To suport >4KB page sizes, the loader may extend LOAD segment mappings
to be contiguous in the virtual address space. This is done in order to
reduce the use of unreclaimable kernel slab memory for the otherwise
necessary gap VMAs (when the runtime-page-size < ELF-segment-p_align).
Such mappings may beyond the end of the backing file when extended;
which breaks the common userspace assumption that file memory maps are
entirely backed by the underlying file.
Existing apps (not yet updated for larger page size support) may
encounter breakages if they parse /proc/self/[s]maps and use the
[start, end] addresses to operate on system libraries that have
crt_pad_segment optimization (VMA extension) [1].
In order to avoid breaking exisiting apps, update the build system to
ensure the platform ELFs' segments are entirely backed by the file
even when the VMA is extended to be contiguous with the subsequent
segment's. This is achieved using the linker flag
-z separate-loadable-segments, which inserts enough padding (zeros)
to also align each segment's offset on file by it's p_align
(max-page-size).
Although laying out the ELF segments on disk to respect the p_align
causes an increase in the file's apparent size (i_size), on Android the
actual disk usage increase is not significant due to most of the padding
being zero blocks which don't get alloacted in the read-only partitions.
The following results were obtained on an ARM64 device on a recent
git_main build:
No Separate Separate Delta Delta % Partition
Loadable Loadable
Segments Segments
4208.90MB 4214.6MB 5.70MB 0.14% All RO Partitions
Note: The overhead of -z separate-loadable-segments is minimized by the
fact that ARM64 android already builds with -z separate-code. [2]
[1] https://cs.android.com/android/platform/superproject/main/+/main:bionic/libc/arch-common/bionic/crt_pad_segment.S
[2] 733198152d:build/soong/cc/config/arm64_device.go;l=53
Bug: 328797737
Test: Manually test previously crashing application
Change-Id: Icb14ad10b5c9282855d54c7945b065b7b4184163
Signed-off-by: Kalesh Singh <kaleshsingh@google.com>
Latest compiler has merged both unaligned access flags into one.
TODO: This is needed until we change clang driver to enable -munaligned access by default.for Android b/327307773
Bug: 326790418
Change-Id: Ia8c29dc56104d2cffb8ac41aae6eeacccae68e61
Bug: http://b/323152930
Bug: http://b/324323434
The ambiguity in frameworks/av/media/codec2 is resolved.
Test: build partner branch, and presubmit
Change-Id: Ie1aa0b6c965b1a9076333d7d25fd9f9664467a31
Our mingw doesn't use this. (It does however still use
__STDC_CONSTANT_MACROS and __STDC_LIMIT_MACROS.)
Test: treehugger
Change-Id: I415017093b9ccffd21fd41f17a2f4387839a0f53
crt_pad_segment adds a NOTE to the ELF which is used by the binoic
loader to determine whether it should pad segments when mapping them
into the virtual address space, such that there are no gaps between
mappings of consecutive segments. This avoids an increase in
unreclaimable kernel slab memory usage for VMAs on devices where the
runtime-page-size > elf-segment-p_align.
Since -fandroid-pad-segment [1] respects -nostdlib used in android
platform builds, soong must link in crt_pad_segment to platform shared
libraries.
For simplicity, link crt_pad_segment everywhere that crtend_so is
applicable, ignoring nocrt property, as there is no other reason
to track these separately.
Example:
❯ readelf -WS /system/lib64/libc++.so [Output simplified]
...
Section Headers:
[Nr] Name Type Address Off Size ES Flg Lk Inf Al
...
[ 2] .note.android.pad_segment NOTE 0000000000000288 000288 000018 00 A 0 0 4
...
[1] https://github.com/llvm/llvm-project/pull/77244
Bug: 316403210
Test: readelf -WS <lib>.so
Change-Id: Icc06611376cfd5ee4de7281b4134f9f8ffe7ca60
Signed-off-by: Kalesh Singh <kaleshsingh@google.com>
Some clang unused-* warnings are not covered by "-Wno-unused". Add them
explicitly to avoid noise from these warnings, given that they don't
generally get fixed.
Test: Local clean builds.
Change-Id: I13f5089fd68f1a67a11be14fa41807688d803ba3
For chromebooks, we need a new build variant whose target arch is
goldmont, but without SHA and XSAVES support (b/314243939#comment21)
Bug: 314243939
Test: Build an image with TARGET_ARCH_VARIANT:=goldmont-without-sha-xsaves
Change-Id: If73660b515b443d5c138ca367fa3d1c6f18485b5
No-one ever fixes unused-* or deprecated-* warnings in external/ (nor should they!).
-Wno-deprecated-enum-enum-conversion should probably have been added
next to -Wno-deprecated-anon-enum-enum-conversion when it was added,
since it's a major spam contributor, and equally unlikely that anyone
will clean them up.
-Wno-deprecated-dynamic-exception-spec is a weird special case because
it should be irrelevant after our next libc++ update, but in the
meantime -- because libc++ is used everywhere -- this is a massive spam
contributor that no-one on Android cares about because we don't use
exceptions (and the people seeing this can't fix libc++ anyway).
Test: local clean builds
Change-Id: I098202337e9e5026c4c5215dbf5a2abf1fbbdca8
The previous patch was insufficient. Craig Topper explains:
-fno-vectorize only disables the loop vectorizer
-fno-slp-vectorize only disables the SLP vectorizer
The backend can also use vector instructions for memcpy/memset or
combining multiple scalar loads and/or stores. That is independent
of -fno-vectorize.
-mllvm -vector-bits-min=0 will disable any use of fixed vectors. And
will make attribute(vector_size) get scalarized.
-mno-implicit-float will disable both vectorizers and
prevent the backend for using vectors for memcpy/memset
oor multiple scalar loads/stores. It will not affect
attribute(vector_size). -mno-implicit-float also prevents scalar
floating point instructions from being used for anything that didn’t
use float/double/_Float16/etc. type in source, but I don’t think that
happens on RISC-V today. 32-bit X86 can use a 64-bit x87 FP load as an
atomic load for uint64_t, for example. Basically -mno-implicit-float
is supposed to prevent the compiler from using FP or vectors when the
source doesn’t explicitly use FP or vectors.
So -mno-implicit-float was what we were actually looking for here. I've
done a clean build with this change, and see only the expected
(hand-written assembler) vector code in bionic, and the ART
ClassLinker::LoadClass() issue is gone too. As far as I can tell, the
remaining vector code is all deliberate in that sense.
We may still end up back here again, to change "gcv" to "gc", but that
still requires some code changes just to build, and still makes it less
obvious that this is just a temporary workaround for a qemu bug
(specifically https://gitlab.com/qemu-project/qemu/-/issues/1976).
Bug: http://b/320416684
Test: objdump
Change-Id: Ibd104e4289d6d1aaf441efa0440fedc90e3da29a
* changes:
Add script to run Soong tests with go tools
Disable TestVariantSingletonModule when go test -short is used
Fix data race in propagateRROEnforcementMutator
Fix data race in finder_test.go
Fix data race in dex_bootjars
Fix race CommonGlobalCflags when running tests in parallel.
Fix data race in snapshot singletons when running parallel tests
Until https://gitlab.com/qemu-project/qemu/-/issues/1976 is fixed, this
is just causing too much confusion and wasted time.
Removing V from the architecture string has the same effect, but
requires some code changes (to be able to compile code that's
_explicitly_ using V), and makes it less obvious that this is just a
temporary workaround for a qemu bug.
Bug: http://b/320416684
Test: objdump
Change-Id: Iec7d63ab227b31f82f530dabbdc4f3aa54277f04
Clone the commonGlobalCflags global variable when modifying it so that
it gets a unique object per context.
Test: go test -race ./...
Change-Id: Ia4b91c40939d4b4d338bd11e4d4f523521874548
Its target arch is goldmont, but without xsaves support.
This ensures efficient execution on a broad range of Intel/AMD CPUs used
in Chromebooks, including those lacking xsaves support.
(e.g. Kaby Lake, Gemini Lake, Alder Lake and AMD Zen series)
Bug: 314243939
Test: Build an image with TARGET_ARCH_VARIANT:=goldmont-without-xsaves
Change-Id: I9963cc3356394815f068fc998a9708bdb30c8266
Remove the ConvertWithBp2build implementations from all the module
types, along with the related code.
Bug: 315353489
Test: m blueprint_tests
Change-Id: I212672286686a318893bc7348ddd5a5ec51e77a7
This is part of a larger change to update the flag
PRODUCT_PAGE_SIZE_AGNOSTIC to PRODUCT_NO_BIONIC_PAGE_SIZE_MACRO.
This is to help clarify that this flag doesn't mean the device
will actually work end-to-end with any page size.
Bug: 312541564
Test:
```
source build/envsetup.sh
lunch aosp_cf_arm64_phone_pgagnostic-trunk-userdebug
m
cat out/soong/build.aosp_cf_arm64_phone_pgagnostic.ninja | grep __BIONIC_NO_PAGE_SIZE_MACRO
```
Change-Id: I33033876d0d4a276d1bb962d40315b71a3968c66
By default clang assumes that on riscv64 jumps are really expensive.
That's probably not true for the kind of SoCs we're dealing with, but
more importantly (see the bug) it causes clang to do loads from the
right hand side of a `&&` before the left hand side has been evaluated.
This found one latent bug in libcore, and it doesn't seem like the best
use of anyone's time to have to chase similar issues if they're going
to be similarly latent for riscv64 when we get clang's default fixed.
Bug: https://github.com/google/android-riscv64/issues/124
Test: treehugger
Change-Id: I640f1b43ea3d2452366ab86e97a9189fa9f5326c
We should update gnu++2a to gnu++2b too, but not in this CL.
Bug: 311052584
Test: treehugger
Test: m checkbuild
Change-Id: I3637444d74868b3a29b425fd749ecbbe1a45fbf7
To simulate 16k page size in x86-64 cuttefish, it is necessary
to compile the shared libraries and binaries with max-page-size greater
than 4096.
Bug: 309816695
Test: source build/envsetup.sh
lunch aosp_cf_x86_64_phone_pgagnostic-trunk-userdebug
m
Change-Id: I32670ef2c25fdcefec11bd07ba41cd0ea96c92bd
To simulate 16k page size in x86-64 cuttefish, it is necessary
to compile the shared libraries and binaries with max-page-size greater
than 4096.
Bug: 309816695
Test: source build/envsetup.sh
lunch aosp_cf_x86_64_phone_pgagnostic-trunk-userdebug
m
readelf -l out/target/product/vsoc_x86_64_only/system/bin/toybox
Change-Id: I38e4449a7826a2eca0c1a1cccc916af21f0b3a93
The MLGO model we embed in the Android Clang is trained on ARM64
ThinLTO artefacts. Applying the model across architectures does not
necessarily translate to performance improvement. Therefore only enable
the optimisation on ARM64 ThinLTO for now.
Test: presubmit
Bug: 293827654
Change-Id: Ie64a65c11191cf700a463637a0746c8470d3aa3c
Also add -Wno-everything as suggested by Pirama
Bug: b/308179696
Follow up: https://b.corp.google.com/issues/308179380
Change-Id: Iaa73f2b52363562b2391346d8b2e5dc73bab4f2d
