asi_series cluster_6 X1 · vCPU state window and its consumers  (d01–d07) cluster_9 X2 · Register state behind the window  (d08–d11) cluster_7 X3 · Multi-page transient maps  (d12–d14) cluster_8 X4 · FPU / XSAVE area handling  (d15–d21) cluster_g5 VH · Per-vCPU area and mapcache for HVM  (d22–d30) cluster_0 Per-domain L3 at domain creation cluster_1 vCPU-scoped perdomain interfaces cluster_2 asi= option and per-vCPU L3 cluster_3 Per-vCPU mapcache cluster_g6 DV · Sparse directmap view  (d31–d39) cluster_4 Sparse directmap view cluster_5 Xenheap frees on the sparse view cluster_20 vmap-strict sub-option cluster_12 PH1 · Rendezvous / callfunc off-stack  (d40–d50) cluster_g8 PH2 · Per-pCPU stacks for HVM  (d51–d66) cluster_13 Per-pCPU stacks cluster_14 Stack checks (droppable extras) cluster_19 Stack-overflow report cluster_15 flushtlb RFC cluster_g10 O · Performance optimizations  (d68–d74, d77) cluster_10 Paging-pool pages in the sparse view cluster_11 Fewer guest page walks cluster_16 Shadow pool RFC cluster_18 Flush mask RFC cluster_17 Debug · Flush counters and audit key  (d75–d76) p01 d01  introduce populate_perdoma in_mapping() p02 d02  introduce a per-vCPU state window in the perdomain area p01->p02 p06 d06  allocate VMCBs from the domain heap behind the vCPU state... p02->p06 p07 d07  move the guest MSR load/save list off the xenheap p02->p07 p09 d09  move the saved register frame behind the vCPU state window p02->p09 p21 d21  move guest XSAVE areas behind the vCPU state window p02->p21 p03 d03  introduce SVM_MAP_VMCB() and SVM_UNMAP_VMCB() p03->p06 p04 d04  restore guest debug registers in svm_do_resume() p04->p06 p05 d05  Pass struct vcpu to svm_vmcb_dump() p05->p06 p08 d08  access a vCPU's saved register frame through map/unmap he... g_state Guest register state not always-mapped p07->g_state p08->p09 p10 d10  move the emulator's operand caches behind the vCPU state ... p11 d11  clear the vCPU's copy of an emulated I/O request once con... p11->g_state p12 d12  introduce map_domain_pages() p13 d13  use map_domain_pages() for cross-page emulated writes p12->p13 p14 d14  use map_domain_pages() for multi-frame emulated accesses p12->p14 p12->p21 p38 d38  add the vmap-strict sub- option of asi= p12->p38 g_map Transient guest mappings per-vCPU private p13->g_map p14->g_map p15 d15  Refactor FXSAVE_AREA to use wrappers p15->p21 p16 d16  Map/unmap xsave area in read_bndcfgu() p17 d17  Pass explicit xsave areas to fpu_(f)xsave() p17->p21 p18 d18  Pass explicit xsave areas to fpu_(f)xrstor() p18->p21 p19 d19  classify xsave-area seam accesses by failure mode p19->p21 p20 d20  restore FPU state after switching to the new page tables p20->p21 p21->g_state c61c1d7aafc 61c1d7aafc  allocate the per-domain L3 in arch_domain_create() c6d59811c3e 6d59811c3e  drop redundant create_perdomain_mapping() call c61c1d7aafc->c6d59811c3e p22 d22  prepare create_perdomain_mapping() for per-vCPU perdomain... c61c1d7aafc->p22 caf8ab775f3 af8ab775f3  purge unneeded destroy_per domain_mapping() cd4f928aaa3 d4f928aaa3  prepare destroy_perdomain_ mapping() for per-vCPU perdomai... caf8ab775f3->cd4f928aaa3 p25 d25  introduce per-vCPU L3 page-table cd4f928aaa3->p25 c6d59811c3e->p22 p22->p25 p23 d23  introduce Address Space Isolation command line option p23->p25 p24 d24  have init_xen_l4_slots() take a vCPU in place of the domain p24->p25 p30 d30  introduce per-vCPU mapcache for vCPU-PT domains p25->p30 p59 d59  map every CPU's stack at a per-CPU address in the per-dom... p25->p59 p26 d26  drop redundant is_pv_vcpu() checks p26->p30 p27 d27  prepare the mapcache for per-vCPU accounting p27->p30 p28 d28  ✓ move struct mapcache_vcpu to struct arch_vcpu p28->p30 p29 d29  ✓ don't unmap a mapcache entry that is not present p29->p30 p34 d34  key the directmap fast paths on the loaded view p30->p34 p30->g_map p31 d31  let Xen's page-table walker operate on a given root p32 d32  add hooks for a sparse view of the directmap p31->p32 p33 d33  build and maintain a sparse view of the directmap p32->p33 p33->p34 p36 d36  run guest contexts on the sparse directmap view p33->p36 p39 d39  split a sparse directmap superpage with a page being freed p33->p39 p37 d37  wire the sparse directmap view into the asi= option p34->p37 p35 d35  clear a mapcache slot at its last unmap p35->p37 p36->p37 p67 d67  [RFC] leave nothing of the old page-tables usable at the CR3 write p36->p67 p62 d62  allocate per-CPU stacks from the domain heap p37->p62 p68 d68  allow domain heap pages in the sparse directmap view p37->p68 p77 d77  [RFC] track the CPUs to flush for the sparse directmap view p37->p77 g_dmap Guest contexts off the full directmap p37->g_dmap caa0b5c26fe aa0b5c26fe  process pending softirqs once per deflate block caa0b5c26fe->p33 ce33b6d11fe e33b6d11fe  return an emptied region after dropping the pool lock ce33b6d11fe->p33 p40 d40  move the set_mtrr() rendezvous data off the stack p50 d50  assert that by-pointer callfunc payloads are not on the s... p40->p50 p41 d41  move smp_send_stop()'s AP- stop flag off the stack p41->p50 p42 d42  move the calibration rendezvous data off the stack p42->p50 p43 d43  move the stop_machine payload off the stack p43->p50 p44 d44  pass NULL to x86_mc_mceinject() p44->p50 p45 d45  introduce by-value payloads for call-function IPIs p47 d47  pass stack-allocated callfunc payloads by value p45->p47 p48 d48  pass the error-injection callfunc payload by value p45->p48 p49 d49  pass remaining stack- allocated callfunc payloads by value p45->p49 p46 d46  move the CPU mask out of struct drv_cmd p46->p47 p47->p50 p48->p50 p49->p50 p61 d61  map only its own CPU's stack in a cpu-stack vCPU's page-t... p50->p61 p53 d53  track per-CPU stack pages alongside stack_base p53->p59 p54 d54  keep a per-CPU mirror of current for cross-CPU readers p54->p62 p55 d55  add the cpu-stack sub- option of asi= p56 d56  keep cpu-stack off while XPTI is enabled p55->p56 p55->p59 p56->p59 p57 d57  write shadow stack tokens for the address the stack is us... p60 d60  run every CPU on a per-CPU stack address p57->p60 p58 d58  allocate the BSP's stack after the speculation mitigations p58->p59 p59->p60 p64 d64  assert that data handed to other CPUs is not on a CPU's s... p59->p64 p65 d65  name the CPU whose stack a fatal page fault hit p59->p65 p60->p61 p61->p62 p63 d63  zero stack on context switch p61->p63 g_stack Xen stacks per-CPU private and scrubbed p62->g_stack p63->g_stack p66 d66  crash the domain if the saved stack is not this CPU's p51 d51  refresh the BSP's rsp0 under FRED when its stack moves p52 d52  have show_stack_overflow() expect the stack the TSS names p51->p52 p52->p60 p69 d69  keep the HAP pool in the sparse directmap view p68->p69 p74 d74  [RFC] keep the shadow pool in the sparse directmap view p68->p74 p76 d76  [DEBUG] add a debug key auditing the sparse directmap view p68->p76 p70 d70  batch the pool's removal from the sparse directmap view p69->p70 g_perf Performance p69->g_perf p75 d75  [DEBUG] count the sparse directmap view's flushes and view switches p70->p75 p70->g_perf p71 d71  don't look up the final frame of a linear translation twice p71->g_perf p72 d72  write a virtual-address runstate area through one transla... p72->g_perf p73 d73  write a virtual-address time area through one translation p73->g_perf p74->g_perf p77->g_perf

Xen ASI series — 77 patches, and 6 in staging

Selected patch

Click a patch. Blue outline = what it needs (ancestors); vermilion = what needs it (descendants). Click again or press Clear to reset.

Legend

verified dependency
inferred (verify during group review)
dotted purple arrow, hollow head: no regression — the later patch works without the earlier one, but without it would degrade another feature (named on hover), so the earlier one goes first
gold arrow: completes an end goal (hexagons). Each goal is drawn beside the group that achieves it; together the end goals constitute the ASI end state (guest-owned data not reachable from other contexts). The gold octagon, Performance, is what group O achieves; it is not part of the ASI end state.
green node border: vetted so far (status shown on selection)
double border: posted to xen-devel (thread link on selection)
gray node, labelled with its commit: in staging (the commit on xenbits on selection); gray arrows: what later patches need from it
blue border and ✓ in the label: carries a Reviewed-by (tags shown on selection)
Dependencies already implied by a drawn path are recorded but not drawn; the click-cone includes them. An arrow from or to a box summarises; its tooltip says what it stands for.

Groups

X = key per-vCPU structures off the xenheap, into a per-vCPU state window in the per-domain area, in four groups: X1 the window and its first consumers, the SVM VMCBs and the VMX guest MSR lists; X2 every guest vCPU's saved register frame and the HVM emulator's operand caches; X3 map_domain_pages(), which maps several pages in the per-domain area of the calling vCPU's domain, used in place of vmap() by two emulation paths; X4 the guest XSAVE areas, with the FPU changes ahead of them, VH = per-vCPU area and mapcache for HVM domains, DV = a sparse view of the directmap, alongside the full one, chosen per domain at creation, PH = per-pCPU stacks for HVM domains, in two groups: PH1 the preparation, rendezvous data and call-function IPI payloads off the stack; PH2 the stacks themselves: first two changes to the stack-overflow report (the BSP's rsp0 under FRED, an upstream fix, and the stack the report expects), then each pCPU's stack at a per-CPU address in the per-domain area, a cpu-stack vCPU's page-tables mapping only its own pCPU's, the stacks allocated from the domain heap and zeroed on context switch, and three stack checks that could be dropped, O = performance optimizations: paging-pool pages kept in the sparse view (the HAP pool; the shadow pool as an [RFC]), their removal batched; fewer guest page walks and p2m lookups for HVM guests; and, as an [RFC], the sparse view's flushes sent only to the CPUs that may hold its translations; the two [RFC]s are at the end of the series, Debug = aids for testing: [DEBUG] patches, not for committing, that count the sparse view's flushes and audit it with a debug key. Numbered in series order within a letter; the series runs X1 X2 X3 X4 VH DV PH1 PH2 O Debug. Ranges are positions in this revision of the page.

Postings

Open items

Source of truth: gen-depgraph.py (same directory); positions read live from stg series; regenerate with python3 gen-depgraph.py.