On Wed, Mar 20, 2024 at 5:21 PM Atish Patra <ati...@rivosinc.com> wrote:
>
>
> On 3/19/24 21:54, Alistair Francis wrote:
>
> On Thu, Mar 7, 2024 at 7:26 PM Atish Patra <ati...@rivosinc.com> wrote:
>
> On 3/4/24 22:47, LIU Zhiwei wrote:
>
> On 2024/2/29 2:51, Atish Patra wrote:
>
> Privilege mode filtering can also be emulated for cycle/instret by
> tracking host_ticks/icount during each privilege mode switch. This
> patch implements that for both cycle/instret and mhpmcounters. The
> first one requires Smcntrpmf while the other one requires Sscofpmf
> to be enabled.
>
> The cycle/instret are still computed using host ticks when icount
> is not enabled. Otherwise, they are computed using raw icount which
> is more accurate in icount mode.
>
> Reviewed-by: Daniel Henrique Barboza <dbarb...@ventanamicro.com>
> Signed-off-by: Atish Patra <ati...@rivosinc.com>
> ---
> target/riscv/cpu.h | 11 +++++
> target/riscv/cpu_bits.h | 5 ++
> target/riscv/cpu_helper.c | 17 ++++++-
> target/riscv/csr.c | 96 ++++++++++++++++++++++++++++++---------
> target/riscv/pmu.c | 64 ++++++++++++++++++++++++++
> target/riscv/pmu.h | 2 +
> 6 files changed, 171 insertions(+), 24 deletions(-)
>
> diff --git a/target/riscv/cpu.h b/target/riscv/cpu.h
> index 174e8ba8e847..9e21d7f7d635 100644
> --- a/target/riscv/cpu.h
> +++ b/target/riscv/cpu.h
> @@ -157,6 +157,15 @@ typedef struct PMUCTRState {
> target_ulong irq_overflow_left;
> } PMUCTRState;
> +typedef struct PMUFixedCtrState {
> + /* Track cycle and icount for each privilege mode */
> + uint64_t counter[4];
> + uint64_t counter_prev[4];
> + /* Track cycle and icount for each privilege mode when V = 1*/
> + uint64_t counter_virt[2];
> + uint64_t counter_virt_prev[2];
> +} PMUFixedCtrState;
> +
> struct CPUArchState {
> target_ulong gpr[32];
> target_ulong gprh[32]; /* 64 top bits of the 128-bit registers */
> @@ -353,6 +362,8 @@ struct CPUArchState {
> /* PMU event selector configured values for RV32 */
> target_ulong mhpmeventh_val[RV_MAX_MHPMEVENTS];
> + PMUFixedCtrState pmu_fixed_ctrs[2];
> +
> target_ulong sscratch;
> target_ulong mscratch;
> diff --git a/target/riscv/cpu_bits.h b/target/riscv/cpu_bits.h
> index e866c60a400c..5fe349e313dc 100644
> --- a/target/riscv/cpu_bits.h
> +++ b/target/riscv/cpu_bits.h
> @@ -920,6 +920,11 @@ typedef enum RISCVException {
> #define MHPMEVENT_BIT_VUINH BIT_ULL(58)
> #define MHPMEVENTH_BIT_VUINH BIT(26)
> +#define MHPMEVENT_FILTER_MASK (MHPMEVENT_BIT_MINH | \
> + MHPMEVENT_BIT_SINH | \
> + MHPMEVENT_BIT_UINH | \
> + MHPMEVENT_BIT_VSINH | \
> + MHPMEVENT_BIT_VUINH)
> #define MHPMEVENT_SSCOF_MASK _ULL(0xFFFF000000000000)
> #define MHPMEVENT_IDX_MASK 0xFFFFF
> #define MHPMEVENT_SSCOF_RESVD 16
> diff --git a/target/riscv/cpu_helper.c b/target/riscv/cpu_helper.c
> index d462d95ee165..33965d843d46 100644
> --- a/target/riscv/cpu_helper.c
> +++ b/target/riscv/cpu_helper.c
> @@ -718,8 +718,21 @@ void riscv_cpu_set_mode(CPURISCVState *env,
> target_ulong newpriv)
> {
> g_assert(newpriv <= PRV_M && newpriv != PRV_RESERVED);
> - if (icount_enabled() && newpriv != env->priv) {
> - riscv_itrigger_update_priv(env);
> + /*
> + * Invoke cycle/instret update between priv mode changes or
> + * VS->HS mode transition is SPV bit must be set
> + * HS->VS mode transition where virt_enabled must be set
> + * In both cases, priv will S mode only.
> + */
> + if (newpriv != env->priv ||
> + (env->priv == PRV_S && newpriv == PRV_S &&
> + (env->virt_enabled || get_field(env->hstatus, HSTATUS_SPV)))) {
> + if (icount_enabled()) {
> + riscv_itrigger_update_priv(env);
> + riscv_pmu_icount_update_priv(env, newpriv);
> + } else {
> + riscv_pmu_cycle_update_priv(env, newpriv);
> + }
> }
> /* tlb_flush is unnecessary as mode is contained in mmu_idx */
> env->priv = newpriv;
> diff --git a/target/riscv/csr.c b/target/riscv/csr.c
> index ff9bac537593..482e212c5f74 100644
> --- a/target/riscv/csr.c
> +++ b/target/riscv/csr.c
> @@ -788,32 +788,16 @@ static RISCVException write_vcsr(CPURISCVState
> *env, int csrno,
> return RISCV_EXCP_NONE;
> }
> +#if defined(CONFIG_USER_ONLY)
> /* User Timers and Counters */
> static target_ulong get_ticks(bool shift)
> {
> - int64_t val;
> - target_ulong result;
> -
> -#if !defined(CONFIG_USER_ONLY)
> - if (icount_enabled()) {
> - val = icount_get();
> - } else {
> - val = cpu_get_host_ticks();
> - }
> -#else
> - val = cpu_get_host_ticks();
> -#endif
> -
> - if (shift) {
> - result = val >> 32;
> - } else {
> - result = val;
> - }
> + int64_t val = cpu_get_host_ticks();
> + target_ulong result = shift ? val >> 32 : val;
> return result;
> }
> -#if defined(CONFIG_USER_ONLY)
> static RISCVException read_time(CPURISCVState *env, int csrno,
> target_ulong *val)
> {
> @@ -952,6 +936,71 @@ static RISCVException
> write_mhpmeventh(CPURISCVState *env, int csrno,
> return RISCV_EXCP_NONE;
> }
> +static target_ulong
> riscv_pmu_ctr_get_fixed_counters_val(CPURISCVState *env,
> + int
> counter_idx,
> + bool
> upper_half)
> +{
> + uint64_t curr_val = 0;
> + target_ulong result = 0;
> + uint64_t *counter_arr = icount_enabled() ?
> env->pmu_fixed_ctrs[1].counter :
> + env->pmu_fixed_ctrs[0].counter;
> + uint64_t *counter_arr_virt = icount_enabled() ?
> + env->pmu_fixed_ctrs[1].counter_virt :
> + env->pmu_fixed_ctrs[0].counter_virt;
> + uint64_t cfg_val = 0;
> +
> + if (counter_idx == 0) {
> + cfg_val = upper_half ? ((uint64_t)env->mcyclecfgh << 32) :
> + env->mcyclecfg;
> + } else if (counter_idx == 2) {
> + cfg_val = upper_half ? ((uint64_t)env->minstretcfgh << 32) :
> + env->minstretcfg;
> + } else {
> + cfg_val = upper_half ?
> + ((uint64_t)env->mhpmeventh_val[counter_idx] << 32) :
> + env->mhpmevent_val[counter_idx];
> + cfg_val &= MHPMEVENT_FILTER_MASK;
> + }
> +
> + if (!cfg_val) {
> + if (icount_enabled()) {
> + curr_val = icount_get_raw();
> + } else {
> + curr_val = cpu_get_host_ticks();
> + }
> + goto done;
> + }
> +
> + if (!(cfg_val & MCYCLECFG_BIT_MINH)) {
> + curr_val += counter_arr[PRV_M];
> + }
> +
> + if (!(cfg_val & MCYCLECFG_BIT_SINH)) {
> + curr_val += counter_arr[PRV_S];
> + }
> +
> + if (!(cfg_val & MCYCLECFG_BIT_UINH)) {
> + curr_val += counter_arr[PRV_U];
> + }
> +
> + if (!(cfg_val & MCYCLECFG_BIT_VSINH)) {
> + curr_val += counter_arr_virt[PRV_S];
> + }
> +
> + if (!(cfg_val & MCYCLECFG_BIT_VUINH)) {
> + curr_val += counter_arr_virt[PRV_U];
> + }
> +
> +done:
> + if (riscv_cpu_mxl(env) == MXL_RV32) {
> + result = upper_half ? curr_val >> 32 : curr_val;
> + } else {
> + result = curr_val;
> + }
> +
> + return result;
> +}
> +
> static RISCVException write_mhpmcounter(CPURISCVState *env, int csrno,
> target_ulong val)
> {
> @@ -962,7 +1011,8 @@ static RISCVException
> write_mhpmcounter(CPURISCVState *env, int csrno,
> counter->mhpmcounter_val = val;
> if (riscv_pmu_ctr_monitor_cycles(env, ctr_idx) ||
> riscv_pmu_ctr_monitor_instructions(env, ctr_idx)) {
> - counter->mhpmcounter_prev = get_ticks(false);
> + counter->mhpmcounter_prev =
> riscv_pmu_ctr_get_fixed_counters_val(env,
> + ctr_idx, false);
> if (ctr_idx > 2) {
> if (riscv_cpu_mxl(env) == MXL_RV32) {
> mhpmctr_val = mhpmctr_val |
> @@ -990,7 +1040,8 @@ static RISCVException
> write_mhpmcounterh(CPURISCVState *env, int csrno,
> mhpmctr_val = mhpmctr_val | (mhpmctrh_val << 32);
> if (riscv_pmu_ctr_monitor_cycles(env, ctr_idx) ||
> riscv_pmu_ctr_monitor_instructions(env, ctr_idx)) {
> - counter->mhpmcounterh_prev = get_ticks(true);
> + counter->mhpmcounterh_prev =
> riscv_pmu_ctr_get_fixed_counters_val(env,
> + ctr_idx, true);
> if (ctr_idx > 2) {
> riscv_pmu_setup_timer(env, mhpmctr_val, ctr_idx);
> }
> @@ -1031,7 +1082,8 @@ static RISCVException
> riscv_pmu_read_ctr(CPURISCVState *env, target_ulong *val,
> */
> if (riscv_pmu_ctr_monitor_cycles(env, ctr_idx) ||
> riscv_pmu_ctr_monitor_instructions(env, ctr_idx)) {
> - *val = get_ticks(upper_half) - ctr_prev + ctr_val;
> + *val = riscv_pmu_ctr_get_fixed_counters_val(env, ctr_idx,
> upper_half) -
> + ctr_prev + ctr_val;
> } else {
> *val = ctr_val;
> }
> diff --git a/target/riscv/pmu.c b/target/riscv/pmu.c
> index 0e7d58b8a5c2..37309ff64cb6 100644
> --- a/target/riscv/pmu.c
> +++ b/target/riscv/pmu.c
> @@ -19,6 +19,7 @@
> #include "qemu/osdep.h"
> #include "qemu/log.h"
> #include "qemu/error-report.h"
> +#include "qemu/timer.h"
> #include "cpu.h"
> #include "pmu.h"
> #include "sysemu/cpu-timers.h"
> @@ -176,6 +177,69 @@ static int riscv_pmu_incr_ctr_rv64(RISCVCPU
> *cpu, uint32_t ctr_idx)
> return 0;
> }
> +void riscv_pmu_icount_update_priv(CPURISCVState *env, target_ulong
> newpriv)
> +{
> + uint64_t delta;
> + uint64_t *counter_arr;
> + uint64_t *counter_arr_prev;
> + uint64_t current_icount = icount_get_raw();
> +
> + if (env->virt_enabled) {
> + counter_arr = env->pmu_fixed_ctrs[1].counter_virt;
> + counter_arr_prev = env->pmu_fixed_ctrs[1].counter_virt_prev;
> + } else {
> + counter_arr = env->pmu_fixed_ctrs[1].counter;
> + counter_arr_prev = env->pmu_fixed_ctrs[1].counter_prev;
> + }
> +
> + if (newpriv != env->priv) {
> + delta = current_icount - counter_arr_prev[env->priv];
> + counter_arr_prev[newpriv] = current_icount;
> + } else {
> + delta = current_icount - counter_arr_prev[env->priv];
> + if (env->virt_enabled) {
> + /* HS->VS transition.The previous value should
> correspond to HS. */
>
> Hi Atish,
>
> Dose env->virt_enabled ensure HS->VS transition?
>
> As per my understanding, riscv_cpu_set_virt_enabled always invoked
> before riscv_cpu_set_mode.
>
> In helper_mret() we call riscv_cpu_set_mode() before
> riscv_cpu_set_virt_enabled().
>
> Ahh yes. I missed the helper_mret condition.
>
> I don't think there is any requirement on which order we call the functions
>
> Indeed. helper_mret and helper_sret invokes them in different order.
>
> That means env->virt_enabled must be enabled before we enter into this
> function. Let me know if I missed
>
> env->virt_enabled will be true if we are in HS or VS mode, but you
> don't know the transition from just env->virt_enabled being set.
>
> Hmm. In riscv_cpu_do_interrupt(), virt_enabled is set to 0 if a trap is taken
> to HS mode
> from VS mode. Am I missing something ?
>
Whoops, I meant VU and VS mode.
Alistair
