Branch: refs/heads/wip/ivshmem2
Home: https://github.com/siemens/jailhouse
Commit: d5776fcc19ea6a8b125581b78806b25b87c8ff6b
https://github.com/siemens/jailhouse/commit/d5776fcc19ea6a8b125581b78806b25b87c8ff6b
Author: Jan Kiszka <[email protected]>
Date: 2017-02-07 (Tue, 07 Feb 2017)
Changed paths:
M hypervisor/include/jailhouse/ivshmem.h
M hypervisor/include/jailhouse/pci.h
M hypervisor/ivshmem.c
Log Message:
-----------
core: ivshmem: Move shared memory parameters into vendor specific capability
This gives us more flexibility in extending the parameters we expose via
the config space. The capability starts with its ID, the next pointer, a
one-byte length field (these parameters are mandated by the PCI spec),
then a one-byte flags field (currently RsvdZ), and finally the shared
memory region address and size. Proper specification of all this is work
in progress.
Signed-off-by: Jan Kiszka <[email protected]>
Commit: 01c6447fc9edd6a0d71b3701552a492f38227d85
https://github.com/siemens/jailhouse/commit/01c6447fc9edd6a0d71b3701552a492f38227d85
Author: Jan Kiszka <[email protected]>
Date: 2017-02-07 (Tue, 07 Feb 2017)
Changed paths:
M hypervisor/arch/arm-common/ivshmem.c
M hypervisor/arch/x86/ivshmem.c
M hypervisor/include/jailhouse/ivshmem.h
M hypervisor/ivshmem.c
Log Message:
-----------
core: ivshmem: Move INTx control into PCI config space
This prepares for unprivileged access to the MMIO register space inside
cells. It uses the flags field of the new vendor capability.
To avoid affecting the arch_ivshmem_update_intx handler from further
changes in the enabling logic, pass the new state as parameter.
Signed-off-by: Jan Kiszka <[email protected]>
Commit: 98ab740a2a7959bef09274172b477ffcc796c001
https://github.com/siemens/jailhouse/commit/98ab740a2a7959bef09274172b477ffcc796c001
Author: Jan Kiszka <[email protected]>
Date: 2017-02-07 (Tue, 07 Feb 2017)
Changed paths:
M hypervisor/ivshmem.c
Log Message:
-----------
core: ivshmem: Derive MMIO register region size from cell config
The BAR mask encodes the size of a PCI device resource. Use this to
allow a target-dependent setting via the cell configuration so that
alignment to the cell's page size become feasible. This will enable the
cell OS to map the MMIO region as a whole into user space.
Signed-off-by: Jan Kiszka <[email protected]>
Commit: c61851398334e75cc266a9ac6aec4b330c6308ec
https://github.com/siemens/jailhouse/commit/c61851398334e75cc266a9ac6aec4b330c6308ec
Author: Jan Kiszka <[email protected]>
Date: 2017-02-07 (Tue, 07 Feb 2017)
Changed paths:
M hypervisor/ivshmem.c
Log Message:
-----------
core: ivshmem: Enable unprivileged MMIO register access
Make sure that unsupported accesses to the MMIO register region do not
raise immediate panic. We should rather ignored them. This allows the
cell OS to hand out the region to unprivileged users.
Signed-off-by: Jan Kiszka <[email protected]>
Commit: 20bd8751dea38f7a6b843ff09c993d055d663325
https://github.com/siemens/jailhouse/commit/20bd8751dea38f7a6b843ff09c993d055d663325
Author: Jan Kiszka <[email protected]>
Date: 2017-02-07 (Tue, 07 Feb 2017)
Changed paths:
M configs/amd-seattle-linux-demo.c
M configs/amd-seattle.c
M configs/bananapi-linux-demo.c
M configs/bananapi.c
M configs/hikey-linux-demo.c
M configs/hikey.c
M configs/ivshmem-demo.c
M configs/jetson-tk1-linux-demo.c
M configs/jetson-tk1.c
M configs/linux-x86-demo.c
M configs/orangepi0-linux-demo.c
M configs/orangepi0.c
M configs/qemu-vm.c
M configs/zynqmp-zcu102-linux-demo-2.c
M configs/zynqmp-zcu102-linux-demo.c
M configs/zynqmp-zcu102.c
Log Message:
-----------
config: Adjust ivshmem MMIO region to 4K size
Required so that the cell OS can map the region as a whole to its user
space.
If a cell may use larger minimal page sizes, the mask for BAR 0 has to
be adjusted accordingly. For now we assume they are all on 4K.
Signed-off-by: Jan Kiszka <[email protected]>
Commit: 9d33ac25427632bcf657dbecdf29010a6bcaa62c
https://github.com/siemens/jailhouse/commit/9d33ac25427632bcf657dbecdf29010a6bcaa62c
Author: Jan Kiszka <[email protected]>
Date: 2017-02-07 (Tue, 07 Feb 2017)
Changed paths:
M hypervisor/ivshmem.c
Log Message:
-----------
core: ivshmem: Pack MMIO registers
Reorder the layout and fill gaps that were left due to unsupported
registers of the original ivshmem device. We are breaking with that
design anyway.
Signed-off-by: Jan Kiszka <[email protected]>
Commit: 69242c7441cba6293438de67e66bd8817df863ee
https://github.com/siemens/jailhouse/commit/69242c7441cba6293438de67e66bd8817df863ee
Author: Jan Kiszka <[email protected]>
Date: 2017-02-07 (Tue, 07 Feb 2017)
Changed paths:
M hypervisor/include/jailhouse/ivshmem.h
M hypervisor/ivshmem.c
Log Message:
-----------
core: ivshmem: Rename IVPOS register to ID
>From user pespective, there is no specific "position" encoded into
"ivpos". It's just a unique identifier on an ivshmem link.
Signed-off-by: Jan Kiszka <[email protected]>
Commit: 4531ad0d7068bce5c8b89d85fadc4bf44969a227
https://github.com/siemens/jailhouse/commit/4531ad0d7068bce5c8b89d85fadc4bf44969a227
Author: Jan Kiszka <[email protected]>
Date: 2017-02-07 (Tue, 07 Feb 2017)
Changed paths:
M hypervisor/ivshmem.c
Log Message:
-----------
core: ivshmem: Rename ivshmem_data to ivshmem_link
Represents more clearly what the structure is about: meta data
describing the link between two ivshmem endpoints.
Signed-off-by: Jan Kiszka <[email protected]>
Commit: b209c5b21489ae05368656b3bbd0a7a534cc5546
https://github.com/siemens/jailhouse/commit/b209c5b21489ae05368656b3bbd0a7a534cc5546
Author: Jan Kiszka <[email protected]>
Date: 2017-02-07 (Tue, 07 Feb 2017)
Changed paths:
M hypervisor/include/jailhouse/ivshmem.h
M hypervisor/ivshmem.c
Log Message:
-----------
core: ivshmem: Introduce link pointer to ivshmem_endpoint
Will be used when moving the remote lock to link level.
Signed-off-by: Jan Kiszka <[email protected]>
Commit: 023e9198eb686cbdbdf8e5015a3e89de2a3d2f2f
https://github.com/siemens/jailhouse/commit/023e9198eb686cbdbdf8e5015a3e89de2a3d2f2f
Author: Jan Kiszka <[email protected]>
Date: 2017-02-07 (Tue, 07 Feb 2017)
Changed paths:
M hypervisor/include/jailhouse/ivshmem.h
M hypervisor/ivshmem.c
Log Message:
-----------
core: ivshmem: Promote remote lock to a link level lock
This is required when synchronizing activities (see upcoming remote
state write-back) that require references to both sides: Taking both
remote locks would required lock nesting and creates deadlock risks. It
is simpler to validate that a single lock at link level is acquired and
released correctly.
Signed-off-by: Jan Kiszka <[email protected]>
Commit: 82eb974e294056f8ef00bbd1484ba74123c82b48
https://github.com/siemens/jailhouse/commit/82eb974e294056f8ef00bbd1484ba74123c82b48
Author: Jan Kiszka <[email protected]>
Date: 2017-02-07 (Tue, 07 Feb 2017)
Changed paths:
M hypervisor/arch/arm-common/ivshmem.c
M hypervisor/arch/x86/ivshmem.c
M hypervisor/arch/x86/vtd.c
M hypervisor/include/jailhouse/ivshmem.h
M hypervisor/ivshmem.c
M hypervisor/pci.c
Log Message:
-----------
core: ivshmem: Reintroduce ivshmem_update_msix
This function consolidates the check for num_msix_vectors > 0 and
ivshmem_is_msix_masked, leaving arch_ivshmem_update_msix with less work.
Signed-off-by: Jan Kiszka <[email protected]>
Commit: ed89ae135a96345bbfdbc253c657b62e1e6b0c62
https://github.com/siemens/jailhouse/commit/ed89ae135a96345bbfdbc253c657b62e1e6b0c62
Author: Jan Kiszka <[email protected]>
Date: 2017-02-07 (Tue, 07 Feb 2017)
Changed paths:
M hypervisor/arch/arm-common/ivshmem.c
M hypervisor/ivshmem.c
Log Message:
-----------
core: ivshmem: Add PCI-conforming INTx mask
Simple enough to add, and newer PCI specs demand this feature anyway:
allow to mask the INTx line via the command register.
For this purpose, factor out ivshmem_update_intx that determines the
state of the line prior to calling arch_ivshmem_update_intx. It also
skips over this call in case num_msix_vectors is non-null, offloading
this check from the arch function.
Signed-off-by: Jan Kiszka <[email protected]>
Commit: 41e6ab0fefdd53d65de2bca4e06a755b19918182
https://github.com/siemens/jailhouse/commit/41e6ab0fefdd53d65de2bca4e06a755b19918182
Author: Jan Kiszka <[email protected]>
Date: 2017-02-07 (Tue, 07 Feb 2017)
Changed paths:
M hypervisor/arch/x86/ivshmem.c
M hypervisor/ivshmem.c
Log Message:
-----------
core: ivshmem: Move arch_ivshmem_update_msix/intx under link lock
This both ensure atomicity /wrt arch_ivshmem_trigger_interrupt and
provides a barrier so that no further interrupts are in flight (provided
the hardware does not delay IPI delivery) after returning from an
interrupt masking operation.
Signed-off-by: Jan Kiszka <[email protected]>
Commit: 1f10fe3b23664bd0e38789b4900b1d94523f55d1
https://github.com/siemens/jailhouse/commit/1f10fe3b23664bd0e38789b4900b1d94523f55d1
Author: Jan Kiszka <[email protected]>
Date: 2017-02-07 (Tue, 07 Feb 2017)
Changed paths:
M hypervisor/include/jailhouse/ivshmem.h
M hypervisor/ivshmem.c
Log Message:
-----------
core: ivshmem: Implement remote state write-back to shared memory
Checking the remote state currently requires a MMIO register access
which implies a vmexit. As we only have one interrupt vector so far (and
there won't be more with INTx), this affects even the hot-path of
exchanging I/O-related events with the peer. Also when doing polling,
the check is unnecessary expensive.
Reduce the costs by writing state changes to the shared memory at a
location the peer can decide. This, of course, requires coordination
between both sides to avoid using the location also for different
purposes, but that is out of scope for the shared memory device.
Enabling or updating the write-back address triggers immediate setting
of the target memory so that the user will have the latest state
available when checking the field afterwards.
Signed-off-by: Jan Kiszka <[email protected]>
Commit: 621212999e1e5c296a42b68f02414bfd9a2cfc58
https://github.com/siemens/jailhouse/commit/621212999e1e5c296a42b68f02414bfd9a2cfc58
Author: Jan Kiszka <[email protected]>
Date: 2017-02-07 (Tue, 07 Feb 2017)
Changed paths:
M configs/amd-seattle-linux-demo.c
M configs/amd-seattle.c
M configs/bananapi-linux-demo.c
M configs/bananapi.c
M configs/hikey-linux-demo.c
M configs/hikey.c
M configs/ivshmem-demo.c
M configs/jetson-tk1-linux-demo.c
M configs/jetson-tk1.c
M configs/linux-x86-demo.c
M configs/orangepi0-linux-demo.c
M configs/orangepi0.c
M configs/qemu-vm.c
M configs/zynqmp-zcu102-linux-demo-2.c
M configs/zynqmp-zcu102-linux-demo.c
M configs/zynqmp-zcu102.c
M hypervisor/include/jailhouse/cell-config.h
M hypervisor/include/jailhouse/ivshmem.h
M hypervisor/ivshmem.c
Log Message:
-----------
core: ivshmem: Add support for unidirectional memory regions
Extend the ivshmem device to map up to 3 memory regions. Conceptually,
all regions are optional, thus can be empty but have to be listed as
such in the cell configuration. If present, the first region shall be
read/write for both sides. The second shall be read/write for the local
side and read-only for the remote. The third one shall be read-only
locally and read/writable for the remote. Thus, second and third region
need to be specified reversely in the configuration of both peers.
All three regions are reported via the vendor-specific capability of the
ivshmem device. The capability is extended accordingly.
For the shared-memory network devices, we only use the second and third
region, thus the unidirectional ones. Update the cell configurations
accordingly by splitting the existing one into two halves.
Signed-off-by: Jan Kiszka <[email protected]>
Commit: f07f6924aec9b0a8e6444d4813917fcece7a20aa
https://github.com/siemens/jailhouse/commit/f07f6924aec9b0a8e6444d4813917fcece7a20aa
Author: Jan Kiszka <[email protected]>
Date: 2017-02-07 (Tue, 07 Feb 2017)
Changed paths:
M hypervisor/ivshmem.c
Log Message:
-----------
core: ivshmem: Reindent defines
Pure styling, no functional changes.
Signed-off-by: Jan Kiszka <[email protected]>
Commit: 3fc56e77ee58f84c31a6d39407619731bc918dec
https://github.com/siemens/jailhouse/commit/3fc56e77ee58f84c31a6d39407619731bc918dec
Author: Jan Kiszka <[email protected]>
Date: 2017-02-07 (Tue, 07 Feb 2017)
Changed paths:
M hypervisor/arch/arm-common/include/asm/ivshmem.h
M hypervisor/arch/x86/include/asm/ivshmem.h
M scripts/header_check
Log Message:
-----------
core: Restrict asm/ivshmem.h to inclusion by jailhouse/ivshmem.h
The asm header will gain a dependency on the generic one and, thus,
should no longer be considered for direct inclusion. Adjust the header
check accordingly.
Signed-off-by: Jan Kiszka <[email protected]>
Commit: c7caf8cc638b68d1562840947098531e9ec4c354
https://github.com/siemens/jailhouse/commit/c7caf8cc638b68d1562840947098531e9ec4c354
Author: Jan Kiszka <[email protected]>
Date: 2017-02-07 (Tue, 07 Feb 2017)
Changed paths:
M configs/amd-seattle-linux-demo.c
M configs/amd-seattle.c
M configs/ivshmem-demo.c
M configs/linux-x86-demo.c
M configs/qemu-vm.c
M hypervisor/arch/arm-common/include/asm/ivshmem.h
M hypervisor/arch/arm-common/ivshmem.c
M hypervisor/arch/x86/include/asm/ivshmem.h
M hypervisor/arch/x86/ivshmem.c
M hypervisor/arch/x86/vtd.c
M hypervisor/include/jailhouse/ivshmem.h
M hypervisor/ivshmem.c
Log Message:
-----------
core, configs: ivshmem: Add support for up to 16 MSI-X vectors
This allows ivshmem devices with MSI-X support to expose up to 16
(PCI_EMBEDDED_MSIX_VECTS) vectors to its users. The doorbell register is
now interpreting the written value as vector number. It is possible to
restrict the number of available vectors via the cell configuration. Out
of range writes to the doorbell register will simply be ignored.
In case the cell uses less than the exposed vectors, it is now important
to correctly mask them initially in the MSI-X table as the PCI spec
demands, or the hypervisor may find invalid vectors and complain.
With the number of vectors increasing, we also need to adjust the size
of the related resource region in the cell configs accordingly.
Signed-off-by: Jan Kiszka <[email protected]>
Commit: 4777d95a48b67ae6c3c31637c0a2e370a0a700ee
https://github.com/siemens/jailhouse/commit/4777d95a48b67ae6c3c31637c0a2e370a0a700ee
Author: Jan Kiszka <[email protected]>
Date: 2017-02-07 (Tue, 07 Feb 2017)
Changed paths:
M hypervisor/include/jailhouse/pci.h
M hypervisor/ivshmem.c
Log Message:
-----------
core: ivshmem: Fold ivshmem_write_msix_control into caller
Using pci_msix_registers to model the update of the MSI-X control
register does not really simplify the code. Rather use a plain mask that
contains all modifiable bits, PCI_MSIX_CTRL_RW_MASK, and perform the
update in ivshmem_pci_cfg_write directly, analogously to the vendor
capability.
Signed-off-by: Jan Kiszka <[email protected]>
Commit: bf69ce0f1ced6e0805b3b50fb2eec73a31bf313d
https://github.com/siemens/jailhouse/commit/bf69ce0f1ced6e0805b3b50fb2eec73a31bf313d
Author: Jan Kiszka <[email protected]>
Date: 2017-02-07 (Tue, 07 Feb 2017)
Changed paths:
M configs/amd-seattle-linux-demo.c
M configs/amd-seattle.c
M configs/ivshmem-demo.c
M configs/linux-x86-demo.c
M configs/qemu-vm.c
M hypervisor/include/jailhouse/ivshmem.h
M hypervisor/ivshmem.c
Log Message:
-----------
core: ivshmem: Compress BAR usage
Move BAR 4 to BAR 2, closing the legacy gap. Pure cosmetic.
Signed-off-by: Jan Kiszka <[email protected]>
Commit: 241b300e7cb0cd81b3bade9811a514d493b96f8a
https://github.com/siemens/jailhouse/commit/241b300e7cb0cd81b3bade9811a514d493b96f8a
Author: Jan Kiszka <[email protected]>
Date: 2017-02-07 (Tue, 07 Feb 2017)
Changed paths:
A Documentation/ivshmem-v2-specification.md
Log Message:
-----------
Documentation: Add specification of IVSHMEM v2 device
Compare:
https://github.com/siemens/jailhouse/compare/943bcc6f15b0...241b300e7cb0
--
You received this message because you are subscribed to the Google Groups
"Jailhouse" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
