Re: [PATCH v5 13/26] powerpc/book3s64/pkeys: Enable MMU_FTR_PKEY

["Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com>]

On 7/6/20 6:40 PM, Michael Ellerman wrote:
> "Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com> writes:
>> Parse storage keys related device tree entry in early_init_devtree
>> and enable MMU feature MMU_FTR_PKEY if pkeys are supported.
>>
>> MMU feature is used instead of CPU feature because this enables us
>> to group MMU_FTR_KUAP and MMU_FTR_PKEY in asm feature fixup code.
> 
> Subject should probably be "Add MMU_FTR_PKEY".
> 
>> diff --git a/arch/powerpc/include/asm/mmu.h b/arch/powerpc/include/asm/mmu.h
>> index f4ac25d4df05..72966d3d8f64 100644
>> --- a/arch/powerpc/include/asm/mmu.h
>> +++ b/arch/powerpc/include/asm/mmu.h
>> @@ -23,6 +23,7 @@
>>   
>>   /* Radix page table supported and enabled */
>>   #define MMU_FTR_TYPE_RADIX		ASM_CONST(0x00000040)
>> +#define MMU_FTR_PKEY			ASM_CONST(0x00000080)
> 
> It's not a type, so it should be with the individual feature bits below:
> 


We don't have free bit in the other group. For now i will move this to

modified   arch/powerpc/include/asm/mmu.h
@@ -23,12 +23,15 @@

  /* Radix page table supported and enabled */
  #define MMU_FTR_TYPE_RADIX		ASM_CONST(0x00000040)
-#define MMU_FTR_PKEY			ASM_CONST(0x00000080)

  /*
   * Individual features below.
   */

+/*
+ * Support for memory protection keys.
+ */
+#define MMU_FTR_PKEY			ASM_CONST(0x00001000)
  /*
   * Support for 68 bit VA space. We added that from ISA 2.05
   */



>>   /*
>>    * Individual features below.
> 
>> diff --git a/arch/powerpc/kernel/prom.c b/arch/powerpc/kernel/prom.c
>> index 6a3bac357e24..6d70797352d8 100644
>> --- a/arch/powerpc/kernel/prom.c
>> +++ b/arch/powerpc/kernel/prom.c
>> @@ -815,6 +815,11 @@ void __init early_init_devtree(void *params)
>>   	/* Now try to figure out if we are running on LPAR and so on */
>>   	pseries_probe_fw_features();
>>   
>> +	/*
>> +	 * Initialize pkey features and default AMR/IAMR values
>> +	 */
>> +	pkey_early_init_devtree();
> 
> Not your fault, but the fact that we're having to do more and more
> initialisation this early, based on the flat device tree, makes me
> wonder if we need to rethink how we're doing the CPU/MMU feature setup.
> 
>> diff --git a/arch/powerpc/mm/book3s64/pkeys.c b/arch/powerpc/mm/book3s64/pkeys.c
>> index 0ff59acdbb84..bbba9c601e14 100644
>> --- a/arch/powerpc/mm/book3s64/pkeys.c
>> +++ b/arch/powerpc/mm/book3s64/pkeys.c
>> @@ -38,38 +39,45 @@ static int execute_only_key = 2;
>>   #define PKEY_REG_BITS (sizeof(u64) * 8)
>>   #define pkeyshift(pkey) (PKEY_REG_BITS - ((pkey+1) * AMR_BITS_PER_PKEY))
>>   
>> +static int __init dt_scan_storage_keys(unsigned long node,
>> +				       const char *uname, int depth,
>> +				       void *data)
>> +{
>> +	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
>> +	const __be32 *prop;
>> +	int pkeys_total;
>> +
>> +	/* We are scanning "cpu" nodes only */
>> +	if (type == NULL || strcmp(type, "cpu") != 0)
>> +		return 0;
>> +
>> +	prop = of_get_flat_dt_prop(node, "ibm,processor-storage-keys", NULL);
>> +	if (!prop)
>> +		return 0;
>> +	pkeys_total = be32_to_cpu(prop[0]);
>> +	return pkeys_total;
> 
> That's not really the way the return value is meant to be used for these
> scanning functions.
> 
> The usual return values are 0 to keep scanning and !0 means we've found
> the node we're looking for and we should stop scanning.
> 
> Doing it this way means if you find 0 pkeys it will keep scanning.
> 
> Instead you should pass &pkeys_total as the data pointer and set that.
> 
>> +}
>> +


done

modified   arch/powerpc/mm/book3s64/pkeys.c
@@ -52,7 +52,7 @@ static int __init dt_scan_storage_keys(unsigned long node,
  {
  	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
  	const __be32 *prop;
-	int pkeys_total;
+	int *pkeys_total = (int *) data;

  	/* We are scanning "cpu" nodes only */
  	if (type == NULL || strcmp(type, "cpu") != 0)
@@ -61,12 +61,13 @@ static int __init dt_scan_storage_keys(unsigned long 
node,
  	prop = of_get_flat_dt_prop(node, "ibm,processor-storage-keys", NULL);
  	if (!prop)
  		return 0;
-	pkeys_total = be32_to_cpu(prop[0]);
-	return pkeys_total;
+	*pkeys_total = be32_to_cpu(prop[0]);
+	return 1;
  }

  static int scan_pkey_feature(void)
  {
+	int ret;
  	int pkeys_total;

  	/*
@@ -75,8 +76,8 @@ static int scan_pkey_feature(void)
  	if (early_radix_enabled())
  		return 0;

-	pkeys_total = of_scan_flat_dt(dt_scan_storage_keys, NULL);
-	if (pkeys_total == 0) {
+	ret = of_scan_flat_dt(dt_scan_storage_keys, &pkeys_total);
+	if (ret == 0) {

  		/*
  		 * Let's assume 32 pkeys on P8/P9 bare metal, if its not defined by 
device
@@ -87,7 +88,6 @@ static int scan_pkey_feature(void)
  		    (early_cpu_has_feature(CPU_FTR_ARCH_207S) ||
  		     early_cpu_has_feature(CPU_FTR_ARCH_300)))
  			pkeys_total = 32;
-
  	}

  	/*


>>   static int scan_pkey_feature(void)
>>   {
>> -	u32 vals[2];
>> -	int pkeys_total = 0;
>> -	struct device_node *cpu;
>> +	int pkeys_total;
>>   
>>   	/*
>>   	 * Pkey is not supported with Radix translation.
>>   	 */
>> -	if (radix_enabled())
>> +	if (early_radix_enabled())
>>   		return 0;
>>   
>> -	cpu = of_find_node_by_type(NULL, "cpu");
>> -	if (!cpu)
>> -		return 0;
>> +	pkeys_total = of_scan_flat_dt(dt_scan_storage_keys, NULL);
>> +	if (pkeys_total == 0) {
>>   
>> -	if (of_property_read_u32_array(cpu,
>> -				       "ibm,processor-storage-keys", vals, 2) == 0) {
>> -		/*
>> -		 * Since any pkey can be used for data or execute, we will
>> -		 * just treat all keys as equal and track them as one entity.
>> -		 */
>> -		pkeys_total = vals[0];
>> -		/*  Should we check for IAMR support FIXME!! */
> 
> ???

The device tree allows us to have different count for both AMR and IAMR.
The current code skip that. I guess i added a comment in earlier patch 
to check that whether we need to handle different AMR and IAMR counts. 
The same comment get dropped here.


> 
>> -	} else {
> 
> This loses the ability to differentiate between no storage-keys property
> at all vs a property that specifies zero keys, which I don't think is a
> good change.
> 
>>   		/*
>>   		 * Let's assume 32 pkeys on P8 bare metal, if its not defined by device
>>   		 * tree. We make this exception since skiboot forgot to expose this
>>   		 * property on power8.
>>   		 */
>>   		if (!firmware_has_feature(FW_FEATURE_LPAR) &&
>> -		    cpu_has_feature(CPU_FTRS_POWER8))
>> +		    early_cpu_has_feature(CPU_FTRS_POWER8))
>>   			pkeys_total = 32;
> 
> That's not how cpu_has_feature() works, we'll need to fix that.
> 
> cheers
> 

I did a separate patch to handle that which switch the above to

		/*
		 * Let's assume 32 pkeys on P8/P9 bare metal, if its not defined by device
		 * tree. We make this exception since skiboot forgot to expose this
		 * property on power8/9.
		 */
		if (!firmware_has_feature(FW_FEATURE_LPAR) &&
		    (early_cpu_has_feature(CPU_FTR_ARCH_207S) ||
		     early_cpu_has_feature(CPU_FTR_ARCH_300)))
			pkeys_total = 32;