Replies: 15 comments 25 replies
-
Sure, I will need the following output to verify how CPPC has been discovered. |
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
|
Is your kernel built with ACPI CPPC support ? |
Beta Was this translation helpful? Give feedback.
-
|
I believe so. |
Beta Was this translation helpful? Give feedback.
-
PrerequisitesDebuggingI have instrumented driver code with some
signed int Read_ACPI_CPPC_Registers(unsigned int cpu, void *arg)
{
#ifdef CONFIG_ACPI_CPPC_LIB
struct cppc_perf_fb_ctrs CPPC_Perf;
struct cppc_perf_caps CPPC_Caps;
CORE_RO *Core = (CORE_RO *) PUBLIC(RO(Core, AT(cpu)));
signed int rc = 0;
if ((rc = cppc_get_perf_ctrs(Core->Bind, &CPPC_Perf)) == 0) {
rc = cppc_get_perf_caps(Core->Bind, &CPPC_Caps);
}
if (rc == 0) {
#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 1, 0)
unsigned long long desired_perf = 0;
#endif
Core->PowerThermal.ACPI_CPPC = (struct ACPI_CPPC_STRUCT) {
.Highest = CPPC_Caps.highest_perf,
.Guaranteed = CPPC_Caps.nominal_perf,
#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 18, 0)
.Efficient = CPPC_Caps.nominal_freq,
.Lowest = CPPC_Caps.lowest_freq,
.Minimum = CPPC_Caps.lowest_freq,
#else
.Efficient = CPPC_Caps.nominal_perf,
.Lowest = CPPC_Caps.lowest_perf,
.Minimum = CPPC_Caps.lowest_perf,
#endif
.Maximum = CPPC_Caps.highest_perf,
.Desired = CPPC_Perf.reference_perf,
.Energy = 0
};
#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 1, 0)
rc = cppc_get_desired_perf(Core->Bind, &desired_perf);
if (rc == 0) {
Core->PowerThermal.ACPI_CPPC.Desired = desired_perf;
}
#endif
} else {
pr_warn("CoreFreq: 'cppc' Execution failure code %d\n", rc);
}
return rc;
#else
pr_warn("CoreFreq: No CONFIG_ACPI_CPPC_LIB support\n");
return -ENODEV;
#endif /* CONFIG_ACPI_CPPC_LIB */
}
void For_All_ACPI_CPPC(signed int(*CPPC_Func)(unsigned int, void*), void *arg)
{
#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 11, 0)
signed int rc = acpi_cpc_valid() == false;
#else
signed int rc = acpi_disabled;
#endif
unsigned int cpu;
pr_info("CoreFreq: 'acpi_cpc' Execution failure code %d\n", rc);
for (cpu = 0; (cpu < PUBLIC(RO(Proc))->CPU.Count) && (rc == 0); cpu++)
{
rc = CPPC_Func(cpu, arg);
}
PUBLIC(RO(Proc))->Features.ACPI_CPPC = (rc == 0);
}
EDIT: I have rebooting with the passing case: ACPI CPPC firmware is now implementing and giving service.
|
Beta Was this translation helpful? Give feedback.
-
|
Using commit deee601 we will now track the BIOS support of _CPC disabled
_CPC enabled but another feature ...
_CPC enabled without any feature preventing Firmware
|
Beta Was this translation helpful? Give feedback.
-
Zen2/LucienneThis Initial state as disabled_thus CPC does not matter in this case Request to enable CPPC
CPPC activated
|
Beta Was this translation helpful? Give feedback.
-
|
Holy crap, god bless you. My BIOS does not have any setting for X2APIC/Local APIC, sadly ;-; (ASRock B450M Steel Legend) Do you know of any other names for this setting? |
Beta Was this translation helpful? Give feedback.
-
|
Yes please.
signed int Read_ACPI_CPPC_Registers(unsigned int cpu, void *arg)
{
#ifdef CONFIG_ACPI_CPPC_LIB
struct cppc_perf_fb_ctrs CPPC_Perf;
struct cppc_perf_caps CPPC_Caps;
CORE_RO *Core = (CORE_RO *) PUBLIC(RO(Core, AT(cpu)));
signed int rc = 0;
if ((rc = cppc_get_perf_ctrs(Core->Bind, &CPPC_Perf)) == 0) {
if ((rc = cppc_get_perf_caps(Core->Bind, &CPPC_Caps)) != 0)
pr_err("CoreFreq: cppc_get_perf_caps() error %d\n", rc);
} else {
pr_err("CoreFreq: cppc_get_perf_ctrs() error %d\n", rc);
}
if (rc == 0) {
#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 1, 0)
unsigned long long desired_perf = 0;
#endif
Core->PowerThermal.ACPI_CPPC = (struct ACPI_CPPC_STRUCT) {
.Highest = CPPC_Caps.highest_perf,
.Guaranteed = CPPC_Caps.nominal_perf,
#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 18, 0)
.Efficient = CPPC_Caps.nominal_freq,
.Lowest = CPPC_Caps.lowest_freq,
.Minimum = CPPC_Caps.lowest_freq,
#else
.Efficient = CPPC_Caps.nominal_perf,
.Lowest = CPPC_Caps.lowest_perf,
.Minimum = CPPC_Caps.lowest_perf,
#endif
.Maximum = CPPC_Caps.highest_perf,
.Desired = CPPC_Perf.reference_perf,
.Energy = 0
};
#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 1, 0)
rc = cppc_get_desired_perf(Core->Bind, &desired_perf);
if (rc == 0) {
Core->PowerThermal.ACPI_CPPC.Desired = desired_perf;
} else {
pr_err("CoreFreq: cppc_get_desired_perf() error %d\n", rc);
}
#endif
}
return rc;
#else
return -ENODEV;
#endif /* CONFIG_ACPI_CPPC_LIB */
}
|
Beta Was this translation helpful? Give feedback.
-
|
Yep, here's the error.
(My performance monitoring window) |
Beta Was this translation helpful? Give feedback.
-
|
Call flow should have ended in kernel function cppc_get_perf() but I don't find why an EINVAL exit happened. |
Beta Was this translation helpful? Give feedback.
-
Do you want more info? I believe I have ftrace/audit disabled. |
Beta Was this translation helpful? Give feedback.
-
|
I'm not sure if we have diverging kernel built but I'm developing on ArchLinux. www.cyring.fr , link DEVELOPMENT, I'm providing an ISO image with ArchLinux + CoreFreq. It's an autorun image, you just have to If you have CPPC enabled then issue is within kernel. If not then ACPI CPPC could be out of specs inside Asrock BIOS implementation and kernel is failing at |
Beta Was this translation helpful? Give feedback.
-
|
I just tried the Live ISO and it works. It is just an issue with my kernel, I have linux-cachyos-bore with the amd_pstate_epp patches so that's the problem. |
Beta Was this translation helpful? Give feedback.
-
|
Took the amd_pstate patch out of the kernel, recompiled, and now it works. God bless you cyring |
Beta Was this translation helpful? Give feedback.
-
|
Your welcome. Feel free to update the CLI output and screenshots. I have added the on-going into the Wiki CPU Support, processor Ryzen 5 5600X I was not aware of the BORE scheduler, but those look like major kernel patches. The same we, Contributers, did for MUQSS and BMQ; we can add BORE to kernel flavors supported by CoreFreq Regards |
Beta Was this translation helpful? Give feedback.
-
I recommend linux-cachyos-bore, I've used it for a long time and it is the default CachyOS kernel. Also, the amd_pstate patch I was referring to (amd_pstate_epp v4) is going to be merged into the kernel relatively soon, either 6.1 or 6.2, it's the newest version of the amd_pstate driver essentially. It is in linux-cachyos-bore by default for 6.0 and 6.1-rc though. |
Beta Was this translation helpful? Give feedback.
-
I'm choosing it but I have added the key
It is clear now that I've to make CoreFreq compatible with those patches. I will come back here when I'm done experimenting BORE and the preliminary developments for the driver. |
Beta Was this translation helpful? Give feedback.
-
|
Oh yeah, those can happen with linux-cachyos-bore because the patches get updated frequently. I just pass --skipinteg to makepkg as the patches are pulled from the CachyOS Github anyway. Also, I started using CoreFreq on a Ryzen 5 2400G due to the TSC being unreliable, so I know for a fact CoreFreq works on that CPU. |
Beta Was this translation helpful? Give feedback.
-
|
Looking at kernel, Patch has changed code as below: static int cppc_get_perf(int cpunum, enum cppc_regs reg_idx, u64 *perf)
{
struct cpc_desc *cpc_desc = per_cpu(cpc_desc_ptr, cpunum);
struct cpc_register_resource *reg;
int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpunum);
struct cppc_pcc_data *pcc_ss_data = NULL;
int ret = -EINVAL;
if (!cpc_desc) {
pr_debug("No CPC descriptor for CPU:%d\n", cpunum);
return -ENODEV;
}
reg = &cpc_desc->cpc_regs[reg_idx];
if (CPC_IN_PCC(reg)) {
if (pcc_ss_id < 0)
return -EIO;
pcc_ss_data = pcc_data[pcc_ss_id];
down_write(&pcc_ss_data->pcc_lock);
if (send_pcc_cmd(pcc_ss_id, CMD_READ) >= 0)
cpc_read(cpunum, reg, perf);
else
ret = -EIO;
up_write(&pcc_ss_data->pcc_lock);
return ret;
}
cpc_read(cpunum, reg, perf);
return ret;
}
I can change CoreFreq to handle |
Beta Was this translation helpful? Give feedback.
-
|
Based on your indication, I have search and found this patch
If all theses changes are going into future mainstream 6.1 or later versions, it's worth working on CoreFreq driver adaptions when release candidates will show up. EDIT: See below for the committed workaround |
Beta Was this translation helpful? Give feedback.
-
|
Meanwhile I have made a working fix for |
Beta Was this translation helpful? Give feedback.
-
|
Commit 6b80b85 is adding the support of the build directive You should now be able to run CoreFreq with no change in your kernel setup. |
Beta Was this translation helpful? Give feedback.
-
|
Wow. Thank you so much. Do you have a donation link? Also, the CachyOS repos now have corefreq-git! |
Beta Was this translation helpful? Give feedback.
-
I'm still asking myself about that. CoreFreq motivations are: my Engineer story ...
Love this. |
Beta Was this translation helpful? Give feedback.
-
|
Please come back here if you are aware of any change in this CPPC lib implementation. |
Beta Was this translation helpful? Give feedback.
-
|
Thank you!! BORE was just updated from 1.7.0 to 1.7.1 on linux-cachyos-bore, I just updated, corefreqk doesn't give any errors or anything so I assume everything's fine but just wanted to let you know. CachyOS/kernel-patches@cb044be https://github.com/CachyOS |
Beta Was this translation helpful? Give feedback.
-
|
Thank you. I wanted to try But Based on its source code I've added Fyi my full command line which works perfectly with the mainstream I've also "fixed" the CPPC lib to return zero rather than Does |
Beta Was this translation helpful? Give feedback.
-
Can you make the patch available through the AUR package ? |
Beta Was this translation helpful? Give feedback.
-
Commit 4bf9f02 within
I've altered the AUR |
Beta Was this translation helpful? Give feedback.
-
|
@gel-crabs Hello,
Usually when a driver has dependencies, it has to be specified. But here, none. |
Beta Was this translation helpful? Give feedback.
-
|
I found driver Now getting these results: Not very detailed, isn't it. What I find significant is within function Here's the instruction to print values in CoreFreq: signed int Read_ACPI_CPPC_Registers(unsigned int cpu, void *arg)
{
...
printk("CPU#%d\tACPI_CPPC.Highest=%d\n", cpu, CPPC_Caps.highest_perf);
}
return rc;
#else
return -ENODEV;
#endif /* CONFIG_ACPI_CPPC_LIB */
} |
Beta Was this translation helpful? Give feedback.
-
|
Since commit 80d6e11 HWP or CPPC capabilities are refactored to print values per CPU Intel Hardware HWPBefore enabling HWP
HWP enabled
Remark (EDIT)
AMD Firmware CPPC
Remarks
Please try and let me know ... |
Beta Was this translation helpful? Give feedback.
-
|
Beta Was this translation helpful? Give feedback.
-
|
TODO when those get merged into kernel: Will impact this line: Line 3911 in 6a4b450 |
Beta Was this translation helpful? Give feedback.
-
|
Since kernel
|
Beta Was this translation helpful? Give feedback.
-
Hello, I have a Ryzen 5 5600X and I am unable to enable CPPC (CPPC/CPPC-PC is enabled in BIOS). acpi_cppc gives the preferred cores information however. I've gathered that the MSRs are not known for this model of CPU; is there any way I can help? I would be glad to provide any information.
Beta Was this translation helpful? Give feedback.
All reactions