From mboxrd@z Thu Jan 1 00:00:00 1970 Message-ID: <3CF4EAE9.1080505@embeddededge.com> Date: Wed, 29 May 2002 10:51:21 -0400 From: Dan Malek MIME-Version: 1.0 To: David Gibson Cc: Armin , Tom Rini , Armin Kuster , linuxppc-embedded@lists.linuxppc.org, Paul Mackerras Subject: Re: Another OCP enet patch References: <20020527040330.GH16537@zax> <20020527162323.GB32718@opus.bloom.county> <20020528005728.GO16537@zax> <20020528012516.GE1295@opus.bloom.county> <3CF32B87.4010908@pacbell.net> <20020528065027.GT16537@zax> <3CF3612B.8020102@embeddededge.com> <20020529034854.GC16537@zax> Content-Type: text/plain; charset=us-ascii; format=flowed Sender: owner-linuxppc-embedded@lists.linuxppc.org List-Id: David Gibson wrote: > Well, actually, dma_cache_wback() and friends still appear to be more > widely used than consistent_sync(). AFAICT only PPC and ARM use > consistent_sync(). Just for a historical note.....these functions were implemented about the same time to support non-PCI USB controllers. They have found other uses since then. Other architectures seem to assume PCI is always present, and will call the dma_cache_* functions within the pci functions to get the same effect. For non-PCI devices, they still call the pci functions with a null pointer for the pci_dev. On PPC and ARM we had systems that wouldn't compile properly with PCI enabled, since there wasn't any PCI bridge support (even fake ones :-). We just added the consistent_sync() to be orthogonal with the other archtecture independent consistent_* functions. The other architectures are moving this way as they are finding it more difficult to continue to fake a PCI on systems that really don't have it. -- Dan ** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/