Linaro-mm-sig

linaro-mm-sig@lists.linaro.org

20 participants
2956 discussions

[RFC 0/3] Unify IOMMU-based DMA-mapping code for ARM and ARM64

by Marek Szyprowski

Dear All, This is an initial RFC on the unification of IOMMU-based DMA-mapping code for ARM and ARM64 architectures. Right now ARM architecture still use my old code for IOMMU-based DMA-mapping glue, initially merged in commit 4ce63fcd919c32d22528e54dcd89506962933719 ("ARM: dma-mapping: add support for IOMMU mapper"). In meantime ARM64 got a new, slightly improved implementation provided by Robin Murphy in commit 13b8629f651164d71f4d38b821925f93ba4236c8 ("arm64: Add IOMMU dma_ops"). Both implementations are very similar thus their unification is desired to avoid duplicating future works and simplify code, which uses this layer on both architectures. In this patchset I've selected the new implementation (from ARM64 architecture) as a base. This means that ARM-specific, old interface (arm_iommu_* functions) for configuring IOMMU domains will be no longer available and its users have to be converted to new API. Besides lack of old interface, the second difference is additional requirements for IOMMU drivers. New code relies on the support for IOMMU_DOMAIN_DMA and default IOMMU domain, which is automatically attached by the IOMMU core. The new code also assumes that the IOMMU-based DMA-mapping ops are mainly configured from arch_setup_dma_ops() function, which means that the IOMMU driver should provide needed of_xlate callbacks and initialize IOMMU ops for device nodes. However it should be also possible to initialize IOMMU-based DMA-mapping ops for client devices directly from IOMMU drivers by calling common_iommu_setup_dma_ops() (some drivers used such approach). IOMMU drivers should be also aware of the fact that the default domain is attached via device_attach and then device_attach callback can be called once again with different domain without previous detach from default domain. For more information on this issue, see the following thread: https://lists.linaro.org/pipermail/linaro-mm-sig/2016-February/004625.html Currently there are 4 users of the old arm_iommu_* interface: 1. Exynos DRM driver 2. Rockchip DRM driver 3. OMAP3 ISP camera driver 4. Renesas VMSA-compatible IPMMU driver In this patchset I've converted Exynos DRM driver for the new API (patch 1). This required some changes in the memory management model inside the driver and removal of some hacks, which were used to setup IOMMU-based DMA-mapping ops on the 'exynos-drm' virtual device and common IOMMU domain for all Exynos DRM sub-devices, those changes have been posted separately here: http://www.spinics.net/lists/dri-devel/msg100861.html Rockchip DRM driver requires similar conversion. Converting OMAP3 ISP camera driver to new API requires adding support for IOMMU groups to OMAP IOMMU driver, because the new DMA/IOMMU code used IOMMU_DOMAIN_DMA type domains and default groups. Renesas IPMMU driver needs also to be extended with IOMMU_DOMAIN_DMA domain type support. It can also be prepared for IOMMU_OF_DECLARE and of_xlate callback-based initialization to let core to automatically setup of IOMMU-based DMA mapping implementation. Patch 2 moves existing code from arch/arm64 to drivers/iommu and introduces some minor changes in function names - mainly adding arch_ prefix to some dma-mapping internal functions, which stay in arch/arm64/ (functions of similar names are present in arch/arm). Patch 3 adapts ARM architecture for the common code. I would like to get your comments on the proposed approach. There is still some work that need to be done to convert remaining users of the old API and updating IOMMU drivers to the new API requirements. This change need to be tested on the all affected ARM sub-architectures. Right now patches were tested on only Exynos based boards: ARM 32bit: Exynos4412 and Exynos5422 boards and ARM 64 bit Exnyos 5433 (with some out-of-tree DTS). To ease testing I've prepared a branch with all the patches needed (there are all needed patches for Exynos subarch, which have been posted as separate patchsets): https://git.linaro.org/people/marek.szyprowski/linux-srpol.git v4.5-dma-iommu-unification Patches are based on Linux v4.5-rc4 vanilla tree. Best regards Marek Szyprowski Samsung R&D Institute Poland Patch summary: Marek Szyprowski (3): drm/exynos: rewrite IOMMU support code iommu: dma-iommu: move IOMMU/DMA-mapping code from ARM64 arch to drivers iommu: dma-iommu: use common implementation also on ARM architecture arch/arm/Kconfig | 22 +- arch/arm/include/asm/device.h | 9 - arch/arm/include/asm/dma-iommu.h | 37 - arch/arm/include/asm/dma-mapping.h | 59 +- arch/arm/mm/dma-mapping.c | 1158 +---------------------------- arch/arm64/include/asm/dma-mapping.h | 39 +- arch/arm64/mm/dma-mapping.c | 491 +----------- drivers/gpu/drm/exynos/Kconfig | 2 +- drivers/gpu/drm/exynos/exynos_drm_drv.c | 7 +- drivers/gpu/drm/exynos/exynos_drm_drv.h | 2 +- drivers/gpu/drm/exynos/exynos_drm_iommu.c | 91 ++- drivers/gpu/drm/exynos/exynos_drm_iommu.h | 2 +- drivers/gpu/drm/rockchip/Kconfig | 1 + drivers/iommu/Kconfig | 1 + drivers/iommu/Makefile | 2 +- drivers/iommu/dma-iommu-ops.c | 471 ++++++++++++ drivers/media/platform/Kconfig | 1 + include/linux/dma-iommu.h | 14 + 18 files changed, 679 insertions(+), 1730 deletions(-) delete mode 100644 arch/arm/include/asm/dma-iommu.h create mode 100644 drivers/iommu/dma-iommu-ops.c -- 1.9.2

9 years, 2 months

Re: [Linaro-mm-sig] [RESEND] fence: add missing descriptions for fence

by Sumit Semwal

Hi Luis, On 12 April 2016 at 04:03, Luis de Bethencourt <luisbg(a)osg.samsung.com> wrote: > On 11/04/16 21:09, Gustavo Padovan wrote: >> Hi Luis, >> >> 2016-04-11 Luis de Bethencourt <luisbg(a)osg.samsung.com>: >> >>> The members child_list and active_list were added to the fence struct >>> without descriptions for the Documentation. Adding these. >>> Thanks for the patch; will get it queued for for-next. >>> Fixes: b55b54b5db33 ("staging/android: remove struct sync_pt") >>> Signed-off-by: Luis de Bethencourt <luisbg(a)osg.samsung.com> >>> Reviewed-by: Javier Martinez Canillas <javier(a)osg.samsung.com> >>> --- >>> Hi, >>> >>> Just resending this patch since it hasn't had any reviews in since >>> March 21st. >>> >>> Thanks, >>> Luis >>> >>> include/linux/fence.h | 2 ++ >>> 1 file changed, 2 insertions(+) >> >> Reviewed-by: Gustavo Padovan <gustavo.padovan(a)collabora.co.uk> >> >> Gustavo >> > > Thank you Gustavo. > > Nice seeing you around here :) > > Luis BR, Sumit.

9 years, 3 months

Re: [Linaro-mm-sig] [RESEND] fence: add missing descriptions for fence

by Gustavo Padovan

Hi Luis, 2016-04-11 Luis de Bethencourt <luisbg(a)osg.samsung.com>: > The members child_list and active_list were added to the fence struct > without descriptions for the Documentation. Adding these. > > Fixes: b55b54b5db33 ("staging/android: remove struct sync_pt") > Signed-off-by: Luis de Bethencourt <luisbg(a)osg.samsung.com> > Reviewed-by: Javier Martinez Canillas <javier(a)osg.samsung.com> > --- > Hi, > > Just resending this patch since it hasn't had any reviews in since > March 21st. > > Thanks, > Luis > > include/linux/fence.h | 2 ++ > 1 file changed, 2 insertions(+) Reviewed-by: Gustavo Padovan <gustavo.padovan(a)collabora.co.uk> Gustavo

9 years, 3 months

[PATCH 0/2] drm: add generic zpos property

by Benjamin Gaignard

The original patches have been done by Marek: https://lists.freedesktop.org/archives/dri-devel/2016-January/099572.html I have just adapt them to make zpos depend on plane and no more on drm core. Since zpos range can be define per plane normalize function takes care of the min and max before compute normalized zpos value. Benjamin Gaignard (2): drm: add generic zpos property drm: sti: use generic zpos for plane Documentation/DocBook/gpu.tmpl | 10 ++ drivers/gpu/drm/Makefile | 2 +- drivers/gpu/drm/drm_atomic.c | 4 + drivers/gpu/drm/drm_atomic_helper.c | 6 + drivers/gpu/drm/drm_blend.c | 242 ++++++++++++++++++++++++++++++++++++ drivers/gpu/drm/drm_crtc_internal.h | 3 + drivers/gpu/drm/sti/sti_mixer.c | 2 +- drivers/gpu/drm/sti/sti_plane.c | 82 +++++------- drivers/gpu/drm/sti/sti_plane.h | 2 - include/drm/drm_crtc.h | 17 +++ 10 files changed, 314 insertions(+), 56 deletions(-) create mode 100644 drivers/gpu/drm/drm_blend.c -- 1.9.1

9 years, 3 months

[PATCH] dma-buf: Release module reference on creation failure

by Chris Wilson

If we fail to create the anon file, we need to remember to release the module reference on the owner. Signed-off-by: Chris Wilson <chris(a)chris-wilson.co.uk> Cc: Sumit Semwal <sumit.semwal(a)linaro.org> Cc: Daniel Vetter <daniel.vetter(a)ffwll.ch> Cc: linux-media(a)vger.kernel.org Cc: dri-devel(a)lists.freedesktop.org Cc: linaro-mm-sig(a)lists.linaro.org --- drivers/dma-buf/dma-buf.c | 15 +++++++++++---- 1 file changed, 11 insertions(+), 4 deletions(-) diff --git a/drivers/dma-buf/dma-buf.c b/drivers/dma-buf/dma-buf.c index 4a2c07ee6677..6f0f0b10a241 100644 --- a/drivers/dma-buf/dma-buf.c +++ b/drivers/dma-buf/dma-buf.c @@ -333,6 +333,7 @@ struct dma_buf *dma_buf_export(const struct dma_buf_export_info *exp_info) struct reservation_object *resv = exp_info->resv; struct file *file; size_t alloc_size = sizeof(struct dma_buf); + int ret; if (!exp_info->resv) alloc_size += sizeof(struct reservation_object); @@ -356,8 +357,8 @@ struct dma_buf *dma_buf_export(const struct dma_buf_export_info *exp_info) dmabuf = kzalloc(alloc_size, GFP_KERNEL); if (!dmabuf) { - module_put(exp_info->owner); - return ERR_PTR(-ENOMEM); + ret = -ENOMEM; + goto free_module; } dmabuf->priv = exp_info->priv; @@ -378,8 +379,8 @@ struct dma_buf *dma_buf_export(const struct dma_buf_export_info *exp_info) file = anon_inode_getfile("dmabuf", &dma_buf_fops, dmabuf, exp_info->flags); if (IS_ERR(file)) { - kfree(dmabuf); - return ERR_CAST(file); + ret = PTR_ERR(file); + goto free_dmabuf; } file->f_mode |= FMODE_LSEEK; @@ -393,6 +394,12 @@ struct dma_buf *dma_buf_export(const struct dma_buf_export_info *exp_info) mutex_unlock(&db_list.lock); return dmabuf; + +free_dmabuf: + kfree(dmabuf); +free_module: + module_put(exp_info->owner); + return ERR_PTR(ret); } EXPORT_SYMBOL_GPL(dma_buf_export); -- 2.8.0.rc3

9 years, 3 months

Re: [Linaro-mm-sig] [PATCH] dma-buf: Update docs for SYNC ioctl

by David Herrmann

Hi On Mon, Mar 28, 2016 at 9:42 PM, Tiago Vignatti <tiago.vignatti(a)intel.com> wrote: > Do we have an agreement here after all? David? I need to know whether this > fixup is okay to go cause I'll need to submit to Chrome OS then. Sure it is fine. The code is already there, we cannot change it. Thanks David

9 years, 3 months

[PATCH] dma-buf, drm, ion: Propagate error code from dma_buf_start_cpu_access()

by Chris Wilson

Drivers, especially i915.ko, can fail during the initial migration of a dma-buf for CPU access. However, the error code from the driver was not being propagated back to ioctl and so userspace was blissfully ignorant of the failure. Rendering corruption ensues. Whilst fixing the ioctl to return the error code from dma_buf_start_cpu_access(), also do the same for dma_buf_end_cpu_access(). For most drivers, dma_buf_end_cpu_access() cannot fail. i915.ko however, as most drivers would, wants to avoid being uninterruptible (as would be required to guarrantee no failure when flushing the buffer to the device). As userspace already has to handle errors from the SYNC_IOCTL, take advantage of this to be able to restart the syscall across signals. This fixes a coherency issue for i915.ko as well as reducing the uninterruptible hold upon its BKL, the struct_mutex. Fixes commit c11e391da2a8fe973c3c2398452000bed505851e Author: Daniel Vetter <daniel.vetter(a)ffwll.ch> Date: Thu Feb 11 20:04:51 2016 -0200 dma-buf: Add ioctls to allow userspace to flush Testcase: igt/gem_concurrent_blit/*dmabuf*interruptible Testcase: igt/prime_mmap_coherency/ioctl-errors Signed-off-by: Chris Wilson <chris(a)chris-wilson.co.uk> Cc: Tiago Vignatti <tiago.vignatti(a)intel.com> Cc: Stéphane Marchesin <marcheu(a)chromium.org> Cc: David Herrmann <dh.herrmann(a)gmail.com> Cc: Sumit Semwal <sumit.semwal(a)linaro.org> Cc: Daniel Vetter <daniel.vetter(a)intel.com> CC: linux-media(a)vger.kernel.org Cc: dri-devel(a)lists.freedesktop.org Cc: linaro-mm-sig(a)lists.linaro.org Cc: intel-gfx(a)lists.freedesktop.org Cc: devel(a)driverdev.osuosl.org --- drivers/dma-buf/dma-buf.c | 17 +++++++++++------ drivers/gpu/drm/i915/i915_gem_dmabuf.c | 15 +++++---------- drivers/gpu/drm/omapdrm/omap_gem_dmabuf.c | 5 +++-- drivers/gpu/drm/udl/udl_fb.c | 4 ++-- drivers/staging/android/ion/ion.c | 6 ++++-- include/linux/dma-buf.h | 6 +++--- 6 files changed, 28 insertions(+), 25 deletions(-) diff --git a/drivers/dma-buf/dma-buf.c b/drivers/dma-buf/dma-buf.c index 9810d1df0691..774a60f4309a 100644 --- a/drivers/dma-buf/dma-buf.c +++ b/drivers/dma-buf/dma-buf.c @@ -259,6 +259,7 @@ static long dma_buf_ioctl(struct file *file, struct dma_buf *dmabuf; struct dma_buf_sync sync; enum dma_data_direction direction; + int ret; dmabuf = file->private_data; @@ -285,11 +286,11 @@ static long dma_buf_ioctl(struct file *file, } if (sync.flags & DMA_BUF_SYNC_END) - dma_buf_end_cpu_access(dmabuf, direction); + ret = dma_buf_end_cpu_access(dmabuf, direction); else - dma_buf_begin_cpu_access(dmabuf, direction); + ret = dma_buf_begin_cpu_access(dmabuf, direction); - return 0; + return ret; default: return -ENOTTY; } @@ -613,13 +614,17 @@ EXPORT_SYMBOL_GPL(dma_buf_begin_cpu_access); * * This call must always succeed. */ -void dma_buf_end_cpu_access(struct dma_buf *dmabuf, - enum dma_data_direction direction) +int dma_buf_end_cpu_access(struct dma_buf *dmabuf, + enum dma_data_direction direction) { + int ret = 0; + WARN_ON(!dmabuf); if (dmabuf->ops->end_cpu_access) - dmabuf->ops->end_cpu_access(dmabuf, direction); + ret = dmabuf->ops->end_cpu_access(dmabuf, direction); + + return ret; } EXPORT_SYMBOL_GPL(dma_buf_end_cpu_access); diff --git a/drivers/gpu/drm/i915/i915_gem_dmabuf.c b/drivers/gpu/drm/i915/i915_gem_dmabuf.c index 1f3eef6fb345..0506016e18e0 100644 --- a/drivers/gpu/drm/i915/i915_gem_dmabuf.c +++ b/drivers/gpu/drm/i915/i915_gem_dmabuf.c @@ -228,25 +228,20 @@ static int i915_gem_begin_cpu_access(struct dma_buf *dma_buf, enum dma_data_dire return ret; } -static void i915_gem_end_cpu_access(struct dma_buf *dma_buf, enum dma_data_direction direction) +static int i915_gem_end_cpu_access(struct dma_buf *dma_buf, enum dma_data_direction direction) { struct drm_i915_gem_object *obj = dma_buf_to_obj(dma_buf); struct drm_device *dev = obj->base.dev; - struct drm_i915_private *dev_priv = to_i915(dev); - bool was_interruptible; int ret; - mutex_lock(&dev->struct_mutex); - was_interruptible = dev_priv->mm.interruptible; - dev_priv->mm.interruptible = false; + ret = i915_mutex_lock_interruptible(dev); + if (ret) + return ret; ret = i915_gem_object_set_to_gtt_domain(obj, false); - - dev_priv->mm.interruptible = was_interruptible; mutex_unlock(&dev->struct_mutex); - if (unlikely(ret)) - DRM_ERROR("unable to flush buffer following CPU access; rendering may be corrupt\n"); + return ret; } static const struct dma_buf_ops i915_dmabuf_ops = { diff --git a/drivers/gpu/drm/omapdrm/omap_gem_dmabuf.c b/drivers/gpu/drm/omapdrm/omap_gem_dmabuf.c index 3cf8aab23a39..af267c35d813 100644 --- a/drivers/gpu/drm/omapdrm/omap_gem_dmabuf.c +++ b/drivers/gpu/drm/omapdrm/omap_gem_dmabuf.c @@ -97,11 +97,12 @@ static int omap_gem_dmabuf_begin_cpu_access(struct dma_buf *buffer, return omap_gem_get_pages(obj, &pages, true); } -static void omap_gem_dmabuf_end_cpu_access(struct dma_buf *buffer, - enum dma_data_direction dir) +static int omap_gem_dmabuf_end_cpu_access(struct dma_buf *buffer, + enum dma_data_direction dir) { struct drm_gem_object *obj = buffer->priv; omap_gem_put_pages(obj); + return 0; } diff --git a/drivers/gpu/drm/udl/udl_fb.c b/drivers/gpu/drm/udl/udl_fb.c index c427499133d6..33239a2b264a 100644 --- a/drivers/gpu/drm/udl/udl_fb.c +++ b/drivers/gpu/drm/udl/udl_fb.c @@ -423,8 +423,8 @@ static int udl_user_framebuffer_dirty(struct drm_framebuffer *fb, } if (ufb->obj->base.import_attach) { - dma_buf_end_cpu_access(ufb->obj->base.import_attach->dmabuf, - DMA_FROM_DEVICE); + ret = dma_buf_end_cpu_access(ufb->obj->base.import_attach->dmabuf, + DMA_FROM_DEVICE); } unlock: diff --git a/drivers/staging/android/ion/ion.c b/drivers/staging/android/ion/ion.c index 0754a37c9674..49436b4510f4 100644 --- a/drivers/staging/android/ion/ion.c +++ b/drivers/staging/android/ion/ion.c @@ -1075,14 +1075,16 @@ static int ion_dma_buf_begin_cpu_access(struct dma_buf *dmabuf, return PTR_ERR_OR_ZERO(vaddr); } -static void ion_dma_buf_end_cpu_access(struct dma_buf *dmabuf, - enum dma_data_direction direction) +static int ion_dma_buf_end_cpu_access(struct dma_buf *dmabuf, + enum dma_data_direction direction) { struct ion_buffer *buffer = dmabuf->priv; mutex_lock(&buffer->lock); ion_buffer_kmap_put(buffer); mutex_unlock(&buffer->lock); + + return 0; } static struct dma_buf_ops dma_buf_ops = { diff --git a/include/linux/dma-buf.h b/include/linux/dma-buf.h index 532108ea0c1c..3fe90d494edb 100644 --- a/include/linux/dma-buf.h +++ b/include/linux/dma-buf.h @@ -94,7 +94,7 @@ struct dma_buf_ops { void (*release)(struct dma_buf *); int (*begin_cpu_access)(struct dma_buf *, enum dma_data_direction); - void (*end_cpu_access)(struct dma_buf *, enum dma_data_direction); + int (*end_cpu_access)(struct dma_buf *, enum dma_data_direction); void *(*kmap_atomic)(struct dma_buf *, unsigned long); void (*kunmap_atomic)(struct dma_buf *, unsigned long, void *); void *(*kmap)(struct dma_buf *, unsigned long); @@ -224,8 +224,8 @@ void dma_buf_unmap_attachment(struct dma_buf_attachment *, struct sg_table *, enum dma_data_direction); int dma_buf_begin_cpu_access(struct dma_buf *dma_buf, enum dma_data_direction dir); -void dma_buf_end_cpu_access(struct dma_buf *dma_buf, - enum dma_data_direction dir); +int dma_buf_end_cpu_access(struct dma_buf *dma_buf, + enum dma_data_direction dir); void *dma_buf_kmap_atomic(struct dma_buf *, unsigned long); void dma_buf_kunmap_atomic(struct dma_buf *, unsigned long, void *); void *dma_buf_kmap(struct dma_buf *, unsigned long); -- 2.8.0.rc3

9 years, 3 months

Re: [Linaro-mm-sig] [PATCH] fence: add missing descriptions for fence

by Javier Martinez Canillas

Hello Luis, On Sat, Mar 19, 2016 at 4:50 PM, Luis de Bethencourt <luisbg(a)osg.samsung.com> wrote: > Commit b55b54b5db33 ("staging/android: remove struct sync_pt") > added the members child_list and active_list to the fence struct, but > didn't add descriptions for these. Adding the descriptions. > Patches whose commit message mentions a specific commit that introduced and issue, usually also have a "Fixes:" tag before the S-o-B. For example this patch should have: Fixes: b55b54b5db33 ("staging/android: remove struct sync_pt") > Signed-off-by: Luis de Bethencourt <luisbg(a)osg.samsung.com> > --- > Hi, > > Noticed this missing descriptions when running make htmldocs. > > Got the following warnings: > .//include/linux/fence.h:84: warning: No description found for parameter 'child_list' > .//include/linux/fence.h:84: warning: No description found for parameter 'active_list' > > Thanks :) > Luis > Patch looks good to me. Reviewed-by: Javier Martinez Canillas <javier(a)osg.samsung.com> Best regards, Javier

9 years, 3 months

[PATCH] dma-buf/fence: fix fence_is_later v2

by Alex Deucher

From: Christian König <christian.koenig(a)amd.com> A fence is never later than itself. This caused a bunch of overhead for AMDGPU. v2: simplify check as suggested by Michel. Signed-off-by: Christian König <christian.koenig(a)amd.com> Reviewed-by: Michel Dänzer <michel.daenzer(a)amd.com> Reviewed-by: Alex Deucher <alexander.deucher(a)amd.com> Signed-off-by: Alex Deucher <alexander.deucher(a)amd.com> --- include/linux/fence.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/include/linux/fence.h b/include/linux/fence.h index bb52201..5aa95eb 100644 --- a/include/linux/fence.h +++ b/include/linux/fence.h @@ -292,7 +292,7 @@ static inline bool fence_is_later(struct fence *f1, struct fence *f2) if (WARN_ON(f1->context != f2->context)) return false; - return f1->seqno - f2->seqno < INT_MAX; + return (int)(f1->seqno - f2->seqno) > 0; } /** -- 2.5.0

9 years, 4 months

[PATCH 0/2] iommu: fix handling of attach/detach for default domains

by Marek Szyprowski

Hello, This patchset fixes attaching and detaching devices to a group, which has default domain configured. With both patches applied all calls to iommu's attach/detach functions will be balanced, whenever caller attaches device to the default domain or the custom one. Best regards Marek Szyprowski Samsung R&D Institute Poland Patch summary: Marek Szyprowski (2): iommu: call detach also for default_domain before attaching to new one iommu: fix default domain handling in __iommu_detach_group() drivers/iommu/iommu.c | 48 +++++++++++++++++++++++++++--------------------- 1 file changed, 27 insertions(+), 21 deletions(-) -- 1.9.2

9 years, 4 months

Jump to page:

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

Linaro-mm-sig