Re: [PATCH 2/7] hw/hppa: Make number of TLB and BTLB entries configurable

[Helge Deller]

On 01.09.20 23:36, Richard Henderson wrote:
> On 9/1/20 11:34 AM, Helge Deller wrote:
>> diff --git a/hw/hppa/machine.c b/hw/hppa/machine.c
>> index 90aeefe2a4..e9d84d0f03 100644
>> --- a/hw/hppa/machine.c
>> +++ b/hw/hppa/machine.c
>> @@ -72,6 +72,14 @@ static FWCfgState *create_fw_cfg(MachineState *ms)
>>      fw_cfg_add_file(fw_cfg, "/etc/firmware-min-version",
>>                      g_memdup(&val, sizeof(val)), sizeof(val));
>>
>> +    val = cpu_to_le64(HPPA_TLB_ENTRIES);
>
> I guess you don't have a cpu structure here against which you could apply
> ARRAY_SIZE?

You mean to calculate the number of TLB entries based on the static size
of an array, e.g. ARRAY_SIZE(struct CPUHPPAState.tlb[256]) ?
I've replaced it intentionally by a constant, because in a follow-up
patch to improve the TLB-lookup speed, the array gets replaced by a GTree.
You can see a working version of the patch here:
https://github.com/hdeller/qemu-hppa/commit/644790132b91cdb835c8dd38198e6f6ed0b533a1

In this patch:
-   hppa_tlb_entry tlb[HPPA_TLB_ENTRIES];
+   GTree *tlb;
+   GList *tlb_list;
+   int tlb_entries;

>>      /* ??? The number of entries isn't specified by the architecture.  */
>> +    #define HPPA_TLB_ENTRIES        256
>> +    #define HPPA_BTLB_ENTRIES       0
>
> The indented defines are weird.

How/Why?

> What's a btlb entry?

Block-TLB entries. Most PA1.1-machines have 16 such BTLBs which can
address a range of 4k pages with a single entry.
In a follow-up qemu patch it makes sense to add those TLBs.

Helge