[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH v8 3/8] xen/cpufreq: implement amd-cppc driver for CPPC in passive mode


  • To: "Penny, Zheng" <penny.zheng@xxxxxxx>
  • From: Jan Beulich <jbeulich@xxxxxxxx>
  • Date: Fri, 29 Aug 2025 08:11:46 +0200
  • Autocrypt: addr=jbeulich@xxxxxxxx; keydata= xsDiBFk3nEQRBADAEaSw6zC/EJkiwGPXbWtPxl2xCdSoeepS07jW8UgcHNurfHvUzogEq5xk hu507c3BarVjyWCJOylMNR98Yd8VqD9UfmX0Hb8/BrA+Hl6/DB/eqGptrf4BSRwcZQM32aZK 7Pj2XbGWIUrZrd70x1eAP9QE3P79Y2oLrsCgbZJfEwCgvz9JjGmQqQkRiTVzlZVCJYcyGGsD /0tbFCzD2h20ahe8rC1gbb3K3qk+LpBtvjBu1RY9drYk0NymiGbJWZgab6t1jM7sk2vuf0Py O9Hf9XBmK0uE9IgMaiCpc32XV9oASz6UJebwkX+zF2jG5I1BfnO9g7KlotcA/v5ClMjgo6Gl MDY4HxoSRu3i1cqqSDtVlt+AOVBJBACrZcnHAUSuCXBPy0jOlBhxPqRWv6ND4c9PH1xjQ3NP nxJuMBS8rnNg22uyfAgmBKNLpLgAGVRMZGaGoJObGf72s6TeIqKJo/LtggAS9qAUiuKVnygo 3wjfkS9A3DRO+SpU7JqWdsveeIQyeyEJ/8PTowmSQLakF+3fote9ybzd880fSmFuIEJldWxp Y2ggPGpiZXVsaWNoQHN1c2UuY29tPsJgBBMRAgAgBQJZN5xEAhsDBgsJCAcDAgQVAggDBBYC AwECHgECF4AACgkQoDSui/t3IH4J+wCfQ5jHdEjCRHj23O/5ttg9r9OIruwAn3103WUITZee e7Sbg12UgcQ5lv7SzsFNBFk3nEQQCACCuTjCjFOUdi5Nm244F+78kLghRcin/awv+IrTcIWF hUpSs1Y91iQQ7KItirz5uwCPlwejSJDQJLIS+QtJHaXDXeV6NI0Uef1hP20+y8qydDiVkv6l IreXjTb7DvksRgJNvCkWtYnlS3mYvQ9NzS9PhyALWbXnH6sIJd2O9lKS1Mrfq+y0IXCP10eS FFGg+Av3IQeFatkJAyju0PPthyTqxSI4lZYuJVPknzgaeuJv/2NccrPvmeDg6Coe7ZIeQ8Yj t0ARxu2xytAkkLCel1Lz1WLmwLstV30g80nkgZf/wr+/BXJW/oIvRlonUkxv+IbBM3dX2OV8 AmRv1ySWPTP7AAMFB/9PQK/VtlNUJvg8GXj9ootzrteGfVZVVT4XBJkfwBcpC/XcPzldjv+3 HYudvpdNK3lLujXeA5fLOH+Z/G9WBc5pFVSMocI71I8bT8lIAzreg0WvkWg5V2WZsUMlnDL9 mpwIGFhlbM3gfDMs7MPMu8YQRFVdUvtSpaAs8OFfGQ0ia3LGZcjA6Ik2+xcqscEJzNH+qh8V m5jjp28yZgaqTaRbg3M/+MTbMpicpZuqF4rnB0AQD12/3BNWDR6bmh+EkYSMcEIpQmBM51qM EKYTQGybRCjpnKHGOxG0rfFY1085mBDZCH5Kx0cl0HVJuQKC+dV2ZY5AqjcKwAxpE75MLFkr wkkEGBECAAkFAlk3nEQCGwwACgkQoDSui/t3IH7nnwCfcJWUDUFKdCsBH/E5d+0ZnMQi+G0A nAuWpQkjM1ASeQwSHEeAWPgskBQL
  • Cc: "Huang, Ray" <Ray.Huang@xxxxxxx>, Andrew Cooper <andrew.cooper3@xxxxxxxxxx>, Roger Pau Monné <roger.pau@xxxxxxxxxx>, Anthony PERARD <anthony.perard@xxxxxxxxxx>, "Orzel, Michal" <Michal.Orzel@xxxxxxx>, Julien Grall <julien@xxxxxxx>, Stefano Stabellini <sstabellini@xxxxxxxxxx>, "xen-devel@xxxxxxxxxxxxxxxxxxxx" <xen-devel@xxxxxxxxxxxxxxxxxxxx>, "Andryuk, Jason" <Jason.Andryuk@xxxxxxx>
  • Delivery-date: Fri, 29 Aug 2025 06:11:58 +0000
  • List-id: Xen developer discussion <xen-devel.lists.xenproject.org>

On 29.08.2025 05:30, Penny, Zheng wrote:
>> -----Original Message-----
>> From: Jan Beulich <jbeulich@xxxxxxxx>
>> Sent: Thursday, August 28, 2025 7:23 PM
>>
>> On 28.08.2025 12:03, Penny Zheng wrote:
>>> +static int cf_check amd_cppc_cpufreq_target(struct cpufreq_policy *policy,
>>> +                                            unsigned int target_freq,
>>> +                                            unsigned int relation) {
>>> +    unsigned int cpu = policy->cpu;
>>> +    const struct amd_cppc_drv_data *data = per_cpu(amd_cppc_drv_data,
>>> +cpu);
>>
>> I fear there's a problem here that I so far overlooked. As it happens, just
>> yesterday I made a patch to eliminate cpufreq_drv_data[] global. In the 
>> course of
>> doing so it became clear that in principle the CPU denoted by
>> policy->cpu can be offline. Hence its per-CPU data is also unavailable.
>> policy->See
>> cpufreq_add_cpu()'s invocation of .init() and cpufreq_del_cpu()'s invocation
>> of .exit(). Is there anything well-hidden (and likely lacking some suitable
>> comment) which guarantees that no two CPUs (threads) will be in the same
>> domain? If not, I fear you simply can't use per-CPU data here.
>>
> 
> Correct me if I understand you wrongly:
> No, my env is always per pcpu per cpufreq domain. So it never occurred to me 
> that cpus, other than the first one in domain, will never call .init(), and 
> of course, no per_cpu(amd_cppc_drv_data) ever gets allocated then.

Well, the question is how domains are organized when using the CPPC driver.
Aiui that's still driven by data passed in by Dom0, so in turn the question
is whether there are any constraints on what ACPI may surface. If there are,
all that may be necessary is adding a check. If there aren't, ...

>> Since initially I was thinking of using per-CPU data also in my patch, I'm
>> reproducing this in raw form below, for your reference. It's generally only
>> 4.22 material now, of course. Yet in turn for your driver the new drv_data 
>> field
>> may want to become a union, with an "acpi" and a "cppc" sub-struct.
> 
> How about I embed my new driver data " struct amd_cppc_drv_data * " into 
> cpufreq policy, maybe pointer is enough?
> Later, maybe, all "cppc", "acpi" and "hwp" could constitute an union in 
> policy.

... I'd prefer to go the union approach right away. Whether then to take my
patch as a prereq is tbd; that largely depends on what (if anything) is
needed on the HWP side. If HWP needs fixing, that wants to to come first, as
it would want backporting.

Jan



 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.