target/riscv: corner case fixes
v2 cover target/riscv: corner case fixes
Changes:v1->v2:
* Added a tcg tests build-time check for vector intrinsics support in target compiler before building new tests that require it. ci images may not support these yet unfortunately, but upgrading those will be a separate effort.
There is ongoing effort to run generated test verification on the QEMU riscv CPU which has turned out a few corner cases.
I added some fixes for these, as well as tcg tests. The interrupted vector test also catches a bug in "Generate strided vector loads/stores with tcg nodes." that I referred to in the v5 thread for that series.
Thanks,
Nick
Nicholas Piggin (3):
target/riscv: Fix IALIGN check in misa write target/risvc: Fix vector whole ldst vstart check tests/tcg: Add riscv test for interrupted vector ops
target/riscv/csr.c | 16 +- target/riscv/vector_helper.c | 2 + tests/tcg/riscv64/Makefile.softmmu-target | 5 + tests/tcg/riscv64/Makefile.target | 15 ++ tests/tcg/riscv64/misa-ialign.S | 88 +++++++++ tests/tcg/riscv64/test-interrupted-v.c | 208 ++++++++++++++++++++++ tests/tcg/riscv64/test-vstart-overflow.c | 75 ++++++++ 7 files changed, 406 insertions(+), 3 deletions(-) create mode 100644 tests/tcg/riscv64/misa-ialign.S create mode 100644 tests/tcg/riscv64/test-interrupted-v.c create mode 100644 tests/tcg/riscv64/test-vstart-overflow.c
Signature · 1 line
2.51.0
PATCH v2 1/3 target/riscv: Fix IALIGN check in misa write
The instruction alignment check for the C extension was inverted. The new value should be checked for C bit clear (thus increasing IALIGN). If IALIGN is incompatible, then the write to misa should be suppressed, not just ignoring the update to the C bit.
From the ISA:
Writing misa may increase IALIGN, e.g., by disabling the "C" extension. If an instruction that would write misa increases IALIGN, and the subsequent instruction’s address is not IALIGN-bit aligned, the write to misa is suppressed, leaving misa unchanged.
This was found with a verification test generator based on RiESCUE.
--- target/riscv/csr.c | 16 ++++- tests/tcg/riscv64/Makefile.softmmu-target | 5 ++ tests/tcg/riscv64/misa-ialign.S | 88 +++++++++++++++++++++++ 3 files changed, 106 insertions(+), 3 deletions(-) create mode 100644 tests/tcg/riscv64/misa-ialign.S
target/riscv/csr.c +13 -3
@@ -2140,9 +2140,19 @@ static RISCVException write_misa(CPURISCVState *env, int csrno,/* Mask extensions that are not supported by this hart */ val &= env->misa_ext_mask; -/* Suppress 'C' if next instruction is not aligned. */ -if ((val &RVC ) && (get_next_pc (env, ra) &3 ) !=0 ) { - val &= ~RVC ; +/* +* misa writes that increase IALIGN beyond alignment of the next +* instruction cause the write to misa to be suppressed. Clearing +* "C" extension increases IALIGN. +*/ +if (!(val &RVC ) && (get_next_pc (env, ra) &3 ) !=0 ) { +/* +* If the next instruction is unaligned mod 4 then "C" must be +* set or this instruction could not be executing, so we know +* this is is clearing "C" (and not just keeping it clear). +*/ +g_assert (env->misa_ext &RVC ); +return RISCV_EXCP_NONE ; }/* Disable RVG if any of its dependencies are disabled */
tests/tcg/riscv64/Makefile.softmmu-target +5 -0
@@ -24,5 +24,10 @@ EXTRA_RUNS += run-test-mepc-masking run-test-mepc-masking: test-mepc-masking $(call run-test, $<, $(QEMU) $(QEMU_OPTS)$<) +EXTRA_RUNS += run-misa-ialign +run-misa-ialign: QEMU_OPTS := -cpu rv64,c=true,v=true,x-misa-w=on $(QEMU_OPTS) +run-misa-ialign: misa-ialign + $(call run-test, $<, $(QEMU) $(QEMU_OPTS)$<) + # We don't currently support the multiarch system tests undefine MULTIARCH_TESTS
tests/tcg/riscv64/misa-ialign.S new file +88 -0
88 lines not on this page · raw message
Signature · 1 line
2.51.0
PATCH v2 2/3 target/risvc: Fix vector whole ldst vstart check
In reply to Nicholas PigginThe whole vector ldst instructions do not include a vstart check, so an overflowed vstart can result in an underflowed memory address offset and crash:
accel/tcg/cputlb.c:1465:probe_access_flags:
assertion failed: (-(addr | TARGET_PAGE_MASK) >= size)
Add the VSTART_CHECK_EARLY_EXIT() check for these helpers.
This was found with a verification test generator based on RiESCUE.
--- target/riscv/vector_helper.c | 2 + tests/tcg/riscv64/Makefile.target | 10 ++++ tests/tcg/riscv64/test-vstart-overflow.c | 75 ++++++++++++++++++++++++ 3 files changed, 87 insertions(+) create mode 100644 tests/tcg/riscv64/test-vstart-overflow.c
target/riscv/vector_helper.c +2 -0
@@ -825,6 +825,8 @@ vext_ldst_whole(void *vd, target_ulong base, CPURISCVState *env, uint32_t desc,uint32_t esz =1 << log2_esz;int mmu_index =riscv_env_mmu_index (env, false); +VSTART_CHECK_EARLY_EXIT (env, evl); +/* Calculate the page range of first page */ addr = base + (env->vstart << log2_esz); page_split = -(addr |TARGET_PAGE_MASK );
tests/tcg/riscv64/Makefile.target +10 -0
@@ -1,6 +1,9 @@ # -*- Mode: makefile -*- # RISC-V specific tweaks +HAVE_RISCV_VECTOR_INTRINSICS := $(shell echo '#ifndef __riscv_v_intrinsic\n#error\n#endif' | \ + $(CC) -march=rv64gcv -E -x c - >/dev/null 2>&1 && echo y) + VPATH += $(SRC_PATH)/tests/tcg/riscv64 TESTS += test-div TESTS += noexec
@@ -18,3 +21,10 @@ TESTS += test-fcvtmod test-fcvtmod: CFLAGS += -march=rv64imafdc test-fcvtmod: LDFLAGS += -static run-test-fcvtmod: QEMU_OPTS += -cpu rv64,d=true,zfa=true + +ifeq ($(HAVE_RISCV_VECTOR_INTRINSICS),y) +# Test for vstart >= vl +TESTS += test-vstart-overflow +test-vstart-overflow: CFLAGS += -march=rv64gcv +run-test-vstart-overflow: QEMU_OPTS += -cpu rv64,v=on +endif
tests/tcg/riscv64/test-vstart-overflow.c new file +75 -0
75 lines not on this page · raw message
Signature · 1 line
2.51.0
PATCH v2 3/3 tests/tcg: Add riscv test for interrupted vector ops
In reply to Nicholas Pigginriscv vector instructions can be interrupted with a trap, and partial completion is recorded in the vstart register. Some causes are implementation dependent, for example an asynchronous interrupt (which I don't think TCG allows). Others are architectural, typically memory access faults on vector load/store instructions.
Add some TCG tests for interrupting vector load instructions and resuming partially completed ones.
This would have caught a recent (now reverted) regression in vector stride load implementation, commit 28c12c1f2f50d ("Generate strided vector loads/stores with tcg nodes.")
--- tests/tcg/riscv64/Makefile.target | 5 + tests/tcg/riscv64/test-interrupted-v.c | 208 +++++++++++++++++++++++++ 2 files changed, 213 insertions(+) create mode 100644 tests/tcg/riscv64/test-interrupted-v.c
tests/tcg/riscv64/Makefile.target +5 -0
@@ -27,4 +27,9 @@ ifeq ($(HAVE_RISCV_VECTOR_INTRINSICS),y) TESTS += test-vstart-overflow test-vstart-overflow: CFLAGS += -march=rv64gcv run-test-vstart-overflow: QEMU_OPTS += -cpu rv64,v=on + +# Test for interrupted vector instructions +TESTS += test-interrupted-v +test-interrupted-v: CFLAGS += -march=rv64gcv +run-test-interrupted-v: QEMU_OPTS += -cpu rv64,v=on endif
tests/tcg/riscv64/test-interrupted-v.c new file +208 -0
208 lines not on this page · raw message
Signature · 1 line
2.51.0