Re: [PATCH] cxl/hdm: Add support for 32 switch decoders

[Li Ming <ming.li@zohomail.com>]

在 2026/3/20 10:03, Alison Schofield 写道:
> On Thu, Mar 19, 2026 at 07:11:34PM +0800, Li Ming wrote:
>> 在 2026/3/18 23:40, Dave Jiang 写道:
>>> On 3/18/26 6:00 AM, Li Ming wrote:
>>>> Per CXL r4.0 section 8.2.4.20.1. CXL host bridge and switch ports can
>>>> support 32 HDM decoders. Current implementation misses some decoders on
>>>> CXL host bridge and switch in the case that the value of Decoder Count
>>>> field in CXL HDM decoder Capability Register is greater than or equal to
>>>> 9.
>>>>
>>>> Update calculation implementation to ensure the decoder count calculation
>>>> is correct for CXL host bridge/switch ports.
>>>>
>>>> Signed-off-by: Li Ming <ming.li@zohomail.com>
>>>> ---
>>>>    drivers/cxl/cxl.h | 7 ++++++-
>>>>    1 file changed, 6 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/cxl/cxl.h b/drivers/cxl/cxl.h
>>>> index 9b947286eb9b..466b8eeefed7 100644
>>>> --- a/drivers/cxl/cxl.h
>>>> +++ b/drivers/cxl/cxl.h
>>>> @@ -77,7 +77,12 @@ static inline int cxl_hdm_decoder_count(u32 cap_hdr)
>>>>    {
>>>>    	int val = FIELD_GET(CXL_HDM_DECODER_COUNT_MASK, cap_hdr);
>>>> -	return val ? val * 2 : 1;
>>>> +	if (!val)
>>>> +		return 1;
>>>> +	else if (val <= 8)
>>>> +		return val * 2;
>>>> +	else
>>>> +		return 4 * (val - 4);
>>> That looks a bit messy. How about:
>>>
>>> switch (val) {
>>> case 0:
>>> 	return 1;
>>> case 1..8:
>>> 	return val * 2;
>>> case 9..12:
>>> 	return val 4 * (val - 4);
>>> default:
>>> 	return -ENXIO;
>>> }
>>>
>>> DJ
>> Sure, will do that in v2.
> Hi Ming,
>
> You may have already spied this while implementing. Before this change
> this function could only return 1 or val*2, so never negative. Adding
> -ENXIO changes that and could wrap to huge value in the decoder for
> loops. Change decoder_count field to int to clean that up.
>
> Or, you could do something like this:
> 	WARN_ONCE(1, "reserved HDM decoder count field: %d\n", val);
>          return 0;
>
> I like the ENXIO and type cleanup or maybe you have something else?
>
> --Alison
>
Hi Alison,

I considered returnning 0 before, but both Dave and you prefer 
returnning ENXIO solution, I will do that in V2.


Ming