From mboxrd@z Thu Jan 1 00:00:00 1970 From: john.garry@huawei.com (John Garry) Date: Tue, 10 Jan 2017 10:22:36 +0000 Subject: [QUESTION] Early Write Acknowledge for PCIe configuration space In-Reply-To: <1882721.jqIYnAaWfQ@wuerfel> References: <6092525.UgRCY3dpzP@wuerfel> <1882721.jqIYnAaWfQ@wuerfel> Message-ID: <3295d30c-0b71-d73b-fd53-c75a9e0b2bae@huawei.com> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org On 09/01/2017 11:52, Arnd Bergmann wrote: > On Monday, January 9, 2017 10:59:47 AM CET John Garry wrote: >> On 06/01/2017 11:24, Arnd Bergmann wrote: >>> On Friday, January 6, 2017 11:15:22 AM CET John Garry wrote: >>> >>> Probably nobody thought about this properly in the Linux drivers. The >>> ARMv8 ARM sounds correct here. >>> >>> I/O space may have the same issue, as it also requires non-posted >>> accesses. >> >> Right, so our HW team's recommendation - from ARM's memory model and >> also PCIe order model - is that not only config space but also PCIe >> memory mapped IO has the same attribute (nE). > > Just to be sure we are talking about the same thing: "PCIe memory > mapped IO" could refer to either PCI I/O space or PCI memory space. > > As far as I can tell, PCI memory space should *not* be using the nE > attribute, while PCI I/O space and PCI config space should. > Does this match what your HW team recomments? > Yes, right, the config and IO space recommendation is nE and memory space is E. In response to Will: > What's the rationale behind that recommendation? As Arnd said, the reasoning is that one access type is non-posted (config and IO) and the other (memory) is posted writes. Thanks, John > Arnd > > . >