[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <aa80728e-a2fc-4b2a-b441-fc884abd0535@broadcom.com>
Date: Fri, 3 May 2024 13:08:56 -0700
From: Florian Fainelli <florian.fainelli@...adcom.com>
To: Christian Marangi <ansuelsmth@...il.com>
Cc: Hauke Mehrtens <hauke@...ke-m.de>, Rafał Miłecki
<zajec5@...il.com>, Thomas Bogendoerfer <tsbogend@...ha.franken.de>,
Rob Herring <robh@...nel.org>,
Krzysztof Kozlowski <krzysztof.kozlowski+dt@...aro.org>,
Conor Dooley <conor+dt@...nel.org>,
Broadcom internal kernel review list
<bcm-kernel-feedback-list@...adcom.com>,
Álvaro Fernández Rojas <noltari@...il.com>,
linux-mips@...r.kernel.org, devicetree@...r.kernel.org,
linux-kernel@...r.kernel.org, Daniel González Cabanelas
<dgcbueu@...il.com>
Subject: Re: [PATCH 6/6] bmips: dma: drop redundant boot_cpu_type in
arch_dma_sync
On 5/3/24 12:39, Christian Marangi wrote:
> On Fri, May 03, 2024 at 12:07:45PM -0700, Florian Fainelli wrote:
>> On 5/3/24 06:54, Christian Marangi wrote:
>>> Drop redundant boot_cpu_type in arch_sync_dma_for_cpu_all. These needs
>>> to be parsed only once and we can make use of bmips_rac_flush_disable to
>>> disable RAC flush on unsupported CPU.
>>>
>>> Set this value in bmips_cpu_setup for unsupported CPU to skip this
>>> redundant check every time DMA needs to be synced.
>>>
>>> Signed-off-by: Christian Marangi <ansuelsmth@...il.com>
>>
>> You are taking a shortcut that is reasonable in premise, but keying off the
>> bmips_rac_flush_disable is IMHO misleading. The RAC is enabled in the
>> BMIPS5000 and BMIPS5200 cores, just it does not need SW management unlike
>> earlier cores.
>>
>> If you renamed it to bmips_rac_flush_needed that might be more compelling.
>> Also, the other reason is that on a kernel that was configured for
>> supporting only BMIPS5000 and BMIPS5200 CPUs, I think we could get some
>> decent dead code elimination of the boot_cpu_type() check, which would not
>> be the case.
>
> I was a bit confused by the last part, should I drop this or just rename
> the variable? Cause I think for kernel that support ONLY those CPU I
> guess the DMA function will be optimized anyway since the bool will
> always be false I guess?
I don't think it can be optimized, I would drop that patch. This is a
hot-path and so any optimization is welcome.
--
Florian
Download attachment "smime.p7s" of type "application/pkcs7-signature" (4221 bytes)
Powered by blists - more mailing lists