Re: [PATCH] crypto: x86/crc32c-intel - Don't match some Zhaoxin CPUs

[Tony W Wang-oc <TonyWWang-oc@zhaoxin.com>]

On 11/12/2020 21:00, Peter Zijlstra wrote:
> On Fri, Dec 11, 2020 at 07:29:04PM +0800, Tony W Wang-oc wrote:
>> The driver crc32c-intel match CPUs supporting X86_FEATURE_XMM4_2.
>> On platforms with Zhaoxin CPUs supporting this X86 feature, When
>> crc32c-intel and crc32c-generic are both registered, system will
>> use crc32c-intel because its .cra_priority is greater than
>> crc32c-generic. This case expect to use crc32c-generic driver for
>> some Zhaoxin CPUs to get performance gain, So remove these Zhaoxin
>> CPUs support from crc32c-intel.
>>
>> Signed-off-by: Tony W Wang-oc <TonyWWang-oc@zhaoxin.com>
>> ---
>>  arch/x86/crypto/crc32c-intel_glue.c | 8 ++++++++
>>  1 file changed, 8 insertions(+)
>>
>> diff --git a/arch/x86/crypto/crc32c-intel_glue.c b/arch/x86/crypto/crc32c-intel_glue.c
>> index feccb52..6dafdae 100644
>> --- a/arch/x86/crypto/crc32c-intel_glue.c
>> +++ b/arch/x86/crypto/crc32c-intel_glue.c
>> @@ -222,8 +222,16 @@ MODULE_DEVICE_TABLE(x86cpu, crc32c_cpu_id);
>>  
>>  static int __init crc32c_intel_mod_init(void)
>>  {
>> +	struct cpuinfo_x86 *c = &boot_cpu_data;
>> +
>>  	if (!x86_match_cpu(crc32c_cpu_id))
>>  		return -ENODEV;
>> +
>> +	if (c->x86_vendor == X86_VENDOR_ZHAOXIN || c->x86_vendor == X86_VENDOR_CENTAUR) {
>> +		if (c->x86 == 0x6 || (c->x86 == 0x7 && c->x86_model <= 0x3b))
>> +			return -ENODEV;
>> +	}
> 
> Egads, why can't you use that x86_match_cpu() above, and also this
> really wants a comment on why you're excluding these chips. 

When doing lmbench3 Create and Delete file test on partitions with ext4
enabling metadata checksum, found using crc32c-generic driver could get
about 20% performance gain than using the driver crc32c-intel on these
chips.

Also, since
> (IIRC) ZHAOXIN is basically AND, shouldn't they also be listed?
> 
> That is; write it like:
> 
> 	m = x86_match_cpu(crc32_cpu_id);
> 	if (!m || !m->data)
> 		return -ENODEV;
> 
> That way you can have positive and negative matches in the array
> (obviously the existing FEATURE test would need data=1 and be last).
> .
> 

Lot thanks for you suggestion, will list these chips in crc32c_cpu_id
and use x86_match_cpu:

 static const struct x86_cpu_id crc32c_cpu_id[] = {
+       X86_MATCH_VENDOR_FAM_FEATURE(ZHAOXIN, 0x6, X86_FEATURE_XMM4_2, 1),
+       X86_MATCH_VENDOR_FAM_MODEL_FEATURE(ZHAOXIN, 0x7, 0x1b,
X86_FEATURE_XMM4_2, 1),
+       X86_MATCH_VENDOR_FAM_MODEL_FEATURE(ZHAOXIN, 0x7, 0x3b,
X86_FEATURE_XMM4_2, 1),
+       X86_MATCH_VENDOR_FAM_FEATURE(CENTAUR, 0x6, X86_FEATURE_XMM4_2, 1),
+       X86_MATCH_VENDOR_FAM_MODEL_FEATURE(CENTAUR, 0x7, 0x1b,
X86_FEATURE_XMM4_2, 1),
+       X86_MATCH_VENDOR_FAM_MODEL_FEATURE(CENTAUR, 0x7, 0x3b,
X86_FEATURE_XMM4_2, 1),
        X86_MATCH_FEATURE(X86_FEATURE_XMM4_2, NULL),
        {}
 };
@@ -228,8 +234,10 @@ MODULE_DEVICE_TABLE(x86cpu, crc32c_cpu_id);

 static int __init crc32c_intel_mod_init(void)
 {
-       if (!x86_match_cpu(crc32c_cpu_id))
+       const struct x86_cpu_id *m = x86_match_cpu(crc32c_cpu_id);
+       if (!m || m->driver_data)
                return -ENODEV;


sincerely
TonyWWangoc