| 1 | PCI SR/IOV EMULATION SUPPORT |
| 2 | ============================ |
| 3 | |
| 4 | Description |
| 5 | =========== |
| 6 | SR/IOV (Single Root I/O Virtualization) is an optional extended capability |
| 7 | of a PCI Express device. It allows a single physical function (PF) to appear as multiple |
| 8 | virtual functions (VFs) for the main purpose of eliminating software |
| 9 | overhead in I/O from virtual machines. |
| 10 | |
| 11 | QEMU now implements the basic common functionality to enable an emulated device |
| 12 | to support SR/IOV. |
| 13 | |
| 14 | Implementation |
| 15 | ============== |
| 16 | Implementing emulation of an SR/IOV capable device typically consists of |
| 17 | implementing support for two types of device classes; the "normal" physical device |
| 18 | (PF) and the virtual device (VF). From QEMU's perspective, the VFs are just |
| 19 | like other devices, except that some of their properties are derived from |
| 20 | the PF. |
| 21 | |
| 22 | A virtual function is different from a physical function in that the BAR |
| 23 | space for all VFs are defined by the BAR registers in the PFs SR/IOV |
| 24 | capability. All VFs have the same BARs and BAR sizes. |
| 25 | |
| 26 | Accesses to these virtual BARs then is computed as |
| 27 | |
| 28 | <VF BAR start> + <VF number> * <BAR sz> + <offset> |
| 29 | |
| 30 | From our emulation perspective this means that there is a separate call for |
| 31 | setting up a BAR for a VF. |
| 32 | |
| 33 | 1) To enable SR/IOV support in the PF, it must be a PCI Express device so |
| 34 | you would need to add a PCI Express capability in the normal PCI |
| 35 | capability list. You might also want to add an ARI (Alternative |
| 36 | Routing-ID Interpretation) capability to indicate that your device |
| 37 | supports functions beyond it's "own" function space (0-7), |
| 38 | which is necessary to support more than 7 functions, or |
| 39 | if functions extends beyond offset 7 because they are placed at an |
| 40 | offset > 1 or have stride > 1. |
| 41 | |
| 42 | ... |
| 43 | #include "hw/pci/pcie.h" |
| 44 | #include "hw/pci/pcie_sriov.h" |
| 45 | |
| 46 | pci_your_pf_dev_realize( ... ) |
| 47 | { |
| 48 | ... |
| 49 | int ret = pcie_endpoint_cap_init(d, 0x70); |
| 50 | ... |
| 51 | pcie_ari_init(d, 0x100); |
| 52 | ... |
| 53 | |
| 54 | /* Add and initialize the SR/IOV capability */ |
| 55 | if (!pcie_sriov_pf_init(d, 0x200, "your_virtual_dev", |
| 56 | vf_devid, initial_vfs, total_vfs, |
| 57 | fun_offset, stride, errp)) { |
| 58 | return; |
| 59 | } |
| 60 | |
| 61 | /* Set up individual VF BARs (parameters as for normal BARs) */ |
| 62 | pcie_sriov_pf_init_vf_bar( ... ) |
| 63 | ... |
| 64 | } |
| 65 | |
| 66 | For cleanup, you simply call: |
| 67 | |
| 68 | pcie_sriov_pf_exit(device); |
| 69 | |
| 70 | which will delete all the virtual functions and associated resources. |
| 71 | |
| 72 | 2) Similarly in the implementation of the virtual function, you need to |
| 73 | make it a PCI Express device and add a similar set of capabilities |
| 74 | except for the SR/IOV capability. Then you need to set up the VF BARs as |
| 75 | subregions of the PFs SR/IOV VF BARs by calling pci_register_bar(): |
| 76 | |
| 77 | pci_your_vf_dev_realize( ... ) |
| 78 | { |
| 79 | ... |
| 80 | int ret = pcie_endpoint_cap_init(d, 0x60); |
| 81 | ... |
| 82 | pcie_ari_init(d, 0x100); |
| 83 | ... |
| 84 | memory_region_init(mr, ... ) |
| 85 | pci_register_bar(d, bar_nr, bar_type, mr); |
| 86 | ... |
| 87 | } |
| 88 | |
| 89 | Testing on Linux guest |
| 90 | ====================== |
| 91 | The easiest is if your device driver supports sysfs based SR/IOV |
| 92 | enabling. Support for this was added in kernel v.3.8, so not all drivers |
| 93 | support it yet. |
| 94 | |
| 95 | To enable 4 VFs for a device at 01:00.0: |
| 96 | |
| 97 | modprobe yourdriver |
| 98 | echo 4 > /sys/bus/pci/devices/0000:01:00.0/sriov_numvfs |
| 99 | |
| 100 | You should now see 4 VFs with lspci. |
| 101 | To turn SR/IOV off again - the standard requires you to turn it off before you can enable |
| 102 | another VF count, and the emulation enforces this: |
| 103 | |
| 104 | echo 0 > /sys/bus/pci/devices/0000:01:00.0/sriov_numvfs |
| 105 | |
| 106 | Older drivers typically provide a max_vfs module parameter |
| 107 | to enable it at load time: |
| 108 | |
| 109 | modprobe yourdriver max_vfs=4 |
| 110 | |
| 111 | To disable the VFs again then, you simply have to unload the driver: |
| 112 | |
| 113 | rmmod yourdriver |