gdunlap/xen-dr-smp.hg

log

age author revision description
2008-02-23 George Dunlap 6095:4f3f0cd75283 Changes to allow build to complete on newer versions of gccdefault tip
2008-02-23 George Dunlap 6094:6f403053cc09 Code related to marking specific gpns not to be tracked
2006-08-26 George Dunlap 6093:ad0ae9b5b801 Add delayed_fault_info, so that we can get stats on faults. Also fixed TESTING_STRICTLY_MINIMAL 0 build.
2006-08-26 George Dunlap 6092:1ff762b525b2 Fix deadlock in monitored_evtchn_set_pending().
2006-08-21 George Dunlap 6091:651b87cfbccc First steps towards more than two vcpus. Define everything off CREW_LAST_VCPU. Use actual # of cpus in domain to divvy up the logging pages. And don't enable crew until logging is ready.
2006-08-20 George Dunlap 6090:491ad98ddf2f Added infrastructure to 'filter' out gpfns from CREW protocol, to simulate what a kernel optimized for CREW might look like. Disabled by default.
2006-08-19 George Dunlap 6089:2ac1f6ac7835 Reset timers when last user timer goes to 0.
2006-08-19 George Dunlap 6088:275ab333a5ce Added TESTING_MINIMAL and TESTING_STRICTLY_MINIMAL.
2006-08-19 George Dunlap 6087:3c91355781e7 Added disabling of EXTRAVIRT codepaths. (Disabled for now.)
2006-08-19 George Dunlap 6086:ddf66bcb0aaf Move LOG_IGNORE to testing. Don't send constraints that are going to be ignored. Don't log events that are going to be ignored. (We still need to flush the TLB for events, tho.)
2006-08-19 George Dunlap 6085:133328eeeded Fix single-vcpu logging & replay
2006-08-19 George Dunlap 6084:f8494c6de0c2 Fast-path for report_crew_{event,constraint}
2006-08-19 George Dunlap 6083:cb4aa5e90b89 Replace post lock with replay queue lock. queue_init should now be called with the replay queue lock held.
2006-08-19 George Dunlap 6082:8623b7379465 queue_update_buffer_window optimization. Rather than mapping & unmapping every time, keep track of the last beginning & end; then unmap & map what doesn't overlap. Also, more direct method of cleaning for logging.
2006-08-19 George Dunlap 6081:9c8ace5cb8f0 Move raising of RECORD_SOFTIRQ to queue_init, and only raise on a page boundary.
2006-04-27 dunlapg 6080:81d7847b6881 null merge
2006-04-27 Michael Fetterman 6079:db4543cc5ea3 merge
2006-04-27 Michael Fetterman 6078:34d25423b0a2 Reversing last two commits...
2006-04-27 Michael Fetterman 6077:d3ab8301c57b Created per-vcpu versions of evtchn_mask.
2006-04-27 Michael Fetterman 6076:0f46d03a9f1e Created per-vcpu versions of evtchn_pending...
2006-04-25 Michael Fetterman 6075:9878d6b79264 Fix compilation problem: Move xen/arch/x86/timer.h to xen/include/asm-x86/timer.h
2006-04-25 Michael Fetterman 6074:e68d3d678f9c merge
2006-04-25 Michael Fetterman 6073:b2977e0105ab Update Makefiles so that tools/replay is part of a normal build.
2006-04-27 dunlapg 6072:b233102e8433 Better analysis from dump_log2. Sort per-gpfn info by total fault count. Keep user/kernel fault count. Added 'user_timer' to log so that we could filter out stuff not from the actual run itself.
2006-04-25 dunlapg 6071:8a517608bdfe crew_shadow_grab_gpfn now takes an argument whether to grab the global shared info page along with the per-vcpu shared info page.
2006-04-25 dunlapg 6070:a1c638c93329 Merge shared_info
2006-04-24 Michael Fetterman 6069:cae22307b215 Break a domain's shared_info onto multiple pages, one page for each
2006-04-24 dunlapg 6068:7f41b0d4bfe2 A ton of changes, again. Mostly reducing log size and cleaning out cruft. Added cr2/fault address to crew_constraint log. Added more analysis stuff to dump_log. Clocks for HYPERCALL_LOCK and more. Took out the header stuff from the logfile; now it's straight pages from the replay queue.
2006-04-17 dunlapg 6067:4e49209c950e Multi-vcpu logging and replay seems to work. This is a huge patchset, but does the following: 1) Add the 'ignore' flag to report_crew_X callpath 2) Add functionality to delay constraints and events to the end of a DELAY_SECTION. 3) Crew constraint uses the last event, rather than the current event count. This avoids circular constraints. 4) Add HYPERCALL_LOCK to shadow call, and deal with it properly for replay (eip switches back and forth) and in schedule. 5) A bunch of debugging stuff 6) Fixes a bug in replay_queue that was exhausing the map_cache.
2006-04-15 dunlapg 6066:11d2b0057f74 Always shadow_grab for current, if possible, even if doing something on behalf of someone else.
2006-04-10 dunlapg 6065:475d4137d5f6 Added per-domain hypercall lock. If replay_hypercall_cb can't grab it, it sets the eip back and re-starts.
2006-04-08 dunlapg 6064:d56fab3e9ed9 crew now has stronger expectations that there will be no crew protocol (i.e., page permissiong going back and forth) happening during replay.
2006-04-08 dunlapg 6063:5cc5344e0e57 Added asserts to find_domain_pt_regs() to detect errors sooner.
2006-04-08 dunlapg 6062:fe50bfdb6416 Added REPLAY_FULLCONTEXT option (still on by default).
2006-04-08 dunlapg 6061:78948291c807 Added ignore field.
2006-04-08 dunlapg 6060:9a7af03e8e8b A bunch of cleanup to the record data structures.
2006-04-08 dunlapg 6059:bfcebb5cae06 Added crew_special, a way to use the crew protocol to 'grab' non-gpfn entities. Used this to order checks on the out of sync list.
2006-04-07 dunlapg 6058:ae9feef29ad3 Doubled the size of the buffer it leaves before pausing itself.
2006-04-07 Michael Fetterman 6057:654e351dea47 merge
2006-04-07 Michael Fetterman 6056:c3eeaa0edb4c Fixed bug in shadowmap where it was loosing track of some entries...
2006-04-06 Michael Fetterman 6055:d8fc4ab5dc16 Get the 'd' key to really dump guest stack state correctly.
2006-04-06 Michael Fetterman 6054:a1043b4425f8 Don't re-enable interrupts in a hypercall until after BTS has been disabled.
2006-04-06 dunlapg 6053:0d8be7ab1683 BTS works again. Several changes: if we're logging or replaying and the target of the bts log is too full, copy the rest of the logs down in the buffer rather than clearing it. Made the bts buffers larger (2k entries). Also, added a (disabled) hack to turn on bts once we hit a certain number of crew events. To enable, set the 'once' variable in crew_fault to 1, and the tcount to the target event_count you want.
2006-04-05 Michael Fetterman 6052:e6ba6072f25c BUG FIX for crew: l1pte_read_fault was granting write permissions without
2006-04-05 Michael Fetterman 6051:918f3b4f0be8 Force domU and domT console output to xen's serial console.
2006-04-05 Michael Fetterman 6050:acaefd230392 adding execute bit to extravirt.sh
2006-04-05 Michael Fetterman 6049:d83d3d5d87fa cosmetic rework of logonly.sh and extravirt.sh
2006-04-05 Michael Fetterman 6048:4c9c7197bfae cleanup .hgignore, and delete another generated file...
2006-04-05 Michael Fetterman 6047:afb8e7210953 Moved replay_ht_enable to xen/Rules.mk -- BEWARE HT is ENABLED...
2006-04-05 Michael Fetterman 6046:cb0c7aaa8c7e Moved /dev/test to /tmp/test.
2006-04-05 dunlapg 6045:252fc48b9ec2 Took out perfctr_extra at a page fault. The pcount / eip change is not atomic, so it was causing trouble. We can put it back in later if we need it.
2006-04-04 dunlapg 6044:92d56a33feb4 Initial merge. WARNING -- turned off HT.
2006-04-04 Michael Fetterman 6043:2c1af4fe0c6f Added various ASSERTs that might help track down various bugs that I've seen.
2006-04-04 Michael Fetterman 6042:34f7bc269509 Add explicit calls to crew_shadow_grab_gpfn() throughout the shadow code.
2006-04-03 Michael Fetterman 6041:b0d5f07f1e4e merge
2006-04-03 Michael Fetterman 6040:6eafb24cc30c shadowmap now only maintains one singly-linked list for its entries.
2006-04-03 Michael Fetterman 6039:452df83aaca4 Added some more stats, and no longer release shadow L1 pages back into the general pool.
2006-03-31 Michael Fetterman 6038:a8b72139ca98 Checkpoint. Shadow behavior seems stable, although shadowmap stuff still
2006-03-29 Michael Fetterman 6037:4d9e71461b60 Added tag mainline for changeset 430dc412e4c9263f90a80f14f66dd25feb4e0840
2006-03-28 Michael Fetterman 6036:995fada97484 Bug fix in the CREW's SHADOW_USE_DEBUG code. Free'ing an L1 shadow should