Re: [Qemu-devel] [PATCH v2] tcg: Really fix cpu_io_recompile

[Philippe Mathieu-Daudé]

On 03/19/2018 04:15 AM, Richard Henderson wrote:
> We have confused the number of instructions that have been
> executed in the TB with the number of instructions needed
> to repeat the I/O instruction.
> 
> We have used cpu_restore_state_from_tb, which means that
> the guest pc is pointing to the I/O instruction.  The only
> time the answer to the later question is not 1 is when
> MIPS or SH4 need to re-execute the branch for the delay
> slot as well.
> 
> We must rely on cpu->cflags_next_tb to generate the next TB,
> as otherwise we have a race condition with other guest cpus
> within the TB cache.
> 
> Fixes: 0790f86861079b1932679d0f011e431aaf4ee9e2
> Signed-off-by: Richard Henderson 

Reviewed-by: Philippe Mathieu-Daudé 

> ---
> 
> My v1 raced with Paolo's pull request, so v2 now fixes Pavel's fix.
> 
> 
> r~
> 
> ---
>  accel/tcg/translate-all.c | 37 ++---
>  1 file changed, 10 insertions(+), 27 deletions(-)
> 
> diff --git a/accel/tcg/translate-all.c b/accel/tcg/translate-all.c
> index 5ad1b919bc..d4190602d1 100644
> --- a/accel/tcg/translate-all.c
> +++ b/accel/tcg/translate-all.c
> @@ -1728,8 +1728,7 @@ void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr)
>  CPUArchState *env = cpu->env_ptr;
>  #endif
>  TranslationBlock *tb;
> -uint32_t n, flags;
> -target_ulong pc, cs_base;
> +uint32_t n;
>  
>  tb_lock();
>  tb = tb_find_pc(retaddr);
> @@ -1737,44 +1736,33 @@ void cpu_io_recompile(CPUState *cpu, uintptr_t 
> retaddr)
>  cpu_abort(cpu, "cpu_io_recompile: could not find TB for pc=%p",
>(void *)retaddr);
>  }
> -n = cpu->icount_decr.u16.low + tb->icount;
>  cpu_restore_state_from_tb(cpu, tb, retaddr);
> -/* Calculate how many instructions had been executed before the fault
> -   occurred.  */
> -n = n - cpu->icount_decr.u16.low;
> -/* Generate a new TB ending on the I/O insn.  */
> -n++;
> +
>  /* On MIPS and SH, delay slot instructions can only be restarted if
> they were already the first instruction in the TB.  If this is not
> the first instruction in a TB then re-execute the preceding
> branch.  */
> +n = 1;
>  #if defined(TARGET_MIPS)
> -if ((env->hflags & MIPS_HFLAG_BMASK) != 0 && n > 1) {
> +if ((env->hflags & MIPS_HFLAG_BMASK) != 0
> +&& env->active_tc.PC != tb->pc) {
>  env->active_tc.PC -= (env->hflags & MIPS_HFLAG_B16 ? 2 : 4);
>  cpu->icount_decr.u16.low++;
>  env->hflags &= ~MIPS_HFLAG_BMASK;
> +n = 2;
>  }
>  #elif defined(TARGET_SH4)
>  if ((env->flags & ((DELAY_SLOT | DELAY_SLOT_CONDITIONAL))) != 0
> -&& n > 1) {
> +&& env->pc != tb->pc) {
>  env->pc -= 2;
>  cpu->icount_decr.u16.low++;
>  env->flags &= ~(DELAY_SLOT | DELAY_SLOT_CONDITIONAL);
> +n = 2;
>  }
>  #endif
> -/* This should never happen.  */
> -if (n > CF_COUNT_MASK) {
> -cpu_abort(cpu, "TB too big during recompile");
> -}
>  
> -pc = tb->pc;
> -cs_base = tb->cs_base;
> -flags = tb->flags;
> -tb_phys_invalidate(tb, -1);
> -
> -/* Execute one IO instruction without caching
> -   instead of creating large TB. */
> -cpu->cflags_next_tb = curr_cflags() | CF_LAST_IO | CF_NOCACHE | 1;
> +/* Generate a new TB executing the I/O insn.  */
> +cpu->cflags_next_tb = curr_cflags() | CF_LAST_IO | n;
>  
>  if (tb->cflags & CF_NOCACHE) {
>  if (tb->orig_tb) {
> @@ -1785,11 +1773,6 @@ void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr)
>  tb_remove(tb);
>  }
>  
> -/* Generate new TB instead of the current one. */
> -/* FIXME: In theory this could raise an exception.  In practice
> -   we have already translated the block once so it's probably ok.  */
> -tb_gen_code(cpu, pc, cs_base, flags, curr_cflags() | CF_LAST_IO | n);
> -
>  /* TODO: If env->pc != tb->pc (i.e. the faulting instruction was not
>   * the first in the TB) then we end up generating a whole new TB and
>   *  repeating the fault, which is horribly inefficient.
> 

Re: [Qemu-devel] [PATCH v2] tcg: Really fix cpu_io_recompile

[Richard Henderson]

On 03/19/2018 11:54 PM, Paolo Bonzini wrote:
> On 19/03/2018 04:15, Richard Henderson wrote:
>> We have confused the number of instructions that have been
>> executed in the TB with the number of instructions needed
>> to repeat the I/O instruction.
>>
>> We have used cpu_restore_state_from_tb, which means that
>> the guest pc is pointing to the I/O instruction.  The only
>> time the answer to the later question is not 1 is when
>> MIPS or SH4 need to re-execute the branch for the delay
>> slot as well.
>>
>> We must rely on cpu->cflags_next_tb to generate the next TB,
>> as otherwise we have a race condition with other guest cpus
>> within the TB cache.
>>
>> Fixes: 0790f86861079b1932679d0f011e431aaf4ee9e2
>> Signed-off-by: Richard Henderson 
>> ---
>>
>> My v1 raced with Paolo's pull request, so v2 now fixes Pavel's fix.
> 
> Thanks, let me know if you prefer to send a pull request yourself, or if
> I should include it in the next.

I'm at Linaro Connect this week.  Please include this in your next.


r~

Re: [Qemu-devel] [PATCH v2] tcg: Really fix cpu_io_recompile

[Paolo Bonzini]

On 19/03/2018 04:15, Richard Henderson wrote:
> We have confused the number of instructions that have been
> executed in the TB with the number of instructions needed
> to repeat the I/O instruction.
> 
> We have used cpu_restore_state_from_tb, which means that
> the guest pc is pointing to the I/O instruction.  The only
> time the answer to the later question is not 1 is when
> MIPS or SH4 need to re-execute the branch for the delay
> slot as well.
> 
> We must rely on cpu->cflags_next_tb to generate the next TB,
> as otherwise we have a race condition with other guest cpus
> within the TB cache.
> 
> Fixes: 0790f86861079b1932679d0f011e431aaf4ee9e2
> Signed-off-by: Richard Henderson 
> ---
> 
> My v1 raced with Paolo's pull request, so v2 now fixes Pavel's fix.

Thanks, let me know if you prefer to send a pull request yourself, or if
I should include it in the next.

Thanks,

Paolo

> 
> 
> r~
> 
> ---
>  accel/tcg/translate-all.c | 37 ++---
>  1 file changed, 10 insertions(+), 27 deletions(-)
> 
> diff --git a/accel/tcg/translate-all.c b/accel/tcg/translate-all.c
> index 5ad1b919bc..d4190602d1 100644
> --- a/accel/tcg/translate-all.c
> +++ b/accel/tcg/translate-all.c
> @@ -1728,8 +1728,7 @@ void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr)
>  CPUArchState *env = cpu->env_ptr;
>  #endif
>  TranslationBlock *tb;
> -uint32_t n, flags;
> -target_ulong pc, cs_base;
> +uint32_t n;
>  
>  tb_lock();
>  tb = tb_find_pc(retaddr);
> @@ -1737,44 +1736,33 @@ void cpu_io_recompile(CPUState *cpu, uintptr_t 
> retaddr)
>  cpu_abort(cpu, "cpu_io_recompile: could not find TB for pc=%p",
>(void *)retaddr);
>  }
> -n = cpu->icount_decr.u16.low + tb->icount;
>  cpu_restore_state_from_tb(cpu, tb, retaddr);
> -/* Calculate how many instructions had been executed before the fault
> -   occurred.  */
> -n = n - cpu->icount_decr.u16.low;
> -/* Generate a new TB ending on the I/O insn.  */
> -n++;
> +
>  /* On MIPS and SH, delay slot instructions can only be restarted if
> they were already the first instruction in the TB.  If this is not
> the first instruction in a TB then re-execute the preceding
> branch.  */
> +n = 1;
>  #if defined(TARGET_MIPS)
> -if ((env->hflags & MIPS_HFLAG_BMASK) != 0 && n > 1) {
> +if ((env->hflags & MIPS_HFLAG_BMASK) != 0
> +&& env->active_tc.PC != tb->pc) {
>  env->active_tc.PC -= (env->hflags & MIPS_HFLAG_B16 ? 2 : 4);
>  cpu->icount_decr.u16.low++;
>  env->hflags &= ~MIPS_HFLAG_BMASK;
> +n = 2;
>  }
>  #elif defined(TARGET_SH4)
>  if ((env->flags & ((DELAY_SLOT | DELAY_SLOT_CONDITIONAL))) != 0
> -&& n > 1) {
> +&& env->pc != tb->pc) {
>  env->pc -= 2;
>  cpu->icount_decr.u16.low++;
>  env->flags &= ~(DELAY_SLOT | DELAY_SLOT_CONDITIONAL);
> +n = 2;
>  }
>  #endif
> -/* This should never happen.  */
> -if (n > CF_COUNT_MASK) {
> -cpu_abort(cpu, "TB too big during recompile");
> -}
>  
> -pc = tb->pc;
> -cs_base = tb->cs_base;
> -flags = tb->flags;
> -tb_phys_invalidate(tb, -1);
> -
> -/* Execute one IO instruction without caching
> -   instead of creating large TB. */
> -cpu->cflags_next_tb = curr_cflags() | CF_LAST_IO | CF_NOCACHE | 1;
> +/* Generate a new TB executing the I/O insn.  */
> +cpu->cflags_next_tb = curr_cflags() | CF_LAST_IO | n;
>  
>  if (tb->cflags & CF_NOCACHE) {
>  if (tb->orig_tb) {
> @@ -1785,11 +1773,6 @@ void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr)
>  tb_remove(tb);
>  }
>  
> -/* Generate new TB instead of the current one. */
> -/* FIXME: In theory this could raise an exception.  In practice
> -   we have already translated the block once so it's probably ok.  */
> -tb_gen_code(cpu, pc, cs_base, flags, curr_cflags() | CF_LAST_IO | n);
> -
>  /* TODO: If env->pc != tb->pc (i.e. the faulting instruction was not
>   * the first in the TB) then we end up generating a whole new TB and
>   *  repeating the fault, which is horribly inefficient.
> 

Re: [Qemu-devel] [PATCH v2] tcg: Really fix cpu_io_recompile

[Pavel Dovgalyuk]

> From: Richard Henderson [mailto:richard.hender...@linaro.org]
> We have confused the number of instructions that have been
> executed in the TB with the number of instructions needed
> to repeat the I/O instruction.
> 
> We have used cpu_restore_state_from_tb, which means that
> the guest pc is pointing to the I/O instruction.  The only
> time the answer to the later question is not 1 is when
> MIPS or SH4 need to re-execute the branch for the delay
> slot as well.
> 
> We must rely on cpu->cflags_next_tb to generate the next TB,
> as otherwise we have a race condition with other guest cpus
> within the TB cache.
> 
> Fixes: 0790f86861079b1932679d0f011e431aaf4ee9e2
> Signed-off-by: Richard Henderson 
> ---
> 
> My v1 raced with Paolo's pull request, so v2 now fixes Pavel's fix.
> 

Works for Ciro's ARM sample and doesn't break icount and replay for i386.
Tested-by: Pavel Dovgalyuk 


Pavel Dovgalyuk
> r~
> 
> ---
>  accel/tcg/translate-all.c | 37 ++---
>  1 file changed, 10 insertions(+), 27 deletions(-)
> 
> diff --git a/accel/tcg/translate-all.c b/accel/tcg/translate-all.c
> index 5ad1b919bc..d4190602d1 100644
> --- a/accel/tcg/translate-all.c
> +++ b/accel/tcg/translate-all.c
> @@ -1728,8 +1728,7 @@ void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr)
>  CPUArchState *env = cpu->env_ptr;
>  #endif
>  TranslationBlock *tb;
> -uint32_t n, flags;
> -target_ulong pc, cs_base;
> +uint32_t n;
> 
>  tb_lock();
>  tb = tb_find_pc(retaddr);
> @@ -1737,44 +1736,33 @@ void cpu_io_recompile(CPUState *cpu, uintptr_t 
> retaddr)
>  cpu_abort(cpu, "cpu_io_recompile: could not find TB for pc=%p",
>(void *)retaddr);
>  }
> -n = cpu->icount_decr.u16.low + tb->icount;
>  cpu_restore_state_from_tb(cpu, tb, retaddr);
> -/* Calculate how many instructions had been executed before the fault
> -   occurred.  */
> -n = n - cpu->icount_decr.u16.low;
> -/* Generate a new TB ending on the I/O insn.  */
> -n++;
> +
>  /* On MIPS and SH, delay slot instructions can only be restarted if
> they were already the first instruction in the TB.  If this is not
> the first instruction in a TB then re-execute the preceding
> branch.  */
> +n = 1;
>  #if defined(TARGET_MIPS)
> -if ((env->hflags & MIPS_HFLAG_BMASK) != 0 && n > 1) {
> +if ((env->hflags & MIPS_HFLAG_BMASK) != 0
> +&& env->active_tc.PC != tb->pc) {
>  env->active_tc.PC -= (env->hflags & MIPS_HFLAG_B16 ? 2 : 4);
>  cpu->icount_decr.u16.low++;
>  env->hflags &= ~MIPS_HFLAG_BMASK;
> +n = 2;
>  }
>  #elif defined(TARGET_SH4)
>  if ((env->flags & ((DELAY_SLOT | DELAY_SLOT_CONDITIONAL))) != 0
> -&& n > 1) {
> +&& env->pc != tb->pc) {
>  env->pc -= 2;
>  cpu->icount_decr.u16.low++;
>  env->flags &= ~(DELAY_SLOT | DELAY_SLOT_CONDITIONAL);
> +n = 2;
>  }
>  #endif
> -/* This should never happen.  */
> -if (n > CF_COUNT_MASK) {
> -cpu_abort(cpu, "TB too big during recompile");
> -}
> 
> -pc = tb->pc;
> -cs_base = tb->cs_base;
> -flags = tb->flags;
> -tb_phys_invalidate(tb, -1);
> -
> -/* Execute one IO instruction without caching
> -   instead of creating large TB. */
> -cpu->cflags_next_tb = curr_cflags() | CF_LAST_IO | CF_NOCACHE | 1;
> +/* Generate a new TB executing the I/O insn.  */
> +cpu->cflags_next_tb = curr_cflags() | CF_LAST_IO | n;
> 
>  if (tb->cflags & CF_NOCACHE) {
>  if (tb->orig_tb) {
> @@ -1785,11 +1773,6 @@ void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr)
>  tb_remove(tb);
>  }
> 
> -/* Generate new TB instead of the current one. */
> -/* FIXME: In theory this could raise an exception.  In practice
> -   we have already translated the block once so it's probably ok.  */
> -tb_gen_code(cpu, pc, cs_base, flags, curr_cflags() | CF_LAST_IO | n);
> -
>  /* TODO: If env->pc != tb->pc (i.e. the faulting instruction was not
>   * the first in the TB) then we end up generating a whole new TB and
>   *  repeating the fault, which is horribly inefficient.
> --
> 2.14.3