| Message ID | 20250217220707.1468365-1-superm1@kernel.org |
|---|---|
| Headers | show |
| Series | amd-pstate cleanups | expand |
On Mon, Feb 17, 2025 at 04:06:52PM -0600, Mario Limonciello wrote: > From: Mario Limonciello <mario.limonciello@amd.com> > > Use the perf_to_freq helpers to calculate this on the fly. > > Signed-off-by: Mario Limonciello <mario.limonciello@amd.com> LGTM. Reviewed-by: Gautham R. Shenoy <gautham.shenoy@amd.com>
On 2/18/2025 3:36 AM, Mario Limonciello wrote: > From: Mario Limonciello <mario.limonciello@amd.com> > > During resume it's possible the firmware didn't restore the CPPC request MSR > but the kernel thinks the values line up. This leads to incorrect performance > after resume from suspend. > > To fix the issue invalidate the cached value at suspend. During resume use > the saved values programmed as cached limits. > > Reported-by: Miroslav Pavleski <miroslav@pavleski.net> > Closes: https://bugzilla.kernel.org/show_bug.cgi?id=217931 > Signed-off-by: Mario Limonciello <mario.limonciello@amd.com> > --- > drivers/cpufreq/amd-pstate.c | 5 ++++- > 1 file changed, 4 insertions(+), 1 deletion(-) > > diff --git a/drivers/cpufreq/amd-pstate.c b/drivers/cpufreq/amd-pstate.c > index f425fb7ec77d7..12fb63169a24c 100644 > --- a/drivers/cpufreq/amd-pstate.c > +++ b/drivers/cpufreq/amd-pstate.c > @@ -1611,7 +1611,7 @@ static int amd_pstate_epp_reenable(struct cpufreq_policy *policy) > max_perf, policy->boost_enabled); > } > > - return amd_pstate_update_perf(cpudata, 0, 0, max_perf, cpudata->epp_cached, false); > + return amd_pstate_epp_update_limit(policy); Can we also add the check "if (policy->min != cpudata->min_limit_freq || policy->max != cpudata->max_limit_freq)" in "amd_pstate_epp_update_limit()" before calling "amd_pstate_update_min_max_limit()". I think it would help in avoiding some unnecessary calls to the update_min_max_limit() function. Patch looks good to me otherwise. Thanks, Dhananjay > } > > static int amd_pstate_epp_cpu_online(struct cpufreq_policy *policy) > @@ -1660,6 +1660,9 @@ static int amd_pstate_epp_suspend(struct cpufreq_policy *policy) > if (cppc_state != AMD_PSTATE_ACTIVE) > return 0; > > + /* invalidate to ensure it's rewritten during resume */ > + cpudata->cppc_req_cached = 0; > + > /* set this flag to avoid setting core offline*/ > cpudata->suspended = true; >
On 2/19/2025 11:42 AM, Dhananjay Ugwekar wrote: > On 2/18/2025 3:36 AM, Mario Limonciello wrote: >> From: Mario Limonciello <mario.limonciello@amd.com> >> >> During resume it's possible the firmware didn't restore the CPPC request MSR >> but the kernel thinks the values line up. This leads to incorrect performance >> after resume from suspend. >> >> To fix the issue invalidate the cached value at suspend. During resume use >> the saved values programmed as cached limits. >> >> Reported-by: Miroslav Pavleski <miroslav@pavleski.net> >> Closes: https://bugzilla.kernel.org/show_bug.cgi?id=217931 >> Signed-off-by: Mario Limonciello <mario.limonciello@amd.com> >> --- >> drivers/cpufreq/amd-pstate.c | 5 ++++- >> 1 file changed, 4 insertions(+), 1 deletion(-) >> >> diff --git a/drivers/cpufreq/amd-pstate.c b/drivers/cpufreq/amd-pstate.c >> index f425fb7ec77d7..12fb63169a24c 100644 >> --- a/drivers/cpufreq/amd-pstate.c >> +++ b/drivers/cpufreq/amd-pstate.c >> @@ -1611,7 +1611,7 @@ static int amd_pstate_epp_reenable(struct cpufreq_policy *policy) >> max_perf, policy->boost_enabled); >> } >> >> - return amd_pstate_update_perf(cpudata, 0, 0, max_perf, cpudata->epp_cached, false); >> + return amd_pstate_epp_update_limit(policy); > > Can we also add the check "if (policy->min != cpudata->min_limit_freq || policy->max != cpudata->max_limit_freq)" > in "amd_pstate_epp_update_limit()" before calling "amd_pstate_update_min_max_limit()". I think it would help in > avoiding some unnecessary calls to the update_min_max_limit() function. You can ignore this, I see that you have handled it in the 3rd patch. Reviewed-by: Dhananjay Ugwekar <dhananjay.ugwekar@amd.com> > > Patch looks good to me otherwise. > > Thanks, > Dhananjay > >> } >> >> static int amd_pstate_epp_cpu_online(struct cpufreq_policy *policy) >> @@ -1660,6 +1660,9 @@ static int amd_pstate_epp_suspend(struct cpufreq_policy *policy) >> if (cppc_state != AMD_PSTATE_ACTIVE) >> return 0; >> >> + /* invalidate to ensure it's rewritten during resume */ >> + cpudata->cppc_req_cached = 0; >> + >> /* set this flag to avoid setting core offline*/ >> cpudata->suspended = true; >> >
On 2/18/2025 23:24, Gautham R. Shenoy wrote: > Hello Mario, > > > On Mon, Feb 17, 2025 at 04:06:50PM -0600, Mario Limonciello wrote: >> From: Mario Limonciello <mario.limonciello@amd.com> >> >> During resume it's possible the firmware didn't restore the CPPC request MSR >> but the kernel thinks the values line up. This leads to incorrect performance >> after resume from suspend. >> >> To fix the issue invalidate the cached value at suspend. During resume use >> the saved values programmed as cached limits. >> >> Reported-by: Miroslav Pavleski <miroslav@pavleski.net> >> Closes: https://bugzilla.kernel.org/show_bug.cgi?id=217931 >> Signed-off-by: Mario Limonciello <mario.limonciello@amd.com> >> --- >> drivers/cpufreq/amd-pstate.c | 5 ++++- >> 1 file changed, 4 insertions(+), 1 deletion(-) >> >> diff --git a/drivers/cpufreq/amd-pstate.c b/drivers/cpufreq/amd-pstate.c >> index f425fb7ec77d7..12fb63169a24c 100644 >> --- a/drivers/cpufreq/amd-pstate.c >> +++ b/drivers/cpufreq/amd-pstate.c >> @@ -1611,7 +1611,7 @@ static int amd_pstate_epp_reenable(struct cpufreq_policy *policy) >> max_perf, policy->boost_enabled); >> } > > You can also remove the tracing code from amd_pstate_epp_reenable(), i.e, > > if (trace_amd_pstate_epp_perf_enabled()) { > trace_amd_pstate_epp_perf(cpudata->cpu, cpudata->highest_perf, > cpudata->epp_cached, > FIELD_GET(AMD_CPPC_MIN_PERF_MASK, cpudata->cppc_req_cached), > max_perf, policy->boost_enabled); > } > > Since amd_pstate_epp_update_limit() also has the the tracing code. Yeah; the tracing code gets updated later in the series. My plan is this commit is a minimal fix, and will go to 6.14, the rest will be in 6.15. > > The patch looks good to me otherwise. > > Reviewed-by: Gautham R. Shenoy <gautham.shenoy@amd.com> > Thanks!
On 2/19/2025 02:00, Dhananjay Ugwekar wrote: > On 2/18/2025 3:36 AM, Mario Limonciello wrote: >> From: Mario Limonciello <mario.limonciello@amd.com> >> >> Use the perf_to_freq helpers to calculate this on the fly. > > Can we call out the below change (in the -1550,7 +1525,8 code chunk) in > the commit message, or split it out to different patch > > * Adding the if check in amd_pstate_epp_update_limit() > Yeah will add to commit meesage. >> >> Signed-off-by: Mario Limonciello <mario.limonciello@amd.com> >> -- > We need one more hyphen here I think i.e."---", otherwise the version info is > showing up in the commit message Ack > >> v3: >> * Fix calc error for min_freq >> v2: >> * Keep cached limits >> --- >> drivers/cpufreq/amd-pstate-ut.c | 14 +++---- >> drivers/cpufreq/amd-pstate.c | 70 +++++++++++---------------------- >> drivers/cpufreq/amd-pstate.h | 9 +---- >> 3 files changed, 32 insertions(+), 61 deletions(-) >> >> diff --git a/drivers/cpufreq/amd-pstate-ut.c b/drivers/cpufreq/amd-pstate-ut.c >> index 3a0a380c3590c..445278cf40b61 100644 >> --- a/drivers/cpufreq/amd-pstate-ut.c >> +++ b/drivers/cpufreq/amd-pstate-ut.c >> @@ -214,14 +214,14 @@ static void amd_pstate_ut_check_freq(u32 index) >> break; >> cpudata = policy->driver_data; >> >> - if (!((cpudata->max_freq >= cpudata->nominal_freq) && >> + if (!((policy->cpuinfo.max_freq >= cpudata->nominal_freq) && >> (cpudata->nominal_freq > cpudata->lowest_nonlinear_freq) && >> - (cpudata->lowest_nonlinear_freq > cpudata->min_freq) && >> - (cpudata->min_freq > 0))) { >> + (cpudata->lowest_nonlinear_freq > policy->cpuinfo.min_freq) && >> + (policy->cpuinfo.min_freq > 0))) { >> amd_pstate_ut_cases[index].result = AMD_PSTATE_UT_RESULT_FAIL; >> pr_err("%s cpu%d max=%d >= nominal=%d > lowest_nonlinear=%d > min=%d > 0, the formula is incorrect!\n", >> - __func__, cpu, cpudata->max_freq, cpudata->nominal_freq, >> - cpudata->lowest_nonlinear_freq, cpudata->min_freq); >> + __func__, cpu, policy->cpuinfo.max_freq, cpudata->nominal_freq, >> + cpudata->lowest_nonlinear_freq, policy->cpuinfo.min_freq); >> goto skip_test; >> } >> >> @@ -233,13 +233,13 @@ static void amd_pstate_ut_check_freq(u32 index) >> } >> >> if (cpudata->boost_supported) { >> - if ((policy->max == cpudata->max_freq) || >> + if ((policy->max == policy->cpuinfo.max_freq) || >> (policy->max == cpudata->nominal_freq)) >> amd_pstate_ut_cases[index].result = AMD_PSTATE_UT_RESULT_PASS; >> else { >> amd_pstate_ut_cases[index].result = AMD_PSTATE_UT_RESULT_FAIL; >> pr_err("%s cpu%d policy_max=%d should be equal cpu_max=%d or cpu_nominal=%d !\n", >> - __func__, cpu, policy->max, cpudata->max_freq, >> + __func__, cpu, policy->max, policy->cpuinfo.max_freq, >> cpudata->nominal_freq); >> goto skip_test; >> } >> diff --git a/drivers/cpufreq/amd-pstate.c b/drivers/cpufreq/amd-pstate.c >> index 87c605348a3dc..a7c41f915b46e 100644 >> --- a/drivers/cpufreq/amd-pstate.c >> +++ b/drivers/cpufreq/amd-pstate.c >> @@ -717,7 +717,7 @@ static int amd_pstate_cpu_boost_update(struct cpufreq_policy *policy, bool on) >> int ret = 0; >> >> nominal_freq = READ_ONCE(cpudata->nominal_freq); >> - max_freq = READ_ONCE(cpudata->max_freq); >> + max_freq = perf_to_freq(cpudata, READ_ONCE(cpudata->highest_perf)); >> >> if (on) >> policy->cpuinfo.max_freq = max_freq; >> @@ -901,35 +901,26 @@ static u32 amd_pstate_get_transition_latency(unsigned int cpu) >> static int amd_pstate_init_freq(struct amd_cpudata *cpudata) >> { >> int ret; >> - u32 min_freq, max_freq; >> - u32 nominal_freq, lowest_nonlinear_freq; >> + u32 min_freq, nominal_freq, lowest_nonlinear_freq; >> struct cppc_perf_caps cppc_perf; >> >> ret = cppc_get_perf_caps(cpudata->cpu, &cppc_perf); >> if (ret) >> return ret; >> >> - if (quirks && quirks->lowest_freq) >> - min_freq = quirks->lowest_freq; >> - else >> - min_freq = cppc_perf.lowest_freq; >> - >> if (quirks && quirks->nominal_freq) >> nominal_freq = quirks->nominal_freq; >> else >> nominal_freq = cppc_perf.nominal_freq; >> >> - min_freq *= 1000; >> nominal_freq *= 1000; >> - >> WRITE_ONCE(cpudata->nominal_freq, nominal_freq); >> - WRITE_ONCE(cpudata->min_freq, min_freq); >> - >> - max_freq = perf_to_freq(cpudata, cpudata->highest_perf); >> - lowest_nonlinear_freq = perf_to_freq(cpudata, cpudata->lowest_nonlinear_perf); >> >> - WRITE_ONCE(cpudata->lowest_nonlinear_freq, lowest_nonlinear_freq); >> - WRITE_ONCE(cpudata->max_freq, max_freq); >> + if (quirks && quirks->lowest_freq) { > > We can avoid the "{" for single line if statement, to keep checkpatch happy This is one of the reason that I don't like to treat checkpatch as gospil. I added it specficially to avoid back and forth for the new patch. In this case I plan to keep it for that reason. > >> + min_freq = quirks->lowest_freq; >> + } else >> + min_freq = cppc_perf.lowest_freq; >> + min_freq *= 1000; > > I see that this min_freq part of the code is unchanged, just moved few lines below. > If the moving is unintended can we avoid it, so that the diff is optimal. Sure I'll see if I can re-order it a bit to keep it the same. > >> >> /** >> * Below values need to be initialized correctly, otherwise driver will fail to load >> @@ -937,12 +928,15 @@ static int amd_pstate_init_freq(struct amd_cpudata *cpudata) >> * lowest_nonlinear_freq is a value between [min_freq, nominal_freq] >> * Check _CPC in ACPI table objects if any values are incorrect >> */ >> - if (min_freq <= 0 || max_freq <= 0 || nominal_freq <= 0 || min_freq > max_freq) { > > Shouldn't we retain these sanity checks for min_freq and max_freq? Now that we're using the helpers isn't it impossible to end up with negative values? > > Thanks, > Dhananjay > >> - pr_err("min_freq(%d) or max_freq(%d) or nominal_freq(%d) value is incorrect\n", >> - min_freq, max_freq, nominal_freq); >> + if (nominal_freq <= 0) { >> + pr_err("nominal_freq(%d) value is incorrect\n", >> + nominal_freq); >> return -EINVAL; >> } >> >> + lowest_nonlinear_freq = perf_to_freq(cpudata, cpudata->lowest_nonlinear_perf); >> + WRITE_ONCE(cpudata->lowest_nonlinear_freq, lowest_nonlinear_freq); >> + >> if (lowest_nonlinear_freq <= min_freq || lowest_nonlinear_freq > nominal_freq) { >> pr_err("lowest_nonlinear_freq(%d) value is out of range [min_freq(%d), nominal_freq(%d)]\n", >> lowest_nonlinear_freq, min_freq, nominal_freq); >> @@ -954,9 +948,9 @@ static int amd_pstate_init_freq(struct amd_cpudata *cpudata) >> >> static int amd_pstate_cpu_init(struct cpufreq_policy *policy) >> { >> - int min_freq, max_freq, ret; >> - struct device *dev; >> struct amd_cpudata *cpudata; >> + struct device *dev; >> + int ret; >> >> /* >> * Resetting PERF_CTL_MSR will put the CPU in P0 frequency, >> @@ -987,17 +981,11 @@ static int amd_pstate_cpu_init(struct cpufreq_policy *policy) >> if (ret) >> goto free_cpudata1; >> >> - min_freq = READ_ONCE(cpudata->min_freq); >> - max_freq = READ_ONCE(cpudata->max_freq); >> - >> policy->cpuinfo.transition_latency = amd_pstate_get_transition_latency(policy->cpu); >> policy->transition_delay_us = amd_pstate_get_transition_delay_us(policy->cpu); >> >> - policy->min = min_freq; >> - policy->max = max_freq; >> - >> - policy->cpuinfo.min_freq = min_freq; >> - policy->cpuinfo.max_freq = max_freq; >> + policy->cpuinfo.min_freq = policy->min = perf_to_freq(cpudata, cpudata->lowest_perf); >> + policy->cpuinfo.max_freq = policy->max = perf_to_freq(cpudata, cpudata->highest_perf); >> >> policy->boost_enabled = READ_ONCE(cpudata->boost_supported); >> >> @@ -1021,9 +1009,6 @@ static int amd_pstate_cpu_init(struct cpufreq_policy *policy) >> goto free_cpudata2; >> } >> >> - cpudata->max_limit_freq = max_freq; >> - cpudata->min_limit_freq = min_freq; >> - >> policy->driver_data = cpudata; >> >> if (!current_pstate_driver->adjust_perf) >> @@ -1081,14 +1066,10 @@ static int amd_pstate_cpu_suspend(struct cpufreq_policy *policy) >> static ssize_t show_amd_pstate_max_freq(struct cpufreq_policy *policy, >> char *buf) >> { >> - int max_freq; >> struct amd_cpudata *cpudata = policy->driver_data; >> >> - max_freq = READ_ONCE(cpudata->max_freq); >> - if (max_freq < 0) >> - return max_freq; >> >> - return sysfs_emit(buf, "%u\n", max_freq); >> + return sysfs_emit(buf, "%u\n", perf_to_freq(cpudata, READ_ONCE(cpudata->highest_perf))); >> } >> >> static ssize_t show_amd_pstate_lowest_nonlinear_freq(struct cpufreq_policy *policy, >> @@ -1446,10 +1427,10 @@ static bool amd_pstate_acpi_pm_profile_undefined(void) >> >> static int amd_pstate_epp_cpu_init(struct cpufreq_policy *policy) >> { >> - int min_freq, max_freq, ret; >> struct amd_cpudata *cpudata; >> struct device *dev; >> u64 value; >> + int ret; >> >> /* >> * Resetting PERF_CTL_MSR will put the CPU in P0 frequency, >> @@ -1480,19 +1461,13 @@ static int amd_pstate_epp_cpu_init(struct cpufreq_policy *policy) >> if (ret) >> goto free_cpudata1; >> >> - min_freq = READ_ONCE(cpudata->min_freq); >> - max_freq = READ_ONCE(cpudata->max_freq); >> - >> - policy->cpuinfo.min_freq = min_freq; >> - policy->cpuinfo.max_freq = max_freq; >> + policy->cpuinfo.min_freq = policy->min = perf_to_freq(cpudata, cpudata->lowest_perf); >> + policy->cpuinfo.max_freq = policy->max = perf_to_freq(cpudata, cpudata->highest_perf); >> /* It will be updated by governor */ >> policy->cur = policy->cpuinfo.min_freq; >> >> policy->driver_data = cpudata; >> >> - policy->min = policy->cpuinfo.min_freq; >> - policy->max = policy->cpuinfo.max_freq; >> - >> policy->boost_enabled = READ_ONCE(cpudata->boost_supported); >> >> /* >> @@ -1550,7 +1525,8 @@ static int amd_pstate_epp_update_limit(struct cpufreq_policy *policy) >> struct amd_cpudata *cpudata = policy->driver_data; >> u8 epp; >> >> - amd_pstate_update_min_max_limit(policy); >> + if (policy->min != cpudata->min_limit_freq || policy->max != cpudata->max_limit_freq) >> + amd_pstate_update_min_max_limit(policy); >> >> if (cpudata->policy == CPUFREQ_POLICY_PERFORMANCE) >> epp = 0; >> diff --git a/drivers/cpufreq/amd-pstate.h b/drivers/cpufreq/amd-pstate.h >> index 19d405c6d805e..0149933692458 100644 >> --- a/drivers/cpufreq/amd-pstate.h >> +++ b/drivers/cpufreq/amd-pstate.h >> @@ -46,8 +46,6 @@ struct amd_aperf_mperf { >> * @max_limit_perf: Cached value of the performance corresponding to policy->max >> * @min_limit_freq: Cached value of policy->min (in khz) >> * @max_limit_freq: Cached value of policy->max (in khz) >> - * @max_freq: the frequency (in khz) that mapped to highest_perf >> - * @min_freq: the frequency (in khz) that mapped to lowest_perf >> * @nominal_freq: the frequency (in khz) that mapped to nominal_perf >> * @lowest_nonlinear_freq: the frequency (in khz) that mapped to lowest_nonlinear_perf >> * @cur: Difference of Aperf/Mperf/tsc count between last and current sample >> @@ -77,11 +75,8 @@ struct amd_cpudata { >> u8 prefcore_ranking; >> u8 min_limit_perf; >> u8 max_limit_perf; >> - u32 min_limit_freq; >> - u32 max_limit_freq; >> - >> - u32 max_freq; >> - u32 min_freq; >> + u32 min_limit_freq; >> + u32 max_limit_freq; >> u32 nominal_freq; >> u32 lowest_nonlinear_freq; >> >
On 2/19/2025 09:25, Gautham R. Shenoy wrote: > On Mon, Feb 17, 2025 at 04:07:06PM -0600, Mario Limonciello wrote: >> From: Mario Limonciello <mario.limonciello@amd.com> >> >> The CPPC enable register is configured as "write once". That is >> any future writes don't actually do anything. >> >> Because of this, all the cleanup paths that currently exist for >> CPPC disable are non-effective. >> >> Rework CPPC enable to only enable after all the CAP registers have >> been read to avoid enabling CPPC on CPUs with invalid _CPC or >> unpopulated MSRs. >> >> As the register is write once, remove all cleanup paths as well. >> >> Signed-off-by: Mario Limonciello <mario.limonciello@amd.com> >> --- >> v3: >> * Fixup for suspend/resume issue >> --- > [..snip..] > >> >> -static int shmem_cppc_enable(bool enable) >> +static int shmem_cppc_enable(struct cpufreq_policy *policy) >> { >> - int cpu, ret = 0; >> + struct amd_cpudata *cpudata = policy->driver_data; >> struct cppc_perf_ctrls perf_ctrls; >> + int ret; >> >> - if (enable == cppc_enabled) >> - return 0; >> + ret = cppc_set_enable(cpudata->cpu, 1); >> + if (ret) >> + return ret; >> >> - for_each_present_cpu(cpu) { >> - ret = cppc_set_enable(cpu, enable); >> + /* Enable autonomous mode for EPP */ >> + if (cppc_state == AMD_PSTATE_ACTIVE) { >> + /* Set desired perf as zero to allow EPP firmware control */ >> + perf_ctrls.desired_perf = 0; >> + ret = cppc_set_perf(cpudata->cpu, &perf_ctrls); >> if (ret) >> return ret; > > We don't need the if condition here. There is nothing following this > inside the if block and the function return "ret" soon after coming > out of this if block. > > >> - >> - /* Enable autonomous mode for EPP */ >> - if (cppc_state == AMD_PSTATE_ACTIVE) { >> - /* Set desired perf as zero to allow EPP firmware control */ >> - perf_ctrls.desired_perf = 0; >> - ret = cppc_set_perf(cpu, &perf_ctrls); >> - if (ret) >> - return ret; >> - } >> } >> >> - cppc_enabled = enable; >> return ret; >> } >> >> DEFINE_STATIC_CALL(amd_pstate_cppc_enable, msr_cppc_enable); >> >> -static inline int amd_pstate_cppc_enable(bool enable) > [..snip..] > >> >> -static int amd_pstate_epp_reenable(struct cpufreq_policy *policy) >> +static int amd_pstate_epp_cpu_online(struct cpufreq_policy *policy) >> { >> struct amd_cpudata *cpudata = policy->driver_data; >> union perf_cached perf = READ_ONCE(cpudata->perf); >> int ret; >> >> - ret = amd_pstate_cppc_enable(true); >> - if (ret) >> - pr_err("failed to enable amd pstate during resume, return %d\n", ret); >> - >> - >> - return amd_pstate_epp_update_limit(policy); >> -} >> + pr_debug("AMD CPU Core %d going online\n", cpudata->cpu); >> >> -static int amd_pstate_epp_cpu_online(struct cpufreq_policy *policy) >> -{ >> - struct amd_cpudata *cpudata = policy->driver_data; >> - int ret; >> + ret = amd_pstate_cppc_enable(policy); >> + if (ret) >> + return ret; >> >> - pr_debug("AMD CPU Core %d going online\n", cpudata->cpu); >> >> - ret = amd_pstate_epp_reenable(policy); >> + ret = amd_pstate_update_perf(policy, 0, 0, perf.highest_perf, cpudata->epp_cached, false); > > Previously, when a CPU came online, the callpath would be > amd_pstate_epp_cpu_online(policy) > --> amd_pstate_epp_reenable(policy) > --> amd_pstate_epp_update_limit(policy) > --> amd_pstate_epp_update_limit(policy) > > which reevaluates the min_perf_limit and max_perf_limit based on > policy->min and policy->max and then calls > > amd_pstate_update_perf(policy, min_limit_perf, 0, max_limit_perf, epp, false) > > With this patch, we call > > amd_pstate_update_perf(policy, 0, 0, perf.highest_perf, cpudata->epp_cached, false); > > which would set CPPC.min_perf to 0. > > I guess this should be ok since cpufreq_online() would eventually call > amd_pstate_verify() and amd_pstate_epp_set_policy() which should > re-initialize the the min_limit_perf and max_limit_perf. Though I > haven't verified if the behaviour changes with this patch when the CPU > is offlined and brought back online. I'll double check with removing amd_pstate_update_perf() call from amd_pstate_epp_cpu_online(). I think it will be clearer to let it get set from the amd_pstate_epp_set_policy() call.
From: Mario Limonciello <mario.limonciello@amd.com> This series overhauls locking and drops many unnecessarily cached variables. Debugging messages are also dropped in favor of more ftracing. This series is based off superm1/linux.git bleeding-edge branch. v2->v3: * Mostly pick up tags * Add new patch 1/18 that fixes a KBZ issue (see patch for detail) * Fixup for min_freq issue in dropping cached values patch * Fixup for unit tests to only run on online CPUs Mario Limonciello (18): cpufreq/amd-pstate: Invalidate cppc_req_cached during suspend cpufreq/amd-pstate: Show a warning when a CPU fails to setup cpufreq/amd-pstate: Drop min and max cached frequencies cpufreq/amd-pstate: Move perf values into a union cpufreq/amd-pstate: Overhaul locking cpufreq/amd-pstate: Drop `cppc_cap1_cached` cpufreq/amd-pstate-ut: Use _free macro to free put policy cpufreq/amd-pstate-ut: Allow lowest nonlinear and lowest to be the same cpufreq/amd-pstate-ut: Drop SUCCESS and FAIL enums cpufreq/amd-pstate-ut: Run on all of the correct CPUs cpufreq/amd-pstate-ut: Adjust variable scope for amd_pstate_ut_check_freq() cpufreq/amd-pstate: Replace all AMD_CPPC_* macros with masks cpufreq/amd-pstate: Cache CPPC request in shared mem case too cpufreq/amd-pstate: Move all EPP tracing into *_update_perf and *_set_epp functions cpufreq/amd-pstate: Update cppc_req_cached for shared mem EPP writes cpufreq/amd-pstate: Drop debug statements for policy setting cpufreq/amd-pstate: Rework CPPC enabling cpufreq/amd-pstate: Stop caching EPP arch/x86/include/asm/msr-index.h | 20 +- arch/x86/kernel/acpi/cppc.c | 4 +- drivers/cpufreq/amd-pstate-trace.h | 13 +- drivers/cpufreq/amd-pstate-ut.c | 209 +++++----- drivers/cpufreq/amd-pstate.c | 592 ++++++++++++++--------------- drivers/cpufreq/amd-pstate.h | 61 +-- 6 files changed, 428 insertions(+), 471 deletions(-)