target/riscv/debug: Sdtrig fixes and TT Ascalon support
cover target/riscv/debug: Sdtrig fixes and TT Ascalon support
Hi,
Sorry for the big series. The Ascalon CPU implements Sdtrig with 2 different types of mcontrol6 trigger and the icount trigger, so in the course of testing and bringing up OpenSBI and Linux support for this, I've accumulated quite a lot.
My new year resolution is to start being better upstream contributor, it's taken me a while with changing jobs and architectures. So I don't expect others to drop everything to review this! Joel has been prodding me, and noted there is some other Sdtrig work going on with the v1.0 support patches.
I think the debug v1.0 patches are somewhat orthogonal to this series, but both are addressing aspects of a common problem of Sdtrig implementation specifics. I wonder if these should be reconciled or left separate. Sdtrig v1.00/v0.13 configuration is a single boolean which is feasible as a CPU property. Whereas the entire space of Sdtrig implementation seems like too much to make configurable in that way.
Any thoughts would be welcome.
Thanks,
Nick
Nicholas Piggin (25):
target/riscv/debug: Check only mcontrol triggers for break/watchpoint
matching
target/riscv/debug: Handle changing trigger types target/riscv/debug: Implement permissive type unavailable trigger target/riscv/debug: Fix icount trigger privilege check target/riscv/debug: Update itrigger_enabled after changing privilege target/riscv/debug: Implement get_trigger_action for icount type
trigger
target/riscv/debug: Fix migration post_load icount_enabled() test target/riscv/debug: Fix icount privilege matching icount_enabled()
test
target/riscv/debug: Implement icount trigger textra matching target/riscv/debug: Maintain itrigger_enabled in
helper_itrigger_match()
target/riscv/debug: Fix breakpoint matching action target/riscv/debug: Put mcontrol load/store match address into tval target/riscv/debug: Remove breakpoints on reset target/riscv/debug: Move debug CPU post_load details into debug.c target/riscv/debug: Insert breakpoints after migration target/riscv/debug: Remove itrigger icount-enabled mode target/riscv/debug: Advertise icount trigger type in tinfo target/riscv/debug: Reset trigger type to unavailable target/riscv/debug: Add new debug state format target/riscv/debug: Migrate mcontext using new sdtrig vmstate target/riscv/debug: Implementation specific Sdtrig configuration target/riscv/debug: Support heterogeneous trigger types target/riscv/debug: Support heterogeneous mcontrol access types target/riscv/debug: Emulate TT Ascalon Sdtrig target/riscv/debug: Fix minor comment typos
target/riscv/cpu.c | 65 ++++- target/riscv/cpu.h | 41 ++- target/riscv/cpu_helper.c | 10 +- target/riscv/csr.c | 7 +- target/riscv/debug.c | 571 ++++++++++++++++++++----------------- target/riscv/debug.h | 19 +- target/riscv/machine.c | 96 ++++++- target/riscv/tcg/tcg-cpu.c | 5 +- 8 files changed, 510 insertions(+), 304 deletions(-)
Signature · 1 line
2.51.0
RFC PATCH 1/25 target/riscv/debug: Check only mcontrol triggers for break/watchpoint matching
trigger_common_match() has some warning log messages in it for unsupported triggers. Don't call it in BP/WP checking except for mcontrol/mcontrol6 triggers applicable to BP/WPs.
--- target/riscv/debug.c | 22 ++++++++++++++++++---- 1 file changed, 18 insertions(+), 4 deletions(-)
target/riscv/debug.c +18 -4
@@ -952,6 +952,14 @@ bool riscv_cpu_debug_check_breakpoint(CPUState *cs)for (i =0 ; i <RV_MAX_TRIGGERS ; i++) { trigger_type =get_trigger_type (env, i); +switch (trigger_type) { +case TRIGGER_TYPE_AD_MATCH : +case TRIGGER_TYPE_AD_MATCH6 : +break ; +default : +continue ;/* No other types match breakpoint */ + } +if (!trigger_common_match (env, trigger_type, i)) {continue ; }
@@ -976,8 +984,7 @@ bool riscv_cpu_debug_check_breakpoint(CPUState *cs) }break ;default : -/* other trigger types are not supported or irrelevant */ -break ; +g_assert_not_reached (); } } }
@@ -998,6 +1005,14 @@ bool riscv_cpu_debug_check_watchpoint(CPUState *cs, CPUWatchpoint *wp)for (i =0 ; i <RV_MAX_TRIGGERS ; i++) { trigger_type =get_trigger_type (env, i); +switch (trigger_type) { +case TRIGGER_TYPE_AD_MATCH : +case TRIGGER_TYPE_AD_MATCH6 : +break ; +default : +continue ;/* No other types match watchpoint */ + } +if (!trigger_common_match (env, trigger_type, i)) {continue ; }
@@ -1036,8 +1051,7 @@ bool riscv_cpu_debug_check_watchpoint(CPUState *cs, CPUWatchpoint *wp) }break ;default : -/* other trigger types are not supported */ -break ; +g_assert_not_reached (); } }
Signature · 1 line
2.51.0
Re: [RFC PATCH 01/25] target/riscv/debug: Check only mcontrol triggers for break/watchpoint matching
Quoted 6 lines · Nicholas Piggin
trigger_common_match()has some warning log messages in it for unsupported triggers. Don't call it in BP/WP checking except for mcontrol/mcontrol6 triggers applicable to BP/WPs.Signed-off-by Nicholas Piggin <[email protected]>---
Quoted 57 lines
target/riscv/debug.c | 22 ++++++++++++++++++---- 1 file changed, 18 insertions(+), 4 deletions(-)target/riscv/debug.c +18
-4@@ -952,6 +952,14 @@ bool riscv_cpu_debug_check_breakpoint(CPUState *cs)for (i =0 ; i <RV_MAX_TRIGGERS ; i++) { trigger_type =get_trigger_type (env, i); +switch (trigger_type) { +case TRIGGER_TYPE_AD_MATCH : +case TRIGGER_TYPE_AD_MATCH6 : +break ; +default : +continue ;/* No other types match breakpoint */ + } +if (!trigger_common_match (env, trigger_type, i)) {continue ; }@@ -976,8 +984,7 @@ bool riscv_cpu_debug_check_breakpoint(CPUState *cs) }break ;default : -/* other trigger types are not supported or irrelevant */ -break ; +g_assert_not_reached (); } } }@@ -998,6 +1005,14 @@ bool riscv_cpu_debug_check_watchpoint(CPUState *cs, CPUWatchpoint *wp)for (i =0 ; i <RV_MAX_TRIGGERS ; i++) { trigger_type =get_trigger_type (env, i); +switch (trigger_type) { +case TRIGGER_TYPE_AD_MATCH : +case TRIGGER_TYPE_AD_MATCH6 : +break ; +default : +continue ;/* No other types match watchpoint */ + } +if (!trigger_common_match (env, trigger_type, i)) {continue ; }@@ -1036,8 +1051,7 @@ bool riscv_cpu_debug_check_watchpoint(CPUState *cs, CPUWatchpoint *wp) }break ;default : -/* other trigger types are not supported */ -break ; +g_assert_not_reached (); } }
Re: [RFC PATCH 01/25] target/riscv/debug: Check only mcontrol triggers for break/watchpoint matching
In reply to Nicholas PigginQuoted 5 lines · Nicholas Piggin
trigger_common_match()has some warning log messages in it for unsupported triggers. Don't call it in BP/WP checking except for mcontrol/mcontrol6 triggers applicable to BP/WPs.Signed-off-by Nicholas Piggin <[email protected]>
Quoted 62 lines
--- target/riscv/debug.c | 22 ++++++++++++++++++---- 1 file changed, 18 insertions(+), 4 deletions(-)target/riscv/debug.c +18
-4@@ -952,6 +952,14 @@ bool riscv_cpu_debug_check_breakpoint(CPUState *cs)for (i =0 ; i <RV_MAX_TRIGGERS ; i++) { trigger_type =get_trigger_type (env, i); +switch (trigger_type) { +case TRIGGER_TYPE_AD_MATCH : +case TRIGGER_TYPE_AD_MATCH6 : +break ; +default : +continue ;/* No other types match breakpoint */ + } +if (!trigger_common_match (env, trigger_type, i)) {continue ; }@@ -976,8 +984,7 @@ bool riscv_cpu_debug_check_breakpoint(CPUState *cs) }break ;default : -/* other trigger types are not supported or irrelevant */ -break ; +g_assert_not_reached (); } } }@@ -998,6 +1005,14 @@ bool riscv_cpu_debug_check_watchpoint(CPUState *cs, CPUWatchpoint *wp)for (i =0 ; i <RV_MAX_TRIGGERS ; i++) { trigger_type =get_trigger_type (env, i); +switch (trigger_type) { +case TRIGGER_TYPE_AD_MATCH : +case TRIGGER_TYPE_AD_MATCH6 : +break ; +default : +continue ;/* No other types match watchpoint */ + } +if (!trigger_common_match (env, trigger_type, i)) {continue ; }@@ -1036,8 +1051,7 @@ bool riscv_cpu_debug_check_watchpoint(CPUState *cs, CPUWatchpoint *wp) }break ;default : -/* other trigger types are not supported */ -break ; +g_assert_not_reached (); } }Signature · 1 line
2.51.0
RFC PATCH 2/25 target/riscv/debug: Handle changing trigger types
In reply to Nicholas PigginUpdating debug registers will first remove the existing TCG breakpoint / watchpoint, then adds it back with new values.
Writing TDATA1 with a value that changes the trigger type attempts to remove the facility for the new trigger type rather than the existing one. That is, it will not remove a breakpoint if the type is changed to a non-breakpoint type.
Fix this by removing based on the old trigger type, then inserting based on the new type.
--- target/riscv/debug.c | 64 +++++++++++++++++++++++--------------------- 1 file changed, 33 insertions(+), 31 deletions(-)
target/riscv/debug.c +33 -31
@@ -528,23 +528,12 @@ static void type2_breakpoint_remove(CPURISCVState *env, target_ulong index)static void type2_reg_write (CPURISCVState *env,target_ulong index,int tdata_index,target_ulong val) { -target_ulong new_val; -switch (tdata_index) {case TDATA1 : - new_val =type2_mcontrol_validate (env, val); -if (new_val != env->tdata1[index]) { - env->tdata1[index] = new_val; -type2_breakpoint_remove (env, index); -type2_breakpoint_insert (env, index); - } + env->tdata1[index] =type2_mcontrol_validate (env, val);break ;case TDATA2 : -if (val != env->tdata2[index]) { - env->tdata2[index] = val; -type2_breakpoint_remove (env, index); -type2_breakpoint_insert (env, index); - } + env->tdata2[index] = val;break ;case TDATA3 : env->tdata3[index] =textra_validate (env, val);
@@ -552,6 +541,8 @@ static void type2_reg_write(CPURISCVState *env, target_ulong index,default :g_assert_not_reached (); } + +type2_breakpoint_insert (env, index); }/* type 6 trigger */
@@ -642,23 +633,12 @@ static void type6_breakpoint_remove(CPURISCVState *env, target_ulong index)static void type6_reg_write (CPURISCVState *env,target_ulong index,int tdata_index,target_ulong val) { -target_ulong new_val; -switch (tdata_index) {case TDATA1 : - new_val =type6_mcontrol6_validate (env, val); -if (new_val != env->tdata1[index]) { - env->tdata1[index] = new_val; -type6_breakpoint_remove (env, index); -type6_breakpoint_insert (env, index); - } + env->tdata1[index] =type6_mcontrol6_validate (env, val);break ;case TDATA2 : -if (val != env->tdata2[index]) { - env->tdata2[index] = val; -type6_breakpoint_remove (env, index); -type6_breakpoint_insert (env, index); - } + env->tdata2[index] = val;break ;case TDATA3 : env->tdata3[index] =textra_validate (env, val);
@@ -666,6 +646,7 @@ static void type6_reg_write(CPURISCVState *env, target_ulong index,default :g_assert_not_reached (); } +type6_breakpoint_insert (env, index); }/* icount trigger type */
@@ -831,8 +812,6 @@ static void itrigger_reg_write(CPURISCVState *env, target_ulong index,/* set the count to timer */ timer_mod (env->itrigger_timer[index], env->last_icount +itrigger_get_count (env, index)); - }else { - env->itrigger_enabled =riscv_itrigger_enabled (env); } }break ;
@@ -881,12 +860,30 @@ target_ulong tdata_csr_read(CPURISCVState *env, int tdata_index)void tdata_csr_write (CPURISCVState *env,int tdata_index,target_ulong val) { -int trigger_type; +int trigger_type =get_trigger_type (env, env->trigger_cur); +bool check_itrigger = false; + +switch (trigger_type) { +case TRIGGER_TYPE_AD_MATCH : +type2_breakpoint_remove (env, env->trigger_cur); +break ; +case TRIGGER_TYPE_AD_MATCH6 : +type6_breakpoint_remove (env, env->trigger_cur); +break ; +case TRIGGER_TYPE_INST_CNT : +/* +* itrigger_enabled is the union of all enabled icount triggers, +* so it's easiest to recheck all if any have changed (removed or +* added or modified). +*/ + check_itrigger = true; +break ; +default : +break ; + }if (tdata_index ==TDATA1 ) { trigger_type =extract_trigger_type (env, val); - }else { - trigger_type =get_trigger_type (env, env->trigger_cur); }switch (trigger_type) {
@@ -898,6 +895,7 @@ void tdata_csr_write(CPURISCVState *env, int tdata_index, target_ulong val)break ;case TRIGGER_TYPE_INST_CNT :itrigger_reg_write (env, env->trigger_cur, tdata_index, val); + check_itrigger = true;break ;case TRIGGER_TYPE_INT :case TRIGGER_TYPE_EXCP :
@@ -913,6 +911,10 @@ void tdata_csr_write(CPURISCVState *env, int tdata_index, target_ulong val)default :g_assert_not_reached (); } + +if (check_itrigger && !icount_enabled ()) { + env->itrigger_enabled =riscv_itrigger_enabled (env); + } }target_ulong tinfo_csr_read (CPURISCVState *env)
Signature · 1 line
2.51.0
Re: [RFC PATCH 02/25] target/riscv/debug: Handle changing trigger types
Quoted 8 lines · Nicholas Piggin
Updating debug registers will first remove the existing TCG breakpoint / watchpoint, then adds it back with new values.
Writing TDATA1 with a value that changes the trigger type attempts to remove the facility for the new trigger type rather than the existing one. That is, it will not remove a breakpoint if the type is changed to a non-breakpoint type.
Quoted 4 lines
Fix this by removing based on the old trigger type, then inserting based on the new type.
Signed-off-by Nicholas Piggin <[email protected]>
Quoted 143 lines
--- target/riscv/debug.c | 64 +++++++++++++++++++++++--------------------- 1 file changed, 33 insertions(+), 31 deletions(-)target/riscv/debug.c +33
-31@@ -528,23 +528,12 @@ static void type2_breakpoint_remove(CPURISCVState *env, target_ulong index)static void type2_reg_write (CPURISCVState *env,target_ulong index,int tdata_index,target_ulong val) { -target_ulong new_val; -switch (tdata_index) {case TDATA1 : - new_val =type2_mcontrol_validate (env, val); -if (new_val != env->tdata1[index]) { - env->tdata1[index] = new_val; -type2_breakpoint_remove (env, index); -type2_breakpoint_insert (env, index); - } + env->tdata1[index] =type2_mcontrol_validate (env, val);break ;case TDATA2 : -if (val != env->tdata2[index]) { - env->tdata2[index] = val; -type2_breakpoint_remove (env, index); -type2_breakpoint_insert (env, index); - } + env->tdata2[index] = val;break ;case TDATA3 : env->tdata3[index] =textra_validate (env, val);@@ -552,6 +541,8 @@ static void type2_reg_write(CPURISCVState *env, target_ulong index,default :g_assert_not_reached (); } + +type2_breakpoint_insert (env, index); }/* type 6 trigger */ @@ -642,23 +633,12 @@ static void type6_breakpoint_remove(CPURISCVState *env, target_ulong index)static void type6_reg_write (CPURISCVState *env,target_ulong index,int tdata_index,target_ulong val) { -target_ulong new_val; -switch (tdata_index) {case TDATA1 : - new_val =type6_mcontrol6_validate (env, val); -if (new_val != env->tdata1[index]) { - env->tdata1[index] = new_val; -type6_breakpoint_remove (env, index); -type6_breakpoint_insert (env, index); - } + env->tdata1[index] =type6_mcontrol6_validate (env, val);break ;case TDATA2 : -if (val != env->tdata2[index]) { - env->tdata2[index] = val; -type6_breakpoint_remove (env, index); -type6_breakpoint_insert (env, index); - } + env->tdata2[index] = val;break ;case TDATA3 : env->tdata3[index] =textra_validate (env, val);@@ -666,6 +646,7 @@ static void type6_reg_write(CPURISCVState *env, target_ulong index,default :g_assert_not_reached (); } +type6_breakpoint_insert (env, index); }/* icount trigger type */ @@ -831,8 +812,6 @@ static void itrigger_reg_write(CPURISCVState *env, target_ulong index,/* set the count to timer */ timer_mod (env->itrigger_timer[index], env->last_icount +itrigger_get_count (env, index)); - }else { - env->itrigger_enabled =riscv_itrigger_enabled (env); } }break ;@@ -881,12 +860,30 @@ target_ulong tdata_csr_read(CPURISCVState *env, int tdata_index)void tdata_csr_write (CPURISCVState *env,int tdata_index,target_ulong val) { -int trigger_type; +int trigger_type =get_trigger_type (env, env->trigger_cur); +bool check_itrigger = false; + +switch (trigger_type) { +case TRIGGER_TYPE_AD_MATCH : +type2_breakpoint_remove (env, env->trigger_cur); +break ; +case TRIGGER_TYPE_AD_MATCH6 : +type6_breakpoint_remove (env, env->trigger_cur); +break ; +case TRIGGER_TYPE_INST_CNT : +/* +* itrigger_enabled is the union of all enabled icount triggers, +* so it's easiest to recheck all if any have changed (removed or +* added or modified). +*/ + check_itrigger = true; +break ; +default : +break ; + }if (tdata_index ==TDATA1 ) { trigger_type =extract_trigger_type (env, val); - }else { - trigger_type =get_trigger_type (env, env->trigger_cur); }switch (trigger_type) {@@ -898,6 +895,7 @@ void tdata_csr_write(CPURISCVState *env, int tdata_index, target_ulong val)break ;case TRIGGER_TYPE_INST_CNT :itrigger_reg_write (env, env->trigger_cur, tdata_index, val); + check_itrigger = true;break ;case TRIGGER_TYPE_INT :case TRIGGER_TYPE_EXCP :@@ -913,6 +911,10 @@ void tdata_csr_write(CPURISCVState *env, int tdata_index, target_ulong val)default :g_assert_not_reached (); } + +if (check_itrigger && !icount_enabled ()) { + env->itrigger_enabled =riscv_itrigger_enabled (env); + } }target_ulong tinfo_csr_read (CPURISCVState *env)Signature · 1 line
2.51.0
RFC PATCH 3/25 target/riscv/debug: Implement permissive type unavailable trigger
In reply to Nicholas PigginWhen a trigger type is "disabled", tdata2 and tdata3 must accept values that are valid for some supported trigger type. Additionally, writing 0 to tdata1 must result in the type becoming "disabled". This is important for the prescribed sequences for updating triggers.
Implement write tdata=0 -> disabled behaviour and permissive accepting of tdata2/3 values in disabled state. This implementation could be improved by checking tdata2/3 values against supported trigger types, but it is good enough to be usable by software.
From the RISC-V Debug Specification for tdata1:
Writing 0 to this register must result in a trigger that is disabled. If this trigger supports multiple types, then the hardware should disable it by changing type to 15.
and, when type=15:
This trigger is disabled. In this state, tdata2 and tdata3 can be written with any value that is supported for any of the types this trigger implements.
--- target/riscv/debug.c | 30 +++++++++++++++++++++++++++++- 1 file changed, 29 insertions(+), 1 deletion(-)
target/riscv/debug.c +29 -1
@@ -827,6 +827,28 @@ static void itrigger_reg_write(CPURISCVState *env, target_ulong index, } } +static void anytype_reg_write (CPURISCVState *env,target_ulong index, +int tdata_index,target_ulong val) +{ +/* +* This should check the value is valid for at least one of the supported +* trigger types. +*/ +switch (tdata_index) { +case TDATA1 : + env->tdata1[env->trigger_cur] = val; +break ; +case TDATA2 : + env->tdata2[env->trigger_cur] = val; +break ; +case TDATA3 : + env->tdata3[env->trigger_cur] = val; +break ; +default : +g_assert_not_reached (); + } +} +static int itrigger_get_adjust_count (CPURISCVState *env) {int count =itrigger_get_count (env, env->trigger_cur), executed;
@@ -883,6 +905,10 @@ void tdata_csr_write(CPURISCVState *env, int tdata_index, target_ulong val) }if (tdata_index ==TDATA1 ) { +if (val ==0 ) { +/* special case, writing 0 results in disabled trigger */ + val =build_tdata1 (env,TRIGGER_TYPE_UNAVAIL ,0 ,0 ); + } trigger_type =extract_trigger_type (env, val); }
@@ -897,6 +923,9 @@ void tdata_csr_write(CPURISCVState *env, int tdata_index, target_ulong val)itrigger_reg_write (env, env->trigger_cur, tdata_index, val); check_itrigger = true;break ; +case TRIGGER_TYPE_UNAVAIL : +anytype_reg_write (env, env->trigger_cur, tdata_index, val); +break ;case TRIGGER_TYPE_INT :case TRIGGER_TYPE_EXCP :case TRIGGER_TYPE_EXT_SRC :
@@ -904,7 +933,6 @@ void tdata_csr_write(CPURISCVState *env, int tdata_index, target_ulong val) trigger_type);break ;case TRIGGER_TYPE_NO_EXIST : -case TRIGGER_TYPE_UNAVAIL :qemu_log_mask (LOG_GUEST_ERROR ,"trigger type: %d does not exit\n" , trigger_type);break ;
Signature · 1 line
2.51.0
Re: [RFC PATCH 03/25] target/riscv/debug: Implement permissive type unavailable trigger
Quoted 16 lines · Nicholas Piggin
When a trigger type is "disabled", tdata2 and tdata3 must accept values that are valid for some supported trigger type. Additionally, writing 0 to tdata1 must result in the type becoming "disabled". This is important for the prescribed sequences for updating triggers.
Implement write tdata=0 -> disabled behaviour and permissive accepting of tdata2/3 values in disabled state. This implementation could be improved by checking tdata2/3 values against supported trigger types, but it is good enough to be usable by software.
From the RISC-V Debug Specification for tdata1:
Writing 0 to this register must result in a trigger that is disabled. If this trigger supports multiple types, then the hardware should disable it by changing type to 15.
Quoted 8 lines
and, when type=15:
This trigger is disabled. In this state, tdata2 and tdata3 can be written with any value that is supported for any of the types this trigger implements.
Signed-off-by Nicholas Piggin <[email protected]>---
Quoted 65 lines
target/riscv/debug.c | 30 +++++++++++++++++++++++++++++- 1 file changed, 29 insertions(+), 1 deletion(-)target/riscv/debug.c +29
-1@@ -827,6 +827,28 @@ static void itrigger_reg_write(CPURISCVState *env, target_ulong index, } } +static void anytype_reg_write (CPURISCVState *env,target_ulong index, +int tdata_index,target_ulong val) +{ +/* +* This should check the value is valid for at least one of the supported +* trigger types. +*/ +switch (tdata_index) { +case TDATA1 : + env->tdata1[env->trigger_cur] = val; +break ; +case TDATA2 : + env->tdata2[env->trigger_cur] = val; +break ; +case TDATA3 : + env->tdata3[env->trigger_cur] = val; +break ; +default : +g_assert_not_reached (); + } +} +static int itrigger_get_adjust_count (CPURISCVState *env) {int count =itrigger_get_count (env, env->trigger_cur), executed;@@ -883,6 +905,10 @@ void tdata_csr_write(CPURISCVState *env, int tdata_index, target_ulong val) }if (tdata_index ==TDATA1 ) { +if (val ==0 ) { +/* special case, writing 0 results in disabled trigger */ + val =build_tdata1 (env,TRIGGER_TYPE_UNAVAIL ,0 ,0 ); + } trigger_type =extract_trigger_type (env, val); }@@ -897,6 +923,9 @@ void tdata_csr_write(CPURISCVState *env, int tdata_index, target_ulong val)itrigger_reg_write (env, env->trigger_cur, tdata_index, val); check_itrigger = true;break ; +case TRIGGER_TYPE_UNAVAIL : +anytype_reg_write (env, env->trigger_cur, tdata_index, val); +break ;case TRIGGER_TYPE_INT :case TRIGGER_TYPE_EXCP :case TRIGGER_TYPE_EXT_SRC :@@ -904,7 +933,6 @@ void tdata_csr_write(CPURISCVState *env, int tdata_index, target_ulong val) trigger_type);break ;case TRIGGER_TYPE_NO_EXIST : -case TRIGGER_TYPE_UNAVAIL :qemu_log_mask (LOG_GUEST_ERROR ,"trigger type: %d does not exit\n" , trigger_type);break ;
Re: [RFC PATCH 03/25] target/riscv/debug: Implement permissive type unavailable trigger
In reply to Nicholas PigginQuoted 16 lines · Nicholas Piggin
When a trigger type is "disabled", tdata2 and tdata3 must accept values that are valid for some supported trigger type. Additionally, writing 0 to tdata1 must result in the type becoming "disabled". This is important for the prescribed sequences for updating triggers.
Implement write tdata=0 -> disabled behaviour and permissive accepting of tdata2/3 values in disabled state. This implementation could be improved by checking tdata2/3 values against supported trigger types, but it is good enough to be usable by software.
From the RISC-V Debug Specification for tdata1:
Writing 0 to this register must result in a trigger that is disabled. If this trigger supports multiple types, then the hardware should disable it by changing type to 15.
Quoted 7 lines
and, when type=15:
This trigger is disabled. In this state, tdata2 and tdata3 can be written with any value that is supported for any of the types this trigger implements.
Signed-off-by Nicholas Piggin <[email protected]>
Quoted 70 lines
--- target/riscv/debug.c | 30 +++++++++++++++++++++++++++++- 1 file changed, 29 insertions(+), 1 deletion(-)target/riscv/debug.c +29
-1@@ -827,6 +827,28 @@ static void itrigger_reg_write(CPURISCVState *env, target_ulong index, } } +static void anytype_reg_write (CPURISCVState *env,target_ulong index, +int tdata_index,target_ulong val) +{ +/* +* This should check the value is valid for at least one of the supported +* trigger types. +*/ +switch (tdata_index) { +case TDATA1 : + env->tdata1[env->trigger_cur] = val; +break ; +case TDATA2 : + env->tdata2[env->trigger_cur] = val; +break ; +case TDATA3 : + env->tdata3[env->trigger_cur] = val; +break ; +default : +g_assert_not_reached (); + } +} +static int itrigger_get_adjust_count (CPURISCVState *env) {int count =itrigger_get_count (env, env->trigger_cur), executed;@@ -883,6 +905,10 @@ void tdata_csr_write(CPURISCVState *env, int tdata_index, target_ulong val) }if (tdata_index ==TDATA1 ) { +if (val ==0 ) { +/* special case, writing 0 results in disabled trigger */ + val =build_tdata1 (env,TRIGGER_TYPE_UNAVAIL ,0 ,0 ); + } trigger_type =extract_trigger_type (env, val); }@@ -897,6 +923,9 @@ void tdata_csr_write(CPURISCVState *env, int tdata_index, target_ulong val)itrigger_reg_write (env, env->trigger_cur, tdata_index, val); check_itrigger = true;break ; +case TRIGGER_TYPE_UNAVAIL : +anytype_reg_write (env, env->trigger_cur, tdata_index, val); +break ;case TRIGGER_TYPE_INT :case TRIGGER_TYPE_EXCP :case TRIGGER_TYPE_EXT_SRC :@@ -904,7 +933,6 @@ void tdata_csr_write(CPURISCVState *env, int tdata_index, target_ulong val) trigger_type);break ;case TRIGGER_TYPE_NO_EXIST : -case TRIGGER_TYPE_UNAVAIL :qemu_log_mask (LOG_GUEST_ERROR ,"trigger type: %d does not exit\n" , trigger_type);break ;Signature · 1 line
2.51.0
RFC PATCH 4/25 target/riscv/debug: Fix icount trigger privilege check
In reply to Nicholas PigginThe check_itrigger_priv() did not check privilege bits properly. Move all priv checks into functions and have the icount check follow the same form as the others.
--- target/riscv/debug.c | 82 ++++++++++++++++++++------------------------ 1 file changed, 37 insertions(+), 45 deletions(-)
target/riscv/debug.c +37 -45
@@ -306,48 +306,50 @@ static void do_trigger_action(CPURISCVState *env, target_ulong trigger_index)* Check the privilege level of specific trigger matches CPU's current privilege * level. */ +static bool type2_priv_match (CPURISCVState *env,target_ulong tdata1) +{ +/* type 2 trigger cannot be fired in VU/VS mode */ +if (env->virt_enabled) { +return false; + } +/* check U/S/M bit against current privilege level */ +return (((tdata1 >>3 ) &0b1011 ) &BIT (env->priv)); +} + +static bool type6_priv_match (CPURISCVState *env,target_ulong tdata1) +{ +if (env->virt_enabled) { +/* check VU/VS bit against current privilege level */ +return (((tdata1 >>23 ) &0b11 ) &BIT (env->priv)); + }else { +/* check U/S/M bit against current privilege level */ +return (((tdata1 >>3 ) &0b1011 ) &BIT (env->priv)); + } +} + +static bool icount_priv_match (CPURISCVState *env,target_ulong tdata1) +{ +if (env->virt_enabled) { +/* check VU/VS bit against current privilege level */ +return (((tdata1 >>25 ) &0b11 ) &BIT (env->priv)); + }else { +/* check U/S/M bit against current privilege level */ +return (((tdata1 >>6 ) &0b1011 ) &BIT (env->priv)); + } +} +static bool trigger_priv_match (CPURISCVState *env,trigger_type_t type,int trigger_index) { -target_ulong ctrl = env->tdata1[trigger_index]; +target_ulong tdata1 = env->tdata1[trigger_index];switch (type) {case TRIGGER_TYPE_AD_MATCH : -/* type 2 trigger cannot be fired in VU/VS mode */ -if (env->virt_enabled) { -return false; - } -/* check U/S/M bit against current privilege level */ -if ((ctrl >>3 ) &BIT (env->priv)) { -return true; - } -break ; +return type2_priv_match (env, tdata1);case TRIGGER_TYPE_AD_MATCH6 : -if (env->virt_enabled) { -/* check VU/VS bit against current privilege level */ -if ((ctrl >>23 ) &BIT (env->priv)) { -return true; - } - }else { -/* check U/S/M bit against current privilege level */ -if ((ctrl >>3 ) &BIT (env->priv)) { -return true; - } - } -break ; +return type6_priv_match (env, tdata1);case TRIGGER_TYPE_INST_CNT : -if (env->virt_enabled) { -/* check VU/VS bit against current privilege level */ -if ((ctrl >>25 ) &BIT (env->priv)) { -return true; - } - }else { -/* check U/S/M bit against current privilege level */ -if ((ctrl >>6 ) &BIT (env->priv)) { -return true; - } - } -break ; +return icount_priv_match (env, tdata1);case TRIGGER_TYPE_INT :case TRIGGER_TYPE_EXCP :case TRIGGER_TYPE_EXT_SRC :
@@ -665,17 +667,7 @@ itrigger_set_count(CPURISCVState *env, int index, int value)static bool check_itrigger_priv (CPURISCVState *env,int index) { -target_ulong tdata1 = env->tdata1[index]; -if (env->virt_enabled) { -/* check VU/VS bit against current privilege level */ -return (get_field (tdata1,ITRIGGER_VS ) == env->priv) || - (get_field (tdata1,ITRIGGER_VU ) == env->priv); - }else { -/* check U/S/M bit against current privilege level */ -return (get_field (tdata1,ITRIGGER_M ) == env->priv) || - (get_field (tdata1,ITRIGGER_S ) == env->priv) || - (get_field (tdata1,ITRIGGER_U ) == env->priv); - } +return icount_priv_match (env, index); }bool riscv_itrigger_enabled (CPURISCVState *env)
Signature · 1 line
2.51.0
Re: [RFC PATCH 04/25] target/riscv/debug: Fix icount trigger privilege check
Quoted 6 lines · Nicholas Piggin
The
check_itrigger_priv()did not check privilege bits properly. Move all priv checks into functions and have the icount check follow the same form as the others.Signed-off-by Nicholas Piggin <[email protected]>---
Quoted 111 lines
target/riscv/debug.c | 82 ++++++++++++++++++++------------------------ 1 file changed, 37 insertions(+), 45 deletions(-)target/riscv/debug.c +37
-45@@ -306,48 +306,50 @@ static void do_trigger_action(CPURISCVState *env, target_ulong trigger_index)* Check the privilege level of specific trigger matches CPU's current privilege * level. */ +static bool type2_priv_match (CPURISCVState *env,target_ulong tdata1) +{ +/* type 2 trigger cannot be fired in VU/VS mode */ +if (env->virt_enabled) { +return false; + } +/* check U/S/M bit against current privilege level */ +return (((tdata1 >>3 ) &0b1011 ) &BIT (env->priv)); +} + +static bool type6_priv_match (CPURISCVState *env,target_ulong tdata1) +{ +if (env->virt_enabled) { +/* check VU/VS bit against current privilege level */ +return (((tdata1 >>23 ) &0b11 ) &BIT (env->priv)); + }else { +/* check U/S/M bit against current privilege level */ +return (((tdata1 >>3 ) &0b1011 ) &BIT (env->priv)); + } +} + +static bool icount_priv_match (CPURISCVState *env,target_ulong tdata1) +{ +if (env->virt_enabled) { +/* check VU/VS bit against current privilege level */ +return (((tdata1 >>25 ) &0b11 ) &BIT (env->priv)); + }else { +/* check U/S/M bit against current privilege level */ +return (((tdata1 >>6 ) &0b1011 ) &BIT (env->priv)); + } +} +static bool trigger_priv_match (CPURISCVState *env,trigger_type_t type,int trigger_index) { -target_ulong ctrl = env->tdata1[trigger_index]; +target_ulong tdata1 = env->tdata1[trigger_index];switch (type) {case TRIGGER_TYPE_AD_MATCH : -/* type 2 trigger cannot be fired in VU/VS mode */ -if (env->virt_enabled) { -return false; - } -/* check U/S/M bit against current privilege level */ -if ((ctrl >>3 ) &BIT (env->priv)) { -return true; - } -break ; +return type2_priv_match (env, tdata1);case TRIGGER_TYPE_AD_MATCH6 : -if (env->virt_enabled) { -/* check VU/VS bit against current privilege level */ -if ((ctrl >>23 ) &BIT (env->priv)) { -return true; - } - }else { -/* check U/S/M bit against current privilege level */ -if ((ctrl >>3 ) &BIT (env->priv)) { -return true; - } - } -break ; +return type6_priv_match (env, tdata1);case TRIGGER_TYPE_INST_CNT : -if (env->virt_enabled) { -/* check VU/VS bit against current privilege level */ -if ((ctrl >>25 ) &BIT (env->priv)) { -return true; - } - }else { -/* check U/S/M bit against current privilege level */ -if ((ctrl >>6 ) &BIT (env->priv)) { -return true; - } - } -break ; +return icount_priv_match (env, tdata1);case TRIGGER_TYPE_INT :case TRIGGER_TYPE_EXCP :case TRIGGER_TYPE_EXT_SRC :@@ -665,17 +667,7 @@ itrigger_set_count(CPURISCVState *env, int index, int value)static bool check_itrigger_priv (CPURISCVState *env,int index) { -target_ulong tdata1 = env->tdata1[index]; -if (env->virt_enabled) { -/* check VU/VS bit against current privilege level */ -return (get_field (tdata1,ITRIGGER_VS ) == env->priv) || - (get_field (tdata1,ITRIGGER_VU ) == env->priv); - }else { -/* check U/S/M bit against current privilege level */ -return (get_field (tdata1,ITRIGGER_M ) == env->priv) || - (get_field (tdata1,ITRIGGER_S ) == env->priv) || - (get_field (tdata1,ITRIGGER_U ) == env->priv); - } +return icount_priv_match (env, index); }bool riscv_itrigger_enabled (CPURISCVState *env)
Re: [RFC PATCH 04/25] target/riscv/debug: Fix icount trigger privilege check
In reply to Nicholas PigginQuoted 27 lines · Nicholas Piggin
The
check_itrigger_priv()did not check privilege bits properly. Move all priv checks into functions and have the icount check follow the same form as the others.Signed-off-by Nicholas Piggin <[email protected]>--- target/riscv/debug.c | 82 ++++++++++++++++++++------------------------ 1 file changed, 37 insertions(+), 45 deletions(-)target/riscv/debug.c +10
-0@@ -306,48 +306,50 @@ static void do_trigger_action(CPURISCVState *env, target_ulong trigger_index)* Check the privilege level of specific trigger matches CPU's current privilege * level. */ +static bool type2_priv_match (CPURISCVState *env,target_ulong tdata1) +{ +/* type 2 trigger cannot be fired in VU/VS mode */ +if (env->virt_enabled) { +return false; + } +/* check U/S/M bit against current privilege level */ +return (((tdata1 >>3 ) &0b1011 ) &BIT (env->priv)); +} +
target/riscv/debug.c static void do_trigger_action(CPURISCVState *env, target_ulong trigger_index) · line 328
Quoted 10 lines
target/riscv/debug.c +10
-0+static bool type6_priv_match (CPURISCVState *env,target_ulong tdata1) +{ +if (env->virt_enabled) { +/* check VU/VS bit against current privilege level */ +return (((tdata1 >>23 ) &0b11 ) &BIT (env->priv)); + }else { +/* check U/S/M bit against current privilege level */ +return (((tdata1 >>3 ) &0b1011 ) &BIT (env->priv)); + } +}
The new logic looks correct to me, but now it uses open-coded shifts and masks to check the privilege bits.
Since we already have these bits defined in debug.h, like TYPE2_U/S/M, TYPE6_VU/VS/U/S/M and ITRIGGER_VU/VS/U/S/M, I think it would be better to use those masks through a small helper instead of spelling out >> 3, >> 6, >> 23 and 0b1011 here.
That would make the code easier to read and maintain later.
like:
```
static bool trigger_priv_bit_match(target_ulong tdata1,
target_ulong u_mask,
target_ulong s_mask,
target_ulong m_mask,
target_ulong priv)
{
switch (priv) {
case PRV_U:
return (tdata1 & u_mask) != 0;
case PRV_S:
return (tdata1 & s_mask) != 0;
case PRV_M:
return (tdata1 & m_mask) != 0;
default:
g_assert_not_reached();
}
}
static bool trigger_vpriv_bit_match(target_ulong tdata1,
target_ulong vu_mask,
target_ulong vs_mask,
target_ulong priv)
{
switch (priv) {
case PRV_U:
return (tdata1 & vu_mask) != 0;
case PRV_S:
return (tdata1 & vs_mask) != 0;
default:
return false;
}
}
```So we can call it this way:
```
static bool type2_priv_match(CPURISCVState *env, target_ulong tdata1)
{
/* type 2 trigger cannot be fired in VU/VS mode */
if (env->virt_enabled) {
return false;
}
return trigger_priv_bit_match(tdata1, TYPE2_U, TYPE2_S, TYPE2_M,
env->priv);
}
static bool type6_priv_match(CPURISCVState *env, target_ulong tdata1)
{
if (env->virt_enabled) {
return trigger_vpriv_bit_match(tdata1, TYPE6_VU, TYPE6_VS,
env->priv);
}
return trigger_priv_bit_match(tdata1, TYPE6_U, TYPE6_S, TYPE6_M,
env->priv);
}
static bool icount_priv_match(CPURISCVState *env, target_ulong tdata1)
{
if (env->virt_enabled) {
return trigger_vpriv_bit_match(tdata1, ITRIGGER_VU, ITRIGGER_VS,
env->priv);
}
return trigger_priv_bit_match(tdata1, ITRIGGER_U, ITRIGGER_S, ITRIGGER_M,
env->priv);
}
```And I don't mean going back to get_field(mask) == env->priv; that pattern is easy to get wrong for PRV_U == 0. The helper should just test the mask selected by the current privilege.
Thanks,
Chao
Quoted 84 lines
target/riscv/debug.c +17
-45+ +static bool icount_priv_match (CPURISCVState *env,target_ulong tdata1) +{ +if (env->virt_enabled) { +/* check VU/VS bit against current privilege level */ +return (((tdata1 >>25 ) &0b11 ) &BIT (env->priv)); + }else { +/* check U/S/M bit against current privilege level */ +return (((tdata1 >>6 ) &0b1011 ) &BIT (env->priv)); + } +} +static bool trigger_priv_match (CPURISCVState *env,trigger_type_t type,int trigger_index) { -target_ulong ctrl = env->tdata1[trigger_index]; +target_ulong tdata1 = env->tdata1[trigger_index];switch (type) {case TRIGGER_TYPE_AD_MATCH : -/* type 2 trigger cannot be fired in VU/VS mode */ -if (env->virt_enabled) { -return false; - } -/* check U/S/M bit against current privilege level */ -if ((ctrl >>3 ) &BIT (env->priv)) { -return true; - } -break ; +return type2_priv_match (env, tdata1);case TRIGGER_TYPE_AD_MATCH6 : -if (env->virt_enabled) { -/* check VU/VS bit against current privilege level */ -if ((ctrl >>23 ) &BIT (env->priv)) { -return true; - } - }else { -/* check U/S/M bit against current privilege level */ -if ((ctrl >>3 ) &BIT (env->priv)) { -return true; - } - } -break ; +return type6_priv_match (env, tdata1);case TRIGGER_TYPE_INST_CNT : -if (env->virt_enabled) { -/* check VU/VS bit against current privilege level */ -if ((ctrl >>25 ) &BIT (env->priv)) { -return true; - } - }else { -/* check U/S/M bit against current privilege level */ -if ((ctrl >>6 ) &BIT (env->priv)) { -return true; - } - } -break ; +return icount_priv_match (env, tdata1);case TRIGGER_TYPE_INT :case TRIGGER_TYPE_EXCP :case TRIGGER_TYPE_EXT_SRC :@@ -665,17 +667,7 @@ itrigger_set_count(CPURISCVState *env, int index, int value)static bool check_itrigger_priv (CPURISCVState *env,int index) { -target_ulong tdata1 = env->tdata1[index]; -if (env->virt_enabled) { -/* check VU/VS bit against current privilege level */ -return (get_field (tdata1,ITRIGGER_VS ) == env->priv) || - (get_field (tdata1,ITRIGGER_VU ) == env->priv); - }else { -/* check U/S/M bit against current privilege level */ -return (get_field (tdata1,ITRIGGER_M ) == env->priv) || - (get_field (tdata1,ITRIGGER_S ) == env->priv) || - (get_field (tdata1,ITRIGGER_U ) == env->priv); - } +return icount_priv_match (env, index); }bool riscv_itrigger_enabled (CPURISCVState *env)Signature · 1 line
2.51.0
RFC PATCH 5/25 target/riscv/debug: Update itrigger_enabled after changing privilege
In reply to Nicholas PigginThe !icount_enabled() path should update itrigger_enabled after changing privilege, which can change whether or not an icount trigger is enabled.
--- target/riscv/cpu_helper.c | 7 +++++++ target/riscv/debug.c | 7 +++++++ target/riscv/debug.h | 3 +++ 3 files changed, 17 insertions(+)
target/riscv/cpu_helper.c +7 -0
@@ -1030,9 +1030,12 @@ void riscv_ctr_add_entry(CPURISCVState *env, target_long src, target_long dst,void riscv_cpu_set_mode (CPURISCVState *env,target_ulong newpriv,bool virt_en) { +bool change = false; +g_assert (newpriv <=PRV_M && newpriv !=PRV_RESERVED );if (newpriv != env->priv || env->virt_enabled != virt_en) { + change = true;if (icount_enabled ()) {riscv_itrigger_update_priv (env); }
@@ -1075,6 +1078,10 @@ void riscv_cpu_set_mode(CPURISCVState *env, target_ulong newpriv, bool virt_en)riscv_cpu_update_mip (env,0 ,0 ); } } + +if (change) { +riscv_cpu_debug_change_priv (env); + } }/*
target/riscv/debug.c +7 -0
@@ -1090,6 +1090,13 @@ void riscv_trigger_realize(CPURISCVState *env) } } +void riscv_cpu_debug_change_priv (CPURISCVState *env) +{ +if (!icount_enabled ()) { + env->itrigger_enabled =riscv_itrigger_enabled (env); + } +} +void riscv_trigger_reset_hold (CPURISCVState *env) {target_ulong tdata1 =build_tdata1 (env,TRIGGER_TYPE_AD_MATCH ,0 ,0 );
target/riscv/debug.h +3 -0
@@ -153,4 +153,7 @@ void riscv_trigger_reset_hold(CPURISCVState *env);bool riscv_itrigger_enabled (CPURISCVState *env);void riscv_itrigger_update_priv (CPURISCVState *env); + +void riscv_cpu_debug_change_priv (CPURISCVState *env); +#endif /* RISCV_DEBUG_H */
Signature · 1 line
2.51.0
Re: [RFC PATCH 05/25] target/riscv/debug: Update itrigger_enabled after changing privilege
Quoted 28 lines · Nicholas Piggin
The !icount_enabled() path should update
itrigger_enabledafter changing privilege, which can change whether or not an icount trigger is enabled.Signed-off-by Nicholas Piggin <[email protected]>--- target/riscv/cpu_helper.c | 7 +++++++ target/riscv/debug.c | 7 +++++++ target/riscv/debug.h | 3 +++ 3 files changed, 17 insertions(+)target/riscv/cpu_helper.c +3
-0@@ -1030,9 +1030,12 @@ void riscv_ctr_add_entry(CPURISCVState *env, target_long src, target_long dst,void riscv_cpu_set_mode (CPURISCVState *env,target_ulong newpriv,bool virt_en) { +bool change = false; +g_assert (newpriv <=PRV_M && newpriv !=PRV_RESERVED );if (newpriv != env->priv || env->virt_enabled != virt_en) { + change = true;if (icount_enabled ()) {riscv_itrigger_update_priv (env); }
target/riscv/cpu_helper.c void riscv_cpu_set_mode(CPURISCVState *env, target_ulong newpriv, bool virt_en) · line 1084
Quoted 7 lines
target/riscv/cpu_helper.c +4
-0@@ -1075,6 +1078,10 @@ void riscv_cpu_set_mode(CPURISCVState *env, target_ulong newpriv, bool virt_en)riscv_cpu_update_mip (env,0 ,0 ); } } + +if (change) { +riscv_cpu_debug_change_priv (env); + }
Is there a particular reason to do this change at the end of cpu_set_mode()? Because if there isn't, we can remove the 'change' flag, then just do a:
target/riscv/cpu_helper.c void riscv_ctr_add_entry(CPURISCVState *env, target_long src, target_long dst, · line 1041
Quoted 7 lines
target/riscv/cpu_helper.c +3
-0if (newpriv != env->priv || env->virt_enabled != virt_en) { +if (!icount_enabled ()) { + env->itrigger_enabled =riscv_itrigger_enabled (env); + }if (icount_enabled ()) {riscv_itrigger_update_priv (env); }
You'll remove the icount_enabled() branch in patch 16 anyway so I wouldn't bother do a if/then/else with icount_enabled(), but can be done too.
This will also get rid of the riscv_cpu_debug_change_priv() function added here. Which is fine: from what I can see in the patches this is the only place that it would be used, so no harm done.
Thanks,
Daniel
Quoted 33 lines
}/*
target/riscv/debug.c +7
-0@@ -1090,6 +1090,13 @@ void riscv_trigger_realize(CPURISCVState *env) } } +void riscv_cpu_debug_change_priv (CPURISCVState *env) +{ +if (!icount_enabled ()) { + env->itrigger_enabled =riscv_itrigger_enabled (env); + } +} +void riscv_trigger_reset_hold (CPURISCVState *env) {target_ulong tdata1 =build_tdata1 (env,TRIGGER_TYPE_AD_MATCH ,0 ,0 );target/riscv/debug.h +3
-0@@ -153,4 +153,7 @@ void riscv_trigger_reset_hold(CPURISCVState *env);bool riscv_itrigger_enabled (CPURISCVState *env);void riscv_itrigger_update_priv (CPURISCVState *env); + +void riscv_cpu_debug_change_priv (CPURISCVState *env); +#endif /* RISCV_DEBUG_H */
RFC PATCH 6/25 target/riscv/debug: Implement get_trigger_action for icount type trigger
In reply to Nicholas Piggin--- target/riscv/debug.c | 2 ++ 1 file changed, 2 insertions(+)
target/riscv/debug.c +2 -0
@@ -110,6 +110,8 @@ static trigger_action_t get_trigger_action(CPURISCVState *env, action = (tdata1 &TYPE6_ACTION ) >>12 ;break ;case TRIGGER_TYPE_INST_CNT : + action = (tdata1 &ITRIGGER_ACTION ); +break ;case TRIGGER_TYPE_INT :case TRIGGER_TYPE_EXCP :case TRIGGER_TYPE_EXT_SRC :
Signature · 1 line
2.51.0
Re: [RFC PATCH 06/25] target/riscv/debug: Implement get_trigger_action for icount type trigger
Quoted 2 lines · Nicholas Piggin
Signed-off-by Nicholas Piggin <[email protected]>---
Quoted 16 lines
target/riscv/debug.c | 2 ++ 1 file changed, 2 insertions(+)target/riscv/debug.c +2
-0@@ -110,6 +110,8 @@ static trigger_action_t get_trigger_action(CPURISCVState *env, action = (tdata1 &TYPE6_ACTION ) >>12 ;break ;case TRIGGER_TYPE_INST_CNT : + action = (tdata1 &ITRIGGER_ACTION ); +break ;case TRIGGER_TYPE_INT :case TRIGGER_TYPE_EXCP :case TRIGGER_TYPE_EXT_SRC :
RFC PATCH 7/25 target/riscv/debug: Fix migration post_load icount_enabled() test
In reply to Nicholas Pigginitrigger_enabled is relevant to the !icount_enabled() path, the test in debug_post_load() is inverted.
--- target/riscv/machine.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
target/riscv/machine.c +1 -1
@@ -230,7 +230,7 @@ static int debug_post_load(void *opaque, int version_id)RISCVCPU *cpu = opaque;CPURISCVState *env = &cpu->env; -if (icount_enabled ()) { +if (!icount_enabled ()) { env->itrigger_enabled =riscv_itrigger_enabled (env); }
Signature · 1 line
2.51.0
Re: [RFC PATCH 07/25] target/riscv/debug: Fix migration post_load icount_enabled() test
Quoted 5 lines · Nicholas Piggin
itrigger_enabledis relevant to the !icount_enabled() path, the test indebug_post_load()is inverted.Signed-off-by Nicholas Piggin <[email protected]>---
Quoted 16 lines
target/riscv/machine.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)target/riscv/machine.c +1
-1@@ -230,7 +230,7 @@ static int debug_post_load(void *opaque, int version_id)RISCVCPU *cpu = opaque;CPURISCVState *env = &cpu->env; -if (icount_enabled ()) { +if (!icount_enabled ()) { env->itrigger_enabled =riscv_itrigger_enabled (env); }
RFC PATCH 8/25 target/riscv/debug: Fix icount privilege matching icount_enabled() test
In reply to Nicholas Pigginitrigger_enabled should be set if the privilege does match. The test is inverted.
--- target/riscv/debug.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
target/riscv/debug.c +1 -1
@@ -679,7 +679,7 @@ bool riscv_itrigger_enabled(CPURISCVState *env)if (get_trigger_type (env, i) !=TRIGGER_TYPE_INST_CNT ) {continue ; } -if (check_itrigger_priv (env, i)) { +if (!check_itrigger_priv (env, i)) {continue ; } count =itrigger_get_count (env, i);
Signature · 1 line
2.51.0
Re: [RFC PATCH 08/25] target/riscv/debug: Fix icount privilege matching icount_enabled() test
Quoted 5 lines · Nicholas Piggin
itrigger_enabledshould be set if the privilege does match. The test is inverted.Signed-off-by Nicholas Piggin <[email protected]>---
Quoted 16 lines
target/riscv/debug.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)target/riscv/debug.c +1
-1@@ -679,7 +679,7 @@ bool riscv_itrigger_enabled(CPURISCVState *env)if (get_trigger_type (env, i) !=TRIGGER_TYPE_INST_CNT ) {continue ; } -if (check_itrigger_priv (env, i)) { +if (!check_itrigger_priv (env, i)) {continue ; } count =itrigger_get_count (env, i);
RFC PATCH 9/25 target/riscv/debug: Implement icount trigger textra matching
In reply to Nicholas Pigginicount triggers are subject to textra checks, per spec.
--- target/riscv/debug.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
target/riscv/debug.c +2 -1
@@ -675,11 +675,12 @@ static bool check_itrigger_priv(CPURISCVState *env, int index)bool riscv_itrigger_enabled (CPURISCVState *env) {int count; +for (int i =0 ; i <RV_MAX_TRIGGERS ; i++) {if (get_trigger_type (env, i) !=TRIGGER_TYPE_INST_CNT ) {continue ; } -if (!check_itrigger_priv (env, i)) { +if (!trigger_common_match (env,TRIGGER_TYPE_INST_CNT , i)) {continue ; } count =itrigger_get_count (env, i);
Signature · 1 line
2.51.0
Re: [RFC PATCH 09/25] target/riscv/debug: Implement icount trigger textra matching
Quoted 4 lines · Nicholas Piggin
icount triggers are subject to textra checks, per spec.
Signed-off-by Nicholas Piggin <[email protected]>---
Quoted 21 lines
target/riscv/debug.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)target/riscv/debug.c +2
-1@@ -675,11 +675,12 @@ static bool check_itrigger_priv(CPURISCVState *env, int index)bool riscv_itrigger_enabled (CPURISCVState *env) {int count; +for (int i =0 ; i <RV_MAX_TRIGGERS ; i++) {if (get_trigger_type (env, i) !=TRIGGER_TYPE_INST_CNT ) {continue ; } -if (!check_itrigger_priv (env, i)) { +if (!trigger_common_match (env,TRIGGER_TYPE_INST_CNT , i)) {continue ; } count =itrigger_get_count (env, i);
RFC PATCH 10/25 target/riscv/debug: Maintain itrigger_enabled in helper_itrigger_match()
In reply to Nicholas Pigginhelper_itrigger_match() iterates over all triggers so it can maintain itrigger_enabled rather than calling riscv_itrigger_enabled() which becomes O(N^2) where N is the number of icount triggers.
--- target/riscv/debug.c | 13 ++++++++++++- 1 file changed, 12 insertions(+), 1 deletion(-)
target/riscv/debug.c +12 -1
@@ -693,9 +693,18 @@ bool riscv_itrigger_enabled(CPURISCVState *env)return false; } +/* +* This is called by TCG when an instruction completes. +* TCG runs in single-step mode when itrigger_enabled = true, so +* it can call after each insn. +*/ void helper_itrigger_match (CPURISCVState *env) {int count; +bool enabled = false; + +g_assert (env->itrigger_enabled); +for (int i =0 ; i <RV_MAX_TRIGGERS ; i++) {if (get_trigger_type (env, i) !=TRIGGER_TYPE_INST_CNT ) {continue ;
@@ -709,10 +718,12 @@ void helper_itrigger_match(CPURISCVState *env) }itrigger_set_count (env, i, count--);if (!count) { - env->itrigger_enabled =riscv_itrigger_enabled (env);do_trigger_action (env, i); + }else { + enabled = true; } } + env->itrigger_enabled = enabled; }static void riscv_itrigger_update_count (CPURISCVState *env)
Signature · 1 line
2.51.0
Re: [RFC PATCH 10/25] target/riscv/debug: Maintain itrigger_enabled in helper_itrigger_match()
Quoted 6 lines · Nicholas Piggin
helper_itrigger_match()iterates over all triggers so it can maintainitrigger_enabledrather than callingriscv_itrigger_enabled()which becomes O(N^2) where N is the number of icount triggers.Signed-off-by Nicholas Piggin <[email protected]>---
Quoted 40 lines
target/riscv/debug.c | 13 ++++++++++++- 1 file changed, 12 insertions(+), 1 deletion(-)target/riscv/debug.c +12
-1@@ -693,9 +693,18 @@ bool riscv_itrigger_enabled(CPURISCVState *env)return false; } +/* +* This is called by TCG when an instruction completes. +* TCG runs in single-step mode when itrigger_enabled = true, so +* it can call after each insn. +*/ void helper_itrigger_match (CPURISCVState *env) {int count; +bool enabled = false; + +g_assert (env->itrigger_enabled); +for (int i =0 ; i <RV_MAX_TRIGGERS ; i++) {if (get_trigger_type (env, i) !=TRIGGER_TYPE_INST_CNT ) {continue ;@@ -709,10 +718,12 @@ void helper_itrigger_match(CPURISCVState *env) }itrigger_set_count (env, i, count--);if (!count) { - env->itrigger_enabled =riscv_itrigger_enabled (env);do_trigger_action (env, i); + }else { + enabled = true; } } + env->itrigger_enabled = enabled; }static void riscv_itrigger_update_count (CPURISCVState *env)
RFC PATCH 11/25 target/riscv/debug: Fix breakpoint matching action
In reply to Nicholas PigginThe debug exception callback is too late to find the action for the trigger(s) which caused it, and it is actually passing the wrong thing to do_trigger_action(), which expects a trigger index but is given DBG_ACTION_BP (which will be interpreted as trigger 0 and use the action set for that trigger).
It could be possible to derive the trigger index from the bp/wp address, but that is clunky and it is really the action that determines whether an exception should be raised, also multiple triggers may perform their actions in the same cycle, so it is more consistent to check action during the breakpoint matching phase. If a breakpoint exception is to be taken then that is signaled at that time.
--- target/riscv/debug.c | 35 +++++++++++++++++++++++------------ 1 file changed, 23 insertions(+), 12 deletions(-)
target/riscv/debug.c +23 -12
@@ -280,7 +280,8 @@ static target_ulong textra_validate(CPURISCVState *env, target_ulong tdata3)return textra; } -static void do_trigger_action (CPURISCVState *env,target_ulong trigger_index) +/* Return true if an exception should be raised */ +static bool do_trigger_action (CPURISCVState *env,target_ulong trigger_index) {trigger_action_t action =get_trigger_action (env, trigger_index);
@@ -288,8 +289,7 @@ static void do_trigger_action(CPURISCVState *env, target_ulong trigger_index)case DBG_ACTION_NONE :break ;case DBG_ACTION_BP : -riscv_raise_exception (env,RISCV_EXCP_BREAKPOINT ,0 ); -break ; +return true;case DBG_ACTION_DBG_MODE :case DBG_ACTION_TRACE0 :case DBG_ACTION_TRACE1 :
@@ -302,6 +302,7 @@ static void do_trigger_action(CPURISCVState *env, target_ulong trigger_index)default :g_assert_not_reached (); } +return false; }/*
@@ -718,7 +719,9 @@ void helper_itrigger_match(CPURISCVState *env) }itrigger_set_count (env, i, count--);if (!count) { -do_trigger_action (env, i); +if (do_trigger_action (env, i)) { +riscv_raise_exception (env,RISCV_EXCP_BREAKPOINT ,0 ); + } }else { enabled = true; }
@@ -965,11 +968,11 @@ void riscv_cpu_debug_excp_handler(CPUState *cs)if (cs->watchpoint_hit) {if (cs->watchpoint_hit->flags &BP_CPU ) { -do_trigger_action (env,DBG_ACTION_BP ); +riscv_raise_exception (env,RISCV_EXCP_BREAKPOINT ,0 ); } }else {if (cpu_breakpoint_test (cs, env->pc,BP_CPU )) { -do_trigger_action (env,DBG_ACTION_BP ); +riscv_raise_exception (env,RISCV_EXCP_BREAKPOINT ,0 ); } } }
@@ -1006,8 +1009,10 @@ bool riscv_cpu_debug_check_breakpoint(CPUState *cs) pc = env->tdata2[i];if ((ctrl &TYPE2_EXEC ) && (bp->pc == pc)) { - env->badaddr = pc; -return true; +if (do_trigger_action (env, i)) { + env->badaddr = pc; +return true; + } }break ;case TRIGGER_TYPE_AD_MATCH6 :
@@ -1015,8 +1020,10 @@ bool riscv_cpu_debug_check_breakpoint(CPUState *cs) pc = env->tdata2[i];if ((ctrl &TYPE6_EXEC ) && (bp->pc == pc)) { - env->badaddr = pc; -return true; +if (do_trigger_action (env, i)) { + env->badaddr = pc; +return true; + } }break ;default :
@@ -1067,7 +1074,9 @@ bool riscv_cpu_debug_check_watchpoint(CPUState *cs, CPUWatchpoint *wp) }if ((wp->flags & flags) && (wp->vaddr == addr)) { -return true; +if (do_trigger_action (env, i)) { +return true; + } }break ;case TRIGGER_TYPE_AD_MATCH6 :
@@ -1083,7 +1092,9 @@ bool riscv_cpu_debug_check_watchpoint(CPUState *cs, CPUWatchpoint *wp) }if ((wp->flags & flags) && (wp->vaddr == addr)) { -return true; +if (do_trigger_action (env, i)) { +return true; + } }break ;default :
Signature · 1 line
2.51.0
Re: [RFC PATCH 11/25] target/riscv/debug: Fix breakpoint matching action
Quoted 13 lines · Nicholas Piggin
The debug exception callback is too late to find the action for the trigger(s) which caused it, and it is actually passing the wrong thing to
do_trigger_action(), which expects a trigger index but is givenDBG_ACTION_BP(which will be interpreted as trigger 0 and use the action set for that trigger).It could be possible to derive the trigger index from the bp/wp address, but that is clunky and it is really the action that determines whether an exception should be raised, also multiple triggers may perform their actions in the same cycle, so it is more consistent to check action during the breakpoint matching phase. If a breakpoint exception is to be taken then that is signaled at that time.
Quoted 2 lines
Signed-off-by Nicholas Piggin <[email protected]>---
Quoted 108 lines
target/riscv/debug.c | 35 +++++++++++++++++++++++------------ 1 file changed, 23 insertions(+), 12 deletions(-)target/riscv/debug.c +23
-12@@ -280,7 +280,8 @@ static target_ulong textra_validate(CPURISCVState *env, target_ulong tdata3)return textra; } -static void do_trigger_action (CPURISCVState *env,target_ulong trigger_index) +/* Return true if an exception should be raised */ +static bool do_trigger_action (CPURISCVState *env,target_ulong trigger_index) {trigger_action_t action =get_trigger_action (env, trigger_index);@@ -288,8 +289,7 @@ static void do_trigger_action(CPURISCVState *env, target_ulong trigger_index)case DBG_ACTION_NONE :break ;case DBG_ACTION_BP : -riscv_raise_exception (env,RISCV_EXCP_BREAKPOINT ,0 ); -break ; +return true;case DBG_ACTION_DBG_MODE :case DBG_ACTION_TRACE0 :case DBG_ACTION_TRACE1 :@@ -302,6 +302,7 @@ static void do_trigger_action(CPURISCVState *env, target_ulong trigger_index)default :g_assert_not_reached (); } +return false; }/* @@ -718,7 +719,9 @@ void helper_itrigger_match(CPURISCVState *env) }itrigger_set_count (env, i, count--);if (!count) { -do_trigger_action (env, i); +if (do_trigger_action (env, i)) { +riscv_raise_exception (env,RISCV_EXCP_BREAKPOINT ,0 ); + } }else { enabled = true; }@@ -965,11 +968,11 @@ void riscv_cpu_debug_excp_handler(CPUState *cs)if (cs->watchpoint_hit) {if (cs->watchpoint_hit->flags &BP_CPU ) { -do_trigger_action (env,DBG_ACTION_BP ); +riscv_raise_exception (env,RISCV_EXCP_BREAKPOINT ,0 ); } }else {if (cpu_breakpoint_test (cs, env->pc,BP_CPU )) { -do_trigger_action (env,DBG_ACTION_BP ); +riscv_raise_exception (env,RISCV_EXCP_BREAKPOINT ,0 ); } } }@@ -1006,8 +1009,10 @@ bool riscv_cpu_debug_check_breakpoint(CPUState *cs) pc = env->tdata2[i];if ((ctrl &TYPE2_EXEC ) && (bp->pc == pc)) { - env->badaddr = pc; -return true; +if (do_trigger_action (env, i)) { + env->badaddr = pc; +return true; + } }break ;case TRIGGER_TYPE_AD_MATCH6 :@@ -1015,8 +1020,10 @@ bool riscv_cpu_debug_check_breakpoint(CPUState *cs) pc = env->tdata2[i];if ((ctrl &TYPE6_EXEC ) && (bp->pc == pc)) { - env->badaddr = pc; -return true; +if (do_trigger_action (env, i)) { + env->badaddr = pc; +return true; + } }break ;default :@@ -1067,7 +1074,9 @@ bool riscv_cpu_debug_check_watchpoint(CPUState *cs, CPUWatchpoint *wp) }if ((wp->flags & flags) && (wp->vaddr == addr)) { -return true; +if (do_trigger_action (env, i)) { +return true; + } }break ;case TRIGGER_TYPE_AD_MATCH6 :@@ -1083,7 +1092,9 @@ bool riscv_cpu_debug_check_watchpoint(CPUState *cs, CPUWatchpoint *wp) }if ((wp->flags & flags) && (wp->vaddr == addr)) { -return true; +if (do_trigger_action (env, i)) { +return true; + } }break ;default :
RFC PATCH 12/25 target/riscv/debug: Put mcontrol load/store match address into tval
In reply to Nicholas PigginWhen an mc/mc6 trigger triggers on a load/store and raises a breakpoint exception, put the matching memory address into tval, as is already done for exec addresses matching.
Tval for breakpoint exceptions is optional in the spec, but it seems more consistent to implement it for load/store if exec is implemented.
--- target/riscv/debug.c | 2 ++ 1 file changed, 2 insertions(+)
target/riscv/debug.c +2 -0
@@ -1075,6 +1075,7 @@ bool riscv_cpu_debug_check_watchpoint(CPUState *cs, CPUWatchpoint *wp)if ((wp->flags & flags) && (wp->vaddr == addr)) {if (do_trigger_action (env, i)) { + env->badaddr = wp->vaddr;return true; } }
@@ -1093,6 +1094,7 @@ bool riscv_cpu_debug_check_watchpoint(CPUState *cs, CPUWatchpoint *wp)if ((wp->flags & flags) && (wp->vaddr == addr)) {if (do_trigger_action (env, i)) { + env->badaddr = wp->vaddr;return true; } }
Signature · 1 line
2.51.0
Re: [RFC PATCH 12/25] target/riscv/debug: Put mcontrol load/store match address into tval
Quoted 6 lines · Nicholas Piggin
When an mc/mc6 trigger triggers on a load/store and raises a breakpoint exception, put the matching memory address into tval, as is already done for exec addresses matching.
Tval for breakpoint exceptions is optional in the spec, but it seems more consistent to implement it for load/store if exec is implemented.
There is a non-written standard that we follow in qemu-riscv where we always tend to supply these tval style addresses for exceptions/interrupts, even if the spec says it's s optional. So yeah, you are correct.
Quoted 3 lines
Signed-off-by Nicholas Piggin <[email protected]>---
Quoted 23 lines
target/riscv/debug.c | 2 ++ 1 file changed, 2 insertions(+)target/riscv/debug.c +2
-0@@ -1075,6 +1075,7 @@ bool riscv_cpu_debug_check_watchpoint(CPUState *cs, CPUWatchpoint *wp)if ((wp->flags & flags) && (wp->vaddr == addr)) {if (do_trigger_action (env, i)) { + env->badaddr = wp->vaddr;return true; } }@@ -1093,6 +1094,7 @@ bool riscv_cpu_debug_check_watchpoint(CPUState *cs, CPUWatchpoint *wp)if ((wp->flags & flags) && (wp->vaddr == addr)) {if (do_trigger_action (env, i)) { + env->badaddr = wp->vaddr;return true; } }
RFC PATCH 13/25 target/riscv/debug: Remove breakpoints on reset
In reply to Nicholas PigginRemove TCG breakpoints/watchpoints when the CPU is reset. Currently they get lost.
--- target/riscv/debug.c | 15 +++++++++++++-- 1 file changed, 13 insertions(+), 2 deletions(-)
target/riscv/debug.c +13 -2
@@ -1131,6 +1131,19 @@ void riscv_trigger_reset_hold(CPURISCVState *env)/* init to type 2 triggers */ for (i =0 ; i <RV_MAX_TRIGGERS ; i++) { +int trigger_type =get_trigger_type (env, i); + +switch (trigger_type) { +case TRIGGER_TYPE_AD_MATCH : +type2_breakpoint_remove (env, i); +break ; +case TRIGGER_TYPE_AD_MATCH6 : +type6_breakpoint_remove (env, i); +break ; +default : +break ; + } +/* * type = TRIGGER_TYPE_AD_MATCH * dmode = 0 (both debug and M-mode can write tdata)
@@ -1147,8 +1160,6 @@ void riscv_trigger_reset_hold(CPURISCVState *env) env->tdata1[i] = tdata1; env->tdata2[i] =0 ; env->tdata3[i] =0 ; - env->cpu_breakpoint[i] =NULL ; - env->cpu_watchpoint[i] =NULL ;timer_del (env->itrigger_timer[i]); }
Signature · 1 line
2.51.0
Re: [RFC PATCH 13/25] target/riscv/debug: Remove breakpoints on reset
Quoted 5 lines · Nicholas Piggin
Remove TCG breakpoints/watchpoints when the CPU is reset. Currently they get lost.
Signed-off-by Nicholas Piggin <[email protected]>---
Quoted 36 lines
target/riscv/debug.c | 15 +++++++++++++-- 1 file changed, 13 insertions(+), 2 deletions(-)target/riscv/debug.c +13
-2@@ -1131,6 +1131,19 @@ void riscv_trigger_reset_hold(CPURISCVState *env)/* init to type 2 triggers */ for (i =0 ; i <RV_MAX_TRIGGERS ; i++) { +int trigger_type =get_trigger_type (env, i); + +switch (trigger_type) { +case TRIGGER_TYPE_AD_MATCH : +type2_breakpoint_remove (env, i); +break ; +case TRIGGER_TYPE_AD_MATCH6 : +type6_breakpoint_remove (env, i); +break ; +default : +break ; + } +/* * type = TRIGGER_TYPE_AD_MATCH * dmode = 0 (both debug and M-mode can write tdata) @@ -1147,8 +1160,6 @@ void riscv_trigger_reset_hold(CPURISCVState *env) env->tdata1[i] = tdata1; env->tdata2[i] =0 ; env->tdata3[i] =0 ; - env->cpu_breakpoint[i] =NULL ; - env->cpu_watchpoint[i] =NULL ;timer_del (env->itrigger_timer[i]); }
RFC PATCH 14/25 target/riscv/debug: Move debug CPU post_load details into debug.c
In reply to Nicholas PigginNo need for machine.c to know internals of how debug.c maintains debug state.
--- target/riscv/debug.c | 9 ++++++++- target/riscv/debug.h | 2 +- target/riscv/machine.c | 4 +--- 3 files changed, 10 insertions(+), 5 deletions(-)
target/riscv/debug.c +8 -1
@@ -673,7 +673,7 @@ static bool check_itrigger_priv(CPURISCVState *env, int index)return icount_priv_match (env, index); } -bool riscv_itrigger_enabled (CPURISCVState *env) +static bool riscv_itrigger_enabled (CPURISCVState *env) {int count;
@@ -1124,6 +1124,13 @@ void riscv_cpu_debug_change_priv(CPURISCVState *env) } } +void riscv_cpu_debug_post_load (CPURISCVState *env) +{ +if (!icount_enabled ()) { + env->itrigger_enabled =riscv_itrigger_enabled (env); + } +} +void riscv_trigger_reset_hold (CPURISCVState *env) {target_ulong tdata1 =build_tdata1 (env,TRIGGER_TYPE_AD_MATCH ,0 ,0 );
target/riscv/debug.h +1 -1
@@ -151,9 +151,9 @@ bool riscv_cpu_debug_check_watchpoint(CPUState *cs, CPUWatchpoint *wp);void riscv_trigger_realize (CPURISCVState *env);void riscv_trigger_reset_hold (CPURISCVState *env); -bool riscv_itrigger_enabled (CPURISCVState *env);void riscv_itrigger_update_priv (CPURISCVState *env);void riscv_cpu_debug_change_priv (CPURISCVState *env); +void riscv_cpu_debug_post_load (CPURISCVState *env);#endif /* RISCV_DEBUG_H */
target/riscv/machine.c +1 -3
@@ -230,9 +230,7 @@ static int debug_post_load(void *opaque, int version_id)RISCVCPU *cpu = opaque;CPURISCVState *env = &cpu->env; -if (!icount_enabled ()) { - env->itrigger_enabled =riscv_itrigger_enabled (env); - } +riscv_cpu_debug_post_load (env);return 0 ; }
Signature · 1 line
2.51.0
Re: [RFC PATCH 14/25] target/riscv/debug: Move debug CPU post_load details into debug.c
Quoted 5 lines · Nicholas Piggin
No need for
machine.cto know internals of howdebug.cmaintains debug state.Signed-off-by Nicholas Piggin <[email protected]>---
Quoted 62 lines
target/riscv/debug.c | 9 ++++++++- target/riscv/debug.h | 2 +- target/riscv/machine.c | 4 +--- 3 files changed, 10 insertions(+), 5 deletions(-)target/riscv/debug.c +8
-1@@ -673,7 +673,7 @@ static bool check_itrigger_priv(CPURISCVState *env, int index)return icount_priv_match (env, index); } -bool riscv_itrigger_enabled (CPURISCVState *env) +static bool riscv_itrigger_enabled (CPURISCVState *env) {int count;@@ -1124,6 +1124,13 @@ void riscv_cpu_debug_change_priv(CPURISCVState *env) } } +void riscv_cpu_debug_post_load (CPURISCVState *env) +{ +if (!icount_enabled ()) { + env->itrigger_enabled =riscv_itrigger_enabled (env); + } +} +void riscv_trigger_reset_hold (CPURISCVState *env) {target_ulong tdata1 =build_tdata1 (env,TRIGGER_TYPE_AD_MATCH ,0 ,0 );target/riscv/debug.h +1
-1@@ -151,9 +151,9 @@ bool riscv_cpu_debug_check_watchpoint(CPUState *cs, CPUWatchpoint *wp);void riscv_trigger_realize (CPURISCVState *env);void riscv_trigger_reset_hold (CPURISCVState *env); -bool riscv_itrigger_enabled (CPURISCVState *env);void riscv_itrigger_update_priv (CPURISCVState *env);void riscv_cpu_debug_change_priv (CPURISCVState *env); +void riscv_cpu_debug_post_load (CPURISCVState *env);#endif /* RISCV_DEBUG_H */ target/riscv/machine.c +1
-3@@ -230,9 +230,7 @@ static int debug_post_load(void *opaque, int version_id)RISCVCPU *cpu = opaque;CPURISCVState *env = &cpu->env; -if (!icount_enabled ()) { - env->itrigger_enabled =riscv_itrigger_enabled (env); - } +riscv_cpu_debug_post_load (env);return 0 ; }
RFC PATCH 15/25 target/riscv/debug: Insert breakpoints after migration
In reply to Nicholas PigginAfter migration, TCG breakpoints should be re-established based on debug register state.
icount trigger timer in the icount_enabled() path is also missing, but that path will be removed subsequently so don't bother to fix it.
--- target/riscv/debug.c | 14 ++++++++++++++ 1 file changed, 14 insertions(+)
target/riscv/debug.c +14 -0
@@ -1126,6 +1126,20 @@ void riscv_cpu_debug_change_priv(CPURISCVState *env)void riscv_cpu_debug_post_load (CPURISCVState *env) { +for (int i =0 ; i <RV_MAX_TRIGGERS ; i++) { +int trigger_type =get_trigger_type (env, i); + +switch (trigger_type) { +case TRIGGER_TYPE_AD_MATCH : +type2_breakpoint_insert (env, i); +break ; +case TRIGGER_TYPE_AD_MATCH6 : +type6_breakpoint_insert (env, i); +break ; +default : +break ; + } + }if (!icount_enabled ()) { env->itrigger_enabled =riscv_itrigger_enabled (env); }
Signature · 1 line
2.51.0
Re: [RFC PATCH 15/25] target/riscv/debug: Insert breakpoints after migration
Quoted 9 lines · Nicholas Piggin
After migration, TCG breakpoints should be re-established based on debug register state.
icount trigger timer in the
icount_enabled()path is also missing, but that path will be removed subsequently so don't bother to fix it.Signed-off-by Nicholas Piggin <[email protected]>---
Quoted 28 lines
target/riscv/debug.c | 14 ++++++++++++++ 1 file changed, 14 insertions(+)target/riscv/debug.c +14
-0@@ -1126,6 +1126,20 @@ void riscv_cpu_debug_change_priv(CPURISCVState *env)void riscv_cpu_debug_post_load (CPURISCVState *env) { +for (int i =0 ; i <RV_MAX_TRIGGERS ; i++) { +int trigger_type =get_trigger_type (env, i); + +switch (trigger_type) { +case TRIGGER_TYPE_AD_MATCH : +type2_breakpoint_insert (env, i); +break ; +case TRIGGER_TYPE_AD_MATCH6 : +type6_breakpoint_insert (env, i); +break ; +default : +break ; + } + }if (!icount_enabled ()) { env->itrigger_enabled =riscv_itrigger_enabled (env); }
RFC PATCH 16/25 target/riscv/debug: Remove itrigger icount-enabled mode
In reply to Nicholas PigginWhen QEMU icount is enabled, the riscv icount trigger facility is implemented cleverly using precise counting timers rather than single-stepping TCG.
I found this possibly has some bugs, it is a bit complicated and splits testing between icount and !icount, and icount enabled is not the important case for performance. Therefore remove the separate icount enabled path.
--- target/riscv/cpu.c | 6 -- target/riscv/cpu.h | 2 - target/riscv/cpu_helper.c | 3 - target/riscv/debug.c | 115 ++----------------------------------- target/riscv/debug.h | 3 - target/riscv/tcg/tcg-cpu.c | 2 +- 6 files changed, 5 insertions(+), 126 deletions(-)
target/riscv/cpu.c +0 -6
@@ -942,12 +942,6 @@ static void riscv_cpu_realize(DeviceState *dev, Error **errp)riscv_cpu_register_gdb_regs_for_features (cs); -#ifndef CONFIG_USER_ONLY -if (cpu->cfg.debug) { -riscv_trigger_realize (&cpu->env); - } -#endif -qemu_init_vcpu (cs);cpu_reset (cs);
target/riscv/cpu.h +0 -2
@@ -450,8 +450,6 @@ struct CPUArchState {target_ulong mcontext;struct CPUBreakpoint *cpu_breakpoint[RV_MAX_TRIGGERS ];struct CPUWatchpoint *cpu_watchpoint[RV_MAX_TRIGGERS ]; -QEMUTimer *itrigger_timer[RV_MAX_TRIGGERS ]; -int64_t last_icount;bool itrigger_enabled;/* machine specific rdtime callback */
target/riscv/cpu_helper.c +0 -3
@@ -1036,9 +1036,6 @@ void riscv_cpu_set_mode(CPURISCVState *env, target_ulong newpriv, bool virt_en)if (newpriv != env->priv || env->virt_enabled != virt_en) { change = true; -if (icount_enabled ()) { -riscv_itrigger_update_priv (env); - }riscv_pmu_update_fixed_ctrs (env, newpriv, virt_en); }
target/riscv/debug.c +4 -111
177 lines not on this page · raw message
target/riscv/debug.h +0 -3
@@ -148,11 +148,8 @@ void riscv_cpu_debug_excp_handler(CPUState *cs);bool riscv_cpu_debug_check_breakpoint (CPUState *cs);bool riscv_cpu_debug_check_watchpoint (CPUState *cs,CPUWatchpoint *wp); -void riscv_trigger_realize (CPURISCVState *env);void riscv_trigger_reset_hold (CPURISCVState *env); -void riscv_itrigger_update_priv (CPURISCVState *env); -void riscv_cpu_debug_change_priv (CPURISCVState *env);void riscv_cpu_debug_post_load (CPURISCVState *env);
target/riscv/tcg/tcg-cpu.c +1 -1
@@ -177,7 +177,7 @@ static TCGTBCPUState riscv_get_tb_cpu_state(CPUState *cs) ? EXT_STATUS_DIRTY : EXT_STATUS_DISABLED; } -if (cpu->cfg.debug && !icount_enabled ()) { +if (cpu->cfg.debug) { flags =FIELD_DP32 (flags,TB_FLAGS ,ITRIGGER , env->itrigger_enabled); }#endif
Signature · 1 line
2.51.0
RFC PATCH 17/25 target/riscv/debug: Advertise icount trigger type in tinfo
In reply to Nicholas Pigginicount triggers are supported so tinfo should advertise them.
--- target/riscv/debug.c | 2 ++ 1 file changed, 2 insertions(+)
target/riscv/debug.c +2 -0
@@ -865,7 +865,9 @@ void tdata_csr_write(CPURISCVState *env, int tdata_index, target_ulong val)target_ulong tinfo_csr_read (CPURISCVState *env) {/* assume all triggers support the same types of triggers */ +/* XXX: should we set 1 (version 1.0) in the version field? */ return BIT (TRIGGER_TYPE_AD_MATCH ) | +BIT (TRIGGER_TYPE_INST_CNT ) |BIT (TRIGGER_TYPE_AD_MATCH6 ); }
Signature · 1 line
2.51.0
Re: [RFC PATCH 17/25] target/riscv/debug: Advertise icount trigger type in tinfo
Quoted 12 lines · Nicholas Piggin
icount triggers are supported so tinfo should advertise them.
Signed-off-by Nicholas Piggin <[email protected]>--- target/riscv/debug.c | 2 ++ 1 file changed, 2 insertions(+)target/riscv/debug.c +0
-0
target/riscv/debug.c void tdata_csr_write(CPURISCVState *env, int tdata_index, target_ulong val) · line 868
Quoted 4 lines
target/riscv/debug.c +1
-0@@ -865,7 +865,9 @@ void tdata_csr_write(CPURISCVState *env, int tdata_index, target_ulong val)target_ulong tinfo_csr_read (CPURISCVState *env) {/* assume all triggers support the same types of triggers */ +/* XXX: should we set 1 (version 1.0) in the version field? */
For the 'debug' flag, no. For sdtrig and others, yes.
target/riscv/debug.c void tdata_csr_write(CPURISCVState *env, int tdata_index, target_ulong val) · line 872
Quoted 4 lines
target/riscv/debug.c +1
-0return BIT (TRIGGER_TYPE_AD_MATCH ) | +BIT (TRIGGER_TYPE_INST_CNT ) |BIT (TRIGGER_TYPE_AD_MATCH6 ); }
Quoted 1 line
RFC PATCH 18/25 target/riscv/debug: Reset trigger type to unavailable
In reply to Nicholas PigginReset triggers to type 15 (unavailable). The reset value for tdata1 type field is implementation specific, legal values are permitted. Unavailable is a legal value, and some implementations may not support type2 triggers so use that instead.
--- target/riscv/debug.c | 17 ++--------------- 1 file changed, 2 insertions(+), 15 deletions(-)
target/riscv/debug.c +2 -15
@@ -1043,10 +1043,10 @@ void riscv_cpu_debug_post_load(CPURISCVState *env)void riscv_trigger_reset_hold (CPURISCVState *env) { -target_ulong tdata1 =build_tdata1 (env,TRIGGER_TYPE_AD_MATCH ,0 ,0 ); +target_ulong tdata1 =build_tdata1 (env,TRIGGER_TYPE_UNAVAIL ,0 ,0 );int i; -/* init to type 2 triggers */ +/* init to type 15 (unavailable) triggers */ for (i =0 ; i <RV_MAX_TRIGGERS ; i++) {int trigger_type =get_trigger_type (env, i);
@@ -1061,19 +1061,6 @@ void riscv_trigger_reset_hold(CPURISCVState *env)break ; } -/* -* type = TRIGGER_TYPE_AD_MATCH -* dmode = 0 (both debug and M-mode can write tdata) -* maskmax = 0 (unimplemented, always 0) -* sizehi = 0 (match against any size, RV64 only) -* hit = 0 (unimplemented, always 0) -* select = 0 (always 0, perform match on address) -* timing = 0 (always 0, trigger before instruction) -* sizelo = 0 (match against any size) -* action = 0 (always 0, raise a breakpoint exception) -* chain = 0 (unimplemented, always 0) -* match = 0 (always 0, when any compare value equals tdata2) -*/ env->tdata1[i] = tdata1; env->tdata2[i] =0 ; env->tdata3[i] =0 ;
Signature · 1 line
2.51.0
Re: [RFC PATCH 18/25] target/riscv/debug: Reset trigger type to unavailable
Quoted 7 lines · Nicholas Piggin
Reset triggers to type 15 (unavailable). The reset value for tdata1 type field is implementation specific, legal values are permitted. Unavailable is a legal value, and some implementations may not support type2 triggers so use that instead.
Signed-off-by Nicholas Piggin <[email protected]>---
Quoted 40 lines
target/riscv/debug.c | 17 ++--------------- 1 file changed, 2 insertions(+), 15 deletions(-)target/riscv/debug.c +2
-15@@ -1043,10 +1043,10 @@ void riscv_cpu_debug_post_load(CPURISCVState *env)void riscv_trigger_reset_hold (CPURISCVState *env) { -target_ulong tdata1 =build_tdata1 (env,TRIGGER_TYPE_AD_MATCH ,0 ,0 ); +target_ulong tdata1 =build_tdata1 (env,TRIGGER_TYPE_UNAVAIL ,0 ,0 );int i; -/* init to type 2 triggers */ +/* init to type 15 (unavailable) triggers */ for (i =0 ; i <RV_MAX_TRIGGERS ; i++) {int trigger_type =get_trigger_type (env, i);@@ -1061,19 +1061,6 @@ void riscv_trigger_reset_hold(CPURISCVState *env)break ; } -/* -* type = TRIGGER_TYPE_AD_MATCH -* dmode = 0 (both debug and M-mode can write tdata) -* maskmax = 0 (unimplemented, always 0) -* sizehi = 0 (match against any size, RV64 only) -* hit = 0 (unimplemented, always 0) -* select = 0 (always 0, perform match on address) -* timing = 0 (always 0, trigger before instruction) -* sizelo = 0 (match against any size) -* action = 0 (always 0, raise a breakpoint exception) -* chain = 0 (unimplemented, always 0) -* match = 0 (always 0, when any compare value equals tdata2) -*/ env->tdata1[i] = tdata1; env->tdata2[i] =0 ; env->tdata3[i] =0 ;
RFC PATCH 19/25 target/riscv/debug: Add new debug state format
In reply to Nicholas PigginSome Sdtrig implementations support more than 2 debug triggers, but the Sdtrig vmstate is tied to 2 triggers. Extending this will be a compatibility issue anyway, so move to a nicer layout and provide state for all 32 architecturally possible triggers.
--- target/riscv/cpu.h | 35 ++++++-- target/riscv/csr.c | 7 +- target/riscv/debug.c | 167 +++++++++++++++++++++---------------- target/riscv/debug.h | 3 +- target/riscv/machine.c | 31 ++++++- target/riscv/tcg/tcg-cpu.c | 3 +- 6 files changed, 157 insertions(+), 89 deletions(-)
target/riscv/cpu.h +27 -8
@@ -194,6 +194,28 @@ FIELD(VTYPE, VMA, 7, 1) FIELD(VTYPE, VEDIV, 8, 2) FIELD(VTYPE, RESERVED, 10, sizeof(target_ulong) * 8 - 11) +#ifndef CONFIG_USER_ONLY +typedef struct SdtrigTrigger { +target_ulong tdata1; +target_ulong tdata2; +target_ulong tdata3; +}SdtrigTrigger ; + +typedef struct SdtrigState { +/* Architected state */ +target_ulong trigger_cur;/* tselect */ +SdtrigTrigger triggers[RV_MAX_SDTRIG_TRIGGERS ]; +target_ulong tcontrol; +target_ulong mcontext;/* hcontext */ +target_ulong scontext; + +/* QEMU state */ +struct CPUBreakpoint *cpu_breakpoint[RV_MAX_SDTRIG_TRIGGERS ]; +struct CPUWatchpoint *cpu_watchpoint[RV_MAX_SDTRIG_TRIGGERS ]; +bool itrigger_enabled; +}SdtrigState ; +#endif + typedef struct PMUCTRState { /* Current value of a counter */ uint64_t mhpmcounter_val;
@@ -443,14 +465,11 @@ struct CPUArchState {target_ulong mseccfg;/* trigger module */ -target_ulong trigger_cur; -target_ulong tdata1[RV_MAX_TRIGGERS ]; -target_ulong tdata2[RV_MAX_TRIGGERS ]; -target_ulong tdata3[RV_MAX_TRIGGERS ]; -target_ulong mcontext; -struct CPUBreakpoint *cpu_breakpoint[RV_MAX_TRIGGERS ]; -struct CPUWatchpoint *cpu_watchpoint[RV_MAX_TRIGGERS ]; -bool itrigger_enabled; +SdtrigState sdtrig_state; +/* migration compat */ +target_ulong old_tdata1[RV_DEFAULT_TRIGGERS ]; +target_ulong old_tdata2[RV_DEFAULT_TRIGGERS ]; +target_ulong old_tdata3[RV_DEFAULT_TRIGGERS ];/* machine specific rdtime callback */ uint64_t (*rdtime_fn)(void *);
target/riscv/csr.c +4 -3
@@ -5311,7 +5311,8 @@ static RISCVException read_tdata(CPURISCVState *env, int csrno, target_ulong *val) {/* return 0 in tdata1 to end the trigger enumeration */ -if (env->trigger_cur >=RV_MAX_TRIGGERS && csrno ==CSR_TDATA1 ) { +if (env->sdtrig_state.trigger_cur >=RV_MAX_SDTRIG_TRIGGERS && + csrno ==CSR_TDATA1 ) { *val =0 ;return RISCV_EXCP_NONE ; }
@@ -5345,7 +5346,7 @@ static RISCVException read_tinfo(CPURISCVState *env, int csrno,static RISCVException read_mcontext (CPURISCVState *env,int csrno,target_ulong *val) { - *val = env->mcontext; + *val = env->sdtrig_state.mcontext;return RISCV_EXCP_NONE ; }
@@ -5363,7 +5364,7 @@ static RISCVException write_mcontext(CPURISCVState *env, int csrno, mask = rv32 ?MCONTEXT32 :MCONTEXT64 ; } - env->mcontext = val & mask; + env->sdtrig_state.mcontext = val & mask;return RISCV_EXCP_NONE ; }
target/riscv/debug.c +95 -72
451 lines not on this page · raw message
target/riscv/debug.h +2 -1
@@ -24,7 +24,8 @@#include "exec/breakpoint.h" -#define RV_MAX_TRIGGERS 2 +#define RV_MAX_SDTRIG_TRIGGERS 32 +#define RV_DEFAULT_TRIGGERS 2/* register index of tdata CSRs */ enum {
target/riscv/machine.c +27 -4
@@ -225,10 +225,32 @@ static bool debug_needed(void *opaque)return cpu->cfg.debug; } +static int debug_pre_save (void *opaque) +{ +RISCVCPU *cpu = opaque; +CPURISCVState *env = &cpu->env; +int i; + +for (i =0 ; i <RV_DEFAULT_TRIGGERS ; i++) { + env->old_tdata1[i] = env->sdtrig_state.triggers[i].tdata1; + env->old_tdata2[i] = env->sdtrig_state.triggers[i].tdata2; + env->old_tdata3[i] = env->sdtrig_state.triggers[i].tdata3; + } + +return 0 ; +} +static int debug_post_load (void *opaque,int version_id) {RISCVCPU *cpu = opaque;CPURISCVState *env = &cpu->env; +int i; + +for (i =0 ; i <RV_DEFAULT_TRIGGERS ; i++) { + env->sdtrig_state.triggers[i].tdata1 = env->old_tdata1[i]; + env->sdtrig_state.triggers[i].tdata2 = env->old_tdata2[i]; + env->sdtrig_state.triggers[i].tdata3 = env->old_tdata3[i]; + }riscv_cpu_debug_post_load (env);
@@ -240,12 +262,13 @@ static const VMStateDescription vmstate_debug = { .version_id = 2, .minimum_version_id = 2, .needed = debug_needed, + .pre_save = debug_pre_save, .post_load = debug_post_load, .fields = (const VMStateField []) { -VMSTATE_UINTTL (env.trigger_cur,RISCVCPU ), -VMSTATE_UINTTL_ARRAY (env.tdata1,RISCVCPU ,RV_MAX_TRIGGERS ), -VMSTATE_UINTTL_ARRAY (env.tdata2,RISCVCPU ,RV_MAX_TRIGGERS ), -VMSTATE_UINTTL_ARRAY (env.tdata3,RISCVCPU ,RV_MAX_TRIGGERS ), +VMSTATE_UINTTL (env.sdtrig_state.trigger_cur,RISCVCPU ), +VMSTATE_UINTTL_ARRAY (env.old_tdata1,RISCVCPU ,RV_DEFAULT_TRIGGERS ), +VMSTATE_UINTTL_ARRAY (env.old_tdata2,RISCVCPU ,RV_DEFAULT_TRIGGERS ), +VMSTATE_UINTTL_ARRAY (env.old_tdata3,RISCVCPU ,RV_DEFAULT_TRIGGERS ),VMSTATE_END_OF_LIST () } };
target/riscv/tcg/tcg-cpu.c +2 -1
@@ -178,7 +178,8 @@ static TCGTBCPUState riscv_get_tb_cpu_state(CPUState *cs) }if (cpu->cfg.debug) { - flags =FIELD_DP32 (flags,TB_FLAGS ,ITRIGGER , env->itrigger_enabled); + flags =FIELD_DP32 (flags,TB_FLAGS ,ITRIGGER , + env->sdtrig_state.itrigger_enabled); }#endif
Signature · 1 line
2.51.0
RFC PATCH 20/25 target/riscv/debug: Migrate mcontext using new sdtrig vmstate
In reply to Nicholas PigginThe debug migration state does not include mcontext. Change to using a new sdtrig vmstate format that allows for all architectural state if mcontext != 0.
This means a machine is sometimes-migratable, which on second thoughts is probably bad. Is there a better way to do this, or could we just break compat and move to the new vmstate?
Not yet signed off.
--- target/riscv/machine.c | 58 +++++++++++++++++++++++++++++++++++++++++- 1 file changed, 57 insertions(+), 1 deletion(-)
target/riscv/machine.c +57 -1
@@ -221,8 +221,10 @@ static const VMStateDescription vmstate_kvmtimer = {static bool debug_needed (void *opaque) {RISCVCPU *cpu = opaque; +CPURISCVState *env = &cpu->env; -return cpu->cfg.debug; +return cpu->cfg.debug && + env->sdtrig_state.mcontext ==0 ; }static int debug_pre_save (void *opaque)
@@ -273,6 +275,59 @@ static const VMStateDescription vmstate_debug = { } }; +/* +* This is a newer version of the debug (sdtrig) state, required +* to migrate hcontext/mcontext. +*/ +static bool sdtrig_needed (void *opaque) +{ +RISCVCPU *cpu = opaque; +CPURISCVState *env = &cpu->env; + +return cpu->cfg.debug && + env->sdtrig_state.mcontext !=0 ; +} + +static int sdtrig_post_load (void *opaque,int version_id) +{ +RISCVCPU *cpu = opaque; +CPURISCVState *env = &cpu->env; + +riscv_cpu_debug_post_load (env); + +return 0 ; +} + +static const VMStateDescription vmstate_sdtrig_trigger = { + .name ="cpu/sdtrig/trigger" , + .version_id =1 , + .minimum_version_id =1 , + .fields = (const VMStateField []) { +VMSTATE_UINTTL (tdata1, SdtrigTrigger), +VMSTATE_UINTTL (tdata2, SdtrigTrigger), +VMSTATE_UINTTL (tdata3, SdtrigTrigger), +VMSTATE_END_OF_LIST () + } +}; + +static const VMStateDescription vmstate_sdtrig = { + .name ="cpu/sdtrig" , + .version_id =1 , + .minimum_version_id =1 , + .needed = sdtrig_needed, + .post_load = sdtrig_post_load, + .fields = (const VMStateField []) { +VMSTATE_UINTTL (env.sdtrig_state.trigger_cur,RISCVCPU ), +VMSTATE_UINTTL (env.sdtrig_state.tcontrol,RISCVCPU ), +VMSTATE_UINTTL (env.sdtrig_state.mcontext,RISCVCPU ), +VMSTATE_UINTTL (env.sdtrig_state.scontext,RISCVCPU ), +VMSTATE_STRUCT_ARRAY (env.sdtrig_state.triggers,RISCVCPU , +RV_MAX_SDTRIG_TRIGGERS , +0 , vmstate_sdtrig_trigger, SdtrigTrigger), +VMSTATE_END_OF_LIST () + } +}; +static int riscv_cpu_post_load (void *opaque,int version_id) {RISCVCPU *cpu = opaque;
@@ -514,6 +569,7 @@ const VMStateDescription vmstate_riscv_cpu = {#endif &vmstate_envcfg, &vmstate_debug, + &vmstate_sdtrig, &vmstate_smstateen, &vmstate_jvt, &vmstate_elp,
Signature · 1 line
2.51.0
RFC PATCH 21/25 target/riscv/debug: Implementation specific Sdtrig configuration
In reply to Nicholas PigginImplement a Sdtrig CPU configuration class that provides details about the Sdtrig configuration.
It is stored as a pointer in RISCVCPUClass to avoid issues with nested compound literals inside static initialisers with GCC 11.
For now, the number of supported triggers is configurable. This requires some logic
--- target/riscv/cpu.c | 25 +++++++++++++++++++++++++ target/riscv/cpu.h | 4 ++++ target/riscv/debug.c | 5 ++++- target/riscv/debug.h | 4 ++++ target/riscv/machine.c | 11 ++++++++--- 5 files changed, 45 insertions(+), 4 deletions(-)
target/riscv/cpu.c +25 -0
@@ -23,6 +23,7 @@#include "qemu/log.h" #include "cpu.h" #include "cpu_vendorid.h" +#include "debug.h" #include "internals.h" #include "qapi/error.h" #include "qapi/visitor.h"
@@ -2816,6 +2817,11 @@ static void riscv_cpu_class_base_init(ObjectClass *c, const void *data) mcc->def->vext_spec = def->vext_spec; } mcc->def->misa_ext |= def->misa_ext; +#if !defined(CONFIG_USER_ONLY ) +if (def->debug_cfg) { + mcc->def->debug_cfg = def->debug_cfg; + } +#endif riscv_cpu_cfg_merge (&mcc->def->cfg, &def->cfg);
@@ -2951,6 +2957,18 @@ void riscv_isa_write_fdt(RISCVCPU *cpu, void *fdt, char *nodename)DEFINE_RISCV_CPU (type_name, parent_type_name, \ .profile = &(profile_)) +#if !defined(CONFIG_USER_ONLY ) +/* Sdtrig implementation has 2 triggers that support match, match6, icount */ +static const RISCVSdtrigConfig default_sdtrig_config = { + .nr_triggers =2 , +}; + +bool riscv_sdtrig_default_implementation (const RISCVSdtrigConfig *config) +{ +return config == &default_sdtrig_config; +} +#endif +static const TypeInfo riscv_cpu_type_infos[] = { { .name =TYPE_RISCV_CPU ,
@@ -2968,6 +2986,9 @@ static const TypeInfo riscv_cpu_type_infos[] = { .cfg.mmu = true, .cfg.pmp = true, .priv_spec = PRIV_VERSION_LATEST, +#if !defined(CONFIG_USER_ONLY) + .debug_cfg = &default_sdtrig_config, +#endif ), DEFINE_ABSTRACT_RISCV_CPU(TYPE_RISCV_VENDOR_CPU, TYPE_RISCV_CPU),
@@ -2995,6 +3016,10 @@ static const TypeInfo riscv_cpu_type_infos[] = {#else .cfg.max_satp_mode = VM_1_10_SV57,#endif + +#if !defined(CONFIG_USER_ONLY ) + .debug_cfg = &default_sdtrig_config, +#endif ), DEFINE_RISCV_CPU(TYPE_RISCV_CPU_MAX, TYPE_RISCV_DYNAMIC_CPU,
target/riscv/cpu.h +4 -0
@@ -571,6 +571,9 @@ typedef struct RISCVCSR RISCVCSR; typedef struct RISCVCPUDef { RISCVMXL misa_mxl_max; /* max mxl for this cpu */ RISCVCPUProfile *profile; +#if !defined(CONFIG_USER_ONLY) + const RISCVSdtrigConfig *debug_cfg; +#endif uint32_t misa_ext; int priv_spec; int32_t vext_spec;
@@ -666,6 +669,7 @@ void riscv_cpu_set_aia_ireg_rmw_fn(CPURISCVState *env, uint32_t priv, void *rmw_fn_arg);RISCVException smstateen_acc_ok (CPURISCVState *env,int index,uint64_t bit); +bool riscv_sdtrig_default_implementation (const RISCVSdtrigConfig *config);#endif /* !CONFIG_USER_ONLY */ void riscv_cpu_set_mode (CPURISCVState *env,target_ulong newpriv,bool virt_en);
target/riscv/debug.c +4 -1
@@ -173,7 +173,10 @@ target_ulong tselect_csr_read(CPURISCVState *env)void tselect_csr_write (CPURISCVState *env,target_ulong val) { -if (val <RV_DEFAULT_TRIGGERS ) { +CPUState *cs =env_cpu (env); +RISCVCPUClass *mcc =RISCV_CPU_GET_CLASS (cs); + +if (val < mcc->def->debug_cfg->nr_triggers) { env->sdtrig_state.trigger_cur = val; } }
target/riscv/debug.h +4 -0
@@ -135,6 +135,10 @@ enum {#define MHSELECT_IGNORE 0#define MHSELECT_MCONTEXT 4 +typedef struct RISCVSdtrigConfig { +unsigned int nr_triggers; +}RISCVSdtrigConfig ; +bool tdata_available (CPURISCVState *env,int tdata_index);target_ulong tselect_csr_read (CPURISCVState *env);
target/riscv/machine.c +8 -3
@@ -221,10 +221,12 @@ static const VMStateDescription vmstate_kvmtimer = {static bool debug_needed (void *opaque) {RISCVCPU *cpu = opaque; +RISCVCPUClass *mcc =RISCV_CPU_GET_CLASS (cpu);CPURISCVState *env = &cpu->env;return cpu->cfg.debug && - env->sdtrig_state.mcontext ==0 ; + (riscv_sdtrig_default_implementation (mcc->def->debug_cfg) && + env->sdtrig_state.mcontext ==0 ); }static int debug_pre_save (void *opaque)
@@ -277,15 +279,18 @@ static const VMStateDescription vmstate_debug = {/* * This is a newer version of the debug (sdtrig) state, required -* to migrate hcontext/mcontext. +* to migrate hcontext/mcontext, or machines with non-default +* sdtrig implementation. */ static bool sdtrig_needed (void *opaque) {RISCVCPU *cpu = opaque; +RISCVCPUClass *mcc =RISCV_CPU_GET_CLASS (cpu);CPURISCVState *env = &cpu->env;return cpu->cfg.debug && - env->sdtrig_state.mcontext !=0 ; + !(riscv_sdtrig_default_implementation (mcc->def->debug_cfg) && + env->sdtrig_state.mcontext ==0 ); }static int sdtrig_post_load (void *opaque,int version_id)
Signature · 1 line
2.51.0
RFC PATCH 22/25 target/riscv/debug: Support heterogeneous trigger types
In reply to Nicholas PigginNot all Sdtrig implementations have heterogeneous triggers. Add support to the CPU class Sdtrig config to specify the trigger types that each trigger supports.
--- target/riscv/cpu.c | 8 ++++++++ target/riscv/debug.c | 17 +++++++++++++---- target/riscv/debug.h | 5 +++++ 3 files changed, 26 insertions(+), 4 deletions(-)
target/riscv/cpu.c +8 -0
@@ -2961,6 +2961,14 @@ void riscv_isa_write_fdt(RISCVCPU *cpu, void *fdt, char *nodename)/* Sdtrig implementation has 2 triggers that support match, match6, icount */ static const RISCVSdtrigConfig default_sdtrig_config = { .nr_triggers =2 , + .triggers = { + [0 ...1 ] = { + .type_mask = (1 <<TRIGGER_TYPE_AD_MATCH ) | + (1 <<TRIGGER_TYPE_AD_MATCH6 ) | + (1 <<TRIGGER_TYPE_INST_CNT ) | + (1 <<TRIGGER_TYPE_UNAVAIL ), + }, + }, };bool riscv_sdtrig_default_implementation (const RISCVSdtrigConfig *config)
target/riscv/debug.c +13 -4
@@ -845,11 +845,19 @@ void tdata_csr_write(CPURISCVState *env, int tdata_index, target_ulong val) }if (tdata_index ==TDATA1 ) { +CPUState *cs =env_cpu (env); +RISCVCPUClass *mcc =RISCV_CPU_GET_CLASS (cs); +if (val ==0 ) {/* special case, writing 0 results in disabled trigger */ val =build_tdata1 (env,TRIGGER_TYPE_UNAVAIL ,0 ,0 ); } trigger_type =extract_trigger_type (env, val); +if (!(mcc->def->debug_cfg->triggers[index].type_mask & + (1 << trigger_type))) { + val =build_tdata1 (env,TRIGGER_TYPE_UNAVAIL ,0 ,0 ); + trigger_type =extract_trigger_type (env, val); + } }switch (trigger_type) {
@@ -887,11 +895,12 @@ void tdata_csr_write(CPURISCVState *env, int tdata_index, target_ulong val)target_ulong tinfo_csr_read (CPURISCVState *env) { -/* assume all triggers support the same types of triggers */ +CPUState *cs =env_cpu (env); +RISCVCPUClass *mcc =RISCV_CPU_GET_CLASS (cs); +target_ulong index = env->sdtrig_state.trigger_cur; +/* XXX: should we set 1 (version 1.0) in the version field? */ -return BIT (TRIGGER_TYPE_AD_MATCH ) | -BIT (TRIGGER_TYPE_INST_CNT ) | -BIT (TRIGGER_TYPE_AD_MATCH6 ); +return mcc->def->debug_cfg->triggers[index].type_mask; }void riscv_cpu_debug_excp_handler (CPUState *cs)
target/riscv/debug.h +5 -0
@@ -135,8 +135,13 @@ enum {#define MHSELECT_IGNORE 0#define MHSELECT_MCONTEXT 4 +struct trigger_properties { +uint16_t type_mask;/* Trigger types supported (0 = no trigger here) */ +}; +typedef struct RISCVSdtrigConfig {unsigned int nr_triggers; +struct trigger_properties triggers[RV_MAX_SDTRIG_TRIGGERS ]; }RISCVSdtrigConfig ;bool tdata_available (CPURISCVState *env,int tdata_index);
Signature · 1 line
2.51.0
RFC PATCH 23/25 target/riscv/debug: Support heterogeneous mcontrol access types
In reply to Nicholas PigginSimilarly to the last patch, mcontrol/mcontrol6 trigger types may not implement the same read/write/execute match capability. Add configuration to describe what access type matches are supported.
--- target/riscv/cpu.c | 1 + target/riscv/debug.c | 26 ++++++++++++++++++++------ target/riscv/debug.h | 1 + 3 files changed, 22 insertions(+), 6 deletions(-)
target/riscv/cpu.c +1 -0
@@ -2967,6 +2967,7 @@ static const RISCVSdtrigConfig default_sdtrig_config = { (1 <<TRIGGER_TYPE_AD_MATCH6 ) | (1 <<TRIGGER_TYPE_INST_CNT ) | (1 << TRIGGER_TYPE_UNAVAIL), + .mcontrol_rwx_mask = 0x7, /* WP/BP */ }, }, };
target/riscv/debug.c +20 -6
@@ -449,7 +449,11 @@ static inline bool type2_breakpoint_enabled(target_ulong ctrl)static target_ulong type2_mcontrol_validate (CPURISCVState *env,target_ulong ctrl) { +CPUState *cs =env_cpu (env); +RISCVCPUClass *mcc =RISCV_CPU_GET_CLASS (cs); +target_ulong index = env->sdtrig_state.trigger_cur;target_ulong val; +target_ulong rwx_mask;uint32_t size;/* validate the generic part first */
@@ -475,9 +479,12 @@ static target_ulong type2_mcontrol_validate(CPURISCVState *env, } } -/* keep the mode and attribute bits */ - val |= (ctrl & (TYPE2_U |TYPE2_S |TYPE2_M | -TYPE2_LOAD |TYPE2_STORE |TYPE2_EXEC )); +/* only set supported access (load/store/exec) bits */ + rwx_mask = mcc->def->debug_cfg->triggers[index].mcontrol_rwx_mask; + val |= ctrl & rwx_mask; + +/* keep the mode bits */ + val |= ctrl & (TYPE2_U |TYPE2_S |TYPE2_M );return val; }
@@ -573,7 +580,11 @@ static inline bool type6_breakpoint_enabled(target_ulong ctrl)static target_ulong type6_mcontrol6_validate (CPURISCVState *env,target_ulong ctrl) { +CPUState *cs =env_cpu (env); +RISCVCPUClass *mcc =RISCV_CPU_GET_CLASS (cs); +target_ulong index = env->sdtrig_state.trigger_cur;target_ulong val; +target_ulong rwx_mask;uint32_t size;/* validate the generic part first */
@@ -596,9 +607,12 @@ static target_ulong type6_mcontrol6_validate(CPURISCVState *env, val |= (ctrl &TYPE6_SIZE ); } -/* keep the mode and attribute bits */ - val |= (ctrl & (TYPE6_VU |TYPE6_VS |TYPE6_U |TYPE6_S |TYPE6_M | -TYPE6_LOAD |TYPE6_STORE |TYPE6_EXEC )); +/* only set supported access (load/store/exec) bits */ + rwx_mask = mcc->def->debug_cfg->triggers[index].mcontrol_rwx_mask; + val |= ctrl & rwx_mask; + +/* keep the mode bits */ + val |= (ctrl & (TYPE6_VU |TYPE6_VS |TYPE6_U |TYPE6_S |TYPE6_M ));return val; }
target/riscv/debug.h +1 -0
@@ -137,6 +137,7 @@ enum {struct trigger_properties {uint16_t type_mask;/* Trigger types supported (0 = no trigger here) */ +uint8_t mcontrol_rwx_mask;/* mc/mc6 rwx access match supported */ }; typedef struct RISCVSdtrigConfig {
Signature · 1 line
2.51.0
RFC PATCH 24/25 target/riscv/debug: Emulate TT Ascalon Sdtrig
In reply to Nicholas PigginThis adds initial Tenstorrent Ascalon Sdtrig implementation details. Ascalon has 9 triggers: 4 mcontrol6 triggers that can match exec access, 4 mcontrol6 triggers that can match load / store access, and 1 icount trigger.
--- target/riscv/cpu.c | 25 +++++++++++++++++++++++++ 1 file changed, 25 insertions(+)
target/riscv/cpu.c +25 -0
@@ -2972,6 +2972,28 @@ static const RISCVSdtrigConfig default_sdtrig_config = { }, }; +#if defined(TARGET_RISCV64 ) +static const RISCVSdtrigConfig tt_ascalon_sdtrig_config = { + .nr_triggers =9 , + .triggers = { + [0 ...3 ] = { + .type_mask = (1 <<TRIGGER_TYPE_AD_MATCH6 ) | + (1 <<TRIGGER_TYPE_UNAVAIL ), + .mcontrol_rwx_mask =0x4 ,/* BP */ + }, + [4 ...7 ] = { + .type_mask = (1 <<TRIGGER_TYPE_AD_MATCH6 ) | + (1 <<TRIGGER_TYPE_UNAVAIL ), + .mcontrol_rwx_mask =0x3 ,/* WP */ + }, + [8 ] = { + .type_mask = (1 <<TRIGGER_TYPE_INST_CNT ) | + (1 <<TRIGGER_TYPE_UNAVAIL ), + }, + }, +}; +#endif +bool riscv_sdtrig_default_implementation (const RISCVSdtrigConfig *config) {return config == &default_sdtrig_config;
@@ -3166,6 +3188,9 @@ static const TypeInfo riscv_cpu_type_infos[] = { .misa_ext = RVG | RVC | RVS | RVU | RVH | RVV, .priv_spec = PRIV_VERSION_1_13_0, .vext_spec = VEXT_VERSION_1_00_0, +#if !defined(CONFIG_USER_ONLY ) + .debug_cfg = &tt_ascalon_sdtrig_config, +#endif /* ISA extensions */ .cfg.mmu = true,
Signature · 1 line
2.51.0
RFC PATCH 25/25 target/riscv/debug: Fix minor comment typos
In reply to Nicholas Piggin--- target/riscv/debug.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
target/riscv/debug.c +2 -2
@@ -120,7 +120,7 @@ static trigger_action_t get_trigger_action(CPURISCVState *env,break ;case TRIGGER_TYPE_NO_EXIST :case TRIGGER_TYPE_UNAVAIL : -qemu_log_mask (LOG_GUEST_ERROR ,"trigger type: %d does not exit\n" , +qemu_log_mask (LOG_GUEST_ERROR ,"trigger type: %d does not exist\n" , trigger_type);break ;default :
@@ -895,7 +895,7 @@ void tdata_csr_write(CPURISCVState *env, int tdata_index, target_ulong val) trigger_type);break ;case TRIGGER_TYPE_NO_EXIST : -qemu_log_mask (LOG_GUEST_ERROR ,"trigger type: %d does not exit\n" , +qemu_log_mask (LOG_GUEST_ERROR ,"trigger type: %d does not exist\n" , trigger_type);break ;default :
Signature · 1 line
2.51.0
Re: [RFC PATCH 25/25] target/riscv/debug: Fix minor comment typos
Quoted 2 lines · Nicholas Piggin
Signed-off-by Nicholas Piggin <[email protected]>---
Quoted 25 lines
target/riscv/debug.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)target/riscv/debug.c +2
-2@@ -120,7 +120,7 @@ static trigger_action_t get_trigger_action(CPURISCVState *env,break ;case TRIGGER_TYPE_NO_EXIST :case TRIGGER_TYPE_UNAVAIL : -qemu_log_mask (LOG_GUEST_ERROR ,"trigger type: %d does not exit\n" , +qemu_log_mask (LOG_GUEST_ERROR ,"trigger type: %d does not exist\n" , trigger_type);break ;default :@@ -895,7 +895,7 @@ void tdata_csr_write(CPURISCVState *env, int tdata_index, target_ulong val) trigger_type);break ;case TRIGGER_TYPE_NO_EXIST : -qemu_log_mask (LOG_GUEST_ERROR ,"trigger type: %d does not exit\n" , +qemu_log_mask (LOG_GUEST_ERROR ,"trigger type: %d does not exist\n" , trigger_type);break ;default :
Salutations,
Sorry for the late review. Just got informed in another thread that this was sent back in January and I missed it.
I believe patches 1-18 and 25 can be sent in a separated "debug related fixes" series. They seem nice to have regardless of the other patches.
Patches 19-24 start to touch into a more sensible area that we're discussing in other threads, most notably this one:
("[PATCH v3 0/1] target/riscv: deprecate 'debug' CPU property and Debug 0.13")
https://lore.kernel.org/qemu-devel/[email protected]/
The lore behind that change is that we want to support Debug 1.0 extensions, in particular sdext, because that is a requirement for the Server Reference Board. But 'debug' is based on spec 0.13, and unfortunately these 2 versions can't play ball, so my initial reaction was to move away for 0.13.
From what I can see in this series there is a desire to keep the existing debug 0.13 code base around, and that's fine. I also believe that despite the incompatibility the 0.13 code base can be re-used for the 1.0 sdtrig as well, with the appropriate "if-then-else" to split between the behaviors we want, so it's feasible to do both.
IIUC what we seem to be missing in patches 19-24 is a separated flag, i.e. a new extension flag, for the 1.0 sdtrig. So we would end up with 'debug' being an alias for 'sdtrig 0.13' and a new 'sdtrig' flag that implements debug 1.0.
Chao can then base his 'sdext' work on top of the new sdtrig flag, and then we can upstream the riscv-server-ref board. At least this is how I'm seeing all this pieces playing out ....
Cheers,
Daniel
Quoted 67 lines · Nicholas Piggin
Hi,
Sorry for the big series. The Ascalon CPU implements Sdtrig with 2 different types of mcontrol6 trigger and the icount trigger, so in the course of testing and bringing up OpenSBI and Linux support for this, I've accumulated quite a lot.
My new year resolution is to start being better upstream contributor, it's taken me a while with changing jobs and architectures. So I don't expect others to drop everything to review this! Joel has been prodding me, and noted there is some other Sdtrig work going on with the v1.0 support patches.
I think the debug v1.0 patches are somewhat orthogonal to this series, but both are addressing aspects of a common problem of Sdtrig implementation specifics. I wonder if these should be reconciled or left separate. Sdtrig v1.00/v0.13 configuration is a single boolean which is feasible as a CPU property. Whereas the entire space of Sdtrig implementation seems like too much to make configurable in that way.
Any thoughts would be welcome.
Thanks,
NickNicholas Piggin (25):
target/riscv/debug: Check only mcontrol triggers for break/watchpoint
matching
target/riscv/debug: Handle changing trigger types target/riscv/debug: Implement permissive type unavailable trigger target/riscv/debug: Fix icount trigger privilege check target/riscv/debug: Updateitrigger_enabledafter changing privilege target/riscv/debug: Implementget_trigger_actionfor icount type
trigger
target/riscv/debug: Fix migrationpost_loadicount_enabled()test target/riscv/debug: Fix icount privilege matchingicount_enabled()
test
target/riscv/debug: Implement icount trigger textra matching target/riscv/debug: Maintainitrigger_enabledin
helper_itrigger_match()
target/riscv/debug: Fix breakpoint matching action target/riscv/debug: Put mcontrol load/store match address into tval target/riscv/debug: Remove breakpoints on reset target/riscv/debug: Move debug CPUpost_loaddetails intodebug.ctarget/riscv/debug: Insert breakpoints after migration target/riscv/debug: Remove itrigger icount-enabled mode target/riscv/debug: Advertise icount trigger type in tinfo target/riscv/debug: Reset trigger type to unavailable target/riscv/debug: Add new debug state format target/riscv/debug: Migrate mcontext using new sdtrig vmstate target/riscv/debug: Implementation specific Sdtrig configuration target/riscv/debug: Support heterogeneous trigger types target/riscv/debug: Support heterogeneous mcontrol access types target/riscv/debug: Emulate TT Ascalon Sdtrig target/riscv/debug: Fix minor comment typostarget/riscv/cpu.c | 65 ++++- target/riscv/cpu.h | 41 ++- target/riscv/cpu_helper.c | 10 +- target/riscv/csr.c | 7 +- target/riscv/debug.c | 571 ++++++++++++++++++++----------------- target/riscv/debug.h | 19 +- target/riscv/machine.c | 96 ++++++- target/riscv/tcg/tcg-cpu.c | 5 +- 8 files changed, 510 insertions(+), 304 deletions(-)
Greetings,
My current understanding is that this series is an official dependency of the riscv-server-ref board, currently on v8:
[PATCH v8 0/7] hw/riscv: Server Platform Reference Board
Because we decided that we want to keep 0.13 and 1.0 instead of deprecating debug 1.0 (see [1]).
We missed the 11.1 boat with this work and with riscv-server-ref. No big deal TBH since the server-ref board can use improvements before upstreaming. But I would like to not miss the 11.2 train in the end of the year.
Nick, I already reviewed a bunch of patches that seems to be fixes that we can use right away. If you're ok with it I'll pick them all and re-send in your behalf, rebased and with acks. My hope is that this will reduce the amount of work needed for the rest of the 0.13/1.0 debug code and we can get that sorted it out. Then Chao can use the new 1.0 infrastructure for his 'sdext' patches, and then we can lend the riscv-server-ref board on top of it all.
Let me know if that works for you. Chao, not sure if you were aware of this series so feel free to comment.
[1] https://lore.kernel.org/qemu-devel/[email protected]/
Thanks,
Daniel
Quoted 67 lines · Nicholas Piggin
Hi,
Sorry for the big series. The Ascalon CPU implements Sdtrig with 2 different types of mcontrol6 trigger and the icount trigger, so in the course of testing and bringing up OpenSBI and Linux support for this, I've accumulated quite a lot.
My new year resolution is to start being better upstream contributor, it's taken me a while with changing jobs and architectures. So I don't expect others to drop everything to review this! Joel has been prodding me, and noted there is some other Sdtrig work going on with the v1.0 support patches.
I think the debug v1.0 patches are somewhat orthogonal to this series, but both are addressing aspects of a common problem of Sdtrig implementation specifics. I wonder if these should be reconciled or left separate. Sdtrig v1.00/v0.13 configuration is a single boolean which is feasible as a CPU property. Whereas the entire space of Sdtrig implementation seems like too much to make configurable in that way.
Any thoughts would be welcome.
Thanks,
NickNicholas Piggin (25):
target/riscv/debug: Check only mcontrol triggers for break/watchpoint
matching
target/riscv/debug: Handle changing trigger types target/riscv/debug: Implement permissive type unavailable trigger target/riscv/debug: Fix icount trigger privilege check target/riscv/debug: Updateitrigger_enabledafter changing privilege target/riscv/debug: Implementget_trigger_actionfor icount type
trigger
target/riscv/debug: Fix migrationpost_loadicount_enabled()test target/riscv/debug: Fix icount privilege matchingicount_enabled()
test
target/riscv/debug: Implement icount trigger textra matching target/riscv/debug: Maintainitrigger_enabledin
helper_itrigger_match()
target/riscv/debug: Fix breakpoint matching action target/riscv/debug: Put mcontrol load/store match address into tval target/riscv/debug: Remove breakpoints on reset target/riscv/debug: Move debug CPUpost_loaddetails intodebug.ctarget/riscv/debug: Insert breakpoints after migration target/riscv/debug: Remove itrigger icount-enabled mode target/riscv/debug: Advertise icount trigger type in tinfo target/riscv/debug: Reset trigger type to unavailable target/riscv/debug: Add new debug state format target/riscv/debug: Migrate mcontext using new sdtrig vmstate target/riscv/debug: Implementation specific Sdtrig configuration target/riscv/debug: Support heterogeneous trigger types target/riscv/debug: Support heterogeneous mcontrol access types target/riscv/debug: Emulate TT Ascalon Sdtrig target/riscv/debug: Fix minor comment typostarget/riscv/cpu.c | 65 ++++- target/riscv/cpu.h | 41 ++- target/riscv/cpu_helper.c | 10 +- target/riscv/csr.c | 7 +- target/riscv/debug.c | 571 ++++++++++++++++++++----------------- target/riscv/debug.h | 19 +- target/riscv/machine.c | 96 ++++++- target/riscv/tcg/tcg-cpu.c | 5 +- 8 files changed, 510 insertions(+), 304 deletions(-)
Quoted 15 lines · Daniel Henrique Barboza
Greetings,
My current understanding is that this series is an official dependency of the riscv-server-ref board, currently on v8:
[PATCH v8 0/7] hw/riscv: Server Platform Reference Board
Because we decided that we want to keep 0.13 and 1.0 instead of deprecating debug 1.0 (see [1]).
We missed the 11.1 boat with this work and with riscv-server-ref. No big deal TBH since the server-ref board can use improvements before upstreaming. But I would like to not miss the 11.2 train in the end of the year.
Quoted 6 lines
Nick, I already reviewed a bunch of patches that seems to be fixes that we can use right away. If you're ok with it I'll pick them all and re-send in your behalf, rebased and with acks. My hope is that this will reduce the amount of work needed for the rest of the 0.13/1.0 debug code and we can get that sorted it out. Then Chao can use the new 1.0 infrastructure for his 'sdext' patches, and then we can lend the riscv-server-ref board on
That's great news. I'll follow up on this series next week and start updating my 'sdext' patches.
Quoted 4 lines
top of it all.
Let me know if that works for you. Chao, not sure if you were aware of this series so feel free to comment.
I'm no longer being blocked by anything, and I feel like our work is basically nearing completion.
The main thing missing right now relates to the debug spec features, but I don't think those will be an issue. We should be able to push them upstream before the 11.2 release.
Thanks,
Chao
Quoted 78 lines
[1] https://lore.kernel.org/qemu-devel/[email protected]/
Thanks,
Daniel
Quoted 67 lines · Nicholas Piggin
Hi,
Sorry for the big series. The Ascalon CPU implements Sdtrig with 2 different types of mcontrol6 trigger and the icount trigger, so in the course of testing and bringing up OpenSBI and Linux support for this, I've accumulated quite a lot.
My new year resolution is to start being better upstream contributor, it's taken me a while with changing jobs and architectures. So I don't expect others to drop everything to review this! Joel has been prodding me, and noted there is some other Sdtrig work going on with the v1.0 support patches.
I think the debug v1.0 patches are somewhat orthogonal to this series, but both are addressing aspects of a common problem of Sdtrig implementation specifics. I wonder if these should be reconciled or left separate. Sdtrig v1.00/v0.13 configuration is a single boolean which is feasible as a CPU property. Whereas the entire space of Sdtrig implementation seems like too much to make configurable in that way.
Any thoughts would be welcome.
Thanks,
NickNicholas Piggin (25):
target/riscv/debug: Check only mcontrol triggers for break/watchpoint
matching
target/riscv/debug: Handle changing trigger types target/riscv/debug: Implement permissive type unavailable trigger target/riscv/debug: Fix icount trigger privilege check target/riscv/debug: Updateitrigger_enabledafter changing privilege target/riscv/debug: Implementget_trigger_actionfor icount type
trigger
target/riscv/debug: Fix migrationpost_loadicount_enabled()test target/riscv/debug: Fix icount privilege matchingicount_enabled()
test
target/riscv/debug: Implement icount trigger textra matching target/riscv/debug: Maintainitrigger_enabledin
helper_itrigger_match()
target/riscv/debug: Fix breakpoint matching action target/riscv/debug: Put mcontrol load/store match address into tval target/riscv/debug: Remove breakpoints on reset target/riscv/debug: Move debug CPUpost_loaddetails intodebug.ctarget/riscv/debug: Insert breakpoints after migration target/riscv/debug: Remove itrigger icount-enabled mode target/riscv/debug: Advertise icount trigger type in tinfo target/riscv/debug: Reset trigger type to unavailable target/riscv/debug: Add new debug state format target/riscv/debug: Migrate mcontext using new sdtrig vmstate target/riscv/debug: Implementation specific Sdtrig configuration target/riscv/debug: Support heterogeneous trigger types target/riscv/debug: Support heterogeneous mcontrol access types target/riscv/debug: Emulate TT Ascalon Sdtrig target/riscv/debug: Fix minor comment typostarget/riscv/cpu.c | 65 ++++- target/riscv/cpu.h | 41 ++- target/riscv/cpu_helper.c | 10 +- target/riscv/csr.c | 7 +- target/riscv/debug.c | 571 ++++++++++++++++++++----------------- target/riscv/debug.h | 19 +- target/riscv/machine.c | 96 ++++++- target/riscv/tcg/tcg-cpu.c | 5 +- 8 files changed, 510 insertions(+), 304 deletions(-)