Hi Boris,
On 09.10.2024 09:10, Steven Price wrote:
> On 08/10/2024 09:47, Boris Brezillon wrote:
> > The system and GPU MMU page size might differ, which becomes a
> > problem for FW sections that need to be mapped at explicit address
> > since our PAGE_SIZE alignment might cover a VA range that's
> > expected to be used for another section.
> >
> > Make sure we never map more than we need.
> >
> > Fixes: 2718d91816ee ("drm/panthor: Add the FW logical block")
> > Signed-off-by: Boris Brezillon <boris.brezillon@xxxxxxxxxxxxx>
>
> Reviewed-by: Steven Price <steven.price@xxxxxxx>
>
> > ---
> > drivers/gpu/drm/panthor/panthor_fw.c | 3 +--
> > drivers/gpu/drm/panthor/panthor_gem.c | 11 ++++++++---
> > drivers/gpu/drm/panthor/panthor_mmu.c | 6 +++---
> > 3 files changed, 12 insertions(+), 8 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/panthor/panthor_fw.c b/drivers/gpu/drm/panthor/panthor_fw.c
> > index ef232c0c2049..293846400296 100644
> > --- a/drivers/gpu/drm/panthor/panthor_fw.c
> > +++ b/drivers/gpu/drm/panthor/panthor_fw.c
> > @@ -515,8 +515,7 @@ static int panthor_fw_load_section_entry(struct panthor_device *ptdev,
> > return -EINVAL;
> > }
> >
> > - if ((hdr.va.start & ~PAGE_MASK) != 0 ||
> > - (hdr.va.end & ~PAGE_MASK) != 0) {
> > + if (!IS_ALIGNED(hdr.va.start, SZ_4K) || !IS_ALIGNED(hdr.va.end, SZ_4K)) {
> > drm_err(&ptdev->base, "Firmware corrupted, virtual addresses not page aligned: 0x%x-0x%x\n",
> > hdr.va.start, hdr.va.end);
> > return -EINVAL;
> > diff --git a/drivers/gpu/drm/panthor/panthor_gem.c b/drivers/gpu/drm/panthor/panthor_gem.c
> > index c60b599665d8..2c8d6e2c7232 100644
> > --- a/drivers/gpu/drm/panthor/panthor_gem.c
> > +++ b/drivers/gpu/drm/panthor/panthor_gem.c
> > @@ -44,8 +44,7 @@ void panthor_kernel_bo_destroy(struct panthor_kernel_bo *bo)
> > to_panthor_bo(bo->obj)->exclusive_vm_root_gem != panthor_vm_root_gem(vm)))
> > goto out_free_bo;
> >
> > - ret = panthor_vm_unmap_range(vm, bo->va_node.start,
> > - panthor_kernel_bo_size(bo));
> > + ret = panthor_vm_unmap_range(vm, bo->va_node.start, bo->va_node.size);
> > if (ret)
> > goto out_free_bo;
> >
> > @@ -95,10 +94,16 @@ panthor_kernel_bo_create(struct panthor_device *ptdev, struct panthor_vm *vm,
> > }
> > bo = to_panthor_bo(&obj->base);
> > - size = obj->base.size;
> > kbo->obj = &obj->base;
> > bo->flags = bo_flags;
> >
> > + /* The system and GPU MMU page size might differ, which becomes a
> > + * problem for FW sections that need to be mapped at explicit address
> > + * since our PAGE_SIZE alignment might cover a VA range that's
> > + * expected to be used for another section.
> > + * Make sure we never map more than we need.
> > + */
I was wondering, this must be a relatively common situation for other DRM
drivers, so maybe the DRM shmem core should have a BO creation function that
lets you specify an exact page alignment?
> > + size = ALIGN(size, SZ_4K);
I was thinking, if we know that the FW's page size is always 4KiB, maybe we could do
#define CSF_FW_PAGESIZE SZ_4K
somewhere in panthor_fw.h to make things clearer?
Cheers,
Adrian
Reviewed-by: Adrian Larumbe <adrian.larumbe@xxxxxxxxxxxxx>
> > ret = panthor_vm_alloc_va(vm, gpu_va, size, &kbo->va_node);
> > if (ret)
> > goto err_put_obj;
> > diff --git a/drivers/gpu/drm/panthor/panthor_mmu.c b/drivers/gpu/drm/panthor/panthor_mmu.c
> > index 3cd2bce59edc..e53d131c53cc 100644
> > --- a/drivers/gpu/drm/panthor/panthor_mmu.c
> > +++ b/drivers/gpu/drm/panthor/panthor_mmu.c
> > @@ -1027,10 +1027,10 @@ panthor_vm_alloc_va(struct panthor_vm *vm, u64 va, u64 size,
> > {
> > int ret;
> >
> > - if (!size || (size & ~PAGE_MASK))
> > + if (!size || !IS_ALIGNED(size, SZ_4K))
> > return -EINVAL;
> >
> > - if (va != PANTHOR_VM_KERNEL_AUTO_VA && (va & ~PAGE_MASK))
> > + if (va != PANTHOR_VM_KERNEL_AUTO_VA && !IS_ALIGNED(va, SZ_4K))
> > return -EINVAL;
> >
> > mutex_lock(&vm->mm_lock);
> > @@ -2370,7 +2370,7 @@ panthor_vm_bind_prepare_op_ctx(struct drm_file *file,
> > int ret;
> >
> > /* Aligned on page size. */
> > - if ((op->va | op->size) & ~PAGE_MASK)
> > + if (!IS_ALIGNED(op->va | op->size, SZ_4K))
> > return -EINVAL;
> >
> > switch (op->flags & DRM_PANTHOR_VM_BIND_OP_TYPE_MASK) {
