diff options
| -rw-r--r-- | .rootkeys | 1 | ||||
| -rw-r--r-- | xen/README | 114 | ||||
| -rw-r--r-- | xen/TODO | 129 |
3 files changed, 133 insertions, 111 deletions
@@ -193,6 +193,7 @@ 3ddb79bcbOVHh38VJzc97-JEGD4dJQ xen/Makefile 3ddb79bcCa2VbsMp7mWKlhgwLQUQGA xen/README 3ddb79bcWnTwYsQRWl_PaneJfa6p0w xen/Rules.mk +3e74d2be6ELqhaY1sW0yyHRKhpOvDQ xen/TODO 3ddb79bcZbRBzT3elFWSX7u6NtMagQ xen/arch/i386/Makefile 3ddb79bcBQF85CfLS4i1WGZ4oLLaCA xen/arch/i386/Rules.mk 3e5636e5FAYZ5_vQnmgwFJfSdmO5Mw xen/arch/i386/acpitable.c diff --git a/xen/README b/xen/README index 3518b8254a..ea14e52d86 100644 --- a/xen/README +++ b/xen/README @@ -1,110 +1,17 @@ ***************************************************** - Xeno Hypervisor (18/7/02) + Xeno Hypervisor (16/3/03) -1) Tree layout -Looks rather like a simplified Linux :-) -Headers are in include/xeno and include asm-<arch>. -At build time we create symlinks: - include/linux -> include/xeno - include/asm -> include/asm-<arch> -In this way, Linux device drivers should need less tweaking of -their #include lines. - -For source files, mapping between hypervisor and Linux is: - Linux Hypervisor - ----- ---------- - kernel/init/mm/lib -> common - net/* -> net/* - drivers/* -> drivers/* - arch/* -> arch/* - -Note that the use of #include <asm/...> and #include <linux/...> can -lead to confusion, as such files will often exist on the system include -path, even if a version doesn't exist within the hypervisor tree. -Unfortunately '-nostdinc' cannot be specified to the compiler, as that -prevents us using stdarg.h in the compiler's own header directory. - -We try to not modify things in driver/* as much as possible, so we can -easily take updates from Linux. arch/* is basically straight from -Linux, with fingers in Linux-specific pies hacked off. common/* has -a lot of Linux code in it, but certain subsystems (task maintenance, -low-level memory handling) have been replaced. net/* contains enough -Linux-like gloop to get network drivers to work with little/no -modification. - -2) Building 'make': Builds ELF executable called 'image' in base directory -'make install': gzip-compresses 'image' and copies it to TFTP server 'make clean': removes *all* build and target files -***************************************************** -Random thoughts and stuff from here down... - -Todo list ---------- -* Hypervisor need only directly map its own memory pool - (maybe 128MB, tops). That would need 0x08000000.... - This would allow 512MB Linux with plenty room for vmalloc'ed areas. -* Network device -- port drivers to hypervisor, implement virtual - driver for xeno-linux. Looks like Ethernet. - -- Hypervisor needs to do (at a minimum): - - packet filtering on tx (unicast IP only) - - packet demux on rx (unicast IP only) - - provide DHCP [maybedo something simpler?] - and ARP [at least for hypervisor IP address] - - -Segment descriptor tables -------------------------- -We want to allow guest OSes to specify GDT and LDT tables using their -own pages of memory (just like with page tables). So allow the following: - * new_table_entry(ptr, val) - [Allows insertion of a code, data, or LDT descriptor into given - location. Can simply be checked then poked, with no need to look at - page type.] - * new_GDT() -- relevent virtual pages are resolved to frames. Either - (i) page not present; or (ii) page is only mapped read-only and checks - out okay (then marked as special page). Old table is resolved first, - and the pages are unmarked (no longer special type). - * new_LDT() -- same as for new_GDT(), with same special page type. - -Page table updates must be hooked, so we look for updates to virtual page -addresses in the GDT/LDT range. If map to not present, then old physpage -has type_count decremented. If map to present, ensure read-only, check the -page, and set special type. - -Merge set_{LDT,GDT} into update_baseptr, by passing four args: - update_baseptrs(mask, ptab, gdttab, ldttab); -Update of ptab requires update of gtab (or set to internal default). -Update of gtab requires update of ltab (or set to internal default). - - -The hypervisor page cache -------------------------- -This will allow guest OSes to make use of spare pages in the system, but -allow them to be immediately used for any new domains or memory requests. -The idea is that, when a page is laundered and falls off Linux's clean_LRU -list, rather than freeing it it becomes a candidate for passing down into -the hypervisor. In return, xeno-linux may ask for one of its previously- -cached pages back: - (page, new_id) = cache_query(page, old_id); -If the requested page couldn't be kept, a blank page is returned. -When would Linux make the query? Whenever it wants a page back without -the delay or going to disc. Also, whenever a page would otherwise be -flushed to disc. - -To try and add to the cache: (blank_page, new_id) = cache_query(page, NULL); - [NULL means "give me a blank page"]. -To try and retrieve from the cache: (page, new_id) = cache_query(x_page, id) - [we may request that x_page just be discarded, and therefore not impinge - on this domain's cache quota]. - Booting secondary processors ---------------------------- +It's twisty and turny, so this is (roughly) the code path: + start_of_day (i386/setup.c) smp_boot_cpus (i386/smpboot.c) * initialises boot CPU data @@ -128,18 +35,3 @@ On other processor: * barrier, then write bitmasks to signal back to boot cpu * then barrel into... cpu_idle (i386/process.c) - [THIS IS PROBABLY REASONABLE -- BOOT CPU SHOULD KICK - SECONDARIES TO GET WORK DONE] - - -SMP capabilities ----------------- - -Current intention is to allow hypervisor to schedule on all processors in -SMP boxen, but to tie each domain to a single processor. This simplifies -many SMP intricacies both in terms of correctness and efficiency (eg. -TLB flushing, network packet delivery, ...). - -Clients can still make use of SMP by installing multiple domains on a single -machine, and treating it as a fast cluster (at the very least, the -hypervisor will have fast routing of locally-destined packets). diff --git a/xen/TODO b/xen/TODO new file mode 100644 index 0000000000..e81023e995 --- /dev/null +++ b/xen/TODO @@ -0,0 +1,129 @@ + +This is stuff we probably want to implement in the near future. I +think I have them in a sensible priority order -- the first few would +be nice to fix before a code release. The later ones can be +longer-term goals. + + -- Keir (16/3/03) + + +1. ASSIGNING DOMAINS TO PROCESSORS +---------------------------------- +More intelligent assignment of domains to processors. In +particular, we don't play well with hyperthreading: we will assign +domains to virtual processors on the same package, rather then +spreading them across processor packages. + +What we need to do is port code from Linux which stores information on +relationships between processors in the system (eg. which ones are +siblings in teh same package). We then use this to balance domains +across packages, and across virtual processors within a package. + +2. PROPER DESTRUCTION OF DOMAINS +-------------------------------- +Currently we do not free resources when destroying a domain. This is +because they may be tied up in subsystems, and there is no way of +pulling them back in a safe manner. + +The fix is probably to reference count resources and automatically +free them when the count reaches zero. We may get away with one count +per domain (for all its resources). When this reaches zero we know it +is safe to free everything: block-device rings, network rings, and all +the rest. + +3. FIX HANDLING OF NETWORK RINGS +-------------------------------- +Handling of the transmit rings is currently very broken (for example, +sending an inter-domain packet will wedge the hypervisor). This is +because we may handle packets out of order (eg. inter-domain packets +are handled eagerly, while packets for real interfaces are queued), +but our current ring design really assumes in-order handling. + +A neat fix will be to allow responses to be queued in a different +order to requests, just as we already do with block-device +rings. We'll need to add an opaque identifier to ring entries, +allowing matching of requests and responses, but that's about it. + +4. GDT AND LDT VIRTUALISATION +----------------------------- +We do not allow modification of the GDT, or any use of the LDT. This +is necessary for support of unmodified applications (eg. Linux uses +LDT in threaded applications, while Windows needs to update GDT +entries). + +I have some text on how to do this: +/usr/groups/xeno/discussion-docs/memory_management/segment_tables.txt +It's already half implemented, but the rest is still to do. + +5. DOMAIN 0 MANAGEMENT DAEMON +----------------------------- +A better control daemon is required for domain 0, which keeps proper +track of machine resources and can make sensible policy choices. This +may require support in Xen; for example, notifications (eg. DOMn is +killed), and requests (eg. can DOMn allocate x frames of memory?). + +6. ACCURATE TIMERS AND WALL-CLOCK TIME +-------------------------------------- +Currently our long-term timebase free runs on CPU0, with no external +calibration. We should run ntpd on domain 0 and allow this to warp +Xen's timebase. Once this is done, we can have a timebase per CPU and +not worry about relative drift (since they'll all get sync'ed +periodically by ntp). + +7. NEW DESIGN FEATURES +---------------------- +This includes the last-chance page cache, and the unified buffer cache. + + + +Graveyard +********* + +Following is some description how some of the above might be +implemented. Some of it is superceded and/or out of date, so follow +with caution. + +Segment descriptor tables +------------------------- +We want to allow guest OSes to specify GDT and LDT tables using their +own pages of memory (just like with page tables). So allow the following: + * new_table_entry(ptr, val) + [Allows insertion of a code, data, or LDT descriptor into given + location. Can simply be checked then poked, with no need to look at + page type.] + * new_GDT() -- relevent virtual pages are resolved to frames. Either + (i) page not present; or (ii) page is only mapped read-only and checks + out okay (then marked as special page). Old table is resolved first, + and the pages are unmarked (no longer special type). + * new_LDT() -- same as for new_GDT(), with same special page type. + +Page table updates must be hooked, so we look for updates to virtual page +addresses in the GDT/LDT range. If map to not present, then old physpage +has type_count decremented. If map to present, ensure read-only, check the +page, and set special type. + +Merge set_{LDT,GDT} into update_baseptr, by passing four args: + update_baseptrs(mask, ptab, gdttab, ldttab); +Update of ptab requires update of gtab (or set to internal default). +Update of gtab requires update of ltab (or set to internal default). + + +The hypervisor page cache +------------------------- +This will allow guest OSes to make use of spare pages in the system, but +allow them to be immediately used for any new domains or memory requests. +The idea is that, when a page is laundered and falls off Linux's clean_LRU +list, rather than freeing it it becomes a candidate for passing down into +the hypervisor. In return, xeno-linux may ask for one of its previously- +cached pages back: + (page, new_id) = cache_query(page, old_id); +If the requested page couldn't be kept, a blank page is returned. +When would Linux make the query? Whenever it wants a page back without +the delay or going to disc. Also, whenever a page would otherwise be +flushed to disc. + +To try and add to the cache: (blank_page, new_id) = cache_query(page, NULL); + [NULL means "give me a blank page"]. +To try and retrieve from the cache: (page, new_id) = cache_query(x_page, id) + [we may request that x_page just be discarded, and therefore not impinge + on this domain's cache quota]. |