Return-path: Received: from smtp.codeaurora.org ([198.145.29.96]:60530 "EHLO smtp.codeaurora.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755027AbeCHNvA (ORCPT ); Thu, 8 Mar 2018 08:51:00 -0500 From: Kalle Valo To: Arnd Bergmann Cc: Govind Singh , Govind Singh , linux-wireless , "ath10k\@lists.infradead.org" Subject: Re: [PATCH] ath10k: Enable IOMMU support for WCN3990 target References: <1518685448-15317-1-git-send-email-govinds@codeaurora.org> <877eqwqfml.fsf@kamboji.qca.qualcomm.com> <588e4d951bd84ea29dfc1a68353505e7@aphydexm01b.ap.qualcomm.com> Date: Thu, 08 Mar 2018 15:50:55 +0200 In-Reply-To: (Arnd Bergmann's message of "Thu, 1 Mar 2018 14:27:35 +0100") Message-ID: <87k1umhe0w.fsf@kamboji.qca.qualcomm.com> (sfid-20180308_145112_375501_C1312D03) MIME-Version: 1.0 Content-Type: text/plain Sender: linux-wireless-owner@vger.kernel.org List-ID: Arnd Bergmann writes: > On Thu, Mar 1, 2018 at 2:18 PM, Govind Singh wrote: >>>> The asm/dma-iommu.h header file exsists only on arm32, no other architecture. >>>> I'm not sure about the purpose of the patch to start with: >>>> it's normally up to the platform code to allocate IOMMU domains, >>>> device drivers should only need to manually interact with the >>>> IOMMU layer if they need more than one domain, but this ath10k >>>> patch appears to be using the default domain and should have no >>>> effect as long as the platform code works correctly. >> Thanks Arnd, I have fixed this and migrated to 64bit >> API's(iommu_attach_device/iommu_detach_device/ >> iommu_get_domain_for_dev), will share the next revision. >> I tried using the default domain by adding the stream ID and mask in >> dt and no manual interaction, but it is resulting in TZ error and >> unhandled context fault. >> Seems I need to provide explicit mapping range(aperture_start/ >> aperture_end) as this is only working combination for me.. > > I don't see why you need to do that at all, can you clarify? > > The IOMMU should be set up implicitly for you here based on the iommus > property in DT, with no driver changes at all. This should work on all > architectures/ Maybe Govind is using some out-of-tree tree which is buggy in this regard? -- Kalle Valo