Skip to contentSkip to conversation

cleanup cpu-exec.c: consolidate handle_cpu_signal

RFC PATCH In review Nathan Froyd 6 participants 11 messages Apr 22, 2009 – Aug 3, 2009

lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 cleanup cpu-exec.c: consolidate handle_cpu_signal

handle_cpu_signal is very nearly copy-paste code for each target, with a few minor variations. This patch sets up appropriate defaults for a generic handle_cpu_signal and provides overrides for particular targets that did things differently. Fixing things like the persistent (XXX:
use sigsetjmp) should now become somewhat easier.

I don't understand what the "activate soft MMU for this block" is trying to do. (Especially since handle_cpu_signal and company are under
!defined(CONFIG_SOFTMMU)...) Even though it appears that if
cpu_*_handle_mmu_fault ever returns a value > 0, that value is always one and therefore that block is superfluous, I've left that cleanup for a later time. Likewise for why x86 has a different EXCEPTION_ACTION than everyone else.

---
 cpu-exec.c         |  379 +++-------------------------------------------------
 target-alpha/cpu.h |    1 +
 target-arm/cpu.h   |    1 +
 target-cris/cpu.h  |    1 +
 target-i386/cpu.h  |    1 +
 target-m68k/cpu.h  |    1 +
 target-mips/cpu.h  |    1 +
 target-ppc/cpu.h   |    1 +
 target-sh4/cpu.h   |    1 +
 target-sparc/cpu.h |    1 +
 10 files changed, 31 insertions(+), 357 deletions(-)
cpu-exec.c +22 -357
index 122bdd1..4d316b0 100644
@@ -754,6 +754,24 @@ void cpu_x86_frstor(CPUX86State *s, target_ulong ptr, int data32)
 #if !defined(CONFIG_SOFTMMU)
 
 #if defined(TARGET_I386)
+#define EXCEPTION_ACTION raise_exception_err(env->exception_index, env->error_code)
+#define SOFTMMU_EXTRA_ACTION env->hflags |= HF_SOFTMMU_MASK
+#define ACTIVATE_SOFTMMU 1
+#elif defined(TARGET_PPC)
+#define ACTIVATE_SOFTMMU 1
+#elif defined(TARGET_MIPS)
+#define ACTIVATE_SOFTMMU 1
+#endif
+
+#if !defined(EXCEPTION_ACTION)
+#define EXCEPTION_ACTION cpu_loop_exit()
+#endif
+#if !defined(SOFTMMU_EXTRA_ACTION
+#define SOFTMMU_EXTRA_ACTION
+#endif
+#if !defined(ACTIVATE_SOFTMMU)
+#define ACTIVATE_SOFTMMU 0
+#endif
 
 /* 'pc' is the host PC at which the exception was raised. 'address' is
    the effective address of the memory exception. 'is_write' is 1 if a
@@ -778,7 +796,7 @@ static inline int handle_cpu_signal(unsigned long pc, unsigned long address,
     }
 
     /* see if it is an MMU fault */
-    ret = cpu_x86_handle_mmu_fault(env, address, is_write, MMU_USER_IDX, 0);
+    ret = cpu_handle_mmu_fault(env, address, is_write, MMU_USER_IDX, 0);
     if (ret < 0)
         return 0; /* not an MMU fault */
     if (ret == 0)
@@ -790,373 +808,20 @@ static inline int handle_cpu_signal(unsigned long pc, unsigned long address,
            a virtual CPU fault */
         cpu_restore_state(tb, env, pc, puc);
     }
-    if (ret == 1) {
-#if 0
-        printf("PF exception: EIP=0x%08x CR2=0x%08x error=0x%x\n",
-               env->eip, env->cr[2], env->error_code);
-#endif
+    if (!ACTIVATE_SOFTMMU || ret == 1) {
         /* we restore the process signal mask as the sigreturn should
            do it (XXX: use sigsetjmp) */
         sigprocmask(SIG_SETMASK, old_set, NULL);
-        raise_exception_err(env->exception_index, env->error_code);
-    } else {
-        /* activate soft MMU for this block */
-        env->hflags |= HF_SOFTMMU_MASK;
-        cpu_resume_from_signal(env, puc);
-    }
-    /* never comes here */
-    return 1;
-}
-
-#elif defined(TARGET_ARM)
-static inline int handle_cpu_signal(unsigned long pc, unsigned long address,
-                                    int is_write, sigset_t *old_set,
-                                    void *puc)
-{
-    TranslationBlock *tb;
-    int ret;
-
-    if (cpu_single_env)
-        env = cpu_single_env; /* XXX: find a correct solution for multithread */
-#if defined(DEBUG_SIGNAL)
-    printf("qemu: SIGSEGV pc=0x%08lx address=%08lx w=%d oldset=0x%08lx\n",
-           pc, address, is_write, *(unsigned long *)old_set);
-#endif
-    /* XXX: locking issue */
-    if (is_write && page_unprotect(h2g(address), pc, puc)) {
-        return 1;
-    }
-    /* see if it is an MMU fault */
-    ret = cpu_arm_handle_mmu_fault(env, address, is_write, MMU_USER_IDX, 0);
-    if (ret < 0)
-        return 0; /* not an MMU fault */
-    if (ret == 0)
-        return 1; /* the MMU fault was handled without causing real CPU fault */
-    /* now we have a real cpu fault */
-    tb = tb_find_pc(pc);
-    if (tb) {
-        /* the PC is inside the translated code. It means that we have
-           a virtual CPU fault */
-        cpu_restore_state(tb, env, pc, puc);
-    }
-    /* we restore the process signal mask as the sigreturn should
-       do it (XXX: use sigsetjmp) */
-    sigprocmask(SIG_SETMASK, old_set, NULL);
-    cpu_loop_exit();
-    /* never comes here */
-    return 1;
-}
-#elif defined(TARGET_SPARC)
-static inline int handle_cpu_signal(unsigned long pc, unsigned long address,
-                                    int is_write, sigset_t *old_set,
-                                    void *puc)
-{
-    TranslationBlock *tb;
-    int ret;
-
-    if (cpu_single_env)
-        env = cpu_single_env; /* XXX: find a correct solution for multithread */
-#if defined(DEBUG_SIGNAL)
-    printf("qemu: SIGSEGV pc=0x%08lx address=%08lx w=%d oldset=0x%08lx\n",
-           pc, address, is_write, *(unsigned long *)old_set);
-#endif
-    /* XXX: locking issue */
-    if (is_write && page_unprotect(h2g(address), pc, puc)) {
-        return 1;
-    }
-    /* see if it is an MMU fault */
-    ret = cpu_sparc_handle_mmu_fault(env, address, is_write, MMU_USER_IDX, 0);
-    if (ret < 0)
-        return 0; /* not an MMU fault */
-    if (ret == 0)
-        return 1; /* the MMU fault was handled without causing real CPU fault */
-    /* now we have a real cpu fault */
-    tb = tb_find_pc(pc);
-    if (tb) {
-        /* the PC is inside the translated code. It means that we have
-           a virtual CPU fault */
-        cpu_restore_state(tb, env, pc, puc);
-    }
-    /* we restore the process signal mask as the sigreturn should
-       do it (XXX: use sigsetjmp) */
-    sigprocmask(SIG_SETMASK, old_set, NULL);
-    cpu_loop_exit();
-    /* never comes here */
-    return 1;
-}
-#elif defined (TARGET_PPC)
-static inline int handle_cpu_signal(unsigned long pc, unsigned long address,
-                                    int is_write, sigset_t *old_set,
-                                    void *puc)
-{
-    TranslationBlock *tb;
-    int ret;
-
-    if (cpu_single_env)
-        env = cpu_single_env; /* XXX: find a correct solution for multithread */
-#if defined(DEBUG_SIGNAL)
-    printf("qemu: SIGSEGV pc=0x%08lx address=%08lx w=%d oldset=0x%08lx\n",
-           pc, address, is_write, *(unsigned long *)old_set);
-#endif
-    /* XXX: locking issue */
-    if (is_write && page_unprotect(h2g(address), pc, puc)) {
-        return 1;
-    }
-
-    /* see if it is an MMU fault */
-    ret = cpu_ppc_handle_mmu_fault(env, address, is_write, MMU_USER_IDX, 0);
-    if (ret < 0)
-        return 0; /* not an MMU fault */
-    if (ret == 0)
-        return 1; /* the MMU fault was handled without causing real CPU fault */
-
-    /* now we have a real cpu fault */
-    tb = tb_find_pc(pc);
-    if (tb) {
-        /* the PC is inside the translated code. It means that we have
-           a virtual CPU fault */
-        cpu_restore_state(tb, env, pc, puc);
-    }
-    if (ret == 1) {
-#if 0
-        printf("PF exception: NIP=0x%08x error=0x%x %p\n",
-               env->nip, env->error_code, tb);
-#endif
-    /* we restore the process signal mask as the sigreturn should
-       do it (XXX: use sigsetjmp) */
-        sigprocmask(SIG_SETMASK, old_set, NULL);
-        cpu_loop_exit();
-    } else {
-        /* activate soft MMU for this block */
-        cpu_resume_from_signal(env, puc);
-    }
-    /* never comes here */
-    return 1;
-}
-
-#elif defined(TARGET_M68K)
-static inline int handle_cpu_signal(unsigned long pc, unsigned long address,
-                                    int is_write, sigset_t *old_set,
-                                    void *puc)
-{
-    TranslationBlock *tb;
-    int ret;
-
-    if (cpu_single_env)
-        env = cpu_single_env; /* XXX: find a correct solution for multithread */
-#if defined(DEBUG_SIGNAL)
-    printf("qemu: SIGSEGV pc=0x%08lx address=%08lx w=%d oldset=0x%08lx\n",
-           pc, address, is_write, *(unsigned long *)old_set);
-#endif
-    /* XXX: locking issue */
-    if (is_write && page_unprotect(address, pc, puc)) {
-        return 1;
-    }
-    /* see if it is an MMU fault */
-    ret = cpu_m68k_handle_mmu_fault(env, address, is_write, MMU_USER_IDX, 0);
-    if (ret < 0)
-        return 0; /* not an MMU fault */
-    if (ret == 0)
-        return 1; /* the MMU fault was handled without causing real CPU fault */
-    /* now we have a real cpu fault */
-    tb = tb_find_pc(pc);
-    if (tb) {
-        /* the PC is inside the translated code. It means that we have
-           a virtual CPU fault */
-        cpu_restore_state(tb, env, pc, puc);
-    }
-    /* we restore the process signal mask as the sigreturn should
-       do it (XXX: use sigsetjmp) */
-    sigprocmask(SIG_SETMASK, old_set, NULL);
-    cpu_loop_exit();
-    /* never comes here */
-    return 1;
-}
-
-#elif defined (TARGET_MIPS)
-static inline int handle_cpu_signal(unsigned long pc, unsigned long address,
-                                    int is_write, sigset_t *old_set,
-                                    void *puc)
-{
-    TranslationBlock *tb;
-    int ret;
-
-    if (cpu_single_env)
-        env = cpu_single_env; /* XXX: find a correct solution for multithread */
-#if defined(DEBUG_SIGNAL)
-    printf("qemu: SIGSEGV pc=0x%08lx address=%08lx w=%d oldset=0x%08lx\n",
-           pc, address, is_write, *(unsigned long *)old_set);
-#endif
-    /* XXX: locking issue */
-    if (is_write && page_unprotect(h2g(address), pc, puc)) {
-        return 1;
-    }
-
-    /* see if it is an MMU fault */
-    ret = cpu_mips_handle_mmu_fault(env, address, is_write, MMU_USER_IDX, 0);
-    if (ret < 0)
-        return 0; /* not an MMU fault */
-    if (ret == 0)
-        return 1; /* the MMU fault was handled without causing real CPU fault */
-
-    /* now we have a real cpu fault */
-    tb = tb_find_pc(pc);
-    if (tb) {
-        /* the PC is inside the translated code. It means that we have
-           a virtual CPU fault */
-        cpu_restore_state(tb, env, pc, puc);
-    }
-    if (ret == 1) {
-#if 0
-        printf("PF exception: PC=0x" TARGET_FMT_lx " error=0x%x %p\n",
-               env->PC, env->error_code, tb);
-#endif
-    /* we restore the process signal mask as the sigreturn should
-       do it (XXX: use sigsetjmp) */
-        sigprocmask(SIG_SETMASK, old_set, NULL);
-        cpu_loop_exit();
+        EXCEPTION_ACTION;
     } else {
         /* activate soft MMU for this block */
+        SOFTMMU_EXTRA_ACTION;
         cpu_resume_from_signal(env, puc);
     }
     /* never comes here */
     return 1;
 }
 
-#elif defined (TARGET_SH4)
-static inline int handle_cpu_signal(unsigned long pc, unsigned long address,
-                                    int is_write, sigset_t *old_set,
-                                    void *puc)
-{
-    TranslationBlock *tb;
-    int ret;
-
-    if (cpu_single_env)
-        env = cpu_single_env; /* XXX: find a correct solution for multithread */
-#if defined(DEBUG_SIGNAL)
-    printf("qemu: SIGSEGV pc=0x%08lx address=%08lx w=%d oldset=0x%08lx\n",
-           pc, address, is_write, *(unsigned long *)old_set);
-#endif
-    /* XXX: locking issue */
-    if (is_write && page_unprotect(h2g(address), pc, puc)) {
-        return 1;
-    }
-
-    /* see if it is an MMU fault */
-    ret = cpu_sh4_handle_mmu_fault(env, address, is_write, MMU_USER_IDX, 0);
-    if (ret < 0)
-        return 0; /* not an MMU fault */
-    if (ret == 0)
-        return 1; /* the MMU fault was handled without causing real CPU fault */
-
-    /* now we have a real cpu fault */
-    tb = tb_find_pc(pc);
-    if (tb) {
-        /* the PC is inside the translated code. It means that we have
-           a virtual CPU fault */
-        cpu_restore_state(tb, env, pc, puc);
-    }
-#if 0
-        printf("PF exception: NIP=0x%08x error=0x%x %p\n",
-               env->nip, env->error_code, tb);
-#endif
-    /* we restore the process signal mask as the sigreturn should
-       do it (XXX: use sigsetjmp) */
-    sigprocmask(SIG_SETMASK, old_set, NULL);
-    cpu_loop_exit();
-    /* never comes here */
-    return 1;
-}
-
-#elif defined (TARGET_ALPHA)
-static inline int handle_cpu_signal(unsigned long pc, unsigned long address,
-                                    int is_write, sigset_t *old_set,
-                                    void *puc)
-{
-    TranslationBlock *tb;
-    int ret;
-
-    if (cpu_single_env)
-        env = cpu_single_env; /* XXX: find a correct solution for multithread */
-#if defined(DEBUG_SIGNAL)
-    printf("qemu: SIGSEGV pc=0x%08lx address=%08lx w=%d oldset=0x%08lx\n",
-           pc, address, is_write, *(unsigned long *)old_set);
-#endif
-    /* XXX: locking issue */
-    if (is_write && page_unprotect(h2g(address), pc, puc)) {
-        return 1;
-    }
-
-    /* see if it is an MMU fault */
-    ret = cpu_alpha_handle_mmu_fault(env, address, is_write, MMU_USER_IDX, 0);
-    if (ret < 0)
-        return 0; /* not an MMU fault */
-    if (ret == 0)
-        return 1; /* the MMU fault was handled without causing real CPU fault */
-
-    /* now we have a real cpu fault */
-    tb = tb_find_pc(pc);
-    if (tb) {
-        /* the PC is inside the translated code. It means that we have
-           a virtual CPU fault */
-        cpu_restore_state(tb, env, pc, puc);
-    }
-#if 0
-        printf("PF exception: NIP=0x%08x error=0x%x %p\n",
-               env->nip, env->error_code, tb);
-#endif
-    /* we restore the process signal mask as the sigreturn should
-       do it (XXX: use sigsetjmp) */
-    sigprocmask(SIG_SETMASK, old_set, NULL);
-    cpu_loop_exit();
-    /* never comes here */
-    return 1;
-}
-#elif defined (TARGET_CRIS)
-static inline int handle_cpu_signal(unsigned long pc, unsigned long address,
-                                    int is_write, sigset_t *old_set,
-                                    void *puc)
-{
-    TranslationBlock *tb;
-    int ret;
-
-    if (cpu_single_env)
-        env = cpu_single_env; /* XXX: find a correct solution for multithread */
-#if defined(DEBUG_SIGNAL)
-    printf("qemu: SIGSEGV pc=0x%08lx address=%08lx w=%d oldset=0x%08lx\n",
-           pc, address, is_write, *(unsigned long *)old_set);
-#endif
-    /* XXX: locking issue */
-    if (is_write && page_unprotect(h2g(address), pc, puc)) {
-        return 1;
-    }
-
-    /* see if it is an MMU fault */
-    ret = cpu_cris_handle_mmu_fault(env, address, is_write, MMU_USER_IDX, 0);
-    if (ret < 0)
-        return 0; /* not an MMU fault */
-    if (ret == 0)
-        return 1; /* the MMU fault was handled without causing real CPU fault */
-
-    /* now we have a real cpu fault */
-    tb = tb_find_pc(pc);
-    if (tb) {
-        /* the PC is inside the translated code. It means that we have
-           a virtual CPU fault */
-        cpu_restore_state(tb, env, pc, puc);
-    }
-    /* we restore the process signal mask as the sigreturn should
-       do it (XXX: use sigsetjmp) */
-    sigprocmask(SIG_SETMASK, old_set, NULL);
-    cpu_loop_exit();
-    /* never comes here */
-    return 1;
-}
-
-#else
-#error unsupported target CPU
-#endif
-
 #if defined(__i386__)
 
 #if defined(__APPLE__)
target-alpha/cpu.h +1 -0
index 8c64def..ade180d 100644
@@ -434,6 +434,7 @@ int cpu_alpha_signal_handler(int host_signum, void *pinfo,
                              void *puc);
 int cpu_alpha_handle_mmu_fault (CPUState *env, uint64_t address, int rw,
                                 int mmu_idx, int is_softmmu);
+#define cpu_handle_mmu_fault cpu_alpha_handle_mmu_fault
 void do_interrupt (CPUState *env);
 
 int cpu_alpha_mfpr (CPUState *env, int iprn, uint64_t *valp);
target-arm/cpu.h +1 -0
index f98655f..192a92f 100644
@@ -221,6 +221,7 @@ int cpu_arm_signal_handler(int host_signum, void *pinfo,
                            void *puc);
 int cpu_arm_handle_mmu_fault (CPUARMState *env, target_ulong address, int rw,
                               int mmu_idx, int is_softmuu);
+#define cpu_handle_mmu_fault cpu_arm_handle_mmu_fault
 
 void cpu_lock(void);
 void cpu_unlock(void);
target-cris/cpu.h +1 -0
index e98a48d..e313a0b 100644
@@ -219,6 +219,7 @@ static inline int cpu_mmu_index (CPUState *env)
 
 int cpu_cris_handle_mmu_fault(CPUState *env, target_ulong address, int rw,
                               int mmu_idx, int is_softmmu);
+#define cpu_handle_mmu_fault cpu_cris_handle_mmu_fault
 
 #if defined(CONFIG_USER_ONLY)
 static inline void cpu_clone_regs(CPUState *env, target_ulong newsp)
target-i386/cpu.h +1 -0
index c6bca94..9a6b00c 100644
@@ -774,6 +774,7 @@ int cpu_x86_signal_handler(int host_signum, void *pinfo,
 /* helper.c */
 int cpu_x86_handle_mmu_fault(CPUX86State *env, target_ulong addr,
                              int is_write, int mmu_idx, int is_softmmu);
+#define cpu_handle_mmu_fault cpu_x86_handle_mmu_fault
 void cpu_x86_set_a20(CPUX86State *env, int a20_state);
 void cpu_x86_cpuid(CPUX86State *env, uint32_t index, uint32_t count,
                    uint32_t *eax, uint32_t *ebx,
target-m68k/cpu.h +1 -0
index e2529eb..195cb23 100644
@@ -225,6 +225,7 @@ static inline int cpu_mmu_index (CPUState *env)
 
 int cpu_m68k_handle_mmu_fault(CPUState *env, target_ulong address, int rw,
                               int mmu_idx, int is_softmmu);
+#define cpu_handle_mmu_fault cpu_m68k_handle_mmu_fault
 
 #if defined(CONFIG_USER_ONLY)
 static inline void cpu_clone_regs(CPUState *env, target_ulong newsp)
target-mips/cpu.h +1 -0
index b415dc4..459edb5 100644
@@ -579,6 +579,7 @@ void cpu_mips_update_irq (CPUState *env);
 /* helper.c */
 int cpu_mips_handle_mmu_fault (CPUState *env, target_ulong address, int rw,
                                int mmu_idx, int is_softmmu);
+#define cpu_handle_mmu_fault cpu_mips_handle_mmu_fault
 void do_interrupt (CPUState *env);
 void r4k_invalidate_tlb (CPUState *env, int idx, int use_extra);
 
target-ppc/cpu.h +1 -0
index 87b3460..b4ad998 100644
@@ -703,6 +703,7 @@ int cpu_ppc_signal_handler (int host_signum, void *pinfo,
                             void *puc);
 int cpu_ppc_handle_mmu_fault (CPUPPCState *env, target_ulong address, int rw,
                               int mmu_idx, int is_softmmu);
+#define cpu_handle_mmu_fault cpu_ppc_handle_mmu_fault
 int get_physical_address (CPUPPCState *env, mmu_ctx_t *ctx, target_ulong vaddr,
                           int rw, int access_type);
 void do_interrupt (CPUPPCState *env);
target-sh4/cpu.h +1 -0
index e597f65..3c125c8 100644
@@ -164,6 +164,7 @@ int cpu_sh4_signal_handler(int host_signum, void *pinfo,
                            void *puc);
 int cpu_sh4_handle_mmu_fault(CPUSH4State * env, target_ulong address, int rw,
 			     int mmu_idx, int is_softmmu);
+#define cpu_handle_mmu_fault cpu_sh4_handle_mmu_fault
 void do_interrupt(CPUSH4State * env);
 
 void sh4_cpu_list(FILE *f, int (*cpu_fprintf)(FILE *f, const char *fmt, ...));
target-sparc/cpu.h +1 -0
index 8b84789..2b6c1bb 100644
@@ -349,6 +349,7 @@ void cpu_lock(void);
 void cpu_unlock(void);
 int cpu_sparc_handle_mmu_fault(CPUSPARCState *env1, target_ulong address, int rw,
                                int mmu_idx, int is_softmmu);
+#define cpu_handle_mmu_fault cpu_sparc_handle_mmu_fault
 target_ulong mmu_probe(CPUSPARCState *env, target_ulong address, int mmulev);
 void dump_mmu(CPUSPARCState *env);
 
Signature · 1 line
1.6.0.5
#1 lore raw
Quoted 7 lines

I don't understand what the "activate soft MMU for this block" is trying to do. (Especially since handle_cpu_signal and company are under
!defined(CONFIG_SOFTMMU)...) Even though it appears that if
cpu_*_handle_mmu_fault ever returns a value > 0, that value is always one and therefore that block is superfluous, I've left that cleanup for a later time. Likewise for why x86 has a different EXCEPTION_ACTION than everyone else.

This is most likely a remnant of the long-deceased qemu-fast, which used mmap tricks to avoid MMU emulation in system mode.

Paul

#2 lore raw
Quoted 2 lines · Paul Brook

This is most likely a remnant of the long-deceased qemu-fast, which used mmap tricks to avoid MMU emulation in system mode.

I'd really like to see kvm used to implement something like this (to emulate non-x86 on x86).

Signature · 1 line
error compiling committee.c: too many arguments to function
#3 lore raw
Quoted 6 lines · Avi Kivity
Quoted 2 lines · Paul Brook

This is most likely a remnant of the long-deceased qemu-fast, which used mmap tricks to avoid MMU emulation in system mode.

I'd really like to see kvm used to implement something like this (to emulate non-x86 on x86).

It's something I've considered a few times. It gets hairy fairly quickly though. You're probably also going to hit a world of pain if your host pagesize is larger than your guest pagesize, and for guests with a software managed TLB shadow paging gets much more interesting.

There's very little of the old code left, and it's bitrotten enough that there's no point trying to keep it on the offchance that it'll be useful.

Paul

#4 lore raw
Quoted 5 lines · Paul Brook
Quoted 3 lines

I'd really like to see kvm used to implement something like this (to emulate non-x86 on x86).

It's something I've considered a few times. It gets hairy fairly quickly though. You're probably also going to hit a world of pain if your host pagesize is larger than your guest pagesize,

Yes. But realistically your host is going to be either x86 or x86 (in a few corner cases that no one cares about, x86). Are there targets with page size < 4K? We don't target VAX.

Quoted 3 lines

and for guests with a software
managed TLB shadow paging gets much more interesting.

Hmm, inserting a tlb entry would just create x86 page table entries, no?

Quoted 2 lines

There's very little of the old code left, and it's bitrotten enough that there's no point trying to keep it on the offchance that it'll be useful.

Sure, starting from scratch sounds much better, this is going to be wildly different.

Signature · 1 line
Do not meddle in the internals of kernels, for they are subtle and quick to panic.
#5 lore raw
Quoted 8 lines · Avi Kivity
Quoted 6 lines · Paul Brook
Quoted 5 lines

I'd really like to see kvm used to implement something like this (to emulate non-x86 on x86).

Quoted 8 lines
Quoted 1 line

It's something I've considered a few times. It gets hairy fairly quickly

though. You're probably also going to hit a world of pain if your host pagesize is larger than your guest pagesize,

Quoted 1 line

Yes. But realistically your host is going to be either x86 or x86 (in a
few corner cases that no one cares about, x86). Are there targets with page size < 4K? We don't target VAX.

ARM has 2k pages IIRC. Sparc64 has 8k pages and then we have the same case for x86 target.

Quoted 6 lines
Quoted 1 line

and for guests with a software managed TLB shadow paging gets much more

interesting.

Quoted 2 lines

Hmm, inserting a tlb entry would just create x86 page table entries, no?

Inserting an entry may evict a previous entry. I don't think it's too difficult though.

Quoted 15 lines
Quoted 1 line

There's very little of the old code left, and it's bitrotten enough that

there's no point trying to keep it on the offchance that it'll be useful.

Quoted 2 lines

Sure, starting from scratch sounds much better, this is going to be wildly
different.

--
Do not meddle in the internals of kernels, for they are subtle and quick to
panic.

#6 lore raw
Quoted 9 lines · Blue Swirl
Quoted 7 lines

though. You're probably also going to hit a world of pain if your host pagesize is larger than your guest pagesize,

Yes. But realistically your host is going to be either x86 or x86 (in a
few corner cases that no one cares about, x86). Are there targets with page size < 4K? We don't target VAX.

ARM has 2k pages IIRC.

ARM is one of the most interesting targets, so this is too bad.

Quoted 3 lines

Sparc64 has 8k pages and then we have the same
case for x86 target.

There is no kvm for sparc64.

Signature · 1 line
Do not meddle in the internals of kernels, for they are subtle and quick to panic.
#7 lore raw
Quoted 19 lines · Avi Kivity
Quoted 4 lines · Blue Swirl
Quoted 4 lines

though. You're probably also going to hit a world of pain if your host pagesize is larger than your guest pagesize,
Yes. But realistically your host is going to be either x86 or x86

(in a

Quoted 1 line
Quoted 1 line

few corner cases that no one cares about, x86). Are there targets with

page

Quoted 6 lines
Quoted 3 lines

size < 4K? We don't target VAX.

ARM has 2k pages IIRC.

ARM is one of the most interesting targets, so this is too bad.

Quoted 6 lines
Quoted 4 lines

Sparc64 has 8k pages and then we have the same
case for x86 target.

There is no kvm for sparc64.

There has been some initial discussions, but no code.

#8 lore raw
In reply to Blue Swirl
Quoted 3 lines · Blue Swirl

ARM has 2k pages IIRC. Sparc64 has 8k pages and then we have the same case for x86 target.

pre-ARMv7 MMU had 1K pages, which have been removed in ARMv7; this probably means their use was not widespread enough to keep them.

Laurent

#9 lore raw
In reply to Nathan Froyd
Quoted 14 lines · Nathan Froyd

handle_cpu_signal is very nearly copy-paste code for each target, with a few minor variations.  This patch sets up appropriate defaults for a generic handle_cpu_signal and provides overrides for particular targets that did things differently.  Fixing things like the persistent (XXX:
use sigsetjmp) should now become somewhat easier.

I don't understand what the "activate soft MMU for this block" is trying to do.  (Especially since handle_cpu_signal and company are under !defined(CONFIG_SOFTMMU)...)  Even though it appears that if cpu_*_handle_mmu_fault ever returns a value > 0, that value is always one and therefore that block is superfluous, I've left that cleanup for a later time.  Likewise for why x86 has a different EXCEPTION_ACTION than everyone else.

Quoted 1 line

Any plans on reviving this patch? It's definitely a cleanup I am interested in.

Best regards,
Filip Navara

#10 lore raw
Quoted 9 lines · Filip Navara
Quoted 5 lines · Nathan Froyd

handle_cpu_signal is very nearly copy-paste code for each target, with a few minor variations.  This patch sets up appropriate defaults for a generic handle_cpu_signal and provides overrides for particular targets that did things differently.  Fixing things like the persistent (XXX:
use sigsetjmp) should now become somewhat easier.

Any plans on reviving this patch? It's definitely a cleanup I am interested in.

I will try to get to posting a revised version incorporating Paul's comments in this next week.

-Nathan