From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-7.0 required=3.0 tests=HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH,MAILING_LIST_MULTI,SIGNED_OFF_BY,SPF_PASS,UNPARSEABLE_RELAY, URIBL_BLOCKED autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 24D2FC282D8 for ; Fri, 1 Feb 2019 09:42:48 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.kernel.org (Postfix) with ESMTP id DEBDE20870 for ; Fri, 1 Feb 2019 09:42:47 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1729659AbfBAJmq (ORCPT ); Fri, 1 Feb 2019 04:42:46 -0500 Received: from mailgw02.mediatek.com ([1.203.163.81]:47723 "EHLO mailgw02.mediatek.com" rhost-flags-OK-FAIL-OK-FAIL) by vger.kernel.org with ESMTP id S1727734AbfBAJmp (ORCPT ); Fri, 1 Feb 2019 04:42:45 -0500 X-UUID: 631a032ab24c43b399bb26bd6d02af71-20190201 X-UUID: 631a032ab24c43b399bb26bd6d02af71-20190201 Received: from mtkcas32.mediatek.inc [(172.27.4.250)] by mailgw02.mediatek.com (envelope-from ) (mailgw01.mediatek.com ESMTP with TLS) with ESMTP id 796730003; Fri, 01 Feb 2019 17:42:30 +0800 Received: from MTKCAS32.mediatek.inc (172.27.4.184) by MTKMBS31N2.mediatek.inc (172.27.4.87) with Microsoft SMTP Server (TLS) id 15.0.1395.4; Fri, 1 Feb 2019 17:42:29 +0800 Received: from [10.17.3.153] (10.17.3.153) by MTKCAS32.mediatek.inc (172.27.4.170) with Microsoft SMTP Server id 15.0.1395.4 via Frontend Transport; Fri, 1 Feb 2019 17:42:28 +0800 Message-ID: <1549014148.12663.27.camel@mhfsdcap03> Subject: Re: [PATCH v6 06/20] iommu/io-pgtable-arm-v7s: Extend MediaTek 4GB Mode From: Yong Wu To: Evan Green CC: Joerg Roedel , Matthias Brugger , Robin Murphy , Rob Herring , Tomasz Figa , Will Deacon , , , "open list:OPEN FIRMWARE AND FLATTENED DEVICE TREE BINDINGS" , LKML , , , Arnd Bergmann , , , Nicolas Boichat Date: Fri, 1 Feb 2019 17:42:28 +0800 In-Reply-To: References: <1546314952-15990-1-git-send-email-yong.wu@mediatek.com> <1546314952-15990-7-git-send-email-yong.wu@mediatek.com> <1548917937.3292.15.camel@mhfsdcap03> Content-Type: text/plain; charset="UTF-8" X-Mailer: Evolution 3.2.3-0ubuntu6 Content-Transfer-Encoding: 7bit MIME-Version: 1.0 X-MTK: N Sender: linux-kernel-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Thu, 2019-01-31 at 11:23 -0800, Evan Green wrote: > On Wed, Jan 30, 2019 at 10:59 PM Yong Wu wrote: > > > > On Wed, 2019-01-30 at 10:28 -0800, Evan Green wrote: > > > On Mon, Dec 31, 2018 at 7:57 PM Yong Wu wrote: > > > > > > > > MediaTek extend the arm v7s descriptor to support the dram over 4GB. > > > > > > > > In the mt2712 and mt8173, it's called "4GB mode", the physical address > > > > is from 0x4000_0000 to 0x1_3fff_ffff, but from EMI point of view, it > > > > is remapped to high address from 0x1_0000_0000 to 0x1_ffff_ffff, the > > > > bit32 is always enabled. thus, in the M4U, we always enable the bit9 > > > > for all PTEs which means to enable bit32 of physical address. > > > > > > I got a little lost here. I get that you're trying to explain why you > > > always used to set bit32 of the physical address. But I don't totally > > > get the part about physical addresses being from 0x4000_0000 - > > > 0x1_3fff_ffff, but also from 0x1_0000_0000-0x1_ffff_ffff. Are you > > > saying that the physical addresses from the iommu's perspective were > > > always >0x1_0000_0000? > > > > Yes. From the IOMMU's perspective, the Physical address is from > > 0x1_0000_0000 to 0x1_ffff_ffff. > > > > > But then from whose perspective is it 0x4000_0000? ... > > > > I guess from SW point view. it is from 0x4000_0000 to 0x1_3fff_ffff. > > > > If 4GB mode is enabled, the memory property in dts like this: > > > > memory@40000000 { > > device_type = "memory"; > > reg = <0 0x40000000 0x00000001 0x00000000>; > > }; > > > > > oh, or you're saying there was some sort of remapping > > > facility that moved the physical addresses around? > > > > > > > > > > > but in mt8183, M4U support the dram from 0x4000_0000 to 0x3_ffff_ffff > > > > which isn't remaped. We extend the PTEs: the bit9 represent bit32 of > > > > PA and the bit4 represent bit33 of PA. Meanwhile the iova still is > > > > 32bits. > > > > > > > > In order to unify code, in the "4GB mode", we add the bit32 for the > > > > physical address manually in our driver. > > > > > > > > Correspondingly, Adding bit32 and bit33 for the PA in the iova_to_phys > > > > has to been moved into v7s. > > > > > > > > Regarding whether the pagetable address could be over 4GB, the mt8183 > > > > support it while the previous mt8173 don't. thus keep it as is. > > > > > > > > Signed-off-by: Yong Wu > > > > Reviewed-by: Robin Murphy > > > > --- > > > > drivers/iommu/io-pgtable-arm-v7s.c | 31 ++++++++++++++++++++++++------- > > > > drivers/iommu/io-pgtable.h | 7 +++---- > > > > drivers/iommu/mtk_iommu.c | 14 ++++++++------ > > > > drivers/iommu/mtk_iommu.h | 1 + > > > > 4 files changed, 36 insertions(+), 17 deletions(-) > > > > > > > > diff --git a/drivers/iommu/io-pgtable-arm-v7s.c b/drivers/iommu/io-pgtable-arm-v7s.c > > > > index 11d8505..8803a35 100644 > > > > --- a/drivers/iommu/io-pgtable-arm-v7s.c > > > > +++ b/drivers/iommu/io-pgtable-arm-v7s.c > > > > @@ -124,7 +124,9 @@ > > > > #define ARM_V7S_TEX_MASK 0x7 > > > > #define ARM_V7S_ATTR_TEX(val) (((val) & ARM_V7S_TEX_MASK) << ARM_V7S_TEX_SHIFT) > > > > > > > > -#define ARM_V7S_ATTR_MTK_4GB BIT(9) /* MTK extend it for 4GB mode */ > > > > +/* MediaTek extend the two bits below for over 4GB mode */ > > > > +#define ARM_V7S_ATTR_MTK_PA_BIT32 BIT(9) > > > > +#define ARM_V7S_ATTR_MTK_PA_BIT33 BIT(4) > > > > > > If other vendors start doing stuff like this we'll need a more generic > > > way to handle this... but I guess until we see a pattern this is okay. > > > > > > > > > > > /* *well, except for TEX on level 2 large pages, of course :( */ > > > > #define ARM_V7S_CONT_PAGE_TEX_SHIFT 6 > > > > @@ -183,13 +185,22 @@ static dma_addr_t __arm_v7s_dma_addr(void *pages) > > > > static arm_v7s_iopte paddr_to_iopte(phys_addr_t paddr, int lvl, > > > > struct io_pgtable_cfg *cfg) > > > > { > > > > - return paddr & ARM_V7S_LVL_MASK(lvl); > > > > + arm_v7s_iopte pte = paddr & ARM_V7S_LVL_MASK(lvl); > > > > + > > > > + if (cfg->quirks & IO_PGTABLE_QUIRK_ARM_MTK_4GB) { > > > > + if (paddr & BIT_ULL(32)) > > > > + pte |= ARM_V7S_ATTR_MTK_PA_BIT32; > > > > + if (paddr & BIT_ULL(33)) > > > > + pte |= ARM_V7S_ATTR_MTK_PA_BIT33; > > > > + } > > > > + return pte; > > > > } > > > > > > > > static phys_addr_t iopte_to_paddr(arm_v7s_iopte pte, int lvl, > > > > struct io_pgtable_cfg *cfg) > > > > { > > > > arm_v7s_iopte mask; > > > > + phys_addr_t paddr; > > > > > > > > if (ARM_V7S_PTE_IS_TABLE(pte, lvl)) > > > > mask = ARM_V7S_TABLE_MASK; > > > > @@ -198,7 +209,14 @@ static phys_addr_t iopte_to_paddr(arm_v7s_iopte pte, int lvl, > > > > else > > > > mask = ARM_V7S_LVL_MASK(lvl); > > > > > > > > - return pte & mask; > > > > + paddr = pte & mask; > > > > + if (cfg->quirks & IO_PGTABLE_QUIRK_ARM_MTK_4GB) { > > > > + if (pte & ARM_V7S_ATTR_MTK_PA_BIT32) > > > > + paddr |= BIT_ULL(32); > > > > + if (pte & ARM_V7S_ATTR_MTK_PA_BIT33) > > > > + paddr |= BIT_ULL(33); > > > > + } > > > > + return paddr; > > > > } > > > > > > > > static arm_v7s_iopte *iopte_deref(arm_v7s_iopte pte, int lvl, > > > > @@ -315,9 +333,6 @@ static arm_v7s_iopte arm_v7s_prot_to_pte(int prot, int lvl, > > > > if (lvl == 1 && (cfg->quirks & IO_PGTABLE_QUIRK_ARM_NS)) > > > > pte |= ARM_V7S_ATTR_NS_SECTION; > > > > > > > > - if (cfg->quirks & IO_PGTABLE_QUIRK_ARM_MTK_4GB) > > > > - pte |= ARM_V7S_ATTR_MTK_4GB; > > > > - > > > > > > So despite getting lost in the details, I guess the reason it's okay > > > that this goes from unconditional to conditional on bit32 is that > > > before, with the older chips, all physical addresses were above 4GB, > > > so we'll always see PA's above 4GB? > > > > > > > return pte; > > > > } > > > > > > > > @@ -504,7 +519,9 @@ static int arm_v7s_map(struct io_pgtable_ops *ops, unsigned long iova, > > > > if (!(prot & (IOMMU_READ | IOMMU_WRITE))) > > > > return 0; > > > > > > > > - if (WARN_ON(upper_32_bits(iova) || upper_32_bits(paddr))) > > > > + if (WARN_ON(upper_32_bits(iova)) || > > > > + WARN_ON(upper_32_bits(paddr) && > > > > + !(iop->cfg.quirks & IO_PGTABLE_QUIRK_ARM_MTK_4GB))) > > > > return -ERANGE; > > > > > > > > ret = __arm_v7s_map(data, iova, paddr, size, prot, 1, data->pgd); > > > > diff --git a/drivers/iommu/io-pgtable.h b/drivers/iommu/io-pgtable.h > > > > index 47d5ae5..69db115 100644 > > > > --- a/drivers/iommu/io-pgtable.h > > > > +++ b/drivers/iommu/io-pgtable.h > > > > @@ -62,10 +62,9 @@ struct io_pgtable_cfg { > > > > * (unmapped) entries but the hardware might do so anyway, perform > > > > * TLB maintenance when mapping as well as when unmapping. > > > > * > > > > - * IO_PGTABLE_QUIRK_ARM_MTK_4GB: (ARM v7s format) Set bit 9 in all > > > > - * PTEs, for Mediatek IOMMUs which treat it as a 33rd address bit > > > > - * when the SoC is in "4GB mode" and they can only access the high > > > > - * remap of DRAM (0x1_00000000 to 0x1_ffffffff). > > > > + * IO_PGTABLE_QUIRK_ARM_MTK_4GB: (ARM v7s format) MediaTek IOMMUs extend > > > > + * to support up to 34 bits PA where the bit32 and bit33 are > > > > + * encoded in the bit9 and bit4 of the PTE respectively. > > > > * > > > > * IO_PGTABLE_QUIRK_NO_DMA: Guarantees that the tables will only ever > > > > * be accessed by a fully cache-coherent IOMMU or CPU (e.g. for a > > > > diff --git a/drivers/iommu/mtk_iommu.c b/drivers/iommu/mtk_iommu.c > > > > index 189d1b5..ae1aa5a 100644 > > > > --- a/drivers/iommu/mtk_iommu.c > > > > +++ b/drivers/iommu/mtk_iommu.c > > > > @@ -367,12 +367,16 @@ static int mtk_iommu_map(struct iommu_domain *domain, unsigned long iova, > > > > phys_addr_t paddr, size_t size, int prot) > > > > { > > > > struct mtk_iommu_domain *dom = to_mtk_domain(domain); > > > > + struct mtk_iommu_data *data = mtk_iommu_get_m4u_data(); > > > > unsigned long flags; > > > > int ret; > > > > > > > > + /* The "4GB mode" M4U physically can not use the lower remap of Dram. */ > > > > + if (data->plat_data->has_4gb_mode && data->enable_4GB) > > > > + paddr |= BIT_ULL(32); > > > > + > > > > > > Ok here's where I get lost. How is this okay? Is the same physical RAM > > > accessible at multiple locations in the physical address space? Won't > > > this map an iova to a different pa than the one requested? > > > > In 4GB mode, HW will remap 0x4000_0000-0x1_3fff_ffff to 0x1_0000_0000- > > 0x1_ffff_ffff. M4U help multimedia HW access dram, thus from M4U point > > of view, the dram always is 0x1_0000_0000 to 0x1_ffff_ffff. > > > > The detailed mapping relationship is like this: > > > > 0x4000_0000 -0xffff_ffff map to 0x1_4000_0000 - 0x1_ffff_ffff. > > 0x1_0000_0000-0x1_3fff_ffff map to 0x1_0000_0000 - 0x1_3fff_ffff. > > > > Thus, we can only add bit32 for the PA in the 4GB mode. > > Ok, I think I get it now. This thread really helped: > https://patchwork.kernel.org/patch/8402211/ > > So from what I understand basically the same DRAM exists in two places: > 0000_0000 - ffff_ffff, and is also available in > 1_0000_0000 - 1_ffff_ffff > > ...except that the peripherals are located in 0000_0000 - 3ffff_ffff, > so that first GB of RAM is not visible at the lower address. I'm > gathering this was in fact the motivation for 4GB mode. The important > part is that address 4000_0000 == 1_4000_0000. > > Then there was also some quirk of the IOMMU where it refused to access > addresses below 4GB. But those same addresses are accessible by ORing > in bit 32, so you just always do that and you're good to go. > > Ok so now I can use that to understand this refactoring: > * You used to always return an address above 4GB in > mtk_iommu_iova_to_phys. I don't fully get how that worked, since it > seems like you'd start returning PAs to the rest of the system that > were outside of the range 4000_0000 - 1_3fff_ffff, but okay, you're no I'm not sure I follow this. From the SW point view, the dram is 0x4000_0000 - 0x1_3fff_ffff. there is no memory outside it. But there is really a issue in the mtk_iommu_iova_to_phys in the 4gb_mode. Currently in the 4gb mode, I always add BIT32 for all the memory, then the PA returned by the mtk_iommu_iova_to_phys(in v7s) always is from 0x1_0000_0000 to 0x1_ffff_ffff. But the SW still expect the PA is from 0x4000_0000 - 0x1_3fff_ffff. Thus, I guess I will add a new patch like this: @@ -418,6 +418,7 @@ static phys_addr_t mtk_iommu_iova_to_phys(struct iommu_domain *domain, dma_addr_t iova) { struct mtk_iommu_domain *dom = to_mtk_domain(domain); + struct mtk_iommu_data *data = mtk_iommu_get_m4u_data(); unsigned long flags; phys_addr_t pa; @@ -425,6 +426,11 @@ static phys_addr_t mtk_iommu_iova_to_phys(struct iommu_domain *domain, pa = dom->iop->iova_to_phys(dom->iop, iova); spin_unlock_irqrestore(&dom->pgtlock, flags); + /* Discard bit32 if pa is 0x1_4000_0000 -0x1_ffff_ffff in 4GB mode. */ + if (data->plat_data->has_4gb_mode && data->enable_4GB && + pa >= 0x140000000) + paddr &= ~BIT_ULL(32); + return pa; } > longer doing that there, so I won't worry about it. > * Now, if you're in the 4GB mode, you just slam the bit in the PTE in > mtk_iommu_map, which seems like the right thing to do. > * The general functions in io-pgtable-arm-v7s.c now carefully reflect > bits 32 & 33 in the PTE, since the new IOMMUs don't have the weird > restriction of staying above 4GB, and there's not this weird 4GB > aliasing mode going on (which I think would be a clearer name for this > feature: has_4gb_alias). A more beautiful name. But our internal and all the CODA call this "4GB mode"..thus I'd like to keep it.... > > > > > > > > > Also, you could have rolled the has_4gb_mode check into whether or not > > > you set enable_4GB. Then you're doing the check for has_4gb_mode once, > > > rather than on every map call. > > > > "has_4gb_mode" means this SoC support 4GB mode. > > "enable_4GB" means whether the current dram size is 4GB. > > Right. But your use of the variable as well as it's name suggest that > it really means "is 4GB aliasing mode on", not "does the system have > >=4GB of RAM". You could reduce the map function to one conditional if > you treated the variable that way. Then the only things that would > need to change would be: > * Add an extra conditional in probe that would only set enable_4GB if > has_4gb_mode is set. I guess I still don't get this. the enable_4GB and has_4gb_mode are not the same. Take mt8173 as a example when its dram size is 2G. it has_4gb_mode, but we can not enable_4GB at that time.(if dram size is 2G, the HW will not remap the PA address, we can not add BIT32 at that time.) > * in mtk_iommu_domain_finalize, you could just always set the MTK > quirk, since if you have <4GB of RAM, those bits will never get set in > the PTEs anyway. oh. Yes. this looks right. > * I suspect mtk_iommu_hw_init would continue to work as-is, since > everything that has vld_pa_rng also has has_4gb_mode. mt8173 has 4gb_mode but it doesn't has vld_pa_rng. > > -Evan From mboxrd@z Thu Jan 1 00:00:00 1970 From: Yong Wu Subject: Re: [PATCH v6 06/20] iommu/io-pgtable-arm-v7s: Extend MediaTek 4GB Mode Date: Fri, 1 Feb 2019 17:42:28 +0800 Message-ID: <1549014148.12663.27.camel@mhfsdcap03> References: <1546314952-15990-1-git-send-email-yong.wu@mediatek.com> <1546314952-15990-7-git-send-email-yong.wu@mediatek.com> <1548917937.3292.15.camel@mhfsdcap03> Mime-Version: 1.0 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: Sender: linux-kernel-owner@vger.kernel.org To: Evan Green Cc: Joerg Roedel , Matthias Brugger , Robin Murphy , Rob Herring , Tomasz Figa , Will Deacon , linux-mediatek@lists.infradead.org, srv_heupstream@mediatek.com, "open list:OPEN FIRMWARE AND FLATTENED DEVICE TREE BINDINGS" , LKML , linux-arm-kernel@lists.infradead.org, iommu@lists.linux-foundation.org, Arnd Bergmann , yingjoe.chen@mediatek.com, youlin.pei@mediatek.com, Nicolas Boichat List-Id: devicetree@vger.kernel.org On Thu, 2019-01-31 at 11:23 -0800, Evan Green wrote: > On Wed, Jan 30, 2019 at 10:59 PM Yong Wu wrote: > > > > On Wed, 2019-01-30 at 10:28 -0800, Evan Green wrote: > > > On Mon, Dec 31, 2018 at 7:57 PM Yong Wu wrote: > > > > > > > > MediaTek extend the arm v7s descriptor to support the dram over 4GB. > > > > > > > > In the mt2712 and mt8173, it's called "4GB mode", the physical address > > > > is from 0x4000_0000 to 0x1_3fff_ffff, but from EMI point of view, it > > > > is remapped to high address from 0x1_0000_0000 to 0x1_ffff_ffff, the > > > > bit32 is always enabled. thus, in the M4U, we always enable the bit9 > > > > for all PTEs which means to enable bit32 of physical address. > > > > > > I got a little lost here. I get that you're trying to explain why you > > > always used to set bit32 of the physical address. But I don't totally > > > get the part about physical addresses being from 0x4000_0000 - > > > 0x1_3fff_ffff, but also from 0x1_0000_0000-0x1_ffff_ffff. Are you > > > saying that the physical addresses from the iommu's perspective were > > > always >0x1_0000_0000? > > > > Yes. From the IOMMU's perspective, the Physical address is from > > 0x1_0000_0000 to 0x1_ffff_ffff. > > > > > But then from whose perspective is it 0x4000_0000? ... > > > > I guess from SW point view. it is from 0x4000_0000 to 0x1_3fff_ffff. > > > > If 4GB mode is enabled, the memory property in dts like this: > > > > memory@40000000 { > > device_type = "memory"; > > reg = <0 0x40000000 0x00000001 0x00000000>; > > }; > > > > > oh, or you're saying there was some sort of remapping > > > facility that moved the physical addresses around? > > > > > > > > > > > but in mt8183, M4U support the dram from 0x4000_0000 to 0x3_ffff_ffff > > > > which isn't remaped. We extend the PTEs: the bit9 represent bit32 of > > > > PA and the bit4 represent bit33 of PA. Meanwhile the iova still is > > > > 32bits. > > > > > > > > In order to unify code, in the "4GB mode", we add the bit32 for the > > > > physical address manually in our driver. > > > > > > > > Correspondingly, Adding bit32 and bit33 for the PA in the iova_to_phys > > > > has to been moved into v7s. > > > > > > > > Regarding whether the pagetable address could be over 4GB, the mt8183 > > > > support it while the previous mt8173 don't. thus keep it as is. > > > > > > > > Signed-off-by: Yong Wu > > > > Reviewed-by: Robin Murphy > > > > --- > > > > drivers/iommu/io-pgtable-arm-v7s.c | 31 ++++++++++++++++++++++++------- > > > > drivers/iommu/io-pgtable.h | 7 +++---- > > > > drivers/iommu/mtk_iommu.c | 14 ++++++++------ > > > > drivers/iommu/mtk_iommu.h | 1 + > > > > 4 files changed, 36 insertions(+), 17 deletions(-) > > > > > > > > diff --git a/drivers/iommu/io-pgtable-arm-v7s.c b/drivers/iommu/io-pgtable-arm-v7s.c > > > > index 11d8505..8803a35 100644 > > > > --- a/drivers/iommu/io-pgtable-arm-v7s.c > > > > +++ b/drivers/iommu/io-pgtable-arm-v7s.c > > > > @@ -124,7 +124,9 @@ > > > > #define ARM_V7S_TEX_MASK 0x7 > > > > #define ARM_V7S_ATTR_TEX(val) (((val) & ARM_V7S_TEX_MASK) << ARM_V7S_TEX_SHIFT) > > > > > > > > -#define ARM_V7S_ATTR_MTK_4GB BIT(9) /* MTK extend it for 4GB mode */ > > > > +/* MediaTek extend the two bits below for over 4GB mode */ > > > > +#define ARM_V7S_ATTR_MTK_PA_BIT32 BIT(9) > > > > +#define ARM_V7S_ATTR_MTK_PA_BIT33 BIT(4) > > > > > > If other vendors start doing stuff like this we'll need a more generic > > > way to handle this... but I guess until we see a pattern this is okay. > > > > > > > > > > > /* *well, except for TEX on level 2 large pages, of course :( */ > > > > #define ARM_V7S_CONT_PAGE_TEX_SHIFT 6 > > > > @@ -183,13 +185,22 @@ static dma_addr_t __arm_v7s_dma_addr(void *pages) > > > > static arm_v7s_iopte paddr_to_iopte(phys_addr_t paddr, int lvl, > > > > struct io_pgtable_cfg *cfg) > > > > { > > > > - return paddr & ARM_V7S_LVL_MASK(lvl); > > > > + arm_v7s_iopte pte = paddr & ARM_V7S_LVL_MASK(lvl); > > > > + > > > > + if (cfg->quirks & IO_PGTABLE_QUIRK_ARM_MTK_4GB) { > > > > + if (paddr & BIT_ULL(32)) > > > > + pte |= ARM_V7S_ATTR_MTK_PA_BIT32; > > > > + if (paddr & BIT_ULL(33)) > > > > + pte |= ARM_V7S_ATTR_MTK_PA_BIT33; > > > > + } > > > > + return pte; > > > > } > > > > > > > > static phys_addr_t iopte_to_paddr(arm_v7s_iopte pte, int lvl, > > > > struct io_pgtable_cfg *cfg) > > > > { > > > > arm_v7s_iopte mask; > > > > + phys_addr_t paddr; > > > > > > > > if (ARM_V7S_PTE_IS_TABLE(pte, lvl)) > > > > mask = ARM_V7S_TABLE_MASK; > > > > @@ -198,7 +209,14 @@ static phys_addr_t iopte_to_paddr(arm_v7s_iopte pte, int lvl, > > > > else > > > > mask = ARM_V7S_LVL_MASK(lvl); > > > > > > > > - return pte & mask; > > > > + paddr = pte & mask; > > > > + if (cfg->quirks & IO_PGTABLE_QUIRK_ARM_MTK_4GB) { > > > > + if (pte & ARM_V7S_ATTR_MTK_PA_BIT32) > > > > + paddr |= BIT_ULL(32); > > > > + if (pte & ARM_V7S_ATTR_MTK_PA_BIT33) > > > > + paddr |= BIT_ULL(33); > > > > + } > > > > + return paddr; > > > > } > > > > > > > > static arm_v7s_iopte *iopte_deref(arm_v7s_iopte pte, int lvl, > > > > @@ -315,9 +333,6 @@ static arm_v7s_iopte arm_v7s_prot_to_pte(int prot, int lvl, > > > > if (lvl == 1 && (cfg->quirks & IO_PGTABLE_QUIRK_ARM_NS)) > > > > pte |= ARM_V7S_ATTR_NS_SECTION; > > > > > > > > - if (cfg->quirks & IO_PGTABLE_QUIRK_ARM_MTK_4GB) > > > > - pte |= ARM_V7S_ATTR_MTK_4GB; > > > > - > > > > > > So despite getting lost in the details, I guess the reason it's okay > > > that this goes from unconditional to conditional on bit32 is that > > > before, with the older chips, all physical addresses were above 4GB, > > > so we'll always see PA's above 4GB? > > > > > > > return pte; > > > > } > > > > > > > > @@ -504,7 +519,9 @@ static int arm_v7s_map(struct io_pgtable_ops *ops, unsigned long iova, > > > > if (!(prot & (IOMMU_READ | IOMMU_WRITE))) > > > > return 0; > > > > > > > > - if (WARN_ON(upper_32_bits(iova) || upper_32_bits(paddr))) > > > > + if (WARN_ON(upper_32_bits(iova)) || > > > > + WARN_ON(upper_32_bits(paddr) && > > > > + !(iop->cfg.quirks & IO_PGTABLE_QUIRK_ARM_MTK_4GB))) > > > > return -ERANGE; > > > > > > > > ret = __arm_v7s_map(data, iova, paddr, size, prot, 1, data->pgd); > > > > diff --git a/drivers/iommu/io-pgtable.h b/drivers/iommu/io-pgtable.h > > > > index 47d5ae5..69db115 100644 > > > > --- a/drivers/iommu/io-pgtable.h > > > > +++ b/drivers/iommu/io-pgtable.h > > > > @@ -62,10 +62,9 @@ struct io_pgtable_cfg { > > > > * (unmapped) entries but the hardware might do so anyway, perform > > > > * TLB maintenance when mapping as well as when unmapping. > > > > * > > > > - * IO_PGTABLE_QUIRK_ARM_MTK_4GB: (ARM v7s format) Set bit 9 in all > > > > - * PTEs, for Mediatek IOMMUs which treat it as a 33rd address bit > > > > - * when the SoC is in "4GB mode" and they can only access the high > > > > - * remap of DRAM (0x1_00000000 to 0x1_ffffffff). > > > > + * IO_PGTABLE_QUIRK_ARM_MTK_4GB: (ARM v7s format) MediaTek IOMMUs extend > > > > + * to support up to 34 bits PA where the bit32 and bit33 are > > > > + * encoded in the bit9 and bit4 of the PTE respectively. > > > > * > > > > * IO_PGTABLE_QUIRK_NO_DMA: Guarantees that the tables will only ever > > > > * be accessed by a fully cache-coherent IOMMU or CPU (e.g. for a > > > > diff --git a/drivers/iommu/mtk_iommu.c b/drivers/iommu/mtk_iommu.c > > > > index 189d1b5..ae1aa5a 100644 > > > > --- a/drivers/iommu/mtk_iommu.c > > > > +++ b/drivers/iommu/mtk_iommu.c > > > > @@ -367,12 +367,16 @@ static int mtk_iommu_map(struct iommu_domain *domain, unsigned long iova, > > > > phys_addr_t paddr, size_t size, int prot) > > > > { > > > > struct mtk_iommu_domain *dom = to_mtk_domain(domain); > > > > + struct mtk_iommu_data *data = mtk_iommu_get_m4u_data(); > > > > unsigned long flags; > > > > int ret; > > > > > > > > + /* The "4GB mode" M4U physically can not use the lower remap of Dram. */ > > > > + if (data->plat_data->has_4gb_mode && data->enable_4GB) > > > > + paddr |= BIT_ULL(32); > > > > + > > > > > > Ok here's where I get lost. How is this okay? Is the same physical RAM > > > accessible at multiple locations in the physical address space? Won't > > > this map an iova to a different pa than the one requested? > > > > In 4GB mode, HW will remap 0x4000_0000-0x1_3fff_ffff to 0x1_0000_0000- > > 0x1_ffff_ffff. M4U help multimedia HW access dram, thus from M4U point > > of view, the dram always is 0x1_0000_0000 to 0x1_ffff_ffff. > > > > The detailed mapping relationship is like this: > > > > 0x4000_0000 -0xffff_ffff map to 0x1_4000_0000 - 0x1_ffff_ffff. > > 0x1_0000_0000-0x1_3fff_ffff map to 0x1_0000_0000 - 0x1_3fff_ffff. > > > > Thus, we can only add bit32 for the PA in the 4GB mode. > > Ok, I think I get it now. This thread really helped: > https://patchwork.kernel.org/patch/8402211/ > > So from what I understand basically the same DRAM exists in two places: > 0000_0000 - ffff_ffff, and is also available in > 1_0000_0000 - 1_ffff_ffff > > ...except that the peripherals are located in 0000_0000 - 3ffff_ffff, > so that first GB of RAM is not visible at the lower address. I'm > gathering this was in fact the motivation for 4GB mode. The important > part is that address 4000_0000 == 1_4000_0000. > > Then there was also some quirk of the IOMMU where it refused to access > addresses below 4GB. But those same addresses are accessible by ORing > in bit 32, so you just always do that and you're good to go. > > Ok so now I can use that to understand this refactoring: > * You used to always return an address above 4GB in > mtk_iommu_iova_to_phys. I don't fully get how that worked, since it > seems like you'd start returning PAs to the rest of the system that > were outside of the range 4000_0000 - 1_3fff_ffff, but okay, you're no I'm not sure I follow this. From the SW point view, the dram is 0x4000_0000 - 0x1_3fff_ffff. there is no memory outside it. But there is really a issue in the mtk_iommu_iova_to_phys in the 4gb_mode. Currently in the 4gb mode, I always add BIT32 for all the memory, then the PA returned by the mtk_iommu_iova_to_phys(in v7s) always is from 0x1_0000_0000 to 0x1_ffff_ffff. But the SW still expect the PA is from 0x4000_0000 - 0x1_3fff_ffff. Thus, I guess I will add a new patch like this: @@ -418,6 +418,7 @@ static phys_addr_t mtk_iommu_iova_to_phys(struct iommu_domain *domain, dma_addr_t iova) { struct mtk_iommu_domain *dom = to_mtk_domain(domain); + struct mtk_iommu_data *data = mtk_iommu_get_m4u_data(); unsigned long flags; phys_addr_t pa; @@ -425,6 +426,11 @@ static phys_addr_t mtk_iommu_iova_to_phys(struct iommu_domain *domain, pa = dom->iop->iova_to_phys(dom->iop, iova); spin_unlock_irqrestore(&dom->pgtlock, flags); + /* Discard bit32 if pa is 0x1_4000_0000 -0x1_ffff_ffff in 4GB mode. */ + if (data->plat_data->has_4gb_mode && data->enable_4GB && + pa >= 0x140000000) + paddr &= ~BIT_ULL(32); + return pa; } > longer doing that there, so I won't worry about it. > * Now, if you're in the 4GB mode, you just slam the bit in the PTE in > mtk_iommu_map, which seems like the right thing to do. > * The general functions in io-pgtable-arm-v7s.c now carefully reflect > bits 32 & 33 in the PTE, since the new IOMMUs don't have the weird > restriction of staying above 4GB, and there's not this weird 4GB > aliasing mode going on (which I think would be a clearer name for this > feature: has_4gb_alias). A more beautiful name. But our internal and all the CODA call this "4GB mode"..thus I'd like to keep it.... > > > > > > > > > Also, you could have rolled the has_4gb_mode check into whether or not > > > you set enable_4GB. Then you're doing the check for has_4gb_mode once, > > > rather than on every map call. > > > > "has_4gb_mode" means this SoC support 4GB mode. > > "enable_4GB" means whether the current dram size is 4GB. > > Right. But your use of the variable as well as it's name suggest that > it really means "is 4GB aliasing mode on", not "does the system have > >=4GB of RAM". You could reduce the map function to one conditional if > you treated the variable that way. Then the only things that would > need to change would be: > * Add an extra conditional in probe that would only set enable_4GB if > has_4gb_mode is set. I guess I still don't get this. the enable_4GB and has_4gb_mode are not the same. Take mt8173 as a example when its dram size is 2G. it has_4gb_mode, but we can not enable_4GB at that time.(if dram size is 2G, the HW will not remap the PA address, we can not add BIT32 at that time.) > * in mtk_iommu_domain_finalize, you could just always set the MTK > quirk, since if you have <4GB of RAM, those bits will never get set in > the PTEs anyway. oh. Yes. this looks right. > * I suspect mtk_iommu_hw_init would continue to work as-is, since > everything that has vld_pa_rng also has has_4gb_mode. mt8173 has 4gb_mode but it doesn't has vld_pa_rng. > > -Evan From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-7.0 required=3.0 tests=DKIMWL_WL_HIGH,DKIM_SIGNED, DKIM_VALID,HEADER_FROM_DIFFERENT_DOMAINS,INCLUDES_PATCH,MAILING_LIST_MULTI, SIGNED_OFF_BY,SPF_PASS,UNPARSEABLE_RELAY,URIBL_BLOCKED autolearn=unavailable autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 04EE6C282D8 for ; Fri, 1 Feb 2019 09:42:55 +0000 (UTC) Received: from bombadil.infradead.org (bombadil.infradead.org [198.137.202.133]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by mail.kernel.org (Postfix) with ESMTPS id C2F9520870 for ; Fri, 1 Feb 2019 09:42:54 +0000 (UTC) Authentication-Results: mail.kernel.org; dkim=pass (2048-bit key) header.d=lists.infradead.org header.i=@lists.infradead.org header.b="aH29Kzuw" DMARC-Filter: OpenDMARC Filter v1.3.2 mail.kernel.org C2F9520870 Authentication-Results: mail.kernel.org; dmarc=none (p=none dis=none) header.from=mediatek.com Authentication-Results: mail.kernel.org; spf=none smtp.mailfrom=linux-arm-kernel-bounces+infradead-linux-arm-kernel=archiver.kernel.org@lists.infradead.org DKIM-Signature: v=1; a=rsa-sha256; q=dns/txt; c=relaxed/relaxed; d=lists.infradead.org; s=bombadil.20170209; h=Sender: Content-Transfer-Encoding:Content-Type:Cc:List-Subscribe:List-Help:List-Post: List-Archive:List-Unsubscribe:List-Id:MIME-Version:References:In-Reply-To: Date:To:From:Subject:Message-ID:Reply-To:Content-ID:Content-Description: Resent-Date:Resent-From:Resent-Sender:Resent-To:Resent-Cc:Resent-Message-ID: List-Owner; bh=z3LyJh7jYNFRkv+BhWQBMUBzgJV3aU5T0Q/Bq1mJjVo=; b=aH29KzuwjVDwfI vCCYxFSI7gNqnV09+9j1ZwRLaKkldgEHBiftTo885cHX/53ccQx6M2XOpC0dJXbMj5LMjEBWjF6QI gOpa/EsFKF4gmUaZiIGSSbux4R2w8DuGCtHnhhm55sjRcT+Se+WVOLjg9p8DhBN85qt8qLHI5JyOn b0L/4MEEv/X1z/NGiC8izC9xrXsoUyO5FCTwSX+MODANT2nUyXxi7BGtfyflvyRAWbRXq0boB1XIq ngdeoTCdg/fRRRwiychHoUCTU1+LYyp/G5kweY9PU1PKy+yt2Wv2ALEd+nEaVu5mFLsbkLFRMENZ0 UaofWeRbnD231CU/lFXw==; Received: from localhost ([127.0.0.1] helo=bombadil.infradead.org) by bombadil.infradead.org with esmtp (Exim 4.90_1 #2 (Red Hat Linux)) id 1gpVLV-0008OD-Pg; Fri, 01 Feb 2019 09:42:49 +0000 Received: from mailgw02.mediatek.com ([216.200.240.185]) by bombadil.infradead.org with esmtps (Exim 4.90_1 #2 (Red Hat Linux)) id 1gpVLK-00087o-AL; Fri, 01 Feb 2019 09:42:45 +0000 X-UUID: 37deb1eed09a41d3946e1dcbede68d94-20190201 X-UUID: 37deb1eed09a41d3946e1dcbede68d94-20190201 Received: from mtkcas66.mediatek.inc [(172.29.193.44)] by mailgw02.mediatek.com (envelope-from ) (musrelay.mediatek.com ESMTP with TLS) with ESMTP id 844790791; Fri, 01 Feb 2019 01:42:33 -0800 Received: from MTKMBS31N2.mediatek.inc (172.27.4.87) by MTKMBS62N2.mediatek.inc (172.29.193.42) with Microsoft SMTP Server (TLS) id 15.0.1395.4; Fri, 1 Feb 2019 01:42:31 -0800 Received: from MTKCAS32.mediatek.inc (172.27.4.184) by MTKMBS31N2.mediatek.inc (172.27.4.87) with Microsoft SMTP Server (TLS) id 15.0.1395.4; Fri, 1 Feb 2019 17:42:29 +0800 Received: from [10.17.3.153] (10.17.3.153) by MTKCAS32.mediatek.inc (172.27.4.170) with Microsoft SMTP Server id 15.0.1395.4 via Frontend Transport; Fri, 1 Feb 2019 17:42:28 +0800 Message-ID: <1549014148.12663.27.camel@mhfsdcap03> Subject: Re: [PATCH v6 06/20] iommu/io-pgtable-arm-v7s: Extend MediaTek 4GB Mode From: Yong Wu To: Evan Green Date: Fri, 1 Feb 2019 17:42:28 +0800 In-Reply-To: References: <1546314952-15990-1-git-send-email-yong.wu@mediatek.com> <1546314952-15990-7-git-send-email-yong.wu@mediatek.com> <1548917937.3292.15.camel@mhfsdcap03> X-Mailer: Evolution 3.2.3-0ubuntu6 MIME-Version: 1.0 X-MTK: N X-CRM114-Version: 20100106-BlameMichelson ( TRE 0.8.0 (BSD) ) MR-646709E3 X-CRM114-CacheID: sfid-20190201_014238_982599_B65F86A6 X-CRM114-Status: GOOD ( 54.27 ) X-BeenThere: linux-arm-kernel@lists.infradead.org X-Mailman-Version: 2.1.21 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Cc: youlin.pei@mediatek.com, "open list:OPEN FIRMWARE AND FLATTENED DEVICE TREE BINDINGS" , Nicolas Boichat , Arnd Bergmann , srv_heupstream@mediatek.com, Joerg Roedel , Will Deacon , LKML , Tomasz Figa , iommu@lists.linux-foundation.org, Rob Herring , linux-mediatek@lists.infradead.org, Matthias Brugger , yingjoe.chen@mediatek.com, Robin Murphy , linux-arm-kernel@lists.infradead.org Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Sender: "linux-arm-kernel" Errors-To: linux-arm-kernel-bounces+infradead-linux-arm-kernel=archiver.kernel.org@lists.infradead.org On Thu, 2019-01-31 at 11:23 -0800, Evan Green wrote: > On Wed, Jan 30, 2019 at 10:59 PM Yong Wu wrote: > > > > On Wed, 2019-01-30 at 10:28 -0800, Evan Green wrote: > > > On Mon, Dec 31, 2018 at 7:57 PM Yong Wu wrote: > > > > > > > > MediaTek extend the arm v7s descriptor to support the dram over 4GB. > > > > > > > > In the mt2712 and mt8173, it's called "4GB mode", the physical address > > > > is from 0x4000_0000 to 0x1_3fff_ffff, but from EMI point of view, it > > > > is remapped to high address from 0x1_0000_0000 to 0x1_ffff_ffff, the > > > > bit32 is always enabled. thus, in the M4U, we always enable the bit9 > > > > for all PTEs which means to enable bit32 of physical address. > > > > > > I got a little lost here. I get that you're trying to explain why you > > > always used to set bit32 of the physical address. But I don't totally > > > get the part about physical addresses being from 0x4000_0000 - > > > 0x1_3fff_ffff, but also from 0x1_0000_0000-0x1_ffff_ffff. Are you > > > saying that the physical addresses from the iommu's perspective were > > > always >0x1_0000_0000? > > > > Yes. From the IOMMU's perspective, the Physical address is from > > 0x1_0000_0000 to 0x1_ffff_ffff. > > > > > But then from whose perspective is it 0x4000_0000? ... > > > > I guess from SW point view. it is from 0x4000_0000 to 0x1_3fff_ffff. > > > > If 4GB mode is enabled, the memory property in dts like this: > > > > memory@40000000 { > > device_type = "memory"; > > reg = <0 0x40000000 0x00000001 0x00000000>; > > }; > > > > > oh, or you're saying there was some sort of remapping > > > facility that moved the physical addresses around? > > > > > > > > > > > but in mt8183, M4U support the dram from 0x4000_0000 to 0x3_ffff_ffff > > > > which isn't remaped. We extend the PTEs: the bit9 represent bit32 of > > > > PA and the bit4 represent bit33 of PA. Meanwhile the iova still is > > > > 32bits. > > > > > > > > In order to unify code, in the "4GB mode", we add the bit32 for the > > > > physical address manually in our driver. > > > > > > > > Correspondingly, Adding bit32 and bit33 for the PA in the iova_to_phys > > > > has to been moved into v7s. > > > > > > > > Regarding whether the pagetable address could be over 4GB, the mt8183 > > > > support it while the previous mt8173 don't. thus keep it as is. > > > > > > > > Signed-off-by: Yong Wu > > > > Reviewed-by: Robin Murphy > > > > --- > > > > drivers/iommu/io-pgtable-arm-v7s.c | 31 ++++++++++++++++++++++++------- > > > > drivers/iommu/io-pgtable.h | 7 +++---- > > > > drivers/iommu/mtk_iommu.c | 14 ++++++++------ > > > > drivers/iommu/mtk_iommu.h | 1 + > > > > 4 files changed, 36 insertions(+), 17 deletions(-) > > > > > > > > diff --git a/drivers/iommu/io-pgtable-arm-v7s.c b/drivers/iommu/io-pgtable-arm-v7s.c > > > > index 11d8505..8803a35 100644 > > > > --- a/drivers/iommu/io-pgtable-arm-v7s.c > > > > +++ b/drivers/iommu/io-pgtable-arm-v7s.c > > > > @@ -124,7 +124,9 @@ > > > > #define ARM_V7S_TEX_MASK 0x7 > > > > #define ARM_V7S_ATTR_TEX(val) (((val) & ARM_V7S_TEX_MASK) << ARM_V7S_TEX_SHIFT) > > > > > > > > -#define ARM_V7S_ATTR_MTK_4GB BIT(9) /* MTK extend it for 4GB mode */ > > > > +/* MediaTek extend the two bits below for over 4GB mode */ > > > > +#define ARM_V7S_ATTR_MTK_PA_BIT32 BIT(9) > > > > +#define ARM_V7S_ATTR_MTK_PA_BIT33 BIT(4) > > > > > > If other vendors start doing stuff like this we'll need a more generic > > > way to handle this... but I guess until we see a pattern this is okay. > > > > > > > > > > > /* *well, except for TEX on level 2 large pages, of course :( */ > > > > #define ARM_V7S_CONT_PAGE_TEX_SHIFT 6 > > > > @@ -183,13 +185,22 @@ static dma_addr_t __arm_v7s_dma_addr(void *pages) > > > > static arm_v7s_iopte paddr_to_iopte(phys_addr_t paddr, int lvl, > > > > struct io_pgtable_cfg *cfg) > > > > { > > > > - return paddr & ARM_V7S_LVL_MASK(lvl); > > > > + arm_v7s_iopte pte = paddr & ARM_V7S_LVL_MASK(lvl); > > > > + > > > > + if (cfg->quirks & IO_PGTABLE_QUIRK_ARM_MTK_4GB) { > > > > + if (paddr & BIT_ULL(32)) > > > > + pte |= ARM_V7S_ATTR_MTK_PA_BIT32; > > > > + if (paddr & BIT_ULL(33)) > > > > + pte |= ARM_V7S_ATTR_MTK_PA_BIT33; > > > > + } > > > > + return pte; > > > > } > > > > > > > > static phys_addr_t iopte_to_paddr(arm_v7s_iopte pte, int lvl, > > > > struct io_pgtable_cfg *cfg) > > > > { > > > > arm_v7s_iopte mask; > > > > + phys_addr_t paddr; > > > > > > > > if (ARM_V7S_PTE_IS_TABLE(pte, lvl)) > > > > mask = ARM_V7S_TABLE_MASK; > > > > @@ -198,7 +209,14 @@ static phys_addr_t iopte_to_paddr(arm_v7s_iopte pte, int lvl, > > > > else > > > > mask = ARM_V7S_LVL_MASK(lvl); > > > > > > > > - return pte & mask; > > > > + paddr = pte & mask; > > > > + if (cfg->quirks & IO_PGTABLE_QUIRK_ARM_MTK_4GB) { > > > > + if (pte & ARM_V7S_ATTR_MTK_PA_BIT32) > > > > + paddr |= BIT_ULL(32); > > > > + if (pte & ARM_V7S_ATTR_MTK_PA_BIT33) > > > > + paddr |= BIT_ULL(33); > > > > + } > > > > + return paddr; > > > > } > > > > > > > > static arm_v7s_iopte *iopte_deref(arm_v7s_iopte pte, int lvl, > > > > @@ -315,9 +333,6 @@ static arm_v7s_iopte arm_v7s_prot_to_pte(int prot, int lvl, > > > > if (lvl == 1 && (cfg->quirks & IO_PGTABLE_QUIRK_ARM_NS)) > > > > pte |= ARM_V7S_ATTR_NS_SECTION; > > > > > > > > - if (cfg->quirks & IO_PGTABLE_QUIRK_ARM_MTK_4GB) > > > > - pte |= ARM_V7S_ATTR_MTK_4GB; > > > > - > > > > > > So despite getting lost in the details, I guess the reason it's okay > > > that this goes from unconditional to conditional on bit32 is that > > > before, with the older chips, all physical addresses were above 4GB, > > > so we'll always see PA's above 4GB? > > > > > > > return pte; > > > > } > > > > > > > > @@ -504,7 +519,9 @@ static int arm_v7s_map(struct io_pgtable_ops *ops, unsigned long iova, > > > > if (!(prot & (IOMMU_READ | IOMMU_WRITE))) > > > > return 0; > > > > > > > > - if (WARN_ON(upper_32_bits(iova) || upper_32_bits(paddr))) > > > > + if (WARN_ON(upper_32_bits(iova)) || > > > > + WARN_ON(upper_32_bits(paddr) && > > > > + !(iop->cfg.quirks & IO_PGTABLE_QUIRK_ARM_MTK_4GB))) > > > > return -ERANGE; > > > > > > > > ret = __arm_v7s_map(data, iova, paddr, size, prot, 1, data->pgd); > > > > diff --git a/drivers/iommu/io-pgtable.h b/drivers/iommu/io-pgtable.h > > > > index 47d5ae5..69db115 100644 > > > > --- a/drivers/iommu/io-pgtable.h > > > > +++ b/drivers/iommu/io-pgtable.h > > > > @@ -62,10 +62,9 @@ struct io_pgtable_cfg { > > > > * (unmapped) entries but the hardware might do so anyway, perform > > > > * TLB maintenance when mapping as well as when unmapping. > > > > * > > > > - * IO_PGTABLE_QUIRK_ARM_MTK_4GB: (ARM v7s format) Set bit 9 in all > > > > - * PTEs, for Mediatek IOMMUs which treat it as a 33rd address bit > > > > - * when the SoC is in "4GB mode" and they can only access the high > > > > - * remap of DRAM (0x1_00000000 to 0x1_ffffffff). > > > > + * IO_PGTABLE_QUIRK_ARM_MTK_4GB: (ARM v7s format) MediaTek IOMMUs extend > > > > + * to support up to 34 bits PA where the bit32 and bit33 are > > > > + * encoded in the bit9 and bit4 of the PTE respectively. > > > > * > > > > * IO_PGTABLE_QUIRK_NO_DMA: Guarantees that the tables will only ever > > > > * be accessed by a fully cache-coherent IOMMU or CPU (e.g. for a > > > > diff --git a/drivers/iommu/mtk_iommu.c b/drivers/iommu/mtk_iommu.c > > > > index 189d1b5..ae1aa5a 100644 > > > > --- a/drivers/iommu/mtk_iommu.c > > > > +++ b/drivers/iommu/mtk_iommu.c > > > > @@ -367,12 +367,16 @@ static int mtk_iommu_map(struct iommu_domain *domain, unsigned long iova, > > > > phys_addr_t paddr, size_t size, int prot) > > > > { > > > > struct mtk_iommu_domain *dom = to_mtk_domain(domain); > > > > + struct mtk_iommu_data *data = mtk_iommu_get_m4u_data(); > > > > unsigned long flags; > > > > int ret; > > > > > > > > + /* The "4GB mode" M4U physically can not use the lower remap of Dram. */ > > > > + if (data->plat_data->has_4gb_mode && data->enable_4GB) > > > > + paddr |= BIT_ULL(32); > > > > + > > > > > > Ok here's where I get lost. How is this okay? Is the same physical RAM > > > accessible at multiple locations in the physical address space? Won't > > > this map an iova to a different pa than the one requested? > > > > In 4GB mode, HW will remap 0x4000_0000-0x1_3fff_ffff to 0x1_0000_0000- > > 0x1_ffff_ffff. M4U help multimedia HW access dram, thus from M4U point > > of view, the dram always is 0x1_0000_0000 to 0x1_ffff_ffff. > > > > The detailed mapping relationship is like this: > > > > 0x4000_0000 -0xffff_ffff map to 0x1_4000_0000 - 0x1_ffff_ffff. > > 0x1_0000_0000-0x1_3fff_ffff map to 0x1_0000_0000 - 0x1_3fff_ffff. > > > > Thus, we can only add bit32 for the PA in the 4GB mode. > > Ok, I think I get it now. This thread really helped: > https://patchwork.kernel.org/patch/8402211/ > > So from what I understand basically the same DRAM exists in two places: > 0000_0000 - ffff_ffff, and is also available in > 1_0000_0000 - 1_ffff_ffff > > ...except that the peripherals are located in 0000_0000 - 3ffff_ffff, > so that first GB of RAM is not visible at the lower address. I'm > gathering this was in fact the motivation for 4GB mode. The important > part is that address 4000_0000 == 1_4000_0000. > > Then there was also some quirk of the IOMMU where it refused to access > addresses below 4GB. But those same addresses are accessible by ORing > in bit 32, so you just always do that and you're good to go. > > Ok so now I can use that to understand this refactoring: > * You used to always return an address above 4GB in > mtk_iommu_iova_to_phys. I don't fully get how that worked, since it > seems like you'd start returning PAs to the rest of the system that > were outside of the range 4000_0000 - 1_3fff_ffff, but okay, you're no I'm not sure I follow this. From the SW point view, the dram is 0x4000_0000 - 0x1_3fff_ffff. there is no memory outside it. But there is really a issue in the mtk_iommu_iova_to_phys in the 4gb_mode. Currently in the 4gb mode, I always add BIT32 for all the memory, then the PA returned by the mtk_iommu_iova_to_phys(in v7s) always is from 0x1_0000_0000 to 0x1_ffff_ffff. But the SW still expect the PA is from 0x4000_0000 - 0x1_3fff_ffff. Thus, I guess I will add a new patch like this: @@ -418,6 +418,7 @@ static phys_addr_t mtk_iommu_iova_to_phys(struct iommu_domain *domain, dma_addr_t iova) { struct mtk_iommu_domain *dom = to_mtk_domain(domain); + struct mtk_iommu_data *data = mtk_iommu_get_m4u_data(); unsigned long flags; phys_addr_t pa; @@ -425,6 +426,11 @@ static phys_addr_t mtk_iommu_iova_to_phys(struct iommu_domain *domain, pa = dom->iop->iova_to_phys(dom->iop, iova); spin_unlock_irqrestore(&dom->pgtlock, flags); + /* Discard bit32 if pa is 0x1_4000_0000 -0x1_ffff_ffff in 4GB mode. */ + if (data->plat_data->has_4gb_mode && data->enable_4GB && + pa >= 0x140000000) + paddr &= ~BIT_ULL(32); + return pa; } > longer doing that there, so I won't worry about it. > * Now, if you're in the 4GB mode, you just slam the bit in the PTE in > mtk_iommu_map, which seems like the right thing to do. > * The general functions in io-pgtable-arm-v7s.c now carefully reflect > bits 32 & 33 in the PTE, since the new IOMMUs don't have the weird > restriction of staying above 4GB, and there's not this weird 4GB > aliasing mode going on (which I think would be a clearer name for this > feature: has_4gb_alias). A more beautiful name. But our internal and all the CODA call this "4GB mode"..thus I'd like to keep it.... > > > > > > > > > Also, you could have rolled the has_4gb_mode check into whether or not > > > you set enable_4GB. Then you're doing the check for has_4gb_mode once, > > > rather than on every map call. > > > > "has_4gb_mode" means this SoC support 4GB mode. > > "enable_4GB" means whether the current dram size is 4GB. > > Right. But your use of the variable as well as it's name suggest that > it really means "is 4GB aliasing mode on", not "does the system have > >=4GB of RAM". You could reduce the map function to one conditional if > you treated the variable that way. Then the only things that would > need to change would be: > * Add an extra conditional in probe that would only set enable_4GB if > has_4gb_mode is set. I guess I still don't get this. the enable_4GB and has_4gb_mode are not the same. Take mt8173 as a example when its dram size is 2G. it has_4gb_mode, but we can not enable_4GB at that time.(if dram size is 2G, the HW will not remap the PA address, we can not add BIT32 at that time.) > * in mtk_iommu_domain_finalize, you could just always set the MTK > quirk, since if you have <4GB of RAM, those bits will never get set in > the PTEs anyway. oh. Yes. this looks right. > * I suspect mtk_iommu_hw_init would continue to work as-is, since > everything that has vld_pa_rng also has has_4gb_mode. mt8173 has 4gb_mode but it doesn't has vld_pa_rng. > > -Evan _______________________________________________ linux-arm-kernel mailing list linux-arm-kernel@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-arm-kernel