On Thu, 18 Jun 2020 at 04:50, wentongw <wentong...@intel.com> wrote: > > From: Wentong Wu <wentong...@intel.com> > > wrctl instruction on nios2 target will cause checking cpu interrupt, > but tcg_handle_interrupt() will call cpu_abort() if the CPU gets an > interrupt while it's not in a 'can do IO' state, so around wrctl > instruction add gen_io_start/end. > > Signed-off-by: Wentong Wu <wentong...@intel.com> > --- > target/nios2/translate.c | 7 +++++++ > 1 file changed, 7 insertions(+) > > diff --git a/target/nios2/translate.c b/target/nios2/translate.c > index e17656e6..deaefcaf 100644 > --- a/target/nios2/translate.c > +++ b/target/nios2/translate.c > @@ -32,6 +32,7 @@ > #include "exec/cpu_ldst.h" > #include "exec/translator.h" > #include "qemu/qemu-print.h" > +#include "exec/gen-icount.h" > > /* is_jmp field values */ > #define DISAS_JUMP DISAS_TARGET_0 /* only pc was modified dynamically */ > @@ -518,7 +519,13 @@ static void wrctl(DisasContext *dc, uint32_t code, > uint32_t flags) > /* If interrupts were enabled using WRCTL, trigger them. */ > #if !defined(CONFIG_USER_ONLY) > if ((instr.imm5 + CR_BASE) == CR_STATUS) { > + if (tb_cflags(dc->tb) & CF_USE_ICOUNT) { > + gen_io_start(); > + } > gen_helper_check_interrupts(dc->cpu_env); > + if (tb_cflags(dc->tb) & CF_USE_ICOUNT) { > + gen_io_end(); > + }
You also need to end the TB, which we should do whether we're using icount or not. This is awkward because nios2 has not correctly implemented the DISAS_UPDATE that is the right way to do that. So you'll need some preliminary patches: Patch 1: Implement and use DISAS_NORETURN + the current places in t_gen_helper_raise_exception() and gen_exception() that set dc->is_jmp = DISAS_UPDATE should instead set it to DISAS_NORETURN. (This is because they both call gen_helper_raise_exception() and at runtime execution will never return from that helper call.) + In gen_intermediate_code(), in the "switch (dc->is_jmp)" you need a new "case DISAS_NORETURN:" which goes with DISAS_TB_JUMP as another kind of "nothing more to generate". Patch 2: Make DISAS_UPDATE write PC back to CPU state + Move the "case DISAS_UPDATE:" from being like DISAS_JUMP up so it does the same thing as DISAS_NEXT (ie explicitly writes the PC back into the cpu state before doing a tcg_gen_exit_tb()). This is OK because patch 1 removed the only two previous users of it, so we don't break anything by bringing its semantics into line with how most targets use it. Patch 3: this patch to use gen_io_start/end + You can now put "dc->is_jmp = DISAS_UPDATE;" inside the "is this a CR_STATUS write" if() block, which should cause the wrctl to always end the TB. thanks -- PMM