| Message ID | 20210224094910.44986-1-david@redhat.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | virtio-mem: vfio support | expand |
On 24.02.21 10:48, David Hildenbrand wrote: > A virtio-mem device manages a memory region in guest physical address > space, represented as a single (currently large) memory region in QEMU, > mapped into system memory address space. Before the guest is allowed to use > memory blocks, it must coordinate with the hypervisor (plug blocks). After > a reboot, all memory is usually unplugged - when the guest comes up, it > detects the virtio-mem device and selects memory blocks to plug (based on > resize requests from the hypervisor). > > Memory hot(un)plug consists of (un)plugging memory blocks via a virtio-mem > device (triggered by the guest). When unplugging blocks, we discard the > memory - similar to memory balloon inflation. In contrast to memory > ballooning, we always know which memory blocks a guest may actually use - > especially during a reboot, after a crash, or after kexec (and during > hibernation as well). Guests agreed to not access unplugged memory again, > especially not via DMA. > > The issue with vfio is, that it cannot deal with random discards - for this > reason, virtio-mem and vfio can currently only run mutually exclusive. > Especially, vfio would currently map the whole memory region (with possible > only little/no plugged blocks), resulting in all pages getting pinned and > therefore resulting in a higher memory consumption than expected (turning > virtio-mem basically useless in these environments). > > To make vfio work nicely with virtio-mem, we have to map only the plugged > blocks, and map/unmap properly when plugging/unplugging blocks (including > discarding of RAM when unplugging). We achieve that by using a new notifier > mechanism that communicates changes. > > It's important to map memory in the granularity in which we could see > unmaps again (-> virtio-mem block size) - so when e.g., plugging > consecutive 100 MB with a block size of 2 MB, we need 50 mappings. When > unmapping, we can use a single vfio_unmap call for the applicable range. > We expect that the block size of virtio-mem devices will be fairly large > in the future (to not run out of mappings and to improve hot(un)plug > performance), configured by the user, when used with vfio (e.g., 128MB, > 1G, ...), but it will depend on the setup. > > More info regarding virtio-mem can be found at: > https://virtio-mem.gitlab.io/ > > v7 is located at: > git@github.com:davidhildenbrand/qemu.git virtio-mem-vfio-v7 > Gentle ping.
On 02.03.21 17:46, David Hildenbrand wrote: > On 24.02.21 10:48, David Hildenbrand wrote: >> A virtio-mem device manages a memory region in guest physical address >> space, represented as a single (currently large) memory region in QEMU, >> mapped into system memory address space. Before the guest is allowed to use >> memory blocks, it must coordinate with the hypervisor (plug blocks). After >> a reboot, all memory is usually unplugged - when the guest comes up, it >> detects the virtio-mem device and selects memory blocks to plug (based on >> resize requests from the hypervisor). >> >> Memory hot(un)plug consists of (un)plugging memory blocks via a virtio-mem >> device (triggered by the guest). When unplugging blocks, we discard the >> memory - similar to memory balloon inflation. In contrast to memory >> ballooning, we always know which memory blocks a guest may actually use - >> especially during a reboot, after a crash, or after kexec (and during >> hibernation as well). Guests agreed to not access unplugged memory again, >> especially not via DMA. >> >> The issue with vfio is, that it cannot deal with random discards - for this >> reason, virtio-mem and vfio can currently only run mutually exclusive. >> Especially, vfio would currently map the whole memory region (with possible >> only little/no plugged blocks), resulting in all pages getting pinned and >> therefore resulting in a higher memory consumption than expected (turning >> virtio-mem basically useless in these environments). >> >> To make vfio work nicely with virtio-mem, we have to map only the plugged >> blocks, and map/unmap properly when plugging/unplugging blocks (including >> discarding of RAM when unplugging). We achieve that by using a new notifier >> mechanism that communicates changes. >> >> It's important to map memory in the granularity in which we could see >> unmaps again (-> virtio-mem block size) - so when e.g., plugging >> consecutive 100 MB with a block size of 2 MB, we need 50 mappings. When >> unmapping, we can use a single vfio_unmap call for the applicable range. >> We expect that the block size of virtio-mem devices will be fairly large >> in the future (to not run out of mappings and to improve hot(un)plug >> performance), configured by the user, when used with vfio (e.g., 128MB, >> 1G, ...), but it will depend on the setup. >> >> More info regarding virtio-mem can be found at: >> https://virtio-mem.gitlab.io/ >> >> v7 is located at: >> git@github.com:davidhildenbrand/qemu.git virtio-mem-vfio-v7 >> > > Gentle ping. @Paolo, can you have another look? thanks.
On 24.02.21 10:48, David Hildenbrand wrote: > A virtio-mem device manages a memory region in guest physical address > space, represented as a single (currently large) memory region in QEMU, > mapped into system memory address space. Before the guest is allowed to use > memory blocks, it must coordinate with the hypervisor (plug blocks). After > a reboot, all memory is usually unplugged - when the guest comes up, it > detects the virtio-mem device and selects memory blocks to plug (based on > resize requests from the hypervisor). > > Memory hot(un)plug consists of (un)plugging memory blocks via a virtio-mem > device (triggered by the guest). When unplugging blocks, we discard the > memory - similar to memory balloon inflation. In contrast to memory > ballooning, we always know which memory blocks a guest may actually use - > especially during a reboot, after a crash, or after kexec (and during > hibernation as well). Guests agreed to not access unplugged memory again, > especially not via DMA. > > The issue with vfio is, that it cannot deal with random discards - for this > reason, virtio-mem and vfio can currently only run mutually exclusive. > Especially, vfio would currently map the whole memory region (with possible > only little/no plugged blocks), resulting in all pages getting pinned and > therefore resulting in a higher memory consumption than expected (turning > virtio-mem basically useless in these environments). > > To make vfio work nicely with virtio-mem, we have to map only the plugged > blocks, and map/unmap properly when plugging/unplugging blocks (including > discarding of RAM when unplugging). We achieve that by using a new notifier > mechanism that communicates changes. > > It's important to map memory in the granularity in which we could see > unmaps again (-> virtio-mem block size) - so when e.g., plugging > consecutive 100 MB with a block size of 2 MB, we need 50 mappings. When > unmapping, we can use a single vfio_unmap call for the applicable range. > We expect that the block size of virtio-mem devices will be fairly large > in the future (to not run out of mappings and to improve hot(un)plug > performance), configured by the user, when used with vfio (e.g., 128MB, > 1G, ...), but it will depend on the setup. > > More info regarding virtio-mem can be found at: > https://virtio-mem.gitlab.io/ > > v7 is located at: > git@github.com:davidhildenbrand/qemu.git virtio-mem-vfio-v7 > > v6 -> v7: > - s/RamDiscardMgr/RamDiscardManager/ > - "memory: Introduce RamDiscardManager for RAM memory regions" > -- Make RamDiscardManager/RamDiscardListener eat MemoryRegionSections > -- Replace notify_discard_all callback by double_discard_supported > -- Reshuffle the individual hunks in memory.h > -- Provide function wrappers for RamDiscardManager calls > - "memory: Helpers to copy/free a MemoryRegionSection" > -- Added > - "virtio-mem: Implement RamDiscardManager interface" > -- Work on MemoryRegionSections instead of ranges > -- Minor optimizations > - "vfio: Support for RamDiscardManager in the !vIOMMU case" > -- Simplify based on new interfaces / MemoryRegionSections > -- Minor cleanups and optimizations > -- Add a comment regarding dirty bitmap sync. > -- Don't store "offset_within_region" in VFIORamDiscardListener > - "vfio: Support for RamDiscardManager in the vIOMMU case" > -- Adjust to new interface > - "softmmu/physmem: Don't use atomic operations in ..." > -- Rename variables > - "softmmu/physmem: Extend ram_block_discard_(require|disable) ..." > -- Rename variables > - Rebased and retested > > v5 -> v6: > - "memory: Introduce RamDiscardMgr for RAM memory regions" > -- Fix variable names in one prototype. > - "virtio-mem: Don't report errors when ram_block_discard_range() fails" > -- Added > - "virtio-mem: Implement RamDiscardMgr interface" > -- Don't report an error if discarding fails > - Rebased and retested > > v4 -> v5: > - "vfio: Support for RamDiscardMgr in the !vIOMMU case" > -- Added more assertions for granularity vs. iommu supported pagesize > - "vfio: Sanity check maximum number of DMA mappings with RamDiscardMgr" > -- Fix accounting of mappings > - "vfio: Disable only uncoordinated discards for VFIO_TYPE1 iommus" > -- Fence off SPAPR and add some comments regarding future support. > -- Tweak patch description > - Rebase and retest > > v3 -> v4: > - "vfio: Query and store the maximum number of DMA mappings > -- Limit the patch to querying and storing only > -- Renamed to "vfio: Query and store the maximum number of possible DMA > mappings" > - "vfio: Support for RamDiscardMgr in the !vIOMMU case" > -- Remove sanity checks / warning the user > - "vfio: Sanity check maximum number of DMA mappings with RamDiscardMgr" > -- Perform sanity checks by looking at the number of memslots and all > registered RamDiscardMgr sections > - Rebase and retest > - Reshuffled the patches slightly > > v2 -> v3: > - Rebased + retested > - Fixed some typos > - Added RB's > > v1 -> v2: > - "memory: Introduce RamDiscardMgr for RAM memory regions" > -- Fix some errors in the documentation > -- Make register_listener() notify about populated parts and > unregister_listener() notify about discarding populated parts, to > simplify future locking inside virtio-mem, when handling requests via a > separate thread. > - "vfio: Query and store the maximum number of DMA mappings" > -- Query number of mappings and track mappings (except for vIOMMU) > - "vfio: Support for RamDiscardMgr in the !vIOMMU case" > -- Adapt to RamDiscardMgr changes and warn via generic DMA reservation > - "vfio: Support for RamDiscardMgr in the vIOMMU case" > -- Use vmstate priority to handle migration dependencies > > RFC - v1: > - VFIO migration code. Due to missing kernel support, I cannot really test > if that part works. > - Understand/test/document vIOMMU implications, also regarding migration > - Nicer ram_block_discard_disable/require handling. > - s/SparseRAMHandler/RamDiscardMgr/, refactorings, cleanups, documentation, > testing, ... > > David Hildenbrand (13): > memory: Introduce RamDiscardManager for RAM memory regions > memory: Helpers to copy/free a MemoryRegionSection > virtio-mem: Factor out traversing unplugged ranges > virtio-mem: Don't report errors when ram_block_discard_range() fails > virtio-mem: Implement RamDiscardManager interface > vfio: Support for RamDiscardManager in the !vIOMMU case > vfio: Query and store the maximum number of possible DMA mappings > vfio: Sanity check maximum number of DMA mappings with > RamDiscardManager > vfio: Support for RamDiscardManager in the vIOMMU case > softmmu/physmem: Don't use atomic operations in > ram_block_discard_(disable|require) > softmmu/physmem: Extend ram_block_discard_(require|disable) by two > discard types > virtio-mem: Require only coordinated discards > vfio: Disable only uncoordinated discards for VFIO_TYPE1 iommus > > hw/vfio/common.c | 315 +++++++++++++++++++++++++- > hw/virtio/virtio-mem.c | 391 ++++++++++++++++++++++++++++----- > include/exec/memory.h | 324 +++++++++++++++++++++++++-- > include/hw/vfio/vfio-common.h | 12 + > include/hw/virtio/virtio-mem.h | 3 + > include/migration/vmstate.h | 1 + > softmmu/memory.c | 98 +++++++++ > softmmu/physmem.c | 108 ++++++--- > 8 files changed, 1133 insertions(+), 119 deletions(-) > Another gentle ping; it's been almost a month with no feedback. I hope we can get this early into 6.1, because I have other stuff coming up that rely on the RamDiscardManager infrastructure. Feedback/acks are appreciated so I can finally make progress with this.
A virtio-mem device manages a memory region in guest physical address space, represented as a single (currently large) memory region in QEMU, mapped into system memory address space. Before the guest is allowed to use memory blocks, it must coordinate with the hypervisor (plug blocks). After a reboot, all memory is usually unplugged - when the guest comes up, it detects the virtio-mem device and selects memory blocks to plug (based on resize requests from the hypervisor). Memory hot(un)plug consists of (un)plugging memory blocks via a virtio-mem device (triggered by the guest). When unplugging blocks, we discard the memory - similar to memory balloon inflation. In contrast to memory ballooning, we always know which memory blocks a guest may actually use - especially during a reboot, after a crash, or after kexec (and during hibernation as well). Guests agreed to not access unplugged memory again, especially not via DMA. The issue with vfio is, that it cannot deal with random discards - for this reason, virtio-mem and vfio can currently only run mutually exclusive. Especially, vfio would currently map the whole memory region (with possible only little/no plugged blocks), resulting in all pages getting pinned and therefore resulting in a higher memory consumption than expected (turning virtio-mem basically useless in these environments). To make vfio work nicely with virtio-mem, we have to map only the plugged blocks, and map/unmap properly when plugging/unplugging blocks (including discarding of RAM when unplugging). We achieve that by using a new notifier mechanism that communicates changes. It's important to map memory in the granularity in which we could see unmaps again (-> virtio-mem block size) - so when e.g., plugging consecutive 100 MB with a block size of 2 MB, we need 50 mappings. When unmapping, we can use a single vfio_unmap call for the applicable range. We expect that the block size of virtio-mem devices will be fairly large in the future (to not run out of mappings and to improve hot(un)plug performance), configured by the user, when used with vfio (e.g., 128MB, 1G, ...), but it will depend on the setup. More info regarding virtio-mem can be found at: https://virtio-mem.gitlab.io/ v7 is located at: git@github.com:davidhildenbrand/qemu.git virtio-mem-vfio-v7 v6 -> v7: - s/RamDiscardMgr/RamDiscardManager/ - "memory: Introduce RamDiscardManager for RAM memory regions" -- Make RamDiscardManager/RamDiscardListener eat MemoryRegionSections -- Replace notify_discard_all callback by double_discard_supported -- Reshuffle the individual hunks in memory.h -- Provide function wrappers for RamDiscardManager calls - "memory: Helpers to copy/free a MemoryRegionSection" -- Added - "virtio-mem: Implement RamDiscardManager interface" -- Work on MemoryRegionSections instead of ranges -- Minor optimizations - "vfio: Support for RamDiscardManager in the !vIOMMU case" -- Simplify based on new interfaces / MemoryRegionSections -- Minor cleanups and optimizations -- Add a comment regarding dirty bitmap sync. -- Don't store "offset_within_region" in VFIORamDiscardListener - "vfio: Support for RamDiscardManager in the vIOMMU case" -- Adjust to new interface - "softmmu/physmem: Don't use atomic operations in ..." -- Rename variables - "softmmu/physmem: Extend ram_block_discard_(require|disable) ..." -- Rename variables - Rebased and retested v5 -> v6: - "memory: Introduce RamDiscardMgr for RAM memory regions" -- Fix variable names in one prototype. - "virtio-mem: Don't report errors when ram_block_discard_range() fails" -- Added - "virtio-mem: Implement RamDiscardMgr interface" -- Don't report an error if discarding fails - Rebased and retested v4 -> v5: - "vfio: Support for RamDiscardMgr in the !vIOMMU case" -- Added more assertions for granularity vs. iommu supported pagesize - "vfio: Sanity check maximum number of DMA mappings with RamDiscardMgr" -- Fix accounting of mappings - "vfio: Disable only uncoordinated discards for VFIO_TYPE1 iommus" -- Fence off SPAPR and add some comments regarding future support. -- Tweak patch description - Rebase and retest v3 -> v4: - "vfio: Query and store the maximum number of DMA mappings -- Limit the patch to querying and storing only -- Renamed to "vfio: Query and store the maximum number of possible DMA mappings" - "vfio: Support for RamDiscardMgr in the !vIOMMU case" -- Remove sanity checks / warning the user - "vfio: Sanity check maximum number of DMA mappings with RamDiscardMgr" -- Perform sanity checks by looking at the number of memslots and all registered RamDiscardMgr sections - Rebase and retest - Reshuffled the patches slightly v2 -> v3: - Rebased + retested - Fixed some typos - Added RB's v1 -> v2: - "memory: Introduce RamDiscardMgr for RAM memory regions" -- Fix some errors in the documentation -- Make register_listener() notify about populated parts and unregister_listener() notify about discarding populated parts, to simplify future locking inside virtio-mem, when handling requests via a separate thread. - "vfio: Query and store the maximum number of DMA mappings" -- Query number of mappings and track mappings (except for vIOMMU) - "vfio: Support for RamDiscardMgr in the !vIOMMU case" -- Adapt to RamDiscardMgr changes and warn via generic DMA reservation - "vfio: Support for RamDiscardMgr in the vIOMMU case" -- Use vmstate priority to handle migration dependencies RFC - v1: - VFIO migration code. Due to missing kernel support, I cannot really test if that part works. - Understand/test/document vIOMMU implications, also regarding migration - Nicer ram_block_discard_disable/require handling. - s/SparseRAMHandler/RamDiscardMgr/, refactorings, cleanups, documentation, testing, ... David Hildenbrand (13): memory: Introduce RamDiscardManager for RAM memory regions memory: Helpers to copy/free a MemoryRegionSection virtio-mem: Factor out traversing unplugged ranges virtio-mem: Don't report errors when ram_block_discard_range() fails virtio-mem: Implement RamDiscardManager interface vfio: Support for RamDiscardManager in the !vIOMMU case vfio: Query and store the maximum number of possible DMA mappings vfio: Sanity check maximum number of DMA mappings with RamDiscardManager vfio: Support for RamDiscardManager in the vIOMMU case softmmu/physmem: Don't use atomic operations in ram_block_discard_(disable|require) softmmu/physmem: Extend ram_block_discard_(require|disable) by two discard types virtio-mem: Require only coordinated discards vfio: Disable only uncoordinated discards for VFIO_TYPE1 iommus hw/vfio/common.c | 315 +++++++++++++++++++++++++- hw/virtio/virtio-mem.c | 391 ++++++++++++++++++++++++++++----- include/exec/memory.h | 324 +++++++++++++++++++++++++-- include/hw/vfio/vfio-common.h | 12 + include/hw/virtio/virtio-mem.h | 3 + include/migration/vmstate.h | 1 + softmmu/memory.c | 98 +++++++++ softmmu/physmem.c | 108 ++++++--- 8 files changed, 1133 insertions(+), 119 deletions(-)