[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [PATCH for-5.2 2/3] linux-user/sparc: Correct set/get_context handli
From: |
Laurent Vivier |
Subject: |
Re: [PATCH for-5.2 2/3] linux-user/sparc: Correct set/get_context handling of fp and i7 |
Date: |
Tue, 10 Nov 2020 07:53:49 +0100 |
User-agent: |
Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101 Thunderbird/78.4.0 |
Le 05/11/2020 à 22:23, Peter Maydell a écrit :
> Because QEMU's user-mode emulation just directly accesses guest CPU
> state, for SPARC the guest register window state is not the same in
> the sparc64_get_context() and sparc64_set_context() functions as it
> is for the real kernel's versions of those functions. Specifically,
> for the kernel it has saved the user space state such that the O*
> registers go into a pt_regs struct as UREG_I*, and the I* registers
> have been spilled onto the userspace stack. For QEMU, we haven't
> done that, so the guest's O* registers are still in WREG_O* and the
> I* registers in WREG_I*.
>
> The code was already accessing the O* registers correctly for QEMU,
> but had copied the kernel code for accessing the I* registers off the
> userspace stack. Replace this with direct accesses to fp and i7 in
> the CPU state, and add a comment explaining why we differ from the
> kernel code here.
>
> This fix is sufficient to get bash to a shell prompt.
>
> Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
> ---
> I'm really pretty unsure about our handling of SPARC register
> windows here. This fix works, but should we instead be
> ensuring that the flush_windows() call cpu_loop() does
> before handling this trap has written the I* regs to the
> stack ???
> ---
> linux-user/sparc/signal.c | 47 ++++++++++++++++++---------------------
> 1 file changed, 22 insertions(+), 25 deletions(-)
>
> diff --git a/linux-user/sparc/signal.c b/linux-user/sparc/signal.c
> index 57ea1593bfc..c315704b389 100644
> --- a/linux-user/sparc/signal.c
> +++ b/linux-user/sparc/signal.c
> @@ -403,7 +403,6 @@ void sparc64_set_context(CPUSPARCState *env)
> struct target_ucontext *ucp;
> target_mc_gregset_t *grp;
> abi_ulong pc, npc, tstate;
> - abi_ulong fp, i7, w_addr;
> unsigned int i;
>
> ucp_addr = env->regwptr[WREG_O0];
> @@ -447,6 +446,15 @@ void sparc64_set_context(CPUSPARCState *env)
> __get_user(env->gregs[5], (&(*grp)[SPARC_MC_G5]));
> __get_user(env->gregs[6], (&(*grp)[SPARC_MC_G6]));
> __get_user(env->gregs[7], (&(*grp)[SPARC_MC_G7]));
> +
> + /*
> + * Note that unlike the kernel, we didn't need to mess with the
> + * guest register window state to save it into a pt_regs to run
> + * the kernel. So for us the guest's O regs are still in WREG_O*
> + * (unlike the kernel which has put them in UREG_I* in a pt_regs)
> + * and the fp and i7 are still in WREG_I6 and WREG_I7 and don't
> + * need to be written back to userspace memory.
> + */
> __get_user(env->regwptr[WREG_O0], (&(*grp)[SPARC_MC_O0]));
> __get_user(env->regwptr[WREG_O1], (&(*grp)[SPARC_MC_O1]));
> __get_user(env->regwptr[WREG_O2], (&(*grp)[SPARC_MC_O2]));
> @@ -456,18 +464,9 @@ void sparc64_set_context(CPUSPARCState *env)
> __get_user(env->regwptr[WREG_O6], (&(*grp)[SPARC_MC_O6]));
> __get_user(env->regwptr[WREG_O7], (&(*grp)[SPARC_MC_O7]));
>
> - __get_user(fp, &(ucp->tuc_mcontext.mc_fp));
> - __get_user(i7, &(ucp->tuc_mcontext.mc_i7));
> + __get_user(env->regwptr[WREG_FP], &(ucp->tuc_mcontext.mc_fp));
> + __get_user(env->regwptr[WREG_I7], &(ucp->tuc_mcontext.mc_i7));
>
> - w_addr = TARGET_STACK_BIAS + env->regwptr[WREG_O6];
> - if (put_user(fp, w_addr + offsetof(struct target_reg_window, ins[6]),
> - abi_ulong) != 0) {
> - goto do_sigsegv;
> - }
> - if (put_user(i7, w_addr + offsetof(struct target_reg_window, ins[7]),
> - abi_ulong) != 0) {
> - goto do_sigsegv;
> - }
> /* FIXME this does not match how the kernel handles the FPU in
> * its sparc64_set_context implementation. In particular the FPU
> * is only restored if fenab is non-zero in:
> @@ -501,7 +500,6 @@ void sparc64_get_context(CPUSPARCState *env)
> struct target_ucontext *ucp;
> target_mc_gregset_t *grp;
> target_mcontext_t *mcp;
> - abi_ulong fp, i7, w_addr;
> int err;
> unsigned int i;
> target_sigset_t target_set;
> @@ -553,6 +551,15 @@ void sparc64_get_context(CPUSPARCState *env)
> __put_user(env->gregs[5], &((*grp)[SPARC_MC_G5]));
> __put_user(env->gregs[6], &((*grp)[SPARC_MC_G6]));
> __put_user(env->gregs[7], &((*grp)[SPARC_MC_G7]));
> +
> + /*
> + * Note that unlike the kernel, we didn't need to mess with the
> + * guest register window state to save it into a pt_regs to run
> + * the kernel. So for us the guest's O regs are still in WREG_O*
> + * (unlike the kernel which has put them in UREG_I* in a pt_regs)
> + * and the fp and i7 are still in WREG_I6 and WREG_I7 and don't
> + * need to be fished out of userspace memory.
> + */
> __put_user(env->regwptr[WREG_O0], &((*grp)[SPARC_MC_O0]));
> __put_user(env->regwptr[WREG_O1], &((*grp)[SPARC_MC_O1]));
> __put_user(env->regwptr[WREG_O2], &((*grp)[SPARC_MC_O2]));
> @@ -562,18 +569,8 @@ void sparc64_get_context(CPUSPARCState *env)
> __put_user(env->regwptr[WREG_O6], &((*grp)[SPARC_MC_O6]));
> __put_user(env->regwptr[WREG_O7], &((*grp)[SPARC_MC_O7]));
>
> - w_addr = TARGET_STACK_BIAS + env->regwptr[WREG_O6];
> - fp = i7 = 0;
> - if (get_user(fp, w_addr + offsetof(struct target_reg_window, ins[6]),
> - abi_ulong) != 0) {
> - goto do_sigsegv;
> - }
> - if (get_user(i7, w_addr + offsetof(struct target_reg_window, ins[7]),
> - abi_ulong) != 0) {
> - goto do_sigsegv;
> - }
> - __put_user(fp, &(mcp->mc_fp));
> - __put_user(i7, &(mcp->mc_i7));
> + __put_user(env->regwptr[WREG_FP], &(mcp->mc_fp));
> + __put_user(env->regwptr[WREG_I7], &(mcp->mc_i7));
>
> {
> uint32_t *dst = ucp->tuc_mcontext.mc_fpregs.mcfpu_fregs.sregs;
>
Applied to my linux-user-for-5.2 branch.
Thanks,
Laurent
- [PATCH for-5.2 0/3] linux-user: fix various sparc64 guest bugs, Peter Maydell, 2020/11/05
- [PATCH for-5.2 2/3] linux-user/sparc: Correct set/get_context handling of fp and i7, Peter Maydell, 2020/11/05
- [PATCH for-5.2 3/3] linux-user/sparc: Don't zero high half of PC, NPC, PSR in sigreturn, Peter Maydell, 2020/11/05
- Re: [PATCH for-5.2 0/3] linux-user: fix various sparc64 guest bugs, Mark Cave-Ayland, 2020/11/10