pc_sysfw: Pass PCMachineState to pc_system_firmware_init()
pc_system_firmware_init() parameter @isapc_ram_fw is PCMachineState
member pci_enabled negated. The next commit will need more of
PCMachineState. To prepare for that, pass a PCMachineState *, and
drop the now redundant parameter @isapc_ram_fw.
vl: Create block backends before setting machine properties
qemu-system-FOO's main() acts on command line arguments in its own
idiosyncratic order. There's not much method to its madness.
Whenever we find a case where one kind of command line argument needs
to refer to something created for another kind later, we rejigger the
order.
Block devices get created long after machine properties get processed.
Therefore, block device machine properties can be created, but not
set. No such properties exist. But the next commit will create some.
Time to rejigger again: create block devices earlier.
The first call of sysbus_get_default() creates the main system bus and
stores it in QOM as "/machine/unattached/sysbus". This must not
happen before main() creates "/machine", or else container_get() would
"helpfully" create it as "container" object, and the real creation of
"/machine" would later abort with "attempt to add duplicate property
'machine' to object (type 'container')". Has been that way ever since
we wired up busses in QOM (commit f968fc6892d, v1.2.0).
I believe the bug is latent. I got it to bite by trying to
qdev_create() a sysbus device from a machine's .instance_init()
method.
The fix is obvious: store the main system bus in QOM right after
creating "/machine".
vl: Fix latent bug with -global and onboard devices
main() registers the user's -global only after we create the machine
object, i.e. too late for devices created in the machine's
.instance_init().
Fortunately, we know the bug is only latent: the commit before
previous fixed a bug that would've crashed any attempt to create a
device in an .instance_init().
qdev: Fix latent bug with compat_props and onboard devices
Compatibility properties started life as a qdev property thing: we
supported them only for qdev properties, and implemented them with the
machinery backing command line option -global.
Recent commit fa0cb34d221 put them to use (tacitly) with memory
backend objects (subtypes of TYPE_MEMORY_BACKEND). To make that
possible, we first moved the work of applying them from the -global
machinery into TYPE_DEVICE's .instance_post_init() method
device_post_init(), in commits ea9ce8934c5 and b66bbee39f6, then made
it available to TYPE_MEMORY_BACKEND's .instance_post_init() method
host_memory_backend_post_init() as object_apply_compat_props(), in
commit 1c3994f6d2a.
Note the code smell: we now have function name starting with object_
in hw/core/qdev.c. It has to be there rather than in qom/, because it
calls qdev_get_machine() to find the current accelerator's and
machine's compat_props.
Turns out calling qdev_get_machine() there is problematic. If we
qdev_create() from a machine's .instance_init() method, we call
device_post_init() and thus qdev_get_machine() before main() can
create "/machine" in QOM. qdev_get_machine() tries to get it with
container_get(), which "helpfully" creates it as "container" object,
and returns that. object_apply_compat_props() tries to paper over the
problem by doing nothing when the value of qdev_get_machine() isn't a
TYPE_MACHINE. But the damage is done already: when main() later
attempts to create the real "/machine", it fails with "attempt to add
duplicate property 'machine' to object (type 'container')", and
aborts.
Since no machine .instance_init() calls qdev_create() so far, the bug
is latent. But since I want to do that, I get to fix the bug first.
Observe that object_apply_compat_props() doesn't actually need the
MachineState, only its the compat_props member of its MachineClass and
AccelClass. This permits a simple fix: register MachineClass and
AccelClass compat_props with the object_apply_compat_props() machinery
right after these classes get selected.
This is actually similar to how things worked before commits ea9ce8934c5 and b66bbee39f6, except we now register much earlier. The
old code registered them only after the machine's .instance_init()
ran, which would've broken compatibility properties for any devices
created there.
Our pflash devices are simplistically modelled has having
"num-blocks" sectors of equal size "sector-length". Real hardware
commonly has sectors of different sizes. How our "sector-length"
property is related to the physical device's multiple sector sizes
is unclear.
Helper functions pflash_cfi01_register() and pflash_cfi02_register()
create a pflash device, set properties including "sector-length" and
"num-blocks", and realize. They take parameters @size, @sector_len
and @nb_blocs.
QOMification left parameter @size unused. Obviously, @size should
match @sector_len and @nb_blocs, i.e. size == sector_len * nb_blocs.
All callers satisfy this.
Remove @nb_blocs and compute it from @size and @sector_len.
mips_malta: Clean up definition of flash memory size somewhat
pflash_cfi01_register() takes a size in bytes, a block size in bytes
and a number of blocks. mips_malta_init() passes BIOS_SIZE, 65536,
FLASH_SIZE >> 16. Actually consistent only because BIOS_SIZE (defined
in include/hw/mips/bios.h as (4 * MiB)) matches FLASH_SIZE (defined
locally as 0x400000). Confusing all the same.
The 'bios_size' variable is only used in the 'if (!kernel_filename &&
!dinfo)' clause. This is the case when we don't provide -pflash command
line option, and also don't provide a -kernel option. In this case we
will check for the -bios option, or use the default BIOS_FILENAME file.
The 'bios' term is valid in this if statement, but is confuse in the
whole mips_malta_init() scope. Restrict his scope.
Variable fl_sectors is used just once. Since
fl_sectors = bios_size >> 16 and bios_size = FLASH_SIZE there,
we can simply use FLASH_SIZE >> 16, and eliminate variable.
The debug code under DEBUG_BOARD_INIT doesn't compile:
hw/mips/mips_malta.c:1273:16: error: implicit declaration of function ‘blk_name’; did you mean ‘basename’? [-Werror=implicit-function-declaration]
blk_name(dinfo->bdrv), fl_sectors);
^~~~~~~~
hw/mips/mips_malta.c:1273:16: error: nested extern declaration of ‘blk_name’ [-Werror=nested-externs]
hw/mips/mips_malta.c:1273:30: error: ‘DriveInfo’ {aka ‘struct DriveInfo’} has no member named ‘bdrv’
blk_name(dinfo->bdrv), fl_sectors);
^~
r2d: Fix flash memory size, sector size, width, device ID
pflash_cfi02_register() takes a size in bytes, a block size in bytes
and a number of blocks. r2d_init() passes FLASH_SIZE, 16 * KiB,
FLASH_SIZE >> 16. Does not compute: size doesn't match block size *
number of blocks. The latter happens to win: FLASH_SIZE / 4,
i.e. 8MiB.
The best information we have on the physical hardware lists a Cypress
S29PL127J60TFI130 128MiBit NOR flash addressable in words of 16 bits,
in sectors of 4 and 32 Kibiwords. We don't model multiple sector
sizes.
Fix the flash size from 8 to 16MiB, and adjust the sector size from 16
to 64KiB. Fix the width from 4 to 2. While there, supply the real
device IDs 0x0001, 0x227e, 0x2220, 0x2200 instead of zeros.
ppc405_boards: Don't size flash memory to match backing image
Machine "ref405ep" maps its flash memory at address 2^32 - image size.
Image size is rounded up to the next multiple of 64KiB. Useless,
because pflash_cfi02_realize() fails with "failed to read the initial
flash content" unless the rounding is a no-op.
If the image size exceeds 0x80000 Bytes, we overlap first SRAM, then
other stuff. No idea how that would play out, but useful outcomes
seem unlikely.
Map the flash memory at fixed address 0xFFF80000 with size 512KiB,
regardless of image size, to match the physical hardware.
Machine "taihu" maps its boot flash memory similarly. The code even
has a comment /* XXX: should check that size is 2MB */, followed by
disabled code to adjust the size to 2MiB regardless of image size.
Its code to map its application flash memory looks the same, except
there the XXX comment asks for 32MiB, and the code to adjust the size
isn't disabled. Note that pflash_cfi02_realize() fails with "failed
to read the initial flash content" for images smaller than 32MiB.
Map the boot flash memory at fixed address 0xFFE00000 with size 2MiB,
to match the physical hardware. Delete dead code from application
flash mapping, and simplify some.
The disabled DEBUG_BOARD_INIT code goes back to the initial commit 1a6c0886203, and has since seen only mechanical updates. It sure
feels like useless clutter now. Delete it.
sam460ex: Don't size flash memory to match backing image
Machine "sam460ex" maps its flash memory at address 0xFFF00000. When
no image is supplied, its size is 1MiB (0x100000), and 512KiB of ROM
get mapped on top of its second half. Else, it's the size of the
image rounded up to the next multiple of 64KiB.
The rounding is actually useless: pflash_cfi01_realize() fails with
"failed to read the initial flash content" unless it's a no-op.
I have no idea what happens when the pflash's size exceeds 1MiB.
Useful outcomes seem unlikely.
I guess memory at the end of the address space remains unmapped when
it's smaller than 1MiB. Again, useful outcomes seem unlikely.
The physical hardware appears to have 512KiB of flash memory:
https://eu.mouser.com/datasheet/2/268/atmel_AT49BV040B-1180330.pdf
For now, just set the flash memory size to 1MiB regardless of image
size, and document the mess.
hw: Use PFLASH_CFI0{1,2} and TYPE_PFLASH_CFI0{1,2}
We have two open-coded copies of macro PFLASH_CFI01(). Move the macro
to the header, so we can ditch the copies. Move PFLASH_CFI02() to the
header for symmetry.
We define macros TYPE_PFLASH_CFI01 and TYPE_PFLASH_CFI02 for type name
strings, then mostly use the strings. If the macros are worth
defining, they are worth using. Replace the strings by the macros.
pflash_cfi01.c and pflash_cfi02.c start their identifiers with
pflash_cfi01_ and pflash_cfi02_ respectively, except for
CFI_PFLASH01(), TYPE_CFI_PFLASH01, CFI_PFLASH02(), TYPE_CFI_PFLASH02.
Rename for consistency.
pflash_cfi01: Do not exit() on guest aborting "write to buffer"
When a guest tries to abort "write to buffer" (command 0xE8), we print
"PFLASH: Possible BUG - Write block confirm", then exit(1). Letting
the guest terminate QEMU is not a good idea. Instead, LOG_UNIMP we
screwed up, then reset the device.
pflash: Rename pflash_t to PFlashCFI01, PFlashCFI02
flash.h's incomplete struct pflash_t is completed both in
pflash_cfi01.c and in pflash_cfi02.c. The complete types are
incompatible. This can hide type errors, such as passing a pflash_t
created with pflash_cfi02_register() to pflash_cfi01_get_memory().
Furthermore, POSIX reserves typedef names ending with _t.
Rename the two structs to PFlashCFI01 and PFlashCFI02.
* remotes/bonzini/tags/for-upstream: (31 commits)
qemugdb: fix licensing
chardev: add support for authorization for TLS clients
qom: cpu: destroy work_mutex in cpu_common_finalize
exec.c: refactor function flatview_add_to_dispatch()
lsi: 810/895A are always little endian
lsi: return dfifo value
lsi: use SCSI phase names instead of numbers in trace
lsi: use enum type for s->msg_action
lsi: use enum type for s->waiting
lsi: use ldn_le_p()/stn_le_p()
scsi-disk: Fix crash if request is invaild or disk is no medium
configure: Disable W^X on OpenBSD
oslib-posix: Ignore fcntl("/dev/null", F_SETFL, O_NONBLOCK) failure
accel: Allow to build QEMU without TCG or KVM support
build: clean trace/generated-helpers.c
build: remove unnecessary assignments from Makefile.target
build: get rid of target-obj-y
update copyright notice
lsi: check if SIGP bit is already set in Wait reselect
lsi: implement basic SBCL functionality
...
Peter Maydell [Mon, 11 Mar 2019 17:16:38 +0000 (17:16 +0000)]
Merge remote-tracking branch 'remotes/amarkovic/tags/mips-queue-mar-11-2019' into staging
MIPS queue for March 11th, 2019
# gpg: Signature made Mon 11 Mar 2019 14:16:09 GMT
# gpg: using RSA key D4972A8967F75A65
# gpg: Good signature from "Aleksandar Markovic <[email protected]>" [unknown]
# gpg: WARNING: This key is not certified with a trusted signature!
# gpg: There is no indication that the signature belongs to the owner.
# Primary key fingerprint: 8526 FBF1 5DA3 811F 4A01 DD75 D497 2A89 67F7 5A65
* remotes/amarkovic/tags/mips-queue-mar-11-2019:
target/mips: Add tests for a variety of MSA integer subtract instructions
target/mips: Add tests for a variety of MSA integer multiply instructions
target/mips: Add tests for a variety of MSA integer dot product instructions
target/mips: Add tests for a variety of MSA integer divide instructions
target/mips: Add tests for a variety of MSA integer average instructions
tests/tcg: target/mips: Rename two header files for consistency
tests/tcg: target/mips: Correct preambles of test source files
chardev: add support for authorization for TLS clients
Currently any client which can complete the TLS handshake is able to use
a chardev server. The server admin can turn on the 'verify-peer' option
for the x509 creds to require the client to provide a x509
certificate. This means the client will have to acquire a certificate
from the CA before they are permitted to use the chardev server. This is
still a fairly low bar.
This adds a 'tls-authz=OBJECT-ID' option to the socket chardev backend
which takes the ID of a previously added 'QAuthZ' object instance. This
will be used to validate the client's x509 distinguished name. Clients
failing the check will not be permitted to use the chardev server.
For example to setup authorization that only allows connection from a
client whose x509 certificate distinguished name contains 'CN=fred', you
would use:
Li Qiang [Wed, 2 Jan 2019 07:41:14 +0000 (23:41 -0800)]
qom: cpu: destroy work_mutex in cpu_common_finalize
Commit 376692b9dc6(cpus: protect work list with work_mutex)
initialize a work_mutex in cpu_common_initfn, however forget
to destroy it. This will cause resource leak when hotunplug cpu
or hotplug cpu fails.
Wei Yang [Mon, 11 Mar 2019 05:42:52 +0000 (13:42 +0800)]
exec.c: refactor function flatview_add_to_dispatch()
flatview_add_to_dispatch() registers page based on the condition of
*section*, which may looks like this:
|s|PPPPPPP|s|
where s stands for subpage and P for page.
The procedure of this function could be described as:
- register first subpage
- register page
- register last subpage
This means the procedure could be simplified into these three steps
instead of a loop iteration.
This patch refactors the function into three corresponding steps and
adds some comment to clarify it.
Signed-off-by: Wei Yang <[email protected]>
Message-Id: <20190311054252[email protected]>
[Paolo: move exit before adjustment of remain.offset_within_*,
otherwise int128_get64 fails when a region is 2^64 bytes long] Signed-off-by: Paolo Bonzini <[email protected]>
Zhengui Li [Thu, 7 Mar 2019 09:12:46 +0000 (17:12 +0800)]
scsi-disk: Fix crash if request is invaild or disk is no medium
Qemu will crash with the assertion error that "assert(r->req.aiocb !=
NULL)" in scsi_read_complete if request is invaild or disk is no medium.
The error is below:
qemu-kvm: hw/scsi/scsi_disk.c:299: scsi_read_complete: Assertion
`r->req.aiocb != NULL' failed.
This patch add a funtion scsi_read_complete_noio to fix it.
Since OpenBSD 6.0 [1], W^X is enforced by default [2].
TCG requires WX access. Disable W^X if it is available.
This fixes:
# lm32-softmmu/qemu-system-lm32
Could not allocate dynamic translator buffer
# sysctl kern.wxabort=1
kern.wxabort: 0 -> 1
# lm32-softmmu/qemu-system-lm32
mmap: Not supported
Abort trap (core dumped)
# gdb -q lm32-softmmu/qemu-system-lm32 qemu-system-lm32.core
(gdb) bt
#0 0x000017e3c156c50a in _thread_sys___syscall () at {standard input}:5
#1 0x000017e3c15e5d7a in *_libc_mmap (addr=Variable "addr" is not available.) at /usr/src/lib/libc/sys/mmap.c:47
#2 0x000017e17d9abc8b in alloc_code_gen_buffer () at /usr/src/qemu/accel/tcg/translate-all.c:1064
#3 0x000017e17d9abd04 in code_gen_alloc (tb_size=0) at /usr/src/qemu/accel/tcg/translate-all.c:1112
#4 0x000017e17d9abe81 in tcg_exec_init (tb_size=0) at /usr/src/qemu/accel/tcg/translate-all.c:1149
#5 0x000017e17d9897e9 in tcg_init (ms=0x17e45e456800) at /usr/src/qemu/accel/tcg/tcg-all.c:66
#6 0x000017e17d9891b8 in accel_init_machine (acc=0x17e3c3f50800, ms=0x17e45e456800) at /usr/src/qemu/accel/accel.c:63
#7 0x000017e17d989312 in configure_accelerator (ms=0x17e45e456800, progname=0x7f7fffff07b0 "lm32-softmmu/qemu-system-lm32") at /usr/src/qemu/accel/accel.c:111
#8 0x000017e17d9d8616 in main (argc=1, argv=0x7f7fffff06b8, envp=0x7f7fffff06c8) at vl.c:4325
Previous to OpenBSD 6.3 [1], fcntl(F_SETFL) is not permitted on
memory devices.
Trying this call sets errno to ENODEV ("not a memory device"):
19 ENODEV Operation not supported by device.
An attempt was made to apply an inappropriate function to a device,
for example, trying to read a write-only device such as a printer.
Do not assert fcntl failures in this specific case (errno set to ENODEV)
on OpenBSD. This fixes:
$ lm32-softmmu/qemu-system-lm32
assertion "f != -1" failed: file "util/oslib-posix.c", line 247, function "qemu_set_nonblock"
Abort trap (core dumped)
[1] The fix seems https://github.com/openbsd/src/commit/c2a35b387f9d3c
"fcntl(F_SETFL) invokes the FIONBIO and FIOASYNC ioctls internally, so
the memory devices (/dev/null, /dev/zero, etc) need to permit them."
Anthony PERARD [Wed, 16 Jan 2019 17:35:27 +0000 (17:35 +0000)]
accel: Allow to build QEMU without TCG or KVM support
Instead of deny build of QEMU without a default accelerator, simply
report an error when the user haven't passed -accel or -machine accel=
and TCG and KVM isn't builtin.
./configure already check that at least one accelerator is available.
Paolo Bonzini [Fri, 15 Feb 2019 09:15:22 +0000 (10:15 +0100)]
build: remove unnecessary assignments from Makefile.target
It is only necessary to clear block-obj-y because Makefile.objs
uses "+=" instead of "="; fix that and remove the assignment.
The other variables need not be cleared at all.
Sven Schnelle [Sun, 17 Feb 2019 11:37:17 +0000 (12:37 +0100)]
lsi: check if SIGP bit is already set in Wait reselect
If SIGP is set, the 'Wait for Reselection' command should jump
immediately to the address stored in the second DWORD of the
instruction. This fixes spurious hangs in the HP-UX 11.11
installer when the SIGP bit gets set by the kernel before the
'Wait for Reselection' command is executed by SCRIPTS.
Sven Schnelle [Fri, 15 Feb 2019 19:40:21 +0000 (20:40 +0100)]
lsi: implement basic SBCL functionality
HP-UX checks this register after sending data to the target. If there's no valid
information present, it assumes the client disconnected because the kernel sent
to much data. Implement at least some of the SBCL functionality that is possible
without having a real SCSI bus.
Greg Kurz [Thu, 28 Feb 2019 17:59:42 +0000 (18:59 +0100)]
virtio-scsi: Fix build with gcc 9
Build fails with gcc 9:
CC ppc64-softmmu/hw/scsi/virtio-scsi.o
hw/scsi/virtio-scsi.c: In function ‘virtio_scsi_do_tmf’:
hw/scsi/virtio-scsi.c:265:39: error: taking address of packed member of ‘struct virtio_scsi_ctrl_tmf_req’ may result in an unaligned pointer value [-Werror=address-of-packed-member]
265 | virtio_tswap32s(VIRTIO_DEVICE(s), &req->req.tmf.subtype);
| ^~~~~~~~~~~~~~~~~~~~~
cc1: all warnings being treated as errors
All the fields in struct virtio_scsi_ctrl_tmf_req are naturally aligned,
so we could in theory drop QEMU_PACKED. Unfortunately, the header file
is imported from linux which already has the packed attribute. Trying to
fix that in the update-linux-headers.sh script is likely to produce
ugliness. Turn the call to virtio_tswap32s() into an assignment instead.
Paolo Bonzini [Thu, 28 Feb 2019 09:23:18 +0000 (10:23 +0100)]
target-i386: add kvm stubs to user-mode emulators
The CPUID code will call kvm_arch_get_supported_cpuid() and, even though
it is undef kvm_enabled() so it never runs for user-mode emulators,
sometimes clang will not optimize it out at -O0.
That could be considered a compiler bug, however at -O0 we give it
a pass and just add the stubs.
The configure script checks multiple times whether it works in a git
repository and it does this by "test -e "${source_path}/.git" in 4 cases
but in one case where it tries to enable werror "-d" is used there which
fails on git worktrees as .git is a file then and not a directory.
This changes the test to "-e" as other occurrences.
Before this patch, if elf2dmp failed to find NT kernel PE magic in
allowed virtual address range, then it assumes NULL as NT kernel
address and cause segfault.
This patch fix the problem described above by checking NT kernel address
before futher processing.
Some Linux specific code is missing guards, leading to
build failure on OSX:
$ sudo brew install libiscsi
$ ./configure && make
[...]
CC block/iscsi.o
qemu/block/iscsi.c:338:24: error: 'iscsi_aiocb_info' defined but not used [-Werror=unused-const-variable=]
static const AIOCBInfo iscsi_aiocb_info = {
^~~~~~~~~~~~~~~~
qemu/block/iscsi.c:168:1: error: 'iscsi_schedule_bh' defined but not used [-Werror=unused-function]
iscsi_schedule_bh(IscsiAIOCB *acb)
^~~~~~~~~~~~~~~~~
cc1: all warnings being treated as errors
hw/i386/pc: run the multiboot loader before the PVH loader
Some multiboot images could be in the ELF format. In the current
implementation QEMU fails because we try to load these images
as a PVH image.
In order to fix this issue, we should try multiboot first (we
already check the multiboot magic header before to load it).
If it is not a multiboot image, we can try the PVH loader.
Peter Maydell [Fri, 8 Mar 2019 13:57:44 +0000 (13:57 +0000)]
Makefile: Don't install non-sphinx files in sphinx docs install
If we're doing an out-of-tree build of Sphinx, then we
copy some extra spurious files to the install directory
as part of 'make install':
qemu-ga-qapi.texi
qemu-ga-ref.7
qemu-ga-ref.7.pod
qemu-ga-ref.html
qemu-ga-ref.txt
qemu-qmp-qapi.texi
qemu-qmp-ref.7
qemu-qmp-ref.7.pod
qemu-qmp-ref.html
qemu-qmp-ref.txt
because these have been built into build/docs/interop along
with the Sphinx interop documents. Filter them out of the
set of files we install when we're installing the Sphinx-built
manual files. (They are installed into their correct locations
as part of the main install-doc target already.)
Peter Maydell [Fri, 8 Mar 2019 13:57:43 +0000 (13:57 +0000)]
Makefile: Fix 'make distclean'
We forgot the '-r' option on the rm command to clean up the
Sphinx .doctrees working directory, which meant that
"make distclean" fails:
rm: cannot remove '.doctrees': Is a directory
Peter Maydell [Fri, 8 Mar 2019 13:57:42 +0000 (13:57 +0000)]
Makefile: Fix Sphinx documentation builds for in-tree builds
The Sphinx build-sphinx tool does not permit building a manual
into the same directory as its source files. This meant that
commit 5f71eac06e15b9a3fa1134d446f broke QEMU in-source-tree
builds, which would fail with:
Error: source directory and destination directory are same.
Fix this by making in-tree builds build the Sphinx manuals
into a subdirectory of docs/.
Combine all variant in a single handler. As source and destination
have different element sizes, we can't use gvec expansion. Expand
manually. Also watch out for overlapping source and destination
registers. Use a safe evaluation order depending on the operation.
Instead of checking e.g. the first access on every touched page, we should
check the actual access, otherwise we might get false positives when Low
Address Protection (LAP) is active. As probe_write() can only deal with
accesses to one page, we have to loop.
Use i64 for the length, although not needed - easier to reuse
TCG temps we already have in the translation functions where this will
be used. Also allow it to be used from other helpers.
This is a big one. Luckily we only have a limited set of such nasty
instructions.
We'll implement all variants with helpers, except when sources and
the destination don't overlap for VECTOR PACK. Provide different helpers
when the cc is to be modified. We'll return the cc then via env->cc_op.
We cannot use gvec expansion as source and destination elements are
have different element numbers. So we'll expand using a fancy loop.
Also, we have to take care of overlapping source and destination
registers, therefore use a safe evaluation irder depending on the
operation.
s390x/tcg: Implement VECTOR LOAD VR ELEMENT FROM GR
Very similar to VECTOR LOAD GR FROM VR ELEMENT, just the opposite
direction. Also provide a fast path in case we don't care about the
register content.
s390x/tcg: Implement VECTOR LOAD TO BLOCK BOUNDARY
Very similar to LOAD COUNT TO BLOCK BOUNDARY, but instead of only
calculating, the actual vector is loaded. Use a temporary vector to
not modify the real vector on exceptions. Initialize that one to zero,
to not leak any data. Provide a fast path if we're loading a full
vector.
As we don't have gvec ool handlers for single vectors, just calculate
the vector address manually.
We can reuse the helper later on for VECTOR LOAD WITH LENGTH. In fact,
we are going to name it "vll" right from the beginning, because that's
a better match.
Try to load the last element first. Access to the first element will
be checked afterwards. This way, we can guarantee that the vector is
not modified before we checked for all possible exceptions. (16 vectors
cannot cross more than two pages)
s390x/tcg: Implement VECTOR LOAD GR FROM VR ELEMENT
To avoid an helper, we have to do the actual calculation of the element
address (offset in cpu_env + cpu_env) manually. Factor that out into
get_vec_element_ptr_i64(). The same logic will be reused for "VECTOR
LOAD VR ELEMENT FROM GR".
Let's start with a more involved one, but it is the first in the list
of vector support instructions (introduced with the vector facility).
Good thing is, we need a lot of basic infrastructure for this. Reading
and writing vector elements as well as checking element validity.
All vector instruction related translation functions will reside in
translate_vx.inc.c, to be included in translate.c - similar to how
other architectures handle it.
While at it, directly add some documentation (which contains parts about
things added in follow-up patches, but splitting this up does not make
too much sense). Also add ES_* defines heavily used later.
projects / qemu.git / log