.
* i386/i386at/kd.c: Likewise.
* i386/i386at/kd_event.c: Likewise.
* i386/i386at/kd_mouse.c: Likewise.
* i386/i386at/kd_queue.c: Likewise.
* i386/i386at/lpr.c: Likewise.
* i386/i386at/model_dep.c: Likewise.
* i386/i386at/rtc.c: Likewise.
* i386/intel/pmap.c: Likewise.
* i386/intel/read_fault.c: Likewise.
* ipc/ipc_entry.c: Likewise.
* ipc/ipc_hash.c: Likewise.
* ipc/ipc_kmsg.c: Likewise.
* ipc/ipc_marequest.c: Likewise.
* ipc/ipc_mqueue.c: Likewise.
* ipc/ipc_notify.c: Likewise.
* ipc/ipc_port.c: Likewise.
* ipc/ipc_right.c: Likewise.
* ipc/mach_debug.c: Likewise.
* ipc/mach_msg.c: Likewise.
* ipc/mach_port.c: Likewise.
* ipc/mach_rpc.c: Likewise.
* kern/act.c: Likewise.
* kern/exception.c: Likewise.
* kern/ipc_mig.c: Likewise.
* kern/ipc_tt.c: Likewise.
* kern/lock_mon.c: Likewise.
* kern/mach_clock.c: Likewise.
* kern/machine.c: Likewise.
* kern/printf.c: Likewise.
* kern/priority.c: Likewise.
* kern/startup.c: Likewise.
* kern/syscall_emulation.c: Likewise.
* kern/syscall_subr.c: Likewise.
* kern/thread_swap.c: Likewise.
* kern/time_stamp.c: Likewise.
* kern/timer.c: Likewise.
* kern/xpr.c: Likewise.
* vm/memory_object.c: Likewise.
* vm/vm_debug.c: Likewise.
* vm/vm_external.c: Likewise.
* vm/vm_fault.c: Likewise.
* vm/vm_kern.c: Likewise.
* vm/vm_map.c: Likewise.
* vm/vm_pageout.c: Likewise.
* vm/vm_user.c: Likewise.
|
|
In addition to physical pages, the slab allocator also consumes kernel
virtual memory, so reclaim pages on failure to allocate from a kernel map.
This method isn't foolproof but helps alleviate fragmentation.
* vm/vm_kern.c (kmem_alloc): Call slab_collect and retry allocation
once on failure.
(kmem_realloc): Likewise.
(kmem_alloc_wired): Likewise.
(kmem_alloc_wired): Likewise.
(kmem_alloc_aligned): Likewise.
|
|
* vm/vm_kern.c (projected_buffer_deallocate): Look for `map' being NULL
or kernel_map before locking it.
|
|
It is not clear to me why the declaration was put there in the first
place. It is not used anywhere, and it conflicts with the declaration
generated by mig.
* vm/memory_object_proxy.h (memory_object_create_proxy): Remove declaration.
|
|
Previously, the bit field left 31 bits unused. By reducing the size
of wire_count by one bit, the size of the whole struct is reduced by
four bytes.
* vm/vm_page.h (struct vm_page): Reduce the size of wire_count to 15
bits.
|
|
* vm/vm_page.h (struct vm_page): Merge the two bit fields.
|
|
Apparently, the NS32000 was a 32-bit CPU from the 1990ies. The string
"ns32000" appears nowhere else in the source.
* vm/vm_page.h (struct vm_page): Remove NS32000-specific padding.
|
|
Declare void argument lists that were not declared in the first
part of this patch and
* kern/sched_prim.h (recompute_priorities): Fix prototype.
* kern/startup.c (setup_main) (recompute_priorities): Fix call.
|
|
|
|
|
|
Members offset and phys_addr are of vm_offset_t types.
* vm/vm_resident.c (vm_page_print) (offset, phys_addr): Remove unnecessary casts.
|
|
The return value from vm_map_min() is already of vm_offset_t type.
* vm/vm_kern.c (kmem_submap) (addr): Remove unnecessary cast.
|