Message ID | 20210625091818.1047980-1-imammedo@redhat.com (mailing list archive) |
---|---|
Headers | show |
Series | acpi: refactor error prone build_header() and packed structures usage in ACPI tables | expand |
On Fri, Jun 25, 2021 at 05:17:24AM -0400, Igor Mammedov wrote: > Highlights: > * drop pointer arithmetic in ACPI tables code > * use endian agnostic API > * simplifies review of tables. /in most cases just line by line comparision with spec/ A hue amount of work, thank you! To make it easier to merge, how about splitting it up a bit? E.g. I think first 10-11 patches make sense on their own, right? > Series replaces build_header() with acpi_init_table()/acpi_table_composed() > API that hides pointer/offset arithmetic from user, to prevent > errors caused by it [1]. > While doing above, it was necessary to split table header from > packed structure that was describing the table, which is > counter-productive since it still leaves packed structure drawbacks. > So that sort of forced me to rewrite tables that were composed with > help of packed structures to preferred build_append_int_noprefix() API. > In cases where switch to build_append_int_noprefix() was small, it was > combined with acpi_init_table()/acpi_table_composed() patch. > Conversion reduced code size quite a bit despite me adding doc comments > for every table row. > > Series also includes optional qtest patches that add missing acpi > tests for tables that I'm touching to verify conversion changes. > That includes an alternative build time based impl. of > qtest_has_accel() API. So if we start bike-shedding this qtest_has_accel() > we can safely drop all tests included, till the time discussion settles > and some form of a qtest_has_accel() is merged, at which point I'd respin > depended tests. > > 1) commits > bb9feea43179 x86: acpi: use offset instead of pointer when using build_header() > 4d027afeb3a9 Virt: ACPI: fix qemu assert due to re-assigned table data address > > Link to repo: > https://gitlab.com/imammedo/qemu/-/commits/acpi_build_header_refactoring > > CC: mst@redhat.com > > Igor Mammedov (53): > tests: acpi: dump table with failed checksum > tests: qtest: add qtest_has_accel() to check if tested binary supports > accelerator > tests: acpi: whitelist expected tables for acpi/q35/numamem testcase > tests: acpi: q35: test for x2APIC entries in SRAT > tests: acpi: update expected tables blobs > tests: acpi: whitelist new expected table > tests/data/acpi/q35/DMAR.dmar > tests: acpi: add testcase for intel_iommu (DMAR table) > tests: acpi: add expected blob for DMAR table > tests: acpi: whitelist expected blobs for new acpi/q35/ivrs testcase > tests: acpi: add testcase for amd-iommu (IVRS table) > tests: acpi: update expected blobs > tests: acpi: arm/virt: drop redundant test_acpi_one() in > test_acpi_virt_tcg() > tests: acpi: whitelist expected tables for acpi/virt/iort testcase > tests: acpi: arm/virt: use kvm to test IORT table > tests: acpi: add expected IORT table blob > tests: arm-cpu-features: use qtest_has_kvm() API > tests: migration-test: use qtest_has_accel() API > tests: bios-tables-test: use qtest_has_accel() API to register TCG > only tests > acpi: add helper routines to initialize ACPI tables > acpi: build_rsdt: use acpi_init_table()/acpi_table_composed() instead > of build_header() > acpi: build_xsdt: use acpi_init_table()/acpi_table_composed() instead > of build_header() > acpi: build_slit: use acpi_init_table()/acpi_table_composed() instead > of build_header() > acpi: build_fadt: use acpi_init_table()/acpi_table_composed() instead > of build_header() > acpi: build_tpm2: use acpi_init_table()/acpi_table_composed() instead > of build_header() > acpi: acpi_build_hest: use acpi_init_table()/acpi_table_composed() > instead of build_header() > acpi: build_mcfg: use acpi_init_table()/acpi_table_composed() instead > of build_header() > acpi: build_hmat: use acpi_init_table()/acpi_table_composed() instead > of build_header() > acpi: nvdimm_build_nfit: use acpi_init_table()/acpi_table_composed() > instead of build_header() > acpi: nvdimm_build_ssdt: use acpi_init_table()/acpi_table_composed() > instead of build_header() > acpi: vmgenid_build_acpi: use acpi_init_table()/acpi_table_composed() > instead of build_header() > acpi: x86: build_dsdt: use acpi_init_table()/acpi_table_composed() > instead of build_header() > acpi: build_hpet: use acpi_init_table()/acpi_table_composed() instead > of build_header() > acpi: build_tpm_tcpa: use acpi_init_table()/acpi_table_composed() > instead of build_header() > acpi: arm/x86: build_srat: use acpi_init_table()/acpi_table_composed() > instead of build_header() > acpi: use build_append_int_noprefix() API to compose SRAT table > acpi: build_dmar_q35: use acpi_init_table()/acpi_table_composed() > instead of build_header() > acpi: build_waet: use acpi_init_table()/acpi_table_composed() instead > of build_header() > acpi: build_amd_iommu: use acpi_init_table()/acpi_table_composed() > instead of build_header() > acpi: madt: arm/x86: use acpi_init_table()/acpi_table_composed() > instead of build_header() > acpi: x86: remove dead code > acpi: x86: set enabled when composing _MAT entries > acpi: x86: madt: use build_append_int_noprefix() API to compose MADT > table > acpi: arm/virt: madt: use build_append_int_noprefix() API to compose > MADT table > acpi: build_dsdt_microvm: use acpi_init_table()/acpi_table_composed() > instead of build_header() > acpi: arm: virt: build_dsdt: use > acpi_init_table()/acpi_table_composed() instead of build_header() > acpi: arm: virt: build_iort: use > acpi_init_table()/acpi_table_composed() instead of build_header() > acpi: arm/virt: convert build_iort() to endian agnostic > build_append_FOO() API > acpi: arm/virt: build_spcr: fix invalid cast > acpi: arm/virt: build_spcr: use > acpi_init_table()/acpi_table_composed() instead of build_header() > acpi: arm/virt: build_gtdt: use > acpi_init_table()/acpi_table_composed() instead of build_header() > acpi: build_facs: use build_append_int_noprefix() API to compose table > acpi: remove no longer used build_header() > acpi: AcpiGenericAddress no longer used to map/access fields of MMIO, > drop packed attribute > > include/hw/acpi/acpi-defs.h | 528 +------------------------ > include/hw/acpi/acpi_dev_interface.h | 3 +- > include/hw/acpi/aml-build.h | 21 +- > include/hw/i386/pc.h | 7 +- > tests/qtest/libqos/libqtest.h | 8 + > hw/acpi/acpi-x86-stub.c | 3 +- > hw/acpi/aml-build.c | 188 +++++---- > hw/acpi/cpu.c | 17 +- > hw/acpi/ghes.c | 10 +- > hw/acpi/hmat.c | 14 +- > hw/acpi/nvdimm.c | 64 ++-- > hw/acpi/pci.c | 18 +- > hw/acpi/vmgenid.c | 16 +- > hw/arm/virt-acpi-build.c | 550 +++++++++++++++------------ > hw/i386/acpi-build.c | 285 +++++++------- > hw/i386/acpi-common.c | 158 ++++---- > hw/i386/acpi-microvm.c | 13 +- > meson.build | 6 + > tests/data/acpi/q35/APIC.numamem | Bin 0 -> 2686 bytes > tests/data/acpi/q35/DMAR.dmar | Bin 0 -> 80 bytes > tests/data/acpi/q35/DSDT.ivrs | Bin 0 -> 7877 bytes > tests/data/acpi/q35/DSDT.numamem | Bin 7865 -> 35222 bytes > tests/data/acpi/q35/FACP.numamem | Bin 0 -> 244 bytes > tests/data/acpi/q35/IVRS.ivrs | Bin 0 -> 104 bytes > tests/data/acpi/q35/SRAT.numamem | Bin 224 -> 5080 bytes > tests/data/acpi/virt/IORT.iort | Bin 0 -> 124 bytes > tests/qtest/acpi-utils.c | 14 + > tests/qtest/arm-cpu-features.c | 29 +- > tests/qtest/bios-tables-test.c | 85 ++++- > tests/qtest/libqtest.c | 27 ++ > tests/qtest/migration-test.c | 15 +- > 31 files changed, 873 insertions(+), 1206 deletions(-) > create mode 100644 tests/data/acpi/q35/APIC.numamem > create mode 100644 tests/data/acpi/q35/DMAR.dmar > create mode 100644 tests/data/acpi/q35/DSDT.ivrs > create mode 100644 tests/data/acpi/q35/FACP.numamem > create mode 100644 tests/data/acpi/q35/IVRS.ivrs > create mode 100644 tests/data/acpi/virt/IORT.iort > > -- > 2.27.0
On Fri, 2 Jul 2021 10:47:20 -0400 "Michael S. Tsirkin" <mst@redhat.com> wrote: > On Fri, Jun 25, 2021 at 05:17:24AM -0400, Igor Mammedov wrote: > > Highlights: > > * drop pointer arithmetic in ACPI tables code > > * use endian agnostic API > > * simplifies review of tables. /in most cases just line by line comparision with spec/ > > > A hue amount of work, thank you! > To make it easier to merge, how about splitting it up a bit? > E.g. I think first 10-11 patches make sense on their own, right? I think you've meant 01-11 patches, and answer is yes, it's in-depended of actual ACPI refactoring as was mentioned is cover letter, see below. [...] > > > > Series also includes optional qtest patches that add missing acpi > > tests for tables that I'm touching to verify conversion changes. > > That includes an alternative build time based impl. of > > qtest_has_accel() API. So if we start bike-shedding this qtest_has_accel() > > we can safely drop all tests included, till the time discussion settles > > and some form of a qtest_has_accel() is merged, at which point I'd respin > > depended tests. [...] it's ok to split tests into a separate series if that's what you prefer.
On Mon, Jul 05, 2021 at 09:56:18AM +0200, Igor Mammedov wrote: > On Fri, 2 Jul 2021 10:47:20 -0400 > "Michael S. Tsirkin" <mst@redhat.com> wrote: > > > On Fri, Jun 25, 2021 at 05:17:24AM -0400, Igor Mammedov wrote: > > > Highlights: > > > * drop pointer arithmetic in ACPI tables code > > > * use endian agnostic API > > > * simplifies review of tables. /in most cases just line by line comparision with spec/ > > > > > > A hue amount of work, thank you! > > To make it easier to merge, how about splitting it up a bit? > > E.g. I think first 10-11 patches make sense on their own, right? > > I think you've meant 01-11 patches, and answer is yes, it's in-depended > of actual ACPI refactoring as was mentioned is cover letter, see below. > > [...] > > > > > > Series also includes optional qtest patches that add missing acpi > > > tests for tables that I'm touching to verify conversion changes. > > > That includes an alternative build time based impl. of > > > qtest_has_accel() API. So if we start bike-shedding this qtest_has_accel() > > > we can safely drop all tests included, till the time discussion settles > > > and some form of a qtest_has_accel() is merged, at which point I'd respin > > > depended tests. > [...] > > it's ok to split tests into a separate series if that's what you prefer. Let's start with a smaller series that still makes sense. I'll merge that we'll look at the next chunk.
On Mon, 5 Jul 2021 14:00:51 -0400 "Michael S. Tsirkin" <mst@redhat.com> wrote: > On Mon, Jul 05, 2021 at 09:56:18AM +0200, Igor Mammedov wrote: > > On Fri, 2 Jul 2021 10:47:20 -0400 > > "Michael S. Tsirkin" <mst@redhat.com> wrote: > > > > > On Fri, Jun 25, 2021 at 05:17:24AM -0400, Igor Mammedov wrote: > > > > Highlights: > > > > * drop pointer arithmetic in ACPI tables code > > > > * use endian agnostic API > > > > * simplifies review of tables. /in most cases just line by line comparision with spec/ > > > > > > > > > A hue amount of work, thank you! > > > To make it easier to merge, how about splitting it up a bit? > > > E.g. I think first 10-11 patches make sense on their own, right? > > > > I think you've meant 01-11 patches, and answer is yes, it's in-depended > > of actual ACPI refactoring as was mentioned is cover letter, see below. > > > > [...] > > > > > > > > Series also includes optional qtest patches that add missing acpi > > > > tests for tables that I'm touching to verify conversion changes. > > > > That includes an alternative build time based impl. of > > > > qtest_has_accel() API. So if we start bike-shedding this qtest_has_accel() > > > > we can safely drop all tests included, till the time discussion settles > > > > and some form of a qtest_has_accel() is merged, at which point I'd respin > > > > depended tests. > > [...] > > > > it's ok to split tests into a separate series if that's what you prefer. > > Let's start with a smaller series that still makes sense. > I'll merge that we'll look at the next chunk. > Ok, I'll drop tests for now and respin as v2 with TPM fixes.