Re: [PATCH net-next v2 2/7] net: airoha: npu: Add NPU wlan memory initialization commands

[Lorenzo Bianconi <lorenzo@kernel.org>]

[-- Attachment #1: Type: text/plain, Size: 3186 bytes --]

> On 07/07/2025 17:24, Lorenzo Bianconi wrote:
> >> On 07/07/2025 09:24, Lorenzo Bianconi wrote:
> >>>> On Sat, Jul 05, 2025 at 11:09:46PM +0200, Lorenzo Bianconi wrote:
> >>>>> +
> >>>>>  struct airoha_npu *airoha_npu_get(struct device *dev, dma_addr_t *stats_addr)
> >>>>>  {
> >>>>>  	struct platform_device *pdev;
> >>>>> @@ -493,6 +573,7 @@ static int airoha_npu_probe(struct platform_device *pdev)
> >>>>>  	npu->ops.ppe_deinit = airoha_npu_ppe_deinit;
> >>>>>  	npu->ops.ppe_flush_sram_entries = airoha_npu_ppe_flush_sram_entries;
> >>>>>  	npu->ops.ppe_foe_commit_entry = airoha_npu_foe_commit_entry;
> >>>>> +	npu->ops.wlan_init_reserved_memory = airoha_npu_wlan_init_memory;
> >>>>
> >>>> I cannot find in your code single place calling this (later you add a
> >>>> wrapper... which is not called either).
> >>>>
> >>>> All this looks like dead code...
> >>>
> >>> As pointed out in the commit log, these callbacks will be used by MT76 driver
> >>> to initialize the NPU reserved memory and registers during driver probe in
> >>> order to initialize the WiFi offloading. Since MT76 patches are going via
> >>> the wireless tree, I needed to add these callbacks first.
> >>
> >> Cover letter does not link to your NPU patchset. You cannot add dead
> >> code to the kernel and now it is pure dead code. Post your user - in
> >> this or separate patchset.
> > 
> > I guess you mean the related MT76 patches are not linked in the cover-letter,
> > right? I have not posted them yet.
> > 
> >>
> >> Your explanation of dependency is also confusing. If these are added to
> >> wireless tree (considering last experience how they rebase and cannot
> >> easily handle cross tree merges), how does it solve your problem? You
> >> will have it in one tree but not in the other, so still nothing...
> >> That's anyway separate problem, because main issue is you add code which
> >> we cannot even verify how it is being used.
> > 
> > My main point here is wireless tree can't acutally merge the MT76 patches
> > since, without the net-next ones (this series), it will not compile (so I
> 
> This does not explain hiding the other part. Again - there is nothing
> weird in patchset dependencies. Weird is saying there is dependency, so
> I will not post code.

I have a working patchset for MT76 support. I have not posted it yet just
because I need to clean it up.

> 
> > posted net-next patches as preliminary ones for MT76 changes).
> > Moreover, this is the same approach we used when we added WED support to
> > mtk_eth_soc driver and the related MT76 support.
> > However, I am fine to post the MT76 changes as RFC and just refer to it in
> > this series cover-letter. Agree? 
> > 
> >>
> >> So far I see ABI break, but without user cannot judge. And that's the
> >> hard reason this cannot be accepted.
> > 
> > if you mean the dts changes, I will fix them in v3.
> > 
> No, I mean driver.

Sorry, can you please explain what is the ABI break in the driver codebase?
airoha_npu_wlan_init_memory() is executed by MT76 driver and not during NPU
probe phase.

Regards,
Lorenzo

> 
> Best regards,
> Krzysztof

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]