diff options
author | Linus Torvalds <torvalds@linux-foundation.org> | 2019-07-16 05:04:27 +0300 |
---|---|---|
committer | Linus Torvalds <torvalds@linux-foundation.org> | 2019-07-16 05:04:27 +0300 |
commit | be8454afc50f43016ca8b6130d9673bdd0bd56ec (patch) | |
tree | 897e49c1ccadeed9b083a3ffc13f0dd2d6d7d874 /drivers/gpu/drm/i915/gem/i915_gem_mman.c | |
parent | fec88ab0af9706b2201e5daf377c5031c62d11f7 (diff) | |
parent | 3729fe2bc2a01f4cc1aa88be8f64af06084c87d6 (diff) | |
download | linux-be8454afc50f43016ca8b6130d9673bdd0bd56ec.tar.xz |
Merge tag 'drm-next-2019-07-16' of git://anongit.freedesktop.org/drm/drm
Pull drm updates from Dave Airlie:
"The biggest thing in this is the AMD Navi GPU support, this again
contains a bunch of header files that are large. These are the new AMD
RX5700 GPUs that just recently became available.
New drivers:
- ST-Ericsson MCDE driver
- Ingenic JZ47xx SoC
UAPI change:
- HDR source metadata property
Core:
- HDR inforframes and EDID parsing
- drm hdmi infoframe unpacking
- remove prime sg_table caching into dma-buf
- New gem vram helpers to reduce driver code
- Lots of drmP.h removal
- reservation fencing fix
- documentation updates
- drm_fb_helper_connector removed
- mode name command handler rewrite
fbcon:
- Remove the fbcon notifiers
ttm:
- forward progress fixes
dma-buf:
- make mmap call optional
- debugfs refcount fixes
- dma-fence free with pending signals fix
- each dma-buf gets an inode
Panels:
- Lots of additional panel bindings
amdgpu:
- initial navi10 support
- avoid hw reset
- HDR metadata support
- new thermal sensors for vega asics
- RAS fixes
- use HMM rather than MMU notifier
- xgmi topology via kfd
- SR-IOV fixes
- driver reload fixes
- DC use a core bpc attribute
- Aux fixes for DC
- Bandwidth calc updates for DC
- Clock handling refactor
- kfd VEGAM support
vmwgfx:
- Coherent memory support changes
i915:
- HDR Support
- HDMI i2c link
- Icelake multi-segmented gamma support
- GuC firmware update
- Mule Creek Canyon PCH support for EHL
- EHL platform updtes
- move i915.alpha_support to i915.force_probe
- runtime PM refactoring
- VBT parsing refactoring
- DSI fixes
- struct mutex dependency reduction
- GEM code reorg
mali-dp:
- Komeda driver features
msm:
- dsi vs EPROBE_DEFER fixes
- msm8998 snapdragon 835 support
- a540 gpu support
- mdp5 and dpu interconnect support
exynos:
- drmP.h removal
tegra:
- misc fixes
tda998x:
- audio support improvements
- pixel repeated mode support
- quantisation range handling corrections
- HDMI vendor info fix
armada:
- interlace support fix
- overlay/video plane register handling refactor
- add gamma support
rockchip:
- RX3328 support
panfrost:
- expose perf counters via hidden ioctls
vkms:
- enumerate CRC sources list
ast:
- rework BO handling
mgag200:
- rework BO handling
dw-hdmi:
- suspend/resume support
rcar-du:
- R8A774A1 Soc Support
- LVDS dual-link mode support
- Additional formats
- Misc fixes
omapdrm:
- DSI command mode display support
stm
- fb modifier support
- runtime PM support
sun4i:
- use vmap ops
vc4:
- binner bo binding rework
v3d:
- compute shader support
- resync/sync fixes
- job management refactoring
lima:
- NULL pointer in irq handler fix
- scheduler default timeout
virtio:
- fence seqno support
- trace events
bochs:
- misc fixes
tc458767:
- IRQ/HDP handling
sii902x:
- HDMI audio support
atmel-hlcdc:
- misc fixes
meson:
- zpos support"
* tag 'drm-next-2019-07-16' of git://anongit.freedesktop.org/drm/drm: (1815 commits)
Revert "Merge branch 'vmwgfx-next' of git://people.freedesktop.org/~thomash/linux into drm-next"
Revert "mm: adjust apply_to_pfn_range interface for dropped token."
mm: adjust apply_to_pfn_range interface for dropped token.
drm/amdgpu/navi10: add uclk activity sensor
drm/amdgpu: properly guard the generic discovery code
drm/amdgpu: add missing documentation on new module parameters
drm/amdgpu: don't invalidate caches in RELEASE_MEM, only do the writeback
drm/amd/display: avoid 64-bit division
drm/amdgpu/psp11: simplify the ucode register logic
drm/amdgpu: properly guard DC support in navi code
drm/amd/powerplay: vega20: fix uninitialized variable use
drm/amd/display: dcn20: include linux/delay.h
amdgpu: make pmu support optional
drm/amd/powerplay: Zero initialize current_rpm in vega20_get_fan_speed_percent
drm/amd/powerplay: Zero initialize freq in smu_v11_0_get_current_clk_freq
drm/amd/powerplay: Use memset to initialize metrics structs
drm/amdgpu/mes10.1: Fix header guard
drm/amd/powerplay: add temperature sensor support for navi10
drm/amdgpu: fix scheduler timeout calc
drm/amdgpu: Prepare for hmm_range_register API change (v2)
...
Diffstat (limited to 'drivers/gpu/drm/i915/gem/i915_gem_mman.c')
-rw-r--r-- | drivers/gpu/drm/i915/gem/i915_gem_mman.c | 508 |
1 files changed, 508 insertions, 0 deletions
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_mman.c b/drivers/gpu/drm/i915/gem/i915_gem_mman.c new file mode 100644 index 000000000000..391621ee3cbb --- /dev/null +++ b/drivers/gpu/drm/i915/gem/i915_gem_mman.c @@ -0,0 +1,508 @@ +/* + * SPDX-License-Identifier: MIT + * + * Copyright © 2014-2016 Intel Corporation + */ + +#include <linux/mman.h> +#include <linux/sizes.h> + +#include "i915_drv.h" +#include "i915_gem_gtt.h" +#include "i915_gem_ioctls.h" +#include "i915_gem_object.h" +#include "i915_vma.h" +#include "intel_drv.h" + +static inline bool +__vma_matches(struct vm_area_struct *vma, struct file *filp, + unsigned long addr, unsigned long size) +{ + if (vma->vm_file != filp) + return false; + + return vma->vm_start == addr && + (vma->vm_end - vma->vm_start) == PAGE_ALIGN(size); +} + +/** + * i915_gem_mmap_ioctl - Maps the contents of an object, returning the address + * it is mapped to. + * @dev: drm device + * @data: ioctl data blob + * @file: drm file + * + * While the mapping holds a reference on the contents of the object, it doesn't + * imply a ref on the object itself. + * + * IMPORTANT: + * + * DRM driver writers who look a this function as an example for how to do GEM + * mmap support, please don't implement mmap support like here. The modern way + * to implement DRM mmap support is with an mmap offset ioctl (like + * i915_gem_mmap_gtt) and then using the mmap syscall on the DRM fd directly. + * That way debug tooling like valgrind will understand what's going on, hiding + * the mmap call in a driver private ioctl will break that. The i915 driver only + * does cpu mmaps this way because we didn't know better. + */ +int +i915_gem_mmap_ioctl(struct drm_device *dev, void *data, + struct drm_file *file) +{ + struct drm_i915_gem_mmap *args = data; + struct drm_i915_gem_object *obj; + unsigned long addr; + + if (args->flags & ~(I915_MMAP_WC)) + return -EINVAL; + + if (args->flags & I915_MMAP_WC && !boot_cpu_has(X86_FEATURE_PAT)) + return -ENODEV; + + obj = i915_gem_object_lookup(file, args->handle); + if (!obj) + return -ENOENT; + + /* prime objects have no backing filp to GEM mmap + * pages from. + */ + if (!obj->base.filp) { + addr = -ENXIO; + goto err; + } + + if (range_overflows(args->offset, args->size, (u64)obj->base.size)) { + addr = -EINVAL; + goto err; + } + + addr = vm_mmap(obj->base.filp, 0, args->size, + PROT_READ | PROT_WRITE, MAP_SHARED, + args->offset); + if (IS_ERR_VALUE(addr)) + goto err; + + if (args->flags & I915_MMAP_WC) { + struct mm_struct *mm = current->mm; + struct vm_area_struct *vma; + + if (down_write_killable(&mm->mmap_sem)) { + addr = -EINTR; + goto err; + } + vma = find_vma(mm, addr); + if (vma && __vma_matches(vma, obj->base.filp, addr, args->size)) + vma->vm_page_prot = + pgprot_writecombine(vm_get_page_prot(vma->vm_flags)); + else + addr = -ENOMEM; + up_write(&mm->mmap_sem); + if (IS_ERR_VALUE(addr)) + goto err; + + /* This may race, but that's ok, it only gets set */ + WRITE_ONCE(obj->frontbuffer_ggtt_origin, ORIGIN_CPU); + } + i915_gem_object_put(obj); + + args->addr_ptr = (u64)addr; + return 0; + +err: + i915_gem_object_put(obj); + return addr; +} + +static unsigned int tile_row_pages(const struct drm_i915_gem_object *obj) +{ + return i915_gem_object_get_tile_row_size(obj) >> PAGE_SHIFT; +} + +/** + * i915_gem_mmap_gtt_version - report the current feature set for GTT mmaps + * + * A history of the GTT mmap interface: + * + * 0 - Everything had to fit into the GTT. Both parties of a memcpy had to + * aligned and suitable for fencing, and still fit into the available + * mappable space left by the pinned display objects. A classic problem + * we called the page-fault-of-doom where we would ping-pong between + * two objects that could not fit inside the GTT and so the memcpy + * would page one object in at the expense of the other between every + * single byte. + * + * 1 - Objects can be any size, and have any compatible fencing (X Y, or none + * as set via i915_gem_set_tiling() [DRM_I915_GEM_SET_TILING]). If the + * object is too large for the available space (or simply too large + * for the mappable aperture!), a view is created instead and faulted + * into userspace. (This view is aligned and sized appropriately for + * fenced access.) + * + * 2 - Recognise WC as a separate cache domain so that we can flush the + * delayed writes via GTT before performing direct access via WC. + * + * 3 - Remove implicit set-domain(GTT) and synchronisation on initial + * pagefault; swapin remains transparent. + * + * Restrictions: + * + * * snoopable objects cannot be accessed via the GTT. It can cause machine + * hangs on some architectures, corruption on others. An attempt to service + * a GTT page fault from a snoopable object will generate a SIGBUS. + * + * * the object must be able to fit into RAM (physical memory, though no + * limited to the mappable aperture). + * + * + * Caveats: + * + * * a new GTT page fault will synchronize rendering from the GPU and flush + * all data to system memory. Subsequent access will not be synchronized. + * + * * all mappings are revoked on runtime device suspend. + * + * * there are only 8, 16 or 32 fence registers to share between all users + * (older machines require fence register for display and blitter access + * as well). Contention of the fence registers will cause the previous users + * to be unmapped and any new access will generate new page faults. + * + * * running out of memory while servicing a fault may generate a SIGBUS, + * rather than the expected SIGSEGV. + */ +int i915_gem_mmap_gtt_version(void) +{ + return 3; +} + +static inline struct i915_ggtt_view +compute_partial_view(const struct drm_i915_gem_object *obj, + pgoff_t page_offset, + unsigned int chunk) +{ + struct i915_ggtt_view view; + + if (i915_gem_object_is_tiled(obj)) + chunk = roundup(chunk, tile_row_pages(obj)); + + view.type = I915_GGTT_VIEW_PARTIAL; + view.partial.offset = rounddown(page_offset, chunk); + view.partial.size = + min_t(unsigned int, chunk, + (obj->base.size >> PAGE_SHIFT) - view.partial.offset); + + /* If the partial covers the entire object, just create a normal VMA. */ + if (chunk >= obj->base.size >> PAGE_SHIFT) + view.type = I915_GGTT_VIEW_NORMAL; + + return view; +} + +/** + * i915_gem_fault - fault a page into the GTT + * @vmf: fault info + * + * The fault handler is set up by drm_gem_mmap() when a object is GTT mapped + * from userspace. The fault handler takes care of binding the object to + * the GTT (if needed), allocating and programming a fence register (again, + * only if needed based on whether the old reg is still valid or the object + * is tiled) and inserting a new PTE into the faulting process. + * + * Note that the faulting process may involve evicting existing objects + * from the GTT and/or fence registers to make room. So performance may + * suffer if the GTT working set is large or there are few fence registers + * left. + * + * The current feature set supported by i915_gem_fault() and thus GTT mmaps + * is exposed via I915_PARAM_MMAP_GTT_VERSION (see i915_gem_mmap_gtt_version). + */ +vm_fault_t i915_gem_fault(struct vm_fault *vmf) +{ +#define MIN_CHUNK_PAGES (SZ_1M >> PAGE_SHIFT) + struct vm_area_struct *area = vmf->vma; + struct drm_i915_gem_object *obj = to_intel_bo(area->vm_private_data); + struct drm_device *dev = obj->base.dev; + struct drm_i915_private *i915 = to_i915(dev); + struct intel_runtime_pm *rpm = &i915->runtime_pm; + struct i915_ggtt *ggtt = &i915->ggtt; + bool write = area->vm_flags & VM_WRITE; + intel_wakeref_t wakeref; + struct i915_vma *vma; + pgoff_t page_offset; + int srcu; + int ret; + + /* Sanity check that we allow writing into this object */ + if (i915_gem_object_is_readonly(obj) && write) + return VM_FAULT_SIGBUS; + + /* We don't use vmf->pgoff since that has the fake offset */ + page_offset = (vmf->address - area->vm_start) >> PAGE_SHIFT; + + trace_i915_gem_object_fault(obj, page_offset, true, write); + + ret = i915_gem_object_pin_pages(obj); + if (ret) + goto err; + + wakeref = intel_runtime_pm_get(rpm); + + srcu = i915_reset_trylock(i915); + if (srcu < 0) { + ret = srcu; + goto err_rpm; + } + + ret = i915_mutex_lock_interruptible(dev); + if (ret) + goto err_reset; + + /* Access to snoopable pages through the GTT is incoherent. */ + if (obj->cache_level != I915_CACHE_NONE && !HAS_LLC(i915)) { + ret = -EFAULT; + goto err_unlock; + } + + /* Now pin it into the GTT as needed */ + vma = i915_gem_object_ggtt_pin(obj, NULL, 0, 0, + PIN_MAPPABLE | + PIN_NONBLOCK | + PIN_NONFAULT); + if (IS_ERR(vma)) { + /* Use a partial view if it is bigger than available space */ + struct i915_ggtt_view view = + compute_partial_view(obj, page_offset, MIN_CHUNK_PAGES); + unsigned int flags; + + flags = PIN_MAPPABLE; + if (view.type == I915_GGTT_VIEW_NORMAL) + flags |= PIN_NONBLOCK; /* avoid warnings for pinned */ + + /* + * Userspace is now writing through an untracked VMA, abandon + * all hope that the hardware is able to track future writes. + */ + obj->frontbuffer_ggtt_origin = ORIGIN_CPU; + + vma = i915_gem_object_ggtt_pin(obj, &view, 0, 0, flags); + if (IS_ERR(vma) && !view.type) { + flags = PIN_MAPPABLE; + view.type = I915_GGTT_VIEW_PARTIAL; + vma = i915_gem_object_ggtt_pin(obj, &view, 0, 0, flags); + } + } + if (IS_ERR(vma)) { + ret = PTR_ERR(vma); + goto err_unlock; + } + + ret = i915_vma_pin_fence(vma); + if (ret) + goto err_unpin; + + /* Finally, remap it using the new GTT offset */ + ret = remap_io_mapping(area, + area->vm_start + (vma->ggtt_view.partial.offset << PAGE_SHIFT), + (ggtt->gmadr.start + vma->node.start) >> PAGE_SHIFT, + min_t(u64, vma->size, area->vm_end - area->vm_start), + &ggtt->iomap); + if (ret) + goto err_fence; + + /* Mark as being mmapped into userspace for later revocation */ + assert_rpm_wakelock_held(rpm); + if (!i915_vma_set_userfault(vma) && !obj->userfault_count++) + list_add(&obj->userfault_link, &i915->ggtt.userfault_list); + if (CONFIG_DRM_I915_USERFAULT_AUTOSUSPEND) + intel_wakeref_auto(&i915->ggtt.userfault_wakeref, + msecs_to_jiffies_timeout(CONFIG_DRM_I915_USERFAULT_AUTOSUSPEND)); + GEM_BUG_ON(!obj->userfault_count); + + i915_vma_set_ggtt_write(vma); + +err_fence: + i915_vma_unpin_fence(vma); +err_unpin: + __i915_vma_unpin(vma); +err_unlock: + mutex_unlock(&dev->struct_mutex); +err_reset: + i915_reset_unlock(i915, srcu); +err_rpm: + intel_runtime_pm_put(rpm, wakeref); + i915_gem_object_unpin_pages(obj); +err: + switch (ret) { + case -EIO: + /* + * We eat errors when the gpu is terminally wedged to avoid + * userspace unduly crashing (gl has no provisions for mmaps to + * fail). But any other -EIO isn't ours (e.g. swap in failure) + * and so needs to be reported. + */ + if (!i915_terminally_wedged(i915)) + return VM_FAULT_SIGBUS; + /* else: fall through */ + case -EAGAIN: + /* + * EAGAIN means the gpu is hung and we'll wait for the error + * handler to reset everything when re-faulting in + * i915_mutex_lock_interruptible. + */ + case 0: + case -ERESTARTSYS: + case -EINTR: + case -EBUSY: + /* + * EBUSY is ok: this just means that another thread + * already did the job. + */ + return VM_FAULT_NOPAGE; + case -ENOMEM: + return VM_FAULT_OOM; + case -ENOSPC: + case -EFAULT: + return VM_FAULT_SIGBUS; + default: + WARN_ONCE(ret, "unhandled error in %s: %i\n", __func__, ret); + return VM_FAULT_SIGBUS; + } +} + +void __i915_gem_object_release_mmap(struct drm_i915_gem_object *obj) +{ + struct i915_vma *vma; + + GEM_BUG_ON(!obj->userfault_count); + + obj->userfault_count = 0; + list_del(&obj->userfault_link); + drm_vma_node_unmap(&obj->base.vma_node, + obj->base.dev->anon_inode->i_mapping); + + for_each_ggtt_vma(vma, obj) + i915_vma_unset_userfault(vma); +} + +/** + * i915_gem_object_release_mmap - remove physical page mappings + * @obj: obj in question + * + * Preserve the reservation of the mmapping with the DRM core code, but + * relinquish ownership of the pages back to the system. + * + * It is vital that we remove the page mapping if we have mapped a tiled + * object through the GTT and then lose the fence register due to + * resource pressure. Similarly if the object has been moved out of the + * aperture, than pages mapped into userspace must be revoked. Removing the + * mapping will then trigger a page fault on the next user access, allowing + * fixup by i915_gem_fault(). + */ +void i915_gem_object_release_mmap(struct drm_i915_gem_object *obj) +{ + struct drm_i915_private *i915 = to_i915(obj->base.dev); + intel_wakeref_t wakeref; + + /* Serialisation between user GTT access and our code depends upon + * revoking the CPU's PTE whilst the mutex is held. The next user + * pagefault then has to wait until we release the mutex. + * + * Note that RPM complicates somewhat by adding an additional + * requirement that operations to the GGTT be made holding the RPM + * wakeref. + */ + lockdep_assert_held(&i915->drm.struct_mutex); + wakeref = intel_runtime_pm_get(&i915->runtime_pm); + + if (!obj->userfault_count) + goto out; + + __i915_gem_object_release_mmap(obj); + + /* Ensure that the CPU's PTE are revoked and there are not outstanding + * memory transactions from userspace before we return. The TLB + * flushing implied above by changing the PTE above *should* be + * sufficient, an extra barrier here just provides us with a bit + * of paranoid documentation about our requirement to serialise + * memory writes before touching registers / GSM. + */ + wmb(); + +out: + intel_runtime_pm_put(&i915->runtime_pm, wakeref); +} + +static int create_mmap_offset(struct drm_i915_gem_object *obj) +{ + struct drm_i915_private *i915 = to_i915(obj->base.dev); + int err; + + err = drm_gem_create_mmap_offset(&obj->base); + if (likely(!err)) + return 0; + + /* Attempt to reap some mmap space from dead objects */ + do { + err = i915_gem_wait_for_idle(i915, + I915_WAIT_INTERRUPTIBLE, + MAX_SCHEDULE_TIMEOUT); + if (err) + break; + + i915_gem_drain_freed_objects(i915); + err = drm_gem_create_mmap_offset(&obj->base); + if (!err) + break; + + } while (flush_delayed_work(&i915->gem.retire_work)); + + return err; +} + +int +i915_gem_mmap_gtt(struct drm_file *file, + struct drm_device *dev, + u32 handle, + u64 *offset) +{ + struct drm_i915_gem_object *obj; + int ret; + + obj = i915_gem_object_lookup(file, handle); + if (!obj) + return -ENOENT; + + ret = create_mmap_offset(obj); + if (ret == 0) + *offset = drm_vma_node_offset_addr(&obj->base.vma_node); + + i915_gem_object_put(obj); + return ret; +} + +/** + * i915_gem_mmap_gtt_ioctl - prepare an object for GTT mmap'ing + * @dev: DRM device + * @data: GTT mapping ioctl data + * @file: GEM object info + * + * Simply returns the fake offset to userspace so it can mmap it. + * The mmap call will end up in drm_gem_mmap(), which will set things + * up so we can get faults in the handler above. + * + * The fault handler will take care of binding the object into the GTT + * (since it may have been evicted to make room for something), allocating + * a fence register, and mapping the appropriate aperture address into + * userspace. + */ +int +i915_gem_mmap_gtt_ioctl(struct drm_device *dev, void *data, + struct drm_file *file) +{ + struct drm_i915_gem_mmap_gtt *args = data; + + return i915_gem_mmap_gtt(file, dev, args->handle, &args->offset); +} + +#if IS_ENABLED(CONFIG_DRM_I915_SELFTEST) +#include "selftests/i915_gem_mman.c" +#endif |