The PIO Setup FIS is written in the PIO:Entry state, which comes before
the ATA and ATAPI data transfer states. As a result, the PIO Setup FIS
interrupt is now raised before DMA ends for ATAPI commands, and tests have
to be adjusted.
This is also hinted by the description of the command header in the AHCI
specification, where the "A" bit is described as
When ‘1’, indicates that a PIO setup FIS shall be sent by the device
indicating a transfer for the ATAPI command.
and also by the description of the ACMD (ATAPI command region):
The ATAPI command must be either 12 or 16 bytes in length. The length
transmitted by the HBA is determined by the PIO setup FIS that is sent
by the device requesting the ATAPI command.
QEMU, which conflates the "generator" and the "receiver" of the FIS into
one device, always uses ATAPI_PACKET_SIZE, aka 12, for the length.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-id: 20180606190955.20845-3-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>
It's not always 512, and it does wind up mattering for PIO tranfers,
because this means DRQ blocks are four times as big for ATAPI.
Replace an instance of 2048 with the correct define, too.
This patch by itself winds changing no behavior. fis->count is ignored
for CMD_PACKET, and sect_count only gets used in non-ATAPI cases.
Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-id: 20180606190955.20845-2-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>
Specs are available here :
https://www.nxp.com/docs/en/application-note/AN264.pdf
This is a simple model supporting the basic registers for led and GPIO
mode. The device also supports two blinking rates but not the model
yet.
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-id: 20180530064049.27976-7-clg@kaod.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Now that we can safely call QOBJECT() on QObject * as well as its
subtypes, we can have macros qobject_ref() / qobject_unref() that work
everywhere instead of having to use QINCREF() / QDECREF() for QObject
and qobject_incref() / qobject_decref() for its subtypes.
The replacement is mechanical, except I broke a long line, and added a
cast in monitor_qmp_cleanup_req_queue_locked(). Unlike
qobject_decref(), qobject_unref() doesn't accept void *.
Note that the new macros evaluate their argument exactly once, thus no
need to shout them.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180419150145.24795-4-marcandre.lureau@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
[Rebased, semantic conflict resolved, commit message improved]
Signed-off-by: Markus Armbruster <armbru@redhat.com>
dev could be NULL if the PCI device can not be found due to some
reasons, so we must not dereference the pointer in this case.
Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-id: 1519713884-2346-1-git-send-email-thuth@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Drop one more client of global_qtest by teaching all remaining
libqos stragglers to pass in an explicit QTestState. Change the
setting of global_qtest from being implicit in libqos' call to
qtest_start() to instead be explicit in all clients that are
still relying on global_qtest.
Note that qmp_execute() can be greatly simplified in the process,
and that we also get rid of interpolation of a JSON string into a
temporary variable when qtest_qmp() can do it more reliably.
Signed-off-by: Eric Blake <eblake@redhat.com>
Acked-by: Greg Kurz <groug@kaod.org>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Drop one more client of global_qtest by teaching all ahci test
functionality to pass in an explicit QTestState. The state was
already available, so no callers had to be adjusted.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Drop one more client of global_qtest by teaching all i2c test
functionality to pass in an explicit QTestState, adjusting all
callers.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Drop one more client of global_qtest by teaching all rtas test
functionality to pass in an explicit QTestState, adjusting all
callers.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
[thuth: Use nicer indentation in rtas.h]
Signed-off-by: Thomas Huth <thuth@redhat.com>
Drop one more client of global_qtest by teaching all fw_cfg test
functionality (invoked through alloc-pc) to pass in an explicit
QTestState, adjusting all callers. In particular, fw_cfg-test
had to reorder things to create the test state prior to creating
the fw_cfg (and drop a pointless strdup in the meantime), but that
test now no longer depends on global_qtest.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
[thuth: Fixed conflict wrt pc_alloc_init() in vhost-user-test.c]
Signed-off-by: Thomas Huth <thuth@redhat.com>
When initializing a QPCIBus, track which QTestState the bus is
associated with (so that a later patch can then explicitly use
that test state for all communication on the bus, rather than
blindly relying on global_qtest). Update the initialization
functions to take another parameter, and update all callers to
pass in state (for now, most callers get away with passing the
current global_qtest as the current state, although this required
fixing the order of initialization to ensure qtest_start() is
called before qpci_init*() in rtl8139-test, and provided an
opportunity to pass in the allocator in e1000e-test).
Touch up some allocations to use g_new0() rather than g_malloc()
while in the area, and simplify some code (all implementations
of QOSOps provide a .init_allocator() that never fails).
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
[thuth: Removed hunk from vhost-user-test.c that is not required anymore,
fixed conflict in qtest_vboot() and adjusted qpci_init_pc() in sdhci-test]
Signed-off-by: Thomas Huth <thuth@redhat.com>
We still use hacks like qmp("") to wait for an event, even though we
have qmp_eventwait() since commit 8fe941f, and qmp_eventwait_ref()
since commit 7ffe312. Both commits neglected to convert all the
existing hacks. Make up what they missed.
Bonus: gets rid of empty format strings. A step towards compile-time
format string checking without triggering -Wformat-zero-length.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
[thuth: dropped the hunks from the usb tests - not needed anymore]
Signed-off-by: Thomas Huth <thuth@redhat.com>
This cleanup makes the number of objects depending on qapi/qmp/qdict.h
drop from 4550 (out of 4743) to 368 in my "build everything" tree.
For qapi/qmp/qobject.h, the number drops from 4552 to 390.
While there, separate #include from file comment with a blank line.
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20180201111846.21846-13-armbru@redhat.com>
When a 9p request is flushed (ie, cancelled) by the guest, the device
is expected to simply mark the request as used, without sending a 9p
reply (ie, without writing anything into the used buffer).
To be able to test this, we need access to the length written by the
device into the used descriptor. This patch adds a uint32_t * argument
to qvirtqueue_get_buf() and qvirtio_wait_used_elem() for this purpose.
All existing users are updated accordingly.
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
[PMD: split of some files in other commits of the same series, add libqtest.c]
Acked-by: John Snow <jsnow@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
We can use the drive_del test on s390x, too, to check that adding and
deleting also works fine with the virtio-ccw bus. But we have to make
sure that we use the devices with the "-ccw" suffix instead of the
"-pci" suffix for the virtio-ccw transport on s390x. Introduce a helper
function called qvirtio_get_dev_type() that returns the correct string
for the current architecture.
Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <1504190408-11143-1-git-send-email-thuth@redhat.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
Broken with commit b4ba67d9a7 ("libqos: Change PCI accessors to take
opaque BAR handle") a while ago, but nobody noticed since the tests are
not run by default: The msix_pba_bar is not correctly initialized
anymore if bir_pba has the same value as bir_table. With this fix,
"make check SPEED=slow" should work fine again.
Fixes: b4ba67d9a7
Tested-by: Cornelia Huck <cohuck@redhat.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
A lot of tests provide code for adding and removing a device via the
device_add and device_del QMP commands. Maintaining this code in so many
places is cumbersome and error-prone (some of the code parts check the
responses for device deletion in an incorrect way, for example, we've got
to deal with both, error code and DEVICE_DEL event here). So let's provide
some proper generic functions for adding and removing a device instead.
The code for correctly unplugging a device has been taken from a patch
from Peter Xu.
Reviewed-by: Peter Xu <peterx@redhat.com>
Tested-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Existing tests do not touch the virtqueue used ring. Instead they poll
the virtqueue ISR register and peek into their request's device-specific
status field.
It turns out that the virtqueue ISR register can be set to 1 more than
once for a single notification (see commit
83d768b564 "virtio: set ISR on dataplane
notifications"). This causes problems for tests that assume a 1:1
correspondence between the ISR being 1 and request completion.
Peeking at device-specific status fields is also problematic if the
device has no field that can be abused for EINPROGRESS polling
semantics. This is the case if all the field's values may be set by the
device; there's no magic constant left for polling.
It's time to process the used ring for completed requests, just like a
real virtio guest driver. This patch adds the necessary APIs.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Tested-by: Eric Blake <eblake@redhat.com>
Tested-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 20170628184724.21378-3-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Tested-by: Eric Blake <eblake@redhat.com>
Tested-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 20170628184724.21378-2-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Use qvirtio_pci_device_find_slot() to avoid leaking the non-hp
device. Add assert() to avoid further leaks in the future.
Use qvirtio_pci_device_free() to correctly free QVirtioPCIDevice.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Allow specifying which slot to look for the device.
This will be used in the following patch to avoid leaking when multiple
devices exists and we want to lookup the hotplug one.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Required for tray tests once a medium may have changed.
Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1478553214-497-6-git-send-email-jsnow@redhat.com
[Line length edit --js]
Signed-off-by: John Snow <jsnow@redhat.com>
(1) Add START_STOP_UNIT command to ahci-test suite
(2) Add eject/start macro commands; this is not a data transfer
command so it is not well-served by the existing generic pipeline.
Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1478553214-497-5-git-send-email-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>
Sometimes we know we'll get back an error, so let's have the
test framework understand that.
Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1478553214-497-4-git-send-email-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>
Commit 9ef2e93f introduced the concept of tagging ATAPI commands as
NONDATA, but this introduced a regression for certain commands better
described as CONDDATA. read_cd is such a command that both requires
a non-zero BCL if a transfer size is set, but is perfectly content to
accept a zero BCL if the transfer size is 0.
This test adds a regression test for the case where BCL and nb_sectors
are both 0.
Flesh out the CDROM tests by:
(1) Allowing the test to specify a BCL
(2) Allowing the buffer comparison test to compare a 0-size buffer
(3) Fix the BCL specification in libqos (It is LE, not BE)
(4) Add a nice human-readable message for future SCSI command additions
Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1477970211-25754-4-git-send-email-jsnow@redhat.com
[Line length edit --js]
Signed-off-by: John Snow <jsnow@redhat.com>
The usual use model for the libqos PCI functions is to map a specific PCI
BAR using qpci_iomap() then pass the returned token into IO accessor
functions. This, and the fact that iomap() returns a (void *) which
actually contains a PCI space address, kind of suggests that the return
value from iomap is supposed to be an opaque token.
..except that the callers expect to be able to add offsets to it. Which
also assumes the compiler will support pointer arithmetic on a (void *),
and treat it as working with byte offsets.
To clarify this situation change iomap() and the IO accessors to take
a definitely opaque BAR handle (enforced with a wrapper struct) along with
an offset within the BAR. This changes both the functions and all the
callers.
There were a number of places that checked if iomap() returned non-NULL,
and or initialized it to NULL before hand. Since iomap() already assert()s
if it fails to map the BAR, these tests were mostly pointless and are
removed.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Greg Kurz <groug@kaod.org>
In a couple of places ahci-test makes assumptions about how the tokens
returned from qpci_iomap() are formatted in ways it probably shouldn't.
First in verify_state() it uses a non-NULL token to indicate that the AHCI
device has been enabled (part of enabling is to iomap()). This changes it
to use an explicit 'enabled' flag instead.
Second, it uses the fact that the token contains a PCI address, stored when
the BAR is mapped during initialization to check that the BAR has the same
value after a migration. This changes it to explicitly read the BAR
register before and after the migration and compare.
Together, these changes will make the test more robust against changes to
the internals of the libqos PCI layer.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: John Snow <jsnow@redhat.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Currently the libqos PCI layer includes accessor helpers for 8, 16 and 32
bit reads and writes. It's likely that we'll want 64-bit accesses in the
future (plenty of modern peripherals will have 64-bit reigsters). This
adds them.
For PIO (not MMIO) accesses on the PC backend, this is implemented as two
32-bit ins or outs. That's not ideal but AFAICT x86 doesn't have 64-bit
versions of in and out.
This patch also converts the single current user of 64-bit accesses -
virtio-pci.c to use the new mechanism, rather than a sequence of 8 byte
reads.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
In the libqos PCI code we now have accessors both for registers (byte
significance preserving) and for streaming data (byte address order
preserving). These exist in both the interface for qtest drivers and in
the machine specific backends.
However, the register-style accessors aren't actually necessary in the
backend. They can be implemented in terms of the byte address order
preserving accessors by the libqos wrappers. This works because PCI is
always little endian.
This does assume that the back end byte address order preserving accessors
will perform the equivalent of a single bus transaction for short lengths.
This is the case, and in fact they currently end up using the same
cpu_physical_memory_rw() implementation within the qtest accelerator.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Currently PCI memory (aka MMIO) space is accessed via a set of readb/writeb
style accessors. This is what we want for accessing discrete registers of
a certain size. However, there are a few cases where we instead need a
"bag of bytes" style streaming interface to PCI MMIO space. This can be
either for streaming data style registers or when there's actual memory
rather than registers in PCI space, for example frame buffers or ivshmem.
This patch adds backend callbacks, and libqos wrappers for this type of
byte address order preserving accesses.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
The usual model for PCI IO with libqos is to use qpci_iomap() to map a
specific BAR for a PCI device, then perform IOs within that BAR using
qpci_io_{read,write}*().
However, certain devices also have legacy PCI IO. In this case, instead of
(or as well as) being accessed via PCI BARs, the device can be accessed
via certain well-known, fixed addresses in PCI IO space.
Two existing tests use legacy PCI IO, and take different flawed approaches
to it:
* tco-test manually constructs a tco_io_base value instead of calling
qpci_iomap(), which assumes internal knowledge of the structure of
the value it shouldn't have
* ide-test uses direct in*() and out*() calls instead of using
qpci_io_*() accessors, meaning it's not portable to non-x86 machine
types.
This patch implements a new qpci_iomap_legacy() interface which gets a
handle in the same format as qpci_iomap() but refers to a region in
the legacy PIO space. For a device which has the same registers
available both in a BAR and in legacy space (quite common), this
allows the same test code to test both options with just a different
iomap() at the beginning.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
The PCI backends in libqos each supply an iomap() and iounmap() function
which is used to set up a specified PCI BAR. But PCI BAR allocation takes
place entirely within PCI space, so doesn't really need per-backend
versions. For example, Linux includes generic BAR allocation code used on
platforms where that isn't done by firmware.
This patch merges the BAR allocation from the two existing backends into a
single simplified copy. The back ends just need to set up some parameters
describing the window of PCI IO and PCI memory addresses which are
available for allocation. Like both the existing versions the new one uses
a simple bump allocator.
Note that (again like the existing versions) this doesn't really handle
64-bit memory BARs properly. It is actually used for such a BAR by the
ivshmem test, and apparently the 32-bit MMIO BAR logic is close enough to
work, as long as the BAR isn't too big. Fixing that to properly handle
64-bit BAR allocation is a problem for another time.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
The PCI IO space (aka PIO, aka legacy IO) and PCI memory space (aka MMIO)
are distinct address spaces by the PCI spec (although parts of one might be
aliased to parts of the other in some cases).
However, qpci_io_read*() and qpci_io_write*() can perform accesses to
either space depending on parameter. That's convenient for test case
drivers, since there are a fair few devices which can be controlled via
either a PIO or MMIO BAR but with an otherwise identical driver.
This is implemented by having addresses below 64kiB treated as PIO, and
those above treated as MMIO. This works because low addresses in memory
space are generally reserved for DMA rather than MMIO.
At the moment, this demultiplexing must be handled by each PCI backend
(pc and spapr, so far). There's no real reason for this - the current
encoding is likely to work for all platforms, and even if it doesn't we
can still use a more complex common encoding since the value returned from
iomap are semi-opaque.
This patch moves the demultiplexing into the common part of the libqos PCI
code, with the backends having simpler, separate accessors for PIO and
MMIO space. This also means we have a way of explicitly accessing either
space if it's necessary for some special case.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
The 'addr' parameter to qvirtio_config_read*() doesn't have a consistent
meaning: when using the virtio-pci versions, it's a full PCI space address,
but for virtio-mmio, it's an offset from the device's base mmio address.
This means that the callers need to do different things to calculate the
addresses in the two cases, which rather defeats the purpose of function
pointer backends.
All the current users of these functions are using them to retrieve
variables from the device specific portion of the virtio config space.
So, this patch alters the semantics to always be an offset into that
device specific config area.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
but disable MSI-X tests on SPAPR as we can't check the result
(the memory region used on PC is not readable on SPAPR).
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Move the definition to libqos/virtio.h as it must be used
only with virtio functions.
Add a QVirtioDevice parameter as it will be needed to
know if the virtio device is using virtio 1.0 specification
and thus is always little-endian (to do)
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This allows to not have to pass bus and device for every virtio functions.
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Thomas Huth <thuth@redhat.com>
[dwg: Fix style nit]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
qtest_spapr_boot()/qtest_pc_boot()/qtest_boot() call qtest_vboot()
and qtest_vboot() calls g_malloc(),
and g_malloc() never fails:
if memory allocation fails, the application is terminated.
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Currently, the MMIO space for accessing PCI on pseries guests begins at
1 TiB in guest address space. Each PCI host bridge (PHB) has a 64 GiB
chunk of address space in which it places its outbound PIO and 32-bit and
64-bit MMIO windows.
This scheme as several problems:
- It limits guest RAM to 1 TiB (though we have a limited fix for this
now)
- It limits the total MMIO window to 64 GiB. This is not always enough
for some of the large nVidia GPGPU cards
- Putting all the windows into a single 64 GiB area means that naturally
aligning things within there will waste more address space.
In addition there was a miscalculation in some of the defaults, which meant
that the MMIO windows for each PHB actually slightly overran the 64 GiB
region for that PHB. We got away without nasty consequences because
the overrun fit within an unused area at the beginning of the next PHB's
region, but it's not pretty.
This patch implements a new scheme which addresses those problems, and is
also closer to what bare metal hardware and pHyp guests generally use.
Because some guest versions (including most current distro kernels) can't
access PCI MMIO above 64 TiB, we put all the PCI windows between 32 TiB and
64 TiB. This is broken into 1 TiB chunks. The first 1 TiB contains the
PIO (64 kiB) and 32-bit MMIO (2 GiB) windows for all of the PHBs. Each
subsequent TiB chunk contains a naturally aligned 64-bit MMIO window for
one PHB each.
This reduces the number of allowed PHBs (without full manual configuration
of all the windows) from 256 to 31, but this should still be plenty in
practice.
We also change some of the default window sizes for manually configured
PHBs to saner values.
Finally we adjust some tests and libqos so that it correctly uses the new
default locations. Ideally it would parse the device tree given to the
guest, but that's a more complex problem for another time.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Currently the functions in pci-spapr.c (like pci-pc.c on which it's based)
don't distinguish between 32-bit and 64-bit PCI MMIO. At the moment, the
qemu side implementation is a bit weird and has a single MMIO window
straddling 32-bit and 64-bit regions, but we're likely to change that in
future.
In any case, pci-pc.c - and therefore the testcases using PCI - only handle
32-bit MMIOs for now. For spapr despite whatever changes might happen with
the MMIO windows, the 32-bit window is likely to remain at 2..4 GiB in PCI
space.
So, explicitly limit pci-spapr.c to 32-bit MMIOs for now, we can add 64-bit
MMIO support back in when and if we need it.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
In pci-spapr.c (as in pci-pc.c from which it was derived), the
pci_hole_start/pci_hole_size and pci_iohole_start/pci_iohole_size pairs[1]
essentially define the region of PCI (not CPU) addresses in which MMIO
or PIO BARs respectively will be allocated.
The size value is relative to the start value. But in pci-spapr.c it is
set to the entire size of the window supported by the (emulated) hardware,
but the start values are *not* at the beginning of the emulated windows.
That means if you tried to map enough PCI BARs, we'd messily overrun the
IO windows, instead of failing in iomap as we should.
This patch corrects this by calculating the hole sizes from the location
of the window in PCI space and the hole start.
[1] Those are bad names, but that's a problem for another time.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
The libqos code for accessing PCI on the spapr machine type uses IOBASE()
and MMIOBASE() macros to determine the address in the CPU memory map of
the windows to PCI address space.
This is a detail of the implementation of PCI in the machine type, it's not
specified by the PAPR standard. Real guests would get the addresses of the
PCI windows from the device tree.
Finding the device tree in libqos would be awkward, but we can at least
localize this knowledge of the implementation to the init function, saving
it in the QPCIBusSPAPR structure for use by the accessors.
That leaves only one place to fix if we alter the location of the PCI
windows, as we're planning to do.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
The target endianness is not deduced anymore from
the architecture name but asked directly to the guest,
using a new qtest command: "endianness". As it can't
change (this is the value of TARGET_WORDS_BIGENDIAN),
we store it to not have to ask every time we want to
know if we have to byte-swap a value.
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
CC: Greg Kurz <groug@kaod.org>
CC: David Gibson <david@gibson.dropbear.id.au>
CC: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Machine specific shutdown function can be registered by
the machine specific qtest_XXX_boot() if needed.
So we will not have to test twice the architecture (on boot and on
shutdown) if the test can be run on several architectures.
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
