After enabling vPCI in hypervisor for vm0, UOS may not able be launched
successfully. Consider this scenario (take MSI-X for example):
- DM makes hypercall to hypervisor to do MSI-X remapping on behalf of
UOS guests.
- After the hypercall, VHM module in SOS kernel updates the physical
MSI-X table with the physical Message Data/Addr.
- These MMIO write requests are intercepted by hypervisor, which will
call ptdev_msix_remap() to do MSI-S remapping.
It may fail due to 2 possible reasons:
1) wrong target VM because:
hypervisor thinks it's a VM0 MSI-X device but they have been registered
as UOS guests through HC_SET_PTDEV_INTR_INFO hypercall.
2) wrong ptdev_msi_info->vmsi_data because:
The virtual MSI-X table is supposed to hold virtual Message data/addr
but the SOS VHM writes the physical ones to it.
This patch resolves these problems by ignoring the HC_VM_PCI_MSIX_REMAP
hypercall, so virtual and physical Message Data are the same from SOS'
perspective and it won't mess up the virtual PCI device in HV.
Also in HC_SET_PTDEV_INTR_INFO handler, vpci updates the target VM
when the PCI devices are assigned to different VMs.
The UOS' MSI/MSI-X remapping is triggered by hypervisor when SOS (either
DM or VHM) updates the Message Data/Addr.
Tracked-On: #1568
Signed-off-by: dongshen <dongsheng.x.zhang@intel.com>
Signed-off-by: Zide Chen <zide.chen@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
If PBA or other data structures reside in the same BAR with MSI-X,
devicemodel could emulate them and make hypercall SET_MEMORY_REGION
to setup EPT for vm0.
Hypervisor can not intercept the whole range of this BAR, but only
the minimum 4KB pages that cover the MSI-X Table.
Tracked-On: #1568
Signed-off-by: Zide Chen <zide.chen@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
Fixed the following violations:
- Widening cast on complex integer expression
- Casting operation to a pointer
- Static procedure is not explicitly called in code analysed
Tracked-On: #1568
Signed-off-by: Zide Chen <zide.chen@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
Compared with partition_mode.c, the major difference between them is
the list of PCI devices is statically defined or dynamically
enumerated and allocated.
Tracked-On: #1568
Signed-off-by: dongshen <dongsheng.x.zhang@intel.com>
Signed-off-by: Zide Chen <zide.chen@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
Similar to MSI emulation, this patch emulates the Message Control word
within MSI-X Capability Structure.
Also it emulates MSI-X table. MSI-X remapping is triggered when the guest
is changing the Mask bit in Vector Control, or the Message Data/Addr when
MSI-X Enable bit is set.
Tracked-On: #1568
Signed-off-by: dongshen <dongsheng.x.zhang@intel.com>
Signed-off-by: Zide Chen <zide.chen@intel.com>
Reviewed-by: Zhao Yakui <yakui.zhao@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
Emulate MSI Capability structure for vm0 in sharing mode:
- it intercepts the IO requests for MSI Capability structure, emulates
the Message Control word, and bypasses all other I/O requests to the
physical device.
- criteria to trigger MSI remapping: MSI Enable bit is being changed,
Message Data/Addr is being changed when MSI Enable is set.
Tracked-On: #1568
Signed-off-by: dongshen <dongsheng.x.zhang@intel.com>
Signed-off-by: Zide Chen <zide.chen@intel.com>
Reviewed-by: Zhao Yakui <yakui.zhao@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
Add 'struct vpci_ops *ops' to 'struct vpci' so we have clearer structure:
- struct vpci: include struct vpci_ops pointing to different callback
functions for partition or sharing mode repsectively.
- struct pci_vdev: includes struct pci_vdev_ops to handle different vpci
functionalities:
hostbridge emulation
passthrough device BAR emulation
msi/msi-x remapping
This patch moves the code around but doesn't change the underlying logic
in terms of PCI spec handling. More detailed implementation:
- create new file partition_mode.c to house the implementation of partition mode
regarding the vpci layer.
- vpci.c: only keeps the abstract code which calls vpci->ops to functions
in partition_mode.c, and potentially to sharing_mode.c.
- the following functions are moved to partition_mode.c and renamed with
partition_mode prefix.
- vpci_init() -> partition_mode_vpci_init()
- vpci_cleanup() -> partition_mode_vpci_deinit()
- pci_cfg_io_write() -> partition_mode_cfgread()
- pci_cfg_io_read() -> partition_mode_cfgwrite()
Track-On: #1568
Signed-off-by: dongshen <dongsheng.x.zhang@intel.com>
Signed-off-by: Zide Chen <zide.chen@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
Reviewed-by: Li, Fei1 <fei1.li@intel.com>
- It starts from scaning bus 0 and scan other buses only if it is referred
as a Secondary Bus by a PCI-to-PCI bridge.
- Skip to next device if function 0 is not implemented.
- Don't enumerate function 1-7 if a device is not a multi-function device.
Tracked-On: #1568
Signed-off-by: dongshen <dongsheng.x.zhang@intel.com>
Signed-off-by: Zide Chen <zide.chen@intel.com>
Reviewed-by: Zhao Yakui <yakui.zhao@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
In order to allow these functions to be called without an associated
struct pci_pdev (for example, at the time of PCI bus enumeration), these
two functions can not take the struct vdev as input parameter.
Tracked-On: #1568
Signed-off-by: dongshen <dongsheng.x.zhang@intel.com>
Signed-off-by: Zide Chen <zide.chen@intel.com>
Reviewed-by: Zhao Yakui <yakui.zhao@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
- It makes more sense to return bool from pci_bar_access()
- other minor changes for cleanup
Tracked-On: #1568
Signed-off-by: dongshen <dongsheng.x.zhang@intel.com>
Signed-off-by: Zide Chen <zide.chen@intel.com>
Reviewed-by: Li, Fei1 <fei1.li@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
Don't have any logical changes.
- add more common PCI macros
- remove redundant definitions in dmar_parse.c
- move all the common mascos from vpci.h to pci.h
Tracked-On: #1568
Signed-off-by: dongshen <dongsheng.x.zhang@intel.com>
Signed-off-by: Zide Chen <zide.chen@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
- move the functions to access physical PCI configuration space from
pci_pt.c to the new file dm/hw/pci.c, so they can be accessed in
sharing mode as well. The new folder dm/hw is created in order to
move APIs talking to physical PCI devices out of dm/vpci.
- move the common vpci code from header file pci_priv.h to core.c.
- move file include/dm/vpci/vpci.h one level up. It seems the folder
include/dm/vpci is not necessary.
- This patch only moves code around, and doesn't make any logical
changes. Besides removes the static keyword from pci_pdev_read_cfg()
and pci_pdev_write_cfg()
Tracked-On: #1568
Signed-off-by: dongshen <dongsheng.x.zhang@intel.com>
Signed-off-by: Zide Chen <zide.chen@intel.com>
Reviewed-by: Li, Fei1 <fei1.li@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
- move most of the content of pci_priv.h to include/dm/pci.h.
This allows other code outside dm/vpci to be able to share these macros.
- code cleanup: fix alignments etc.
Tracked-On: #1568
Signed-off-by: dongshen <dongsheng.x.zhang@intel.com>
Signed-off-by: Zide Chen <zide.chen@intel.com>
Reviewed-by: Li, Fei1 <fei1.li@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>
- Converted MACROS to functions
- Defined pci_bar_type enum
- Defined pci_bdf as union instead of uint16_t to eliminate macros
- Use L or UL postfix after unsigned integers
Tracked-On: #1126
Signed-off-by: dongshen <dongsheng.x.zhang@intel.com>
V4:
- Moved error checking to vdev_hostbridge_cfgwrite/vdev_hostbridge_cfgread
V3:
- Unified ops calling and implemented deinit/cfgwrite/cfgread ops,
previously only init op is implemented
Reviewed-by: Anthony Xu <anthony.xu@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
Signed-off-by: dongshen <dongsheng.x.zhang@intel.com>
V4:
- Renamed members for struct pcibar and changed code accordingly
V3:
- Do not use ASSERT
- Use EPT_XX defines when claling ept_mr_add
- Report 64-bit MMIO physical bar to UOS as 32-bit virtual bar
(assume bar size is always less than 4GB), which removed quite some of
64-bit bar handling code
Reviewed-by: Anthony Xu <anthony.xu@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
Signed-off-by: dongshen <dongsheng.x.zhang@intel.com>
V4:
- Clear address cache info after a full cf8/cfc access
- Add NULL pointer checking when calling init/deinit ops
V3:
- Do not use ASSERT
- Loop through the vdev list defined in vm_desctiption table to call the vdev init/unit functions
- Make the cached vbdf info struct per vm instead of per pcpu
V2:
- Fixed MISRA violations
Reviewed-by: Anthony Xu <anthony.xu@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
Signed-off-by: dongshen <dongsheng.x.zhang@intel.com>
V4:
- Renamed members for struct pcibar
- License header fix
- Added vpci_vdev_array to struct vm_description
V3:
- Defined the static centralized vpci table to reduce code size,
previously many of the settings are obtained/generated dynamically
Reviewed-by: Anthony Xu <anthony.xu@intel.com>
Acked-by: Anthony Xu <anthony.xu@intel.com>
Signed-off-by: dongshen <dongsheng.x.zhang@intel.com>
