The memory-mapped files implementation (mmap() etc.) is implemented with
the help of the filesystems using the in-VM FS cache. Filesystems tell it
about all cached blocks and their metadata. Metadata is: device offset and,
if any (and known), inode number and in-inode offset. VM can then map in
requested memory-mapped file blocks, and request them if necessary.
A limitation of this system is that filesystem block sizes that are not
a multiple of the VM system (and VM hardware) page size are not possible;
we can't map blocks in partially. (We can copy, but then the benefits of
mapping and sharing the physical pages is gone.) So until before this
commit various pieces of caching code assumed page size multiple
blocksizes. This isn't strictly necessary as long as mmap() needn't be
supported on that FS.
This change allows the in-FS cache code (libminixfs) to allocate any-sized
blocks, and will not interact with the VM cache for non-pagesize-multiple
blocks. In that case it will also signal requestors, by failing 'peek'
requests, that mmap() should not be supported on this FS. VM and VFS
will then gracefully fail all file-mapping mmap() calls, and exec() will
fall back to copying executable blocks instead of mmap()ping executables.
As a result, 3 diagnostics that signal file-mapped mmap()s failing
(hitherto an unusual occurence) are disabled, as ld.so does file-mapped
mmap()s to map in objects it needs. On FSes not supporting it this situation
is legitimate and shouldn't cause so much noise. ld.so will revert to its own
minix-specific allocate+copy style of starting executables if mmap()s fail.
Change-Id: Iecb1c8090f5e0be28da8f5181bb35084eb18f67b
. test74 for mmap functionality
. vm: add a mem_file memory type that specifies an mmap()ped
memory range, backed by a file
. add fdref, an object that keeps track of FD references within
VM per process and so knows how to de-duplicate the use of FD's
by various mmap()ped ranges; there can be many more than there can
be FD's
. turned off for now, enable with 'filemap=1' as boot option
Change-Id: I640b1126cdaa522a0560301cf6732b7661555672
Primary purpose of change: to support the mmap implementation, VM must
know both (a) about some block metadata for FS cache blocks, i.e.
inode numbers and inode offsets where applicable; and (b) know about
*all* cache blocks, i.e. also of the FS primary caches and not just
the blocks that spill into the secondary one. This changes the
interface and VM data structures.
This change is only for the interface (libminixfs) and VM data
structures; the filesystem code is unmodified, so although the
secondary cache will be used as normal, blocks will not be annotated
with inode information until the FS is modified to provide this
information. Until it is modified, mmap of files will fail gracefully
on such filesystems.
This is indicated to VFS/VM by returning ENOSYS for REQ_PEEK.
Change-Id: I1d2df6c485e6c5e89eb28d9055076cc02629594e
Some (backwards-compatible) changes in mmap() call message fields
that allow for a 64-bit offset. minix_mmap() takes an off_t and
minix_mmap64() takes a u64_t. Some mmap() work in VM goes into a
separate function, using the new fields, so that that can be re-used
when files are to be mapped (future commit).
Change-Id: Ifb77a90b593dd3c33cf81b396068e4da1ec5fb1c
Introduce explicit abstractions for different mapping types,
handling the instantiation, forking, pagefaults and freeing of
anonymous memory, direct physical mappings, shared memory and
physically contiguous anonymous memory as separate types, making
region.c more generic.
Also some other genericification like merging the 3 munmap cases
into one.
COW and SMAP safemap code is still implicit in region.c.
complete munmap implementation; single-page references made
a general munmap() implementation possible to write cleanly.
. memory: let the MIOCRAMSIZE ioctl set the imgrd device
size (but only to 0)
. let the ramdisk command set sizes to 0
. use this command to set /dev/imgrd to 0 after mounting /usr
in /etc/rc, so the boot time ramdisk is freed (about 4MB
currently)
This commit removes all traces of Minix segments (the text/data/stack
memory map abstraction in the kernel) and significance of Intel segments
(hardware segments like CS, DS that add offsets to all addressing before
page table translation). This ultimately simplifies the memory layout
and addressing and makes the same layout possible on non-Intel
architectures.
There are only two types of addresses in the world now: virtual
and physical; even the kernel and processes have the same virtual
address space. Kernel and user processes can be distinguished at a
glance as processes won't use 0xF0000000 and above.
No static pre-allocated memory sizes exist any more.
Changes to booting:
. The pre_init.c leaves the kernel and modules exactly as
they were left by the bootloader in physical memory
. The kernel starts running using physical addressing,
loaded at a fixed location given in its linker script by the
bootloader. All code and data in this phase are linked to
this fixed low location.
. It makes a bootstrap pagetable to map itself to a
fixed high location (also in linker script) and jumps to
the high address. All code and data then use this high addressing.
. All code/data symbols linked at the low addresses is prefixed by
an objcopy step with __k_unpaged_*, so that that code cannot
reference highly-linked symbols (which aren't valid yet) or vice
versa (symbols that aren't valid any more).
. The two addressing modes are separated in the linker script by
collecting the unpaged_*.o objects and linking them with low
addresses, and linking the rest high. Some objects are linked
twice, once low and once high.
. The bootstrap phase passes a lot of information (e.g. free memory
list, physical location of the modules, etc.) using the kinfo
struct.
. After this bootstrap the low-linked part is freed.
. The kernel maps in VM into the bootstrap page table so that VM can
begin executing. Its first job is to make page tables for all other
boot processes. So VM runs before RS, and RS gets a fully dynamic,
VM-managed address space. VM gets its privilege info from RS as usual
but that happens after RS starts running.
. Both the kernel loading VM and VM organizing boot processes happen
using the libexec logic. This removes the last reason for VM to
still know much about exec() and vm/exec.c is gone.
Further Implementation:
. All segments are based at 0 and have a 4 GB limit.
. The kernel is mapped in at the top of the virtual address
space so as not to constrain the user processes.
. Processes do not use segments from the LDT at all; there are
no segments in the LDT any more, so no LLDT is needed.
. The Minix segments T/D/S are gone and so none of the
user-space or in-kernel copy functions use them. The copy
functions use a process endpoint of NONE to realize it's
a physical address, virtual otherwise.
. The umap call only makes sense to translate a virtual address
to a physical address now.
. Segments-related calls like newmap and alloc_segments are gone.
. All segments-related translation in VM is gone (vir2map etc).
. Initialization in VM is simpler as no moving around is necessary.
. VM and all other boot processes can be linked wherever they wish
and will be mapped in at the right location by the kernel and VM
respectively.
Other changes:
. The multiboot code is less special: it does not use mb_print
for its diagnostics any more but uses printf() as normal, saving
the output into the diagnostics buffer, only printing to the
screen using the direct print functions if a panic() occurs.
. The multiboot code uses the flexible 'free memory map list'
style to receive the list of free memory if available.
. The kernel determines the memory layout of the processes to
a degree: it tells VM where the kernel starts and ends and
where the kernel wants the top of the process to be. VM then
uses this entire range, i.e. the stack is right at the top,
and mmap()ped bits of memory are placed below that downwards,
and the break grows upwards.
Other Consequences:
. Every process gets its own page table as address spaces
can't be separated any more by segments.
. As all segments are 0-based, there is no distinction between
virtual and linear addresses, nor between userspace and
kernel addresses.
. Less work is done when context switching, leading to a net
performance increase. (8% faster on my machine for 'make servers'.)
. The layout and configuration of the GDT makes sysenter and syscall
possible.
. new mode for sys_memset: include process so memset can be
done in physical or virtual address space.
. add a mode to mmap() that lets a process allocate uninitialized
memory.
. this allows an exec()er (RS, VFS, etc.) to request uninitialized
memory from VM and selectively clear the ranges that don't come
from a file, leaving no uninitialized memory left for the process
to see.
. use callbacks for clearing the process, clearing memory in the
process, and copying into the process; so that the libexec code
can be used from rs, vfs, and in the future, kernel (to load vm)
and vm (to load boot-time processes)
. make exec() callers (i.e. vfs and rs) determine the
memory layout by explicitly reserving regions using
mmap() calls on behalf of the exec()ing process,
i.e. handling all of the exec logic, thereby eliminating
all special exec() knowledge from VM.
. the new procedure is: clear the exec()ing process
first, then call third-party mmap()s to reserve memory, then
copy the executable file section contents in, all using callbacks
tailored to the caller's way of starting an executable
. i.e. no more explicit EXEC_NEWMEM-style calls in PM or VM
as with rigid 2-section arguments
. this naturally allows generalizing exec() by simply loading
all ELF sections
. drop/merge of lots of duplicate exec() code into libexec
. not copying the code sections to vfs and into the executable
again is a measurable performance improvement (about 3.3% faster
for 'make' in src/servers/)
these two functions will be used to support all exec() functionality
going into a single library shared by RS and VFS and exec() knowledge
leaving VM.
. third-party mmap: allow certain processes (VFS, RS) to
do mmap() on behalf of another process
. PROCCTL: used to free and clear a process' address space
Previously, the mmap address (if given) was merely used as a lower
bound, and then possibly overriden with a hint. Now, the mapping is
first tried at the exact given address. If that fails, the start of
the mmap range is used as lower bound (which is then still overridden
by the hint for efficiency).
This allows two pages to be mapped in at predefined addresses, where
the second address is lower than the first. That was not possible.
. MAP_SHARED was used to implement sysv shared memory
. used to signal shareable memory region to VM
. assumptions about this situation break when processes
use MAP_SHARED for its normal, standardised meaning
- RTS_VMINHIBIT flag is used to stop process while VM is fiddling with
its pagetables
- more generic way of sending synchronous scheduling events among cpus
- do the x-cpu smp sched calls only if the target process is runnable.
If it is not, it cannot be running and it cannot become runnable
this CPU holds the BKL
map_copy_ph_block is replaced by map_clone_ph_block, which can
replace a single physical block by multiple physical blocks.
also,
. merge map_mem.c with region.c, as they manipulate the same
data structures
. NOTRUNNABLE removed as sanity check
. use direct functions for ALLOC_MEM and FREE_MEM again
. add some checks to shared memory mapping code
. fix for data structure integrity when using shared memory
. fix sanity checks
this change
- makes panic() variadic, doing full printf() formatting -
no more NO_NUM, and no more separate printf() statements
needed to print extra info (or something in hex) before panicing
- unifies panic() - same panic() name and usage for everyone -
vm, kernel and rest have different names/syntax currently
in order to implement their own luxuries, but no longer
- throws out the 1st argument, to make source less noisy.
the panic() in syslib retrieves the server name from the kernel
so it should be clear enough who is panicing; e.g.
panic("sigaction failed: %d", errno);
looks like:
at_wini(73130): panic: sigaction failed: 0
syslib:panic.c: stacktrace: 0x74dc 0x2025 0x100a
- throws out report() - printf() is more convenient and powerful
- harmonizes/fixes the use of panic() - there were a few places
that used printf-style formatting (didn't work) and newlines
(messes up the formatting) in panic()
- throws out a few per-server panic() functions
- cleans up a tie-in of tty with panic()
merging printf() and panic() statements to be done incrementally.
Main changes:
- COW optimization for safecopy.
- safemap, a grant-based interface for sharing memory regions between processes.
- Integration with safemap and complete rework of DS, supporting new data types
natively (labels, memory ranges, memory mapped ranges).
- For further information:
http://wiki.minix3.org/en/SummerOfCode2009/MemoryGrants
Additional changes not included in the original Wu's branch:
- Fixed unhandled case in VM when using COW optimization for safecopy in case
of a block that has already been shared as SMAP.
- Better interface and naming scheme for sys_saferevmap and ds_retrieve_map
calls.
- Better input checking in syslib: check for page alignment when creating
memory mapping grants.
- DS notifies subscribers when an entry is deleted.
- Documented the behavior of indirect grants in case of memory mapping.
- Test suite in /usr/src/test/safeperf|safecopy|safemap|ds/* reworked
and extended.
- Minor fixes and general cleanup.
- TO-DO: Grant ids should be generated and managed the way endpoints are to make
sure grant slots are never misreused.
told to kernel
- makes VM ask the kernel if a certain process is allowed
to map in a range of physical memory (VM rounds it to page
boundaries afterwards - but it's impossible to map anything
smaller otherwise so I assume this is safe, i.e. there won't
be anything else in that page; certainly no regular memory)
- VM permission check cleanup (no more hardcoded calls, less
hardcoded logic, more readable main loop), a loose end left
by GQ
- remove do_copy warning, as the ipc server triggers this but
it's no more harmful than the special cases already excluded
explicitly (VFS, PM, etc).
shared with the kernel, mapped into kernel address space;
kernel is notified of its location. kernel segment size is
increased to make it fit.
- map in kernel and other processes that don't have their
own page table using single 4MB (global) mapping.
- new sanity check facility: objects that are allocated with
the slab allocator are, when running with sanity checking on,
marked readonly until they are explicitly unlocked using the USE()
macro.
- another sanity check facility: collect all uses of memory and
see if they don't overlap with (a) eachother and (b) free memory
- own munmap() and munmap_text() functions.
- exec() recovers from out-of-memory conditions properly now; this
solves some weird exec() behaviour
- chew off memory from the same side of the chunk as where we
start scanning, solving some memory fragmentation issues
- use avl trees for freelist and phys_ranges in regions
- implement most useful part of munmap()
- remap() stuff is GQ's for shared memory
their own fully fledged virtual address space and freeing
their pre-allocated heap+stack area (necessary to let memory
driver map in arbitrary areas of memory for /dev/mem without
sys_vm_map)
- small optimization preallocating memory on exec
- finished VR_DIRECT physical mapping code
