Intel-IOMMU.txt 3.2 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293
  1. Linux IOMMU Support
  2. ===================
  3. The architecture spec can be obtained from the below location.
  4. http://www.intel.com/technology/virtualization/
  5. This guide gives a quick cheat sheet for some basic understanding.
  6. Some Keywords
  7. DMAR - DMA remapping
  8. DRHD - DMA Engine Reporting Structure
  9. RMRR - Reserved memory Region Reporting Structure
  10. ZLR - Zero length reads from PCI devices
  11. IOVA - IO Virtual address.
  12. Basic stuff
  13. -----------
  14. ACPI enumerates and lists the different DMA engines in the platform, and
  15. device scope relationships between PCI devices and which DMA engine controls
  16. them.
  17. What is RMRR?
  18. -------------
  19. There are some devices the BIOS controls, for e.g USB devices to perform
  20. PS2 emulation. The regions of memory used for these devices are marked
  21. reserved in the e820 map. When we turn on DMA translation, DMA to those
  22. regions will fail. Hence BIOS uses RMRR to specify these regions along with
  23. devices that need to access these regions. OS is expected to setup
  24. unity mappings for these regions for these devices to access these regions.
  25. How is IOVA generated?
  26. ---------------------
  27. Well behaved drivers call pci_map_*() calls before sending command to device
  28. that needs to perform DMA. Once DMA is completed and mapping is no longer
  29. required, device performs a pci_unmap_*() calls to unmap the region.
  30. The Intel IOMMU driver allocates a virtual address per domain. Each PCIE
  31. device has its own domain (hence protection). Devices under p2p bridges
  32. share the virtual address with all devices under the p2p bridge due to
  33. transaction id aliasing for p2p bridges.
  34. IOVA generation is pretty generic. We used the same technique as vmalloc()
  35. but these are not global address spaces, but separate for each domain.
  36. Different DMA engines may support different number of domains.
  37. We also allocate gaurd pages with each mapping, so we can attempt to catch
  38. any overflow that might happen.
  39. Graphics Problems?
  40. ------------------
  41. If you encounter issues with graphics devices, you can try adding
  42. option intel_iommu=igfx_off to turn off the integrated graphics engine.
  43. Some exceptions to IOVA
  44. -----------------------
  45. Interrupt ranges are not address translated, (0xfee00000 - 0xfeefffff).
  46. The same is true for peer to peer transactions. Hence we reserve the
  47. address from PCI MMIO ranges so they are not allocated for IOVA addresses.
  48. Boot Message Sample
  49. -------------------
  50. Something like this gets printed indicating presence of DMAR tables
  51. in ACPI.
  52. ACPI: DMAR (v001 A M I OEMDMAR 0x00000001 MSFT 0x00000097) @ 0x000000007f5b5ef0
  53. When DMAR is being processed and initialized by ACPI, prints DMAR locations
  54. and any RMRR's processed.
  55. ACPI DMAR:Host address width 36
  56. ACPI DMAR:DRHD (flags: 0x00000000)base: 0x00000000fed90000
  57. ACPI DMAR:DRHD (flags: 0x00000000)base: 0x00000000fed91000
  58. ACPI DMAR:DRHD (flags: 0x00000001)base: 0x00000000fed93000
  59. ACPI DMAR:RMRR base: 0x00000000000ed000 end: 0x00000000000effff
  60. ACPI DMAR:RMRR base: 0x000000007f600000 end: 0x000000007fffffff
  61. When DMAR is enabled for use, you will notice..
  62. PCI-DMA: Using DMAR IOMMU
  63. TBD
  64. ----
  65. - For compatibility testing, could use unity map domain for all devices, just
  66. provide a 1-1 for all useful memory under a single domain for all devices.
  67. - API for paravirt ops for abstracting functionlity for VMM folks.