Age | Commit message (Collapse) | Author |
|
The purpose of this system call is to help debugging in situations where
it's desirable to bypass the mach_msg call entirely.
* include/mach/syscall_sw.h (mach_print): Generate system call code.
* kern/syscall_subr.c: Include <kern/printf.h>.
[MACH_KDB] (mach_print): New function.
* kern/syscall_subr.h (mach_print): New prototype.
* kern/syscall_sw.c [MACH_KDB] (mach_trap_table): Declare mach_print
system call.
|
|
|
|
* vm/vm_fault.c (vm_fault_page): Properly increment
current_task()->zero_fills and cow_faults.
|
|
* ipc/ipc_mqueue.c (ipc_mqueue_send, ipc_mqueue_receive): Increment
counters for message sent and received.
* kern/ipc_kobject.c (ipc_kobject_server): Increment sent and received
counters for the kernel task.
* kern/task.c (task_create): Zero statistics.
* kern/task.c (task_info): Add task_events_info call.
* kern/task.h: Add statistics.
* vm/vm_fault.c (vm_fault_page): Increment zero_fills, pageins,
reactivations and cow_faults counters.
* vm/vm_fault.c (vm_fault_wire_fast): Increment faults counters.
* vm/vm_pageout.c (vm_pageout_scan): Increment reactivations counter.
|
|
This problem was overlooked because of simple locks being no-ops.
* kern/slab.c (slab_collect): Fix lock name to kmem_cache_list_lock.
(host_slab_info): Likewise.
|
|
The purpose of this function is to allow kernel code to display the state
of the slab caches in situations where the host_slab_info RPC wouldn't be
available, e.g. before a panic.
* kern/slab.c (slab_info): New function.
* kern/slab.h: Add declaration for slab_info.
|
|
* vm/vm_map.h (KENTRY_DATA_SIZE): Bump from 32 pages to 64 pages.
|
|
When kernel debugger support is requested ipc and vm need
the prototype for the debugger function kdbprintf.
* ddb/db_output.h: Add prototype for kdbprintf.
* ipc/ipc_object.c: Add include file ddb/db_output.h
* ipc/ipc_port.c: Likewise.
* ipc/ipc_pset.c: Likewise.
* vm/vm_map.c: Likewise.
* vm/vm_object.c: Likewise.
* vm/vm_resident.c: Likewise.
|
|
By disabling some of the most slow drivers by default, and giving progress
feedback to the user.
* linux/configfrag.ac (CONFIG_SCSI_NCR53C7xx, CONFIG_SCSI_AIC7XXX,
CONFIG_SCSI_GDTH): Disable by default
* linux/src/drivers/scsi/eata.c (__initfunc): Don't feed line in probe
message.
* linux/src/drivers/scsi/hosts.c (scsi_init): Print SCSI probe progress.
|
|
Cherry picked from Linux kernel commit 76c25284e0d845bff4ee1031721556148af4db1c
* linux/dev/drivers/block/floppy.c (config_types): Clear from
`allowed_drive_mask' floppy drives whose cmos type is 0.
|
|
|
|
0a55db5 made thread_suspend return KERN_FAILURE when the target thread is in
TH_UNINT state. That however is not currently handled by libc, and it's more
useful to just wait for the thread to get interruptible.
* kern/sched_prim.c (thread_invoke): Wake NEW_THREAD->STATE for thread_suspend.
* kern/thread.c (thread_suspend): Wait on NEW_THREAD->STATE as long as THREAD
has TH_UNINT.
|
|
Based upon a patch by Philippe De Muyter, and feedback from Mark
Lord and Robert Hancock.
As noted by Mark Lord, the outdated ATA1 spec specifies a 20msec
timeout for setting DRQ but lots of common devices overshoot this.
Signed-off-by: David S. Miller <davem@davemloft.net>
(cherry picked from Linux kernel commit 602da297e293eb2cbd28dcdbbe247593a46a853a)
* linux/src/drivers/block/ide.h (WAIT_DRQ): Increase to 1 s.
|
|
* linux/src/drivers/block/ide.c: Partially update to Linux 2.0.40 code.
|
|
The mapable_time_init function is called before machine_init, but the
latter sets the system boot time. Fix their calling order.
* kern/startup.c (setup_main): Call machine_init before mapable_time_init.
|
|
* kern/rbtree.h (rbtree_lookup): Prefix local variable names with three
underscores.
(rbtree_lookup_nearest): Likewise.
(rbtree_insert): Likewise.
(rbtree_lookup_slot): Likewise.
(rbtree_insert_slot): Rewrite as an inline function.
|
|
* vm/vm_map.c (vm_map_enter): return KERN_INVALID_ARGUMENT if size is 0.
* vm/vm_user.c (vm_map): Likewise.
|
|
Some versions of GCC will error if you define a type twice (even if
the definition is the same). The NCR53C8XX SCSI driver defines
'vm_offset_t', this is also defined in vm_types.h and will therefore
cause a compilation error depending on the GCC version.
* linux/src/drivers/scsi/ncr53c8xx.c [!MACH] (vm_offset_t, vm_size_t): Don't
define types.
|
|
* kern/ipc_kobject.c (ipc_kobject_server): Set gnumach_server_routine
prototype to mig_routine_t.
|
|
* kern/gnumach.srv: New file.
|
|
* include/mach/vm_cache_statistics.h: New file
|
|
* linux/src/drivers/net/pci-scan.c (pci_drv_register): Skip device
if we are getting an invalid IRQ >= 16 and different from 255 (it
happens in some motherboard).
|
|
* linux/configfrag.ac (--enable-ide-forcedma): Add option.
* linux/src/drivers/block/ide.c (ide_probe_promise_20246): Do not call
ide_init_promise.
(probe_for_hwifs): Add SI, VIA, AL PCI probing.
* linux/src/drivers/block/ide.h (hwif_chipset_t): Add hpt343, udma and
ultra66 chipsets.
* linux/src/drivers/block/triton.c: Update to upstream 1.14 version.
* linux/src/include/linux/hdreg.h (HDIO_GET_IDENTITY): Rename to ...
(HDIO_OBSOLETE_IDENTITY): ... new macro.
(HDIO_GET_IDENTITY): New macro.
(hd_driveid): Add fields for extended identification.
|
|
* linux/dev/glue/block.c
(device_open): Properly return devp on multiple opens.
(device_close_forced): Renamed from device_close, when force parameter is 1, do
not wait for open_count to become 0 before closing.
(device_close): New function.
(device_no_senders): New function.
linux_block_emulation_ops: Use (device_no_senders).
|
|
* Makefrag.am (EXTRA_DIST): Add kern/gnumach.srv.
(include_mach_HEADERS): Add include/mach/gnumach.defs and
include/mach/vm_cache_statistics.h
(nodist_libkernel_a_SOURCES): Add kern/gnumach.server.defs.c,
kern/gnumach.server.h, kern/gnumach.server.c, kern/gnumach.server.msgids,
kern/gnumach.server.defs.
* include/mach/mach_types.h: Add #include <mach/vm_cache_statistics.h>.
* kern/ipc_kobject.c (ipc_kobject_server): Declare and call
gnumach_server_routine.
* vm/vm_object.c (vm_object_cached_pages): New variable.
(vm_object_cached_pages_lock_data): Likewise.
(vm_object_deallocate): Update number of cached pages.
(vm_object_lookup): Likewise.
(vm_object_lookup_name): Likewise.
(vm_object_destroy): Likewise.
(vm_object_enter): Likewise.
* vm/vm_object.h (ref_count): Declare as int.
(resident_page_count): Likewise.
(vm_object_cached_count): Add extern declaration.
(vm_object_cached_pages): Likewise.
(vm_object_cached_pages_lock_data): Likewise.
(vm_object_cached_pages_update): New macro.
* vm/vm_resident.c (vm_page_insert): Assert resident page count doesn't
overflow, update number of cached pages as appropriate.
(vm_page_replace): Likewise.
(vm_page_remove): Update number of cached pages as appropriate.
* vm/vm_user.c: Add #include <mach/vm_cache_statistics.h>.
(vm_cache_statistics): New function.
* vm/vm_user.h: Add #include <mach/mach_types.h>.
(vm_cache_statistics): New declaration.
* include/mach/gnumach.defs: New file.
|
|
* ipc/ipc_thread.h (ipc_thread_enqueue_macro): Insert thread at the
head of the list instead of the tail.
|
|
* i386/intel/pmap.c (pmap_bootstrap): Interate over linear addresses and compute
l3 and l2 offsets from it instead of assuming nul l3 offset.
|
|
The slab garbage collector uses sched_tick as its time reference, which
is increased every seconds, while the interval is expressed in clock
ticks. Use the proper time reference instead.
* kern/slab.c (kmem_gc_last_tick): Declare as unsigned long.
(slab_collect): Use elapsed_ticks instead of sched_tick.
|
|
* kern/slab.c (KMEM_GC_INTERVAL): Increase to 5 seconds.
|
|
The kernel submaps eat most of the available kernel space. Using the
main kernel map for thread stacks sometimes lead to exhaustion when many
threads are created. Use kmem_map instead to increase this limit.
* kern/thread.c (stack_alloc): Use kmem_map instead of kernel_map for
stack allocation.
(stack_collect): Likewise for release.
|
|
* ipc/ipc_table.c: Add #include <kern/slab.h>.
(ipc_table_alloc): Use kmem_map instead of kalloc_map when allocating
a table.
(ipc_table_realloc): Likewise for reallocation.
(ipc_table_free): Likewise for release.
* kern/kalloc.h (kalloc_map): Remove declaration.
* kern/slab.c (KMEM_MAP_SIZE): Increase to 128 MiB.
(KALLOC_MAP_SIZE): Remove macro.
(kalloc_map_store): Remove variable.
(kalloc_map): Likewise.
(kalloc_pagealloc): Use kmem_map instead of kalloc_map for general
purpose allocations.
(kalloc_pagefree): Likewise.
(kalloc_init): Remove the creation of kalloc_map.
|
|
* linux/configfrag.ac (AC_OPTION_Linux_group): New function
(AC_OPTION_Linux_ix86_at): Do not enable driver by default if the group was
explicitly disabled.
(scsi, net, pcmcia, wireless): Define groups.
|
|
* linux/src/arch/i386/kernel/bios32.c (check_pcibios, pcibios_init): Convert
physical bios32 entry address to kernel virtual address.
|
|
Linux does this, there are probably wild BIOSes out there.
* linux/src/arch/i386/kernel/bios32.c (bios32_service, check_pcibios,
pci_bios_find_class, pci_bios_find_device, pci_bios_read_config_byte,
pci_bios_read_config_word, pci_bios_read_config_dword,
pci_bios_write_config_byte, pci_bios_write_config_word,
pci_bios_write_config_dword): Clear direction flag after lcall to bios32.
|
|
This permits better trace support in kdb.
* i386/i386/vm_param.h (VM_MIN_KERNEL_ADDRESS): Set to 0xC0000000
* i386/Makefrag.am (_START): Set to 0xC0100000.
|
|
* i386/i386/vm_param.h (VM_MIN_KERNEL_ADDRESS): Document that it can be
changed, but _START has to be offset too.
(INIT_VM_MIN_KERNEL_ADDRESS): Document that it should remain 0.
|
|
* vm/vm_map.c: Add #include <kern/rbtree.h>.
(vm_map_setup): Initialize the map red-black tree.
(vm_map_entry_cmp_lookup): New function.
(vm_map_entry_cmp_insert): Likewise.
(_vm_map_entry_link): Insert map entry in the red-black tree.
(_vm_map_entry_unlink): Remove map entry from the red-black tree.
(vm_map_lookup_entry): Rework the way the VM map hint is used, and
replace the linear search with a binary search tree lookup.
(vm_map_copy_insert): Move map entries from the map copy tree to the
destination map tree.
(vm_map_copyin): Initialize the map copy red-black tree.
* vm/vm_map.h: Add #include <kern/rbtree.h>.
(vm_map_entry): Add `tree_node' member.
(vm_map_header): Add `tree' member.
|
|
Literature about red-black trees vary concerning the cases numbering,
and the implementation doesn't make all of them clearly appear. Remove
cases numbering references to avoid confusion.
|
|
Maksym and I have assigned copyright to the Free Software Foundation.
In addition, restore the original upstream copyrights for correctness.
* kern/list.h: Fix copyright assignment.
* kern/rbtree.c: Likewise.
* kern/rbtree.h: Likewise.
* kern/rbtree_i.h: Likewise.
* kern/slab.c: Likewise.
* kern/slab.h: Likewise.
|
|
* i386/i386/pcb.c (switch_ktss): Check value returned by hyp_stack_switch.
|
|
* kern/thread.c (host_stack_usage): Set `total' type to natural_t.
|
|
This breaks on x86_64, where it is an array and thus gets bogus results.
* ddb/db_output.c (db_printf, kdbprintf): Pass copy of va_list variable instead
of its address.
* kern/debug.c (panic, log): Likewise.
* kern/printf.c (vprintf, iprintf, sprintf, vsnprintf): Likewise.
(_doprnt): Take va_list instead of va_list *, fix usage and comment accordingly.
* kern/printf.h (_doprnt): Take va_list instead of va_list *.
|
|
* i386/Makefrag.am (_START_MAP): Define symbol to 0x100000.
* i386/xen/Makefrag.am (_START_MAP): Define symbol to 0xC0000000.
* i386/ldscript: Use _START_MAP instead of hardcoding 0x100000.
|
|
Grub is not able to map us at high addresses. We can however make it load us at
low addresses (through the linker script mapping), then use segmentation to move
ourselves to high addresses, and switch to C which uses high addresses
(through _START definition).
* i386/ldscript: Force mapping kernel at 0x100000, regardless of the value
of _START.
* i386/i386at/boothdr.S (boot_entry): Set up initial segments to project the
bootstrap linear space to high addresses.
|
|
* i386/i386at/model_dep.c (i386at_init): Map BIOS memory at 0.
* i386/intel/pmap.c (pmap_create): Do not map BIOS memory in user tasks.
|
|
The former is the initial value set by the bootloader. It may not be the
same as what Mach will set up.
* i386/i386/vm_param.h (INIT_VM_MIN_KERNEL_ADDRESS): New macro.
* i386/i386at/model_dep.c (i386at_init): Use INIT_VM_MIN_KERNEL_ADDRESS
instead of VM_MIN_KERNEL_ADDRESS.
|
|
* i386/i386/vm_param.h [!MACH_XEN]: Do not include <xen/public/xen.h>.
* i386/i386at/model_dep.c (init_alloc_aligned): Use phystokv to compare physical
memory addresses with kernel start, end, and symbol table.
* i386/intel/pmap.c (pmap_enter): Use kvtophys to convert ptp pointer to pte
entry.
* linux/dev/init/main.c (alloc_contig_mem, linux_init): Use phystokv to convert
allocated pages to virtual pointer.
* linux/src/include/asm-i386/io.h: Include <machine/vm_param.h>.
(virt_to_phys): Call _kvtophys.
(phys_to_virt): Call phystokv.
* linux/src/include/linux/compatmac.h: Include <asm/io.h>.
(ioremap): Use phys_to_virt to convert physical address to virtual pointer.
(my_iounmap): Likewise.
* linux/dev/include/asm-i386/page.h: Include <mach/vm_param.h>.
(PAGE_SHIFT, PAGE_SIZE, PAGE_MASK): Remove macros.
* linux/src/drivers/scsi/ncr53c8xx.c (vm_size_t): Remove type.
* linux/dev/glue/net.c: Include <machine/vm_param.h>
(device_write): Call phystokv to convert from physical page address to
virtual pointer.
* linux/dev/glue/block.c (alloc_buffer, free_buffer, rdwr_full): Likewise.
|
|
* linux/dev/glue/misc.c (issig): Return 0 if current_thread() is NULL.
|
|
* i386/intel/pmap.c (pmap_bootstrap): Check against `va' wrap around
0xffffffff.
|
|
When _start is not close to 0, just using init_alloc_aligned() is not enough
to know how much should be duplicated between linear and virtual mappings.
Just mapping everything that can be is not very costly and should just work
when it can work.
* i386/i386at/model_dep.c (i386at_init) [VM_MIN_KERNEL_ADDRESS !=
LINEAR_MIN_KERNEL_ADDRESS]: Extend `nb_direct' to VM_MIN_KERNEL_ADDRESS -
LINEAR_MIN_KERNEL_ADDRESS.
|