nouveau/instmem: use iomapping interface for instmem handling (v3)
Subsystems (4)
Dave Airlie
maintainerPATCH 1/1 nouveau/instmem: use iomapping interface for instmem handling (v3)
From: Dave Airlie <[email protected]>
This avoids constant need to ioremap when instobjs move at least on 64-bit systems.
This create the io mapping on first use, because creating it at init time causes a resource mapping error, because nouveau hasn't kicked simpledrm off the hardware yet, but ioremap_wc the whole BAR causes an overlap with BOOTFB/simpledrm. I think the resource system could do better here, but it's easier to just delay creating the mapping until first use.
---
v2: move the mapping into the main code.
v3: fixup some Sashiko review, just do mapping in boot code which should always be called for bar2.
--- .../nouveau/nvkm/subdev/gsp/rm/r535/fbsr.c | 1 + .../drm/nouveau/nvkm/subdev/instmem/nv50.c | 39 ++++++++++++++----- .../drm/nouveau/nvkm/subdev/instmem/priv.h | 1 + 3 files changed, 32 insertions(+), 9 deletions(-)
drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/fbsr.c +1 -0
@@ -297,6 +297,7 @@ r535_fbsr = {static void *r535_instmem_dtor (struct nvkm_instmem *imem) { +nv50_instmem_dtor (imem);kfree (imem->func);return imem; }
drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv50.c +30 -9
@@ -30,10 +30,14 @@#include <subdev/gsp.h> #include <subdev/mmu.h> +#include <linux/io-mapping.h> +struct nv50_instmem {struct nvkm_instmem base;u64 addr; +struct io_mapping iomap; +/* Mappings that can be evicted when BAR2 space has been exhausted. */ struct list_head lru; };
@@ -124,7 +128,6 @@ nv50_instobj_kmap(struct nv50_instobj *iobj, struct nvkm_vmm *vmm)struct nv50_instobj *eobj;struct nvkm_memory *memory = &iobj->base.memory;struct nvkm_subdev *subdev = &imem->base.subdev; -struct nvkm_device *device = subdev->device;struct nvkm_vma *bar =NULL , *ebar;u64 size =nvkm_memory_size (memory);void *emap;
@@ -155,7 +158,7 @@ nv50_instobj_kmap(struct nv50_instobj *iobj, struct nvkm_vmm *vmm)mutex_unlock (&imem->base.mutex);if (!eobj)break ; -iounmap (emap); +io_mapping_unmap (emap);nvkm_vmm_put (vmm, &ebar); }
@@ -172,8 +175,7 @@ nv50_instobj_kmap(struct nv50_instobj *iobj, struct nvkm_vmm *vmm)/* Make the mapping visible to the host. */ iobj->bar = bar; - iobj->map =ioremap_wc (device->func->resource_addr (device,NVKM_BAR2_INST ) + - (u32 )iobj->bar->addr, size); + iobj->map =io_mapping_map_wc (&imem->iomap, (u32 )iobj->bar->addr, size);if (!iobj->map) {nvkm_warn (subdev,"PRAMIN ioremap failed\n" );nvkm_vmm_put (vmm, &iobj->bar);
@@ -188,6 +190,16 @@ nv50_instobj_map(struct nvkm_memory *memory, u64 offset, struct nvkm_vmm *vmm,return nvkm_memory_map (memory, offset, vmm, vma, argv, argc); } +static bool +check_io_mapping (struct nv50_instmem *imem) +{ +struct nvkm_device *device = imem->base.subdev.device; + +return io_mapping_init_wc (&imem->iomap, + device->func->resource_addr (device,NVKM_BAR2_INST ), + device->func->resource_size (device,NVKM_BAR2_INST )) !=NULL ; +} +static void nv50_instobj_release (struct nvkm_memory *memory) {
@@ -239,7 +251,7 @@ nv50_instobj_acquire(struct nvkm_memory *memory)/* Attempt to get a direct CPU mapping of the object. */ if ((vmm =nvkm_bar_bar2_vmm (imem->subdev.device))) { -if (!iobj->map) +if (!iobj->map && iobj->imem->iomap.size)nv50_instobj_kmap (iobj, vmm); map = iobj->map; }
@@ -277,7 +289,12 @@ nv50_instobj_boot(struct nvkm_memory *memory, struct nvkm_vmm *vmm) iobj->lru.next =NULL ; } -nv50_instobj_kmap (iobj, vmm); +/* +* boot is only called on BAR2, if we can't remap the complete +* BAR it's unlikely things are functioning well. +*/ +if (check_io_mapping (iobj->imem)) +nv50_instobj_kmap (iobj, vmm);nvkm_instmem_boot (imem);mutex_unlock (&imem->mutex); }
@@ -330,7 +347,7 @@ nv50_instobj_dtor(struct nvkm_memory *memory)if (map) {struct nvkm_vmm *vmm =nvkm_bar_bar2_vmm (imem->subdev.device); -iounmap (map); +io_mapping_unmap (map);if (likely (vmm))/* Can be NULL during BAR destructor. */ nvkm_vmm_put (vmm, &bar); }
@@ -406,10 +423,14 @@ nv50_instmem_fini(struct nvkm_instmem *base)nv50_instmem (base)->addr = ~0ULL ; } -static void * +void *nv50_instmem_dtor (struct nvkm_instmem *base) { -return nv50_instmem (base); +struct nv50_instmem *imem =nv50_instmem (base); + +if (imem->iomap.size) +io_mapping_fini (&imem->iomap); +return imem; } static const struct nvkm_instmem_func
drivers/gpu/drm/nouveau/nvkm/subdev/instmem/priv.h +1 -0
@@ -26,6 +26,7 @@ int nv50_instobj_new(struct nvkm_instmem *, u32 size, u32 align, bool zero, struct nvkm_memory **); int nv50_instobj_wrap(struct nvkm_instmem *, struct nvkm_memory *vram,struct nvkm_memory **bar2); +void *nv50_instmem_dtor (struct nvkm_instmem *base);void nvkm_instmem_ctor (const struct nvkm_instmem_func *,struct nvkm_device *,enum nvkm_subdev_type ,int ,struct nvkm_instmem *);
Signature · 1 line
2.54.0
[email protected]
AI reviewThank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [High] Unconditional execution of io_mapping_init_wc() in check_io_mapping() causes severe ioremap_wc…
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] Unconditional execution of io_mapping_init_wc() in check_io_mapping() causes severe ioremap_wc memory leaks during BAR2 bootstrapping. --
commit 17419e048ef493385192b9b14726d1e9ad616fed
Author: Dave Airlie <[email protected]>
nouveau/instmem: use iomapping interface for instmem handling (v3)
This replaces constant ioremapping with the io_mapping interface to avoid the constant need to ioremap when instobjs move. It delays creating the mapping until first use to avoid resource mapping errors with simpledrm.
Quoted 4 lines
drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv50.c +0
-0
[ ... ]
Quoted 4 lines
drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv50.c +0
-0@@ -188,6 +190,16 @@ nv50_instobj_map(struct nvkm_memory *memory, u64 offset, struct nvkm_vmm *vmm,return nvkm_memory_map (memory, offset, vmm, vma, argv, argc); }
drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv50.c nv50_instobj_map(struct nvkm_memory *memory, u64 offset, struct nvkm_vmm *vmm, · line 202
Quoted 10 lines
drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv50.c +10
-0+static bool +check_io_mapping (struct nv50_instmem *imem) +{ +struct nvkm_device *device = imem->base.subdev.device; + +return io_mapping_init_wc (&imem->iomap, + device->func->resource_addr (device,NVKM_BAR2_INST ), + device->func->resource_size (device,NVKM_BAR2_INST )) !=NULL ; +} +
[Severity: High]
Does this unconditionally initialize the mapping and leak the previous ioremap_wc allocation if called more than once?
Since nv50_instobj_boot() acts as the .boot callback for BAR2 page tables, it is invoked for every leaf page table during VMM bootstrap. Because BAR2 typically spans multiple page tables, check_io_mapping() gets executed multiple times in a loop.
Should this check if the mapping is already initialized (e.g., checking if imem->iomap.size is non-zero) before calling io_mapping_init_wc() to avoid leaking the vmalloc space?
Quoted 3 lines
drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv50.c +0
-0static void nv50_instobj_release (struct nvkm_memory *memory) {
[ ... ]
Quoted 14 lines
drivers/gpu/drm/nouveau/nvkm/subdev/instmem/nv50.c +6
-1@@ -277,7 +289,12 @@ nv50_instobj_boot(struct nvkm_memory *memory, struct nvkm_vmm *vmm) iobj->lru.next =NULL ; } -nv50_instobj_kmap (iobj, vmm); +/* +* boot is only called on BAR2, if we can't remap the complete +* BAR it's unlikely things are functioning well. +*/ +if (check_io_mapping (iobj->imem)) +nv50_instobj_kmap (iobj, vmm);nvkm_instmem_boot (imem);mutex_unlock (&imem->mutex); }
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
Quoted 1 line · Dave Airlie
[PATCH] nouveau/instmem: use iomapping interface for instmem handling (v3)
Applied, thanks!
Branch: drm-misc-next
Tree: https://gitlab.freedesktop.org/drm/misc/kernel.git
[1/1] nouveau/instmem: use iomapping interface for instmem handling
commit: 34e27b90552a
The patch will appear in the next linux-next integration (typically within 24 hours on weekdays).
The patch is queued up for the upcoming merge window for the next major kernel release.