Re: [PATCH] target/ppc: Fix mtmsr(d) L=1 variant that loses interrupts

[Nathan Chancellor]

On Tue, Apr 14, 2020 at 09:11:31PM +1000, Nicholas Piggin wrote:
> If mtmsr L=1 sets MSR[EE] while there is a maskable exception pending,
> it does not cause an interrupt. This causes the test case to hang:
> 
> https://lists.gnu.org/archive/html/qemu-ppc/2019-10/msg00826.html
> 
> More recently, Linux reduced the occurance of operations (e.g., rfi)
> which stop translation and allow pending interrupts to be processed.
> This started causing hangs in Linux boot in long-running kernel tests,
> running with '-d int' shows the decrementer stops firing despite DEC
> wrapping and MSR[EE]=1.
> 
> https://lists.ozlabs.org/pipermail/linuxppc-dev/2020-April/208301.html
> 
> The cause is the broken mtmsr L=1 behaviour, which is contrary to the
> architecture. From Power ISA v3.0B, p.977, Move To Machine State Register,
> Programming Note states:
> 
>     If MSR[EE]=0 and an External, Decrementer, or Performance Monitor
>     exception is pending, executing an mtmsrd instruction that sets
>     MSR[EE] to 1 will cause the interrupt to occur before the next
>     instruction is executed, if no higher priority exception exists
> 
> Fix this by handling L=1 exactly the same way as L=0, modulo the MSR
> bits altered.
> 
> The confusion arises from L=0 being "context synchronizing" whereas L=1
> is "execution synchronizing", which is a weaker semantic. However this
> is not a relaxation of the requirement that these exceptions cause
> interrupts when MSR[EE]=1 (e.g., when mtmsr executes to completion as
> TCG is doing here), rather it specifies how a pipelined processor can
> have multiple instructions in flight where one may influence how another
> behaves.
> 
> Cc: address@hidden
> Reported-by: Anton Blanchard <address@hidden>
> Reported-by: Nathan Chancellor <address@hidden>
> Tested-by: Nathan Chancellor <address@hidden>
> Signed-off-by: Nicholas Piggin <address@hidden>
> ---
> Thanks very much to Nathan for reporting and testing it, I added his
> Tested-by tag despite a more polished patch, as the the basics are 
> still the same (and still fixes his test case here).

I did re-run the test with the updated version of your patch and it
passed still so that tag can still stand without any controversy :)

Thank you for the fix again!
Nathan

> This bug possibly goes back to early v2.04 / mtmsrd L=1 support around
> 2007, and the code has been changed several times since then so may
> require some backporting.
> 
> 32-bit / mtmsr untested at the moment, I don't have an environment
> handy.
> 
>  target/ppc/translate.c | 46 +++++++++++++++++++++++++-----------------
>  1 file changed, 27 insertions(+), 19 deletions(-)
> 
> diff --git a/target/ppc/translate.c b/target/ppc/translate.c
> index b207fb5386..9959259dba 100644
> --- a/target/ppc/translate.c
> +++ b/target/ppc/translate.c
> @@ -4361,30 +4361,34 @@ static void gen_mtmsrd(DisasContext *ctx)
>      CHK_SV;
>  
>  #if !defined(CONFIG_USER_ONLY)
> +    if (tb_cflags(ctx->base.tb) & CF_USE_ICOUNT) {
> +        gen_io_start();
> +    }
>      if (ctx->opcode & 0x00010000) {
> -        /* Special form that does not need any synchronisation */
> +        /* L=1 form only updates EE and RI */
>          TCGv t0 = tcg_temp_new();
> +        TCGv t1 = tcg_temp_new();
>          tcg_gen_andi_tl(t0, cpu_gpr[rS(ctx->opcode)],
>                          (1 << MSR_RI) | (1 << MSR_EE));
> -        tcg_gen_andi_tl(cpu_msr, cpu_msr,
> +        tcg_gen_andi_tl(t1, cpu_msr,
>                          ~(target_ulong)((1 << MSR_RI) | (1 << MSR_EE)));
> -        tcg_gen_or_tl(cpu_msr, cpu_msr, t0);
> +        tcg_gen_or_tl(t1, t1, t0);
> +
> +        gen_helper_store_msr(cpu_env, t1);
>          tcg_temp_free(t0);
> +        tcg_temp_free(t1);
> +
>      } else {
>          /*
>           * XXX: we need to update nip before the store if we enter
>           *      power saving mode, we will exit the loop directly from
>           *      ppc_store_msr
>           */
> -        if (tb_cflags(ctx->base.tb) & CF_USE_ICOUNT) {
> -            gen_io_start();
> -        }
>          gen_update_nip(ctx, ctx->base.pc_next);
>          gen_helper_store_msr(cpu_env, cpu_gpr[rS(ctx->opcode)]);
> -        /* Must stop the translation as machine state (may have) changed */
> -        /* Note that mtmsr is not always defined as context-synchronizing */
> -        gen_stop_exception(ctx);
>      }
> +    /* Must stop the translation as machine state (may have) changed */
> +    gen_stop_exception(ctx);
>  #endif /* !defined(CONFIG_USER_ONLY) */
>  }
>  #endif /* defined(TARGET_PPC64) */
> @@ -4394,15 +4398,23 @@ static void gen_mtmsr(DisasContext *ctx)
>      CHK_SV;
>  
>  #if !defined(CONFIG_USER_ONLY)
> -   if (ctx->opcode & 0x00010000) {
> -        /* Special form that does not need any synchronisation */
> +    if (tb_cflags(ctx->base.tb) & CF_USE_ICOUNT) {
> +        gen_io_start();
> +    }
> +    if (ctx->opcode & 0x00010000) {
> +        /* L=1 form only updates EE and RI */
>          TCGv t0 = tcg_temp_new();
> +        TCGv t1 = tcg_temp_new();
>          tcg_gen_andi_tl(t0, cpu_gpr[rS(ctx->opcode)],
>                          (1 << MSR_RI) | (1 << MSR_EE));
> -        tcg_gen_andi_tl(cpu_msr, cpu_msr,
> +        tcg_gen_andi_tl(t1, cpu_msr,
>                          ~(target_ulong)((1 << MSR_RI) | (1 << MSR_EE)));
> -        tcg_gen_or_tl(cpu_msr, cpu_msr, t0);
> +        tcg_gen_or_tl(t1, t1, t0);
> +
> +        gen_helper_store_msr(cpu_env, t1);
>          tcg_temp_free(t0);
> +        tcg_temp_free(t1);
> +
>      } else {
>          TCGv msr = tcg_temp_new();
>  
> @@ -4411,9 +4423,6 @@ static void gen_mtmsr(DisasContext *ctx)
>           *      power saving mode, we will exit the loop directly from
>           *      ppc_store_msr
>           */
> -        if (tb_cflags(ctx->base.tb) & CF_USE_ICOUNT) {
> -            gen_io_start();
> -        }
>          gen_update_nip(ctx, ctx->base.pc_next);
>  #if defined(TARGET_PPC64)
>          tcg_gen_deposit_tl(msr, cpu_msr, cpu_gpr[rS(ctx->opcode)], 0, 32);
> @@ -4422,10 +4431,9 @@ static void gen_mtmsr(DisasContext *ctx)
>  #endif
>          gen_helper_store_msr(cpu_env, msr);
>          tcg_temp_free(msr);
> -        /* Must stop the translation as machine state (may have) changed */
> -        /* Note that mtmsr is not always defined as context-synchronizing */
> -        gen_stop_exception(ctx);
>      }
> +    /* Must stop the translation as machine state (may have) changed */
> +    gen_stop_exception(ctx);
>  #endif
>  }
>  
> -- 
> 2.23.0
>