]> rtime.felk.cvut.cz Git - jailhouse.git/log
jailhouse.git
9 years agoconfigs/tools: Exclude MSI-X table from memory regions
Jan Kiszka [Sat, 16 Aug 2014 07:00:18 +0000 (09:00 +0200)]
configs/tools: Exclude MSI-X table from memory regions

We need to virtualize the access to the MSI-X table, thus we have to
prevent that the cells access them directly. As the table may be
embedded into a larger BAR, specifically followed by the PBA, split up
the BAR region as required.

Update the QEMU config accordingly - currently the only config with
MSI-X devices.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agoconfigs/tools: Provide MSI-X parameters via config
Jan Kiszka [Thu, 14 Aug 2014 17:59:19 +0000 (19:59 +0200)]
configs/tools: Provide MSI-X parameters via config

It is simpler to calculate and validate these offline than to do it
during hypervisor setup.

Add the new information to the QEMU config file.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agoconfigs/x86: Add check for VT-d existence
Jan Kiszka [Tue, 12 Aug 2014 07:40:23 +0000 (09:40 +0200)]
configs/x86: Add check for VT-d existence

Add a check that validates the existence of configured VT-d units.
Require that every config contains at least one unit. Experimental VT-d
emulation, including interrupt remapping support, is now available for
QEMU. So we can set the config fields accordingly. If VT-d is disabled
or a QEMU version without that support is used, the hypervisor continues
to ignore this error.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Perform VT-d interrupt remapping table programming
Jan Kiszka [Thu, 7 Aug 2014 08:28:41 +0000 (10:28 +0200)]
x86: Perform VT-d interrupt remapping table programming

This fills vtd_map_interrupt with the missing logic to set entries in
the interrupt remapping table and programs the hardware to use that
table, blocking all non-remappable interrupts.

vtd_map_interrupt performs the required entry validation and rejects any
improper settings. There is one exception: CPU masks in logical
destination mode will silently be adjusted instead of failing the
service call. The reason is that Linux may have these masks set for
practically inactive interrupt vectors programmed  (e.g. enabled MSI
vectors of PCIe ports without any port service using them) and will only
update them when they may be used again. Failing such settings would
stop Linux prematurely.

We reserve remapping table entries for the IOAPIC upon cell creation. We
also reserve as many entries as MSI vectors are reported for each PCI
device of a cell. IOAPIC reservation happens only once, and the region
is kept over the lifetime of the hypervisor as the IOAPIC may be shared
between cells (though this is not recommended). PCI device reservations
are released again when the device is removed from its owning cell.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Pack jailhouse_system::platform_info
Jan Kiszka [Tue, 26 Aug 2014 08:05:54 +0000 (10:05 +0200)]
core: Pack jailhouse_system::platform_info

Better safe than sorry: the packed attribute does not apply on nested
structs and unions.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Virtualize legacy MSI for interrupt remapping support
Jan Kiszka [Sun, 27 Jul 2014 12:42:56 +0000 (14:42 +0200)]
core: Virtualize legacy MSI for interrupt remapping support

Analogously to edge-triggered IOAPIC interrupts, handover all legacy
MSIs by disabling them first, programming the VT-d remapping table and
then writing remappable parameters into the MSI capability registers.
An additional triggering of active vectors ensures that we do not lose
events during handover.

Disabling is done on x86 via a trick: we program an empty CPU mask in
logical destination mode.

MSI-X remains on the to-do list. Thus, once enabling interrupt
remapping, systems that use MSI-X will become unsupported for the time
being.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Virtualize IOAPIC redir table for interrupt remapping support
Jan Kiszka [Sat, 7 Jun 2014 12:26:34 +0000 (14:26 +0200)]
x86: Virtualize IOAPIC redir table for interrupt remapping support

Handover the IOAPIC on hypervisor setup by first masking all pins, then
requesting interrupt remapping from VT-d and finally reprogramming them
according to the index that VT-d reported. If vtd_map_interrupt returns
-ENOSYS, unconditionally right now due to a missing implementation,
later on when running in QEMU, we continue to write the redirection
table entry unmodified.

As we may lose edge-triggered interrupts while they are makes, we inject
them unconditionally into the target CPU. This may cause one spurious
interrupt per handover to/from the hypervisor, but Linux can deal with
that.

Whenever a cell is added or removed, we rewrite all root cell IOAPIC
redir table entries in order to validate them regarding the target CPU
masks.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Report root cell as added/removed on initial config commit
Jan Kiszka [Fri, 8 Aug 2014 07:23:50 +0000 (09:23 +0200)]
core: Report root cell as added/removed on initial config commit

This aligns the initial config commit with those performed on non-root
cell creation and destruction. We only need to avoid some then redundant
operations on the root cell in x86's arch_config_commit and
vtd_config_commit.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Define access moderation for IOAPIC EOI register
Jan Kiszka [Fri, 8 Aug 2014 06:32:36 +0000 (08:32 +0200)]
x86: Define access moderation for IOAPIC EOI register

There is no point in virtualizating this register. Only avoid that a
cell without any business with the IOAPIC can write to it.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Factor out ioapic_reg_read/write
Jan Kiszka [Fri, 8 Aug 2014 06:31:00 +0000 (08:31 +0200)]
x86: Factor out ioapic_reg_read/write

There will be more spots where we need to talk to the IOAPIC. These
helpers enable and coordinate the access.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agoconfigs: Prepare h87i for interrupt remapping
Jan Kiszka [Thu, 7 Aug 2014 21:06:49 +0000 (23:06 +0200)]
configs: Prepare h87i for interrupt remapping

Add the missing configuration bits that will allow us to remap IOAPIC
interrupts and MSIs.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Provide stub for programming VT-d interrupt remapping table entries
Jan Kiszka [Thu, 7 Aug 2014 10:40:47 +0000 (12:40 +0200)]
x86: Provide stub for programming VT-d interrupt remapping table entries

This allows us to step-wise build up the infrastructure on the IOAPIC
and PCI subsystem side.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Factor out vtd_update_gcmd_reg
Jan Kiszka [Tue, 1 Jul 2014 16:56:50 +0000 (18:56 +0200)]
x86: Factor out vtd_update_gcmd_reg

This encapsulates changes to the global command register that preserve
the current state and wait for the state change to become effective.
Both setting and clearing is supported.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Switch VT-d to queued invalidation
Jan Kiszka [Sat, 7 Jun 2014 07:49:33 +0000 (09:49 +0200)]
x86: Switch VT-d to queued invalidation

Required for interrupt remapping support.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Add support for sending interrupts to CPUs
Jan Kiszka [Tue, 17 Jun 2014 14:19:03 +0000 (16:19 +0200)]
x86: Add support for sending interrupts to CPUs

Will be used to inject events from temporarily masked edge-triggered
sources or to generate interrupts from virtual devices.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agoconfigs/tools: Introduce device and interrupt number limit
Jan Kiszka [Thu, 7 Aug 2014 06:25:17 +0000 (08:25 +0200)]
configs/tools: Introduce device and interrupt number limit

VT-d interrupt remapping but also (one day) AMD IOMMUs require us to
dimension related tables during setup. Introduce two parameters to the
config file, one set an upper limit of interrupts (all types) that the
system may have to control (for all cells) and another one for devices
in the system. The former will be used for VT-d, the latter should once
be helpful for AMD support, thus it can remain unset in Intel configs.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agoconfigs/tools: Provide MSI parameters via config
Jan Kiszka [Mon, 28 Jul 2014 13:46:31 +0000 (15:46 +0200)]
configs/tools: Provide MSI parameters via config

This avoids having to extract these MSI capabilities from hardware in
the hypervisor.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agotools: config-create: Remove write restriction for ACPI regions
Jan Kiszka [Thu, 21 Aug 2014 17:30:39 +0000 (19:30 +0200)]
tools: config-create: Remove write restriction for ACPI regions

We don't depend on ACPI in the hypervisor anymore. Also, the NVS memory
may contain some semaphore the kernel updates to synchronize with the
chipset-embedded controller. So let's give Linux full access to this
memory again.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Fix VT-d Fault Information printing
Jan Kiszka [Sat, 23 Aug 2014 10:10:55 +0000 (12:10 +0200)]
x86: Fix VT-d Fault Information printing

This value has more than 32 bits.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agotools: config-create: Improve error reporting
Jan Kiszka [Wed, 20 Aug 2014 17:55:46 +0000 (19:55 +0200)]
tools: config-create: Improve error reporting

In case an ACPI table file is broken, report what we expected, not only
what we got (can be empty).

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agotools: config-create: Initial support for non-Intel boards
Valentine Sinitsyn [Tue, 19 Aug 2014 19:20:02 +0000 (01:20 +0600)]
tools: config-create: Initial support for non-Intel boards

DMAR is Intel-specific. For AMD, we need to parse IVRS table, but
this code is currently missing since there is no IOMMU support in
AMD-V port.

Initially, we just make DMAR optional and ignore possible errors
on AMD (Intel will still fail at signature check stage if DMAR is
missing, however the error message will be somewhat misleading).
The template was also modified so missing bits of information are
not included (and IOAPIC ID is set to zero).

Signed-off-by: Valentine Sinitsyn <valentine.sinitsyn@gmail.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agotools: config-create: Simplify parse_dmar_devscope
Jan Kiszka [Wed, 20 Aug 2014 05:46:30 +0000 (07:46 +0200)]
tools: config-create: Simplify parse_dmar_devscope

We do not support complex device scope paths yet, code this into the
parsing function. This allows to simplify the call sites as well because
parse_dmar_devscope will now read as many bytes as supported or fail.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agotools: config-create: Sort helper functions to the top
Jan Kiszka [Wed, 20 Aug 2014 05:11:36 +0000 (07:11 +0200)]
tools: config-create: Sort helper functions to the top

Separate them from all the specific parser functions. No functional
changes.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agotools: config-create: Exclude DMAR units from memory regions
Jan Kiszka [Sat, 16 Aug 2014 06:43:03 +0000 (08:43 +0200)]
tools: config-create: Exclude DMAR units from memory regions

The cells must not be able to access the physical units. We will emulate
what is required for interrupt remapping.

At this chance, clarify in a comment that the IOAPIC is also excluded.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agotools: config-create: fix shell template
Henning Schild [Wed, 13 Aug 2014 10:39:34 +0000 (12:39 +0200)]
tools: config-create: fix shell template

Not every shell supports the "-ne" operator in its builtin test, use
string compare instead. Also $UID is not always set.

Signed-off-by: Henning Schild <henning.schild@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agotools: config-create: more fine grained parsing of /proc/iomem
Henning Schild [Tue, 12 Aug 2014 14:56:58 +0000 (16:56 +0200)]
tools: config-create: more fine grained parsing of /proc/iomem

Parse /proc/iomem keeping the tree structure. So far we only looked at
the top level entries, mainly at the type strings. But for more fine
grained selection of memory regions we need to look deeper into the
tree.

Signed-off-by: Henning Schild <henning.schild@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agotools: config-create: make sure we are root when reading input files
Henning Schild [Thu, 7 Aug 2014 12:45:52 +0000 (14:45 +0200)]
tools: config-create: make sure we are root when reading input files

File access permissions allow users and root to read PCI config spaces
of devices via sysfs. But depending on your uid you will read different
content.
That is why we have to make sure we open those files as root and can not
just rely on getting an EPERM.

Signed-off-by: Henning Schild <henning.schild@siemens.com>
[Jan: removed extra blank line]
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Use unsigned counter in vtd_init_fault_nmi
Jan Kiszka [Mon, 18 Aug 2014 06:04:41 +0000 (08:04 +0200)]
x86: Use unsigned counter in vtd_init_fault_nmi

Basically just a style cleanup.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Enforce DWORD-alignment of MMCONFIG accesses
Jan Kiszka [Sat, 16 Aug 2014 09:01:46 +0000 (11:01 +0200)]
core: Enforce DWORD-alignment of MMCONFIG accesses

We only support DWORDs, thus the access address has to be DWORD-aligned
to fulfill the PCI standard - and to match our handling logic.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Rework APIC_INVALID_ID to CPU_ID_INVALID
Jan Kiszka [Wed, 20 Aug 2014 17:08:53 +0000 (19:08 +0200)]
x86: Rework APIC_INVALID_ID to CPU_ID_INVALID

APIC_INVALID_ID is actually an invalid logical CPU ID. Rename the
constant and also ensure that no CPU is registered with this ID. That
way we can drop one extra test from apic_send_ipi as this ID will never
be included in any CPU set.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Introduce and use cell_owns_cpu
Jan Kiszka [Wed, 20 Aug 2014 17:01:01 +0000 (19:01 +0200)]
core: Introduce and use cell_owns_cpu

This helper combines the check for the maximum CPU number with the
consultation of the CPU set's bitmap. Both helps to make the code more
compact and avoid to forget the former test in the future.

We keep it inline as it generates quite a bit boilerplate code
otherwise, and IPI transmission depends on efficient execution.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Prevent out-of-bounds in apic_send_logical_dest_ipi in x2APIC mode
Jan Kiszka [Wed, 20 Aug 2014 16:48:24 +0000 (18:48 +0200)]
x86: Prevent out-of-bounds in apic_send_logical_dest_ipi in x2APIC mode

We may have been given a too large APIC ID (via an invalid cluster ID).
Protect us from out-of-bounds accesses by checking against
APIC_MAX_PHYS_ID.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Check for CPU set limit in apic_send_ipi
Jan Kiszka [Wed, 20 Aug 2014 16:45:30 +0000 (18:45 +0200)]
x86: Check for CPU set limit in apic_send_ipi

Make sure we do not test for CPU IDs beyond the limit of the cell's CPU
set.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agotools: config-create: use os.path.join to construct filenames
Henning Schild [Tue, 5 Aug 2014 12:08:52 +0000 (14:08 +0200)]
tools: config-create: use os.path.join to construct filenames

Use os.path.join instead of + '/'. The script is Linux only but it is
still cleaner.

Signed-off-by: Henning Schild <henning.schild@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agotools: config-create: locate templates next to script or in given dir
Henning Schild [Tue, 5 Aug 2014 12:08:50 +0000 (14:08 +0200)]
tools: config-create: locate templates next to script or in given dir

When the script was called from outside the tools dir the templates
could not be found.
This patch assumes the templates are next to the script, one can also
specify a path.

Signed-off-by: Henning Schild <henning.schild@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Fold ioapic_root_cell_shrink into ioapic_cell_init
Jan Kiszka [Tue, 5 Aug 2014 12:12:46 +0000 (14:12 +0200)]
x86: Fold ioapic_root_cell_shrink into ioapic_cell_init

One IOAPIC interface less to care about.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Privatize IOAPIC defines
Jan Kiszka [Tue, 5 Aug 2014 12:07:20 +0000 (14:07 +0200)]
x86: Privatize IOAPIC defines

They are not part of the IOAPIC subsystem interface.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Reorder root cell init and arch_init_early
Jan Kiszka [Mon, 4 Aug 2014 14:39:38 +0000 (16:39 +0200)]
core: Reorder root cell init and arch_init_early

The latter will depend on the former while there is no dependency the
other way around.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Fold register_root_cpu into caller
Jan Kiszka [Mon, 4 Aug 2014 14:39:05 +0000 (16:39 +0200)]
core: Fold register_root_cpu into caller

Nothing to gain by keeping this in a function.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Adjust max_cpu_id calculation in cell_init
Jan Kiszka [Mon, 4 Aug 2014 14:35:20 +0000 (16:35 +0200)]
core: Adjust max_cpu_id calculation in cell_init

No need to set the limit to what we can hold in the bitmap. The limit is
better defined by the cpu set size in the config file. This not only
simplifies the code but also shortens cpu set iterations.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Enforce online CPUs == expected CPUs
Jan Kiszka [Mon, 4 Aug 2014 12:16:49 +0000 (14:16 +0200)]
core: Enforce online CPUs == expected CPUs

No longer allow the hypervisor to start with less than the number of
expected CPUs. This prevents that we leave (known) CPUs under native
control after the hypervisor started.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Use union for MSI address & data encoding/decoding
Jan Kiszka [Fri, 1 Aug 2014 15:23:26 +0000 (17:23 +0200)]
x86: Use union for MSI address & data encoding/decoding

We will work more intensively with MSIs. Encoding the address and data
word fields as a union with bit fields will simplify those tasks. For
now we can already exploit this in vtd_init_fault_nmi.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore/configs/tools: Remove ACPI support from hypervisor
Jan Kiszka [Wed, 30 Jul 2014 18:54:38 +0000 (20:54 +0200)]
core/configs/tools: Remove ACPI support from hypervisor

The is no more user of the APCI table lookup. Remove this code as well
as the config memory region in the configuration files. Update the
config generator accordingly.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Obtain DMAR unit addresses from system configuration
Jan Kiszka [Wed, 30 Jul 2014 17:27:46 +0000 (19:27 +0200)]
x86: Obtain DMAR unit addresses from system configuration

This removes the last ACPI user and simplifies the hypervisor code.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agoconfigs/tools: Describe DMAR units in config files
Jan Kiszka [Wed, 30 Jul 2014 16:15:43 +0000 (18:15 +0200)]
configs/tools: Describe DMAR units in config files

This prepares to switch from ACPI parsing to config file based DMAR unit
discovery. For simplicity reasons, we limit the number of supported DMAR
units to 8. Can be extended or made dynamic when needed.

Update the h87i config accordingly.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Obtain MMCONFIG region via system configuration
Jan Kiszka [Wed, 30 Jul 2014 16:26:50 +0000 (18:26 +0200)]
core: Obtain MMCONFIG region via system configuration

This is a step towards overcoming ACPI in the hypervisor completely.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agoconfigs/tools: Describe MMCONFIG region in config files
Jan Kiszka [Wed, 30 Jul 2014 14:11:30 +0000 (16:11 +0200)]
configs/tools: Describe MMCONFIG region in config files

This prepares to switch from ACPI parsing to config file based MMCONFIG
discovery.

Update the qemu-vm and h87i configs accordingly.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agoconfigs: Remove Celsius H700 config
Jan Kiszka [Wed, 30 Jul 2014 18:49:54 +0000 (20:49 +0200)]
configs: Remove Celsius H700 config

No longer in use, already started to rot.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agoconfigs: Require Q35 machine model for QEMU-based test setup
Jan Kiszka [Wed, 30 Jul 2014 15:15:46 +0000 (17:15 +0200)]
configs: Require Q35 machine model for QEMU-based test setup

With the introduction of config-based MMCONFIG parameters, it becomes
impossible to have one QEMU config for both its PC machines. Restrict
us to the one that will soon gain VT-d support: Q35.

Update README to reflect these requirements and changes.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agoconfigs: Adjust qemu-vm regarding ACPI region size
Jan Kiszka [Wed, 30 Jul 2014 08:46:10 +0000 (10:46 +0200)]
configs: Adjust qemu-vm regarding ACPI region size

QEMU 2.1 faced another change in the ACPI region size. Account for it.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agotools: config-create: Initialize PCIDevice.num_caps already on construction
Jan Kiszka [Mon, 28 Jul 2014 12:10:11 +0000 (14:10 +0200)]
tools: config-create: Initialize PCIDevice.num_caps already on construction

No need to postpone this until we know where the caps are located -
their number will stay the same.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Disable PCI devices on removal
Jan Kiszka [Mon, 4 Aug 2014 08:41:52 +0000 (10:41 +0200)]
core: Disable PCI devices on removal

Switch off any bus master, MMIO and PIO dispatching when removing a
device from a cell. Also try to suppress INTx signals (not all devices
may respect this).

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Introduce PCI device iterator
Jan Kiszka [Mon, 4 Aug 2014 07:43:06 +0000 (09:43 +0200)]
core: Introduce PCI device iterator

We will have more of these iterations in the future, so let's make them
more handy to use.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Remove presence checks from vtd_add/remove_pci_device
Jan Kiszka [Mon, 4 Aug 2014 08:29:49 +0000 (10:29 +0200)]
x86: Remove presence checks from vtd_add/remove_pci_device

Runtime ownership tracking at generic PCI layer level protects us from
being called redundantly.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Fix PCI device runtime ownership tracking
Jan Kiszka [Mon, 4 Aug 2014 06:29:26 +0000 (08:29 +0200)]
core: Fix PCI device runtime ownership tracking

Trigger PCI device addition and removal from the PCI core and update the
cell field in the device state in order to track active ownership. The
vtd module now only provides callbacks to update its tables when adding
or removing a device.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Introduce PCI device state
Jan Kiszka [Tue, 29 Jul 2014 18:34:24 +0000 (20:34 +0200)]
core: Introduce PCI device state

We will have to store a number of runtime state information for PCI
devices, specifically its owner. Allocate these states as an array
during cell creation and release them on cell destruction.

We can already use the structure to keep a reference to the cell the
device belongs to. This avoids having to pass this around over multiple
hops. It will also be used soon to encode runtime ownership by setting
or clearing the reference.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Only perform PCI config space writes on PCI_ACCESS_PERFORM
Jan Kiszka [Tue, 29 Jul 2014 18:45:23 +0000 (20:45 +0200)]
core: Only perform PCI config space writes on PCI_ACCESS_PERFORM

If we emulate a config space write, we may be able to skip the physical
access completely. To model this, rename PCI_ACCESS_EMULATE to
PCI_ACCESS_DONE which signals to the caller of the moderation functions
that no physical access should be performed.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Pass value directly to pci_cfg_write_moderate
Jan Kiszka [Sun, 3 Aug 2014 19:11:37 +0000 (21:11 +0200)]
core: Pass value directly to pci_cfg_write_moderate

Convert pass-by-reference to pass-by-value for the value
pci_cfg_write_moderate should handle. Reason: either we will emulate and
write in the context of the moderation, or we let the original value
pass as-is.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Introduce generic PCI config space access functions
Jan Kiszka [Mon, 28 Jul 2014 05:52:03 +0000 (07:52 +0200)]
core: Introduce generic PCI config space access functions

If available, these functions use MMIO-based access. Otherwise, they
call the arch-specific access functions.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Clean up PCI defines
Jan Kiszka [Mon, 28 Jul 2014 05:45:04 +0000 (07:45 +0200)]
x86: Clean up PCI defines

Most of them became unused. Use self-describing definitions for the
remaining ones.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Factor out arch_pci_read/write_config
Jan Kiszka [Mon, 28 Jul 2014 05:32:03 +0000 (07:32 +0200)]
x86: Factor out arch_pci_read/write_config

These PIO-based access functions will be reused for providing
hypervisor-internal config access.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Pass combined register address to pci_cfg_read/write_moderate
Jan Kiszka [Mon, 28 Jul 2014 04:47:38 +0000 (06:47 +0200)]
core: Pass combined register address to pci_cfg_read/write_moderate

Fold reg_base and reg_bias parameters of the access moderation functions
into a single address value. It simplifies most callers.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore/configs/tools: Switch PCI configuration format to single BDF value
Jan Kiszka [Sun, 27 Jul 2014 17:38:48 +0000 (19:38 +0200)]
core/configs/tools: Switch PCI configuration format to single BDF value

There is no value in splitting up the PCI device address in the config
format into bus and devfn. Fold them into a single value that can easier
be matched and is also easily be split up again via new helper macros
whenever needed.

This generates some work for locally maintained config file, sorry.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Restrict vtd_init_fault_nmi to 8-bit APIC IDs
Jan Kiszka [Fri, 1 Aug 2014 15:21:22 +0000 (17:21 +0200)]
x86: Restrict vtd_init_fault_nmi to 8-bit APIC IDs

There is more code in Jailhouse that is restricted to 8-bit IDs, so
let's simplify vtd_init_fault_nmi correspondingly.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Micro-refactor vtd_add_device_to_cell/vtd_remove_device_from_cell
Jan Kiszka [Sun, 27 Jul 2014 17:31:12 +0000 (19:31 +0200)]
x86: Micro-refactor vtd_add_device_to_cell/vtd_remove_device_from_cell

Keep the pointer of referenced root table entry low-word instead of its
value. This simplifies its manipulation later on.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Privatize PCI_CONFIG_HEADER_SIZE define
Jan Kiszka [Sun, 27 Jul 2014 17:19:22 +0000 (19:19 +0200)]
core: Privatize PCI_CONFIG_HEADER_SIZE define

No use case outside of hypervisor/pci.c

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Provide MMIO accessors for all sizes
Jan Kiszka [Sun, 27 Jul 2014 16:27:01 +0000 (18:27 +0200)]
core: Provide MMIO accessors for all sizes

Use a macro to define size-specific MMIO accessors and add support for
8 and 16 bit.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Drop useless conditions from data_port_in/out_handler
Jan Kiszka [Sun, 27 Jul 2014 15:59:26 +0000 (17:59 +0200)]
x86: Drop useless conditions from data_port_in/out_handler

Size must be 4 if it's not 1 or 2.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Fix calculation of MMCONFIG region size
Jan Kiszka [Sun, 27 Jul 2014 15:42:24 +0000 (17:42 +0200)]
core: Fix calculation of MMCONFIG region size

The PCI Firmware Specification says: "For PCI-X and PCI Express
platforms utilizing the enhanced configuration access method, the base
address of the memory mapped configuration space always corresponds to
bus number 0 (regardless of the start bus number decoded by the host
bridge) [...]." So drop the start bus from the size calculation.

Moreover, we had an off-by-one regarding end bus to size translation.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agoconfigs: Adjust MSI capability address of HDA devices
Jan Kiszka [Sun, 27 Jul 2014 06:54:06 +0000 (08:54 +0200)]
configs: Adjust MSI capability address of HDA devices

Real hardware locates MSI at 0x60, only current QEMU has this at 0x50
(patch posted).

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agoinmates: pci-demo: Clear STATESTS before triggering the MSI
Jan Kiszka [Sun, 27 Jul 2014 06:41:25 +0000 (08:41 +0200)]
inmates: pci-demo: Clear STATESTS before triggering the MSI

STATESTS may still have pending events which could prevent a MSI
delivery after controller reset. That reset doesn't clear them, so do
this explicitly.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Fix error path in arch_cell_create
Jan Kiszka [Tue, 29 Jul 2014 17:50:29 +0000 (19:50 +0200)]
x86: Fix error path in arch_cell_create

Regression of 46ab6c2f1e: Properly return the error if vtd_cell_init
fails.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agoUpdate TODO
Jan Kiszka [Sat, 26 Jul 2014 06:35:10 +0000 (08:35 +0200)]
Update TODO

Recently added APIC topics are addressed now. Also add new ideas
regarding error handling and cell watchdogs / message timeouts.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Block write access to IA32_APIC_BASE MSR
Jan Kiszka [Sun, 3 Aug 2014 17:55:31 +0000 (19:55 +0200)]
x86: Block write access to IA32_APIC_BASE MSR

The hypervisor depends on a consistent APIC mode. So prevent that a cell
can mess it up. As the APIC is kept in the same state across cell
assignments, no cell has a need to change it.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Filter LVT delivery modes
Jan Kiszka [Sat, 26 Jul 2014 06:17:46 +0000 (08:17 +0200)]
x86: Filter LVT delivery modes

Do not allow cells to program anything else than Fixed or NMI mode. NMIs
will still be swallowed by the hypervisor NMI interception path, so perf
& Co. remain broken.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Filter writes to reserved APIC register bits
Jan Kiszka [Sat, 26 Jul 2014 05:45:22 +0000 (07:45 +0200)]
x86: Filter writes to reserved APIC register bits

Set up a bitmap for all xAPIC/x2APIC register that marks reserved bits
(or complete registers). Check to-be-written values against this bitmap
before executing accesses in hypervisor context.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Move ICR access dispatching into x2apic_handle_write
Jan Kiszka [Sat, 26 Jul 2014 05:44:13 +0000 (07:44 +0200)]
x86: Move ICR access dispatching into x2apic_handle_write

More code sharing, specifically when AMD support will be added.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Refactor x2apic_handle_read/write
Jan Kiszka [Sat, 26 Jul 2014 05:42:03 +0000 (07:42 +0200)]
x86: Refactor x2apic_handle_read/write

Use the relative APIC registers number internally and prepare the
interface of x2apic_handle_write for returning access errors.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Prevent getting stuck while trying to clear the APIC
Jan Kiszka [Thu, 24 Jul 2014 17:16:59 +0000 (19:16 +0200)]
x86: Prevent getting stuck while trying to clear the APIC

If some interrupt source (typically a level-triggered IOAPIC pin)
continuously sends messages to the APIC we are trying to clear from
pending bits in ISR and IRR, we will get stuck in the hypervisor in an
interrupt storm.

Avoid this by limiting the number of handled interrupts to the number of
vectors we have. When reaching this limit, simply raise TPR to break out
of the loop. It's cleared again on exit from apic_clear, and the code
booting the CPU can handle the then pending interrupt itself. That's
almost like real hardware would behave (low-prio IRR bits may remain set
due to a stuck high-prio interrupt). However, only buggy SMP cells will
once be able to trigger this, IOAPIC pins of exitings cells will soon be
masked to prevent this scenario.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Moderate access to PCI capabilities
Jan Kiszka [Wed, 23 Jul 2014 07:19:27 +0000 (09:19 +0200)]
core: Moderate access to PCI capabilities

Make use of the capability configuration and permit write access only to
explicitly configured capabilities. Read access is harmless as it is
free of side effects.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agotools/configs: Describe PCI capabilities in config files
Jan Kiszka [Sun, 20 Jul 2014 09:45:37 +0000 (11:45 +0200)]
tools/configs: Describe PCI capabilities in config files

Instead of parsing the PCI config spaces in the hypervisor, offload this
to the configuration generator. It will produce a (logically) linked
list of capabilities per device, their ID, start and length. Each
capability can furthermore be marked as writable by the cell.

Note that identical capability lists shared between multiple devices
will automatically folded into a single one. The user can duplicate and
customize them individually in a manual post-processing step.

Configurations are updated for QEMU, the H87i and the pci-demo cell.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agotools: config-create: Sort listed directories
Jan Kiszka [Wed, 23 Jul 2014 12:04:27 +0000 (14:04 +0200)]
tools: config-create: Sort listed directories

This will specifically ensure that PCI devices will be added to the
configuration file in a sorted manner which helps with manual
post-processing.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agotools: config-create: Do not bail out on missing ACPI tables
Jan Kiszka [Mon, 21 Jul 2014 07:45:38 +0000 (09:45 +0200)]
tools: config-create: Do not bail out on missing ACPI tables

Rather warn and give the user a chance to post-process the config file.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agotools: config-create: Simplify optional file handling and generator mode
Jan Kiszka [Sun, 20 Jul 2014 22:13:15 +0000 (00:13 +0200)]
tools: config-create: Simplify optional file handling and generator mode

Stop passing exceptions from input_open to its callers: An optional file
can be returned as empty (derived from /dev/null), same on failures
during collector creation. The typical mode for input_open is 'r', so we
can save passing this explicitly in most cases. Same for optional which
is generally False.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agotools: config-create: Differentiate between optional and mandatory files during collect
Jan Kiszka [Sun, 20 Jul 2014 21:25:44 +0000 (23:25 +0200)]
tools: config-create: Differentiate between optional and mandatory files during collect

Bail out if a required file is missing but continue without irritating
warnings if an optional one cannot be found.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agotools: config-create: Cleanup typos and style in collector template
Jan Kiszka [Sun, 20 Jul 2014 21:10:46 +0000 (23:10 +0200)]
tools: config-create: Cleanup typos and style in collector template

No functional changes.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agotools: config-create: Get rid of record switch in input_open
Jan Kiszka [Sun, 20 Jul 2014 18:11:08 +0000 (20:11 +0200)]
tools: config-create: Get rid of record switch in input_open

We no longer invoke input_open if the file is part of a recorded
directory. So this switch became obsolete.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agotools: config-create: Return empty dir list when running in generator mode
Jan Kiszka [Sun, 20 Jul 2014 18:01:10 +0000 (20:01 +0200)]
tools: config-create: Return empty dir list when running in generator mode

No point in returning valid results here. The files in this directory
may not be accessible for normal users, and then the collector
generation will fail prematurely.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agotools: config-create: Account for larger sets of CPUs
Henning Schild [Thu, 24 Jul 2014 14:07:15 +0000 (16:07 +0200)]
tools: config-create: Account for larger sets of CPUs

Reserves as many cpu set words as actually required. Format the
initialization in hex to improve readability with large numbers of CPUs.

Signed-off-by: Henning Schild <henning.schild@siemens.com>
[Jan: fixed trailing whitespace]
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Remove forgotten release of pci_lock
Jan Kiszka [Mon, 28 Jul 2014 04:31:38 +0000 (06:31 +0200)]
x86: Remove forgotten release of pci_lock

Regression of 8c20eff688: lock is taken/released inside the
data_port_in/out_handlers now.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Allow writing to more PCI config space registers
Jan Kiszka [Wed, 23 Jul 2014 18:24:13 +0000 (20:24 +0200)]
core: Allow writing to more PCI config space registers

Enable write access to the Master Latency Timer (non-functional for
PCIe) and the Cacheline Size (no misuse potential identified).

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agoconfigs: Update ACPI location and size of QEMU VM
Jan Kiszka [Wed, 23 Jul 2014 17:56:02 +0000 (19:56 +0200)]
configs: Update ACPI location and size of QEMU VM

The ACPI region size grew to 16K with latest QEMU. Adjust the config.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agoconfigs: Tune memory regions of H87i
Jan Kiszka [Tue, 22 Jul 2014 19:08:13 +0000 (21:08 +0200)]
configs: Tune memory regions of H87i

ACPI was off by one page, and two regions overlapped, exposing too much
access.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agoconfigs: Enable MMCONFIG moderation for QEMU Q35 and H87i
Jan Kiszka [Tue, 22 Jul 2014 12:29:45 +0000 (14:29 +0200)]
configs: Enable MMCONFIG moderation for QEMU Q35 and H87i

Exclude the MMCONFIG regions from memory regions so that Jailhouse can
intercept accesses. Not yet converted is the H700 config.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Increase remapping region size
Jan Kiszka [Tue, 22 Jul 2014 16:57:06 +0000 (18:57 +0200)]
x86: Increase remapping region size

If we have to map the MMCONFIG space of a complete PCI section, we run
out of remapping space. 4 means 128K pages, more than enough for now.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Don't match MMIO accesses if there is no MMCONFIG space
Jan Kiszka [Tue, 22 Jul 2014 16:53:16 +0000 (18:53 +0200)]
core: Don't match MMIO accesses if there is no MMCONFIG space

Without this additional check, we will use addr >= (u64)-1 as upper
boundary if there is no MMCONFIG, which is almost always true.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Fix error handling of MMCONFIG setup
Jan Kiszka [Tue, 22 Jul 2014 16:48:19 +0000 (18:48 +0200)]
core: Fix error handling of MMCONFIG setup

If we have an MMCONFIG region, we must either successfully map it or
fail the initialization. Succeeding without setting up pci_space will
cause crashes later on when accessing it on behalf of a cell.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Enable unwinding from exception handler
Jan Kiszka [Tue, 22 Jul 2014 16:37:40 +0000 (18:37 +0200)]
x86: Enable unwinding from exception handler

Preserve the .eh_frame section for the linked hypervisor objection and
only remove it from the binary. Then add .cfi directives to the
exception entry code. This enables to use a debugger for unwinding from
the exception handler to the causing function and beyond (not perfect
due to missing stack frames, though).

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agox86: Minimize scope of pci_lock
Jan Kiszka [Mon, 21 Jul 2014 21:03:17 +0000 (23:03 +0200)]
x86: Minimize scope of pci_lock

Move PIO accesses and the related spinlock into data_port_in/out_handler
in order to reduce the lock contention time.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
9 years agocore: Rework PCI config space access handling
Jan Kiszka [Mon, 21 Jul 2014 18:22:08 +0000 (20:22 +0200)]
core: Rework PCI config space access handling

Move more logic into generic code by extending the write handler to
pci_cfg_write_moderate and introducing pci_cfg_read_moderate. These
handlers are responsible for any config space access, including to
unowned or non-existent devices. They can reject the access, return an
emulated value on read or a real value to be written to hardware, or
they instruct the caller to perform the access directly.

We already pass a reference to the issuing cell to the access handlers.
It stays unused for now but will be needed by succeeding changes. So
add it now to avoid changing API and callers once again later on.

This commit lays the foundation for capability access moderation and,
specifically, MSI/MSI-X emulation.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>