scsi: cxlflash: Support adapter context mmap and release
authorUma Krishnan <ukrishn@linux.vnet.ibm.com>
Mon, 26 Mar 2018 16:34:20 +0000 (11:34 -0500)
committerMartin K. Petersen <martin.petersen@oracle.com>
Wed, 18 Apr 2018 23:32:49 +0000 (19:32 -0400)
The cxlflash userspace API requires that users be able to mmap and release the
adapter context. Support mapping by implementing the AFU mmap fop to map the
context MMIO space and install the corresponding page table entry upon page
fault. Similarly, implement the AFU release fop to terminate and clean up the
context when invoked.

Signed-off-by: Uma Krishnan <ukrishn@linux.vnet.ibm.com>
Acked-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
drivers/scsi/cxlflash/ocxl_hw.c

index f531403..8a9f2f4 100644 (file)
@@ -1029,10 +1029,80 @@ err:
        goto out;
 }
 
+/**
+ * afu_release() - release and free the context
+ * @inode:     File inode pointer.
+ * @file:      File associated with the context.
+ *
+ * Return: 0 on success, -errno on failure
+ */
+static int afu_release(struct inode *inode, struct file *file)
+{
+       struct ocxlflash_context *ctx = file->private_data;
+       int i;
+
+       /* Unmap and free the interrupts associated with the context */
+       for (i = ctx->num_irqs; i >= 0; i--)
+               afu_unmap_irq(0, ctx, i, ctx);
+       free_afu_irqs(ctx);
+
+       return ocxlflash_release_context(ctx);
+}
+
+/**
+ * ocxlflash_mmap_fault() - mmap fault handler
+ * @vmf:       VM fault associated with current fault.
+ *
+ * Return: 0 on success, -errno on failure
+ */
+static int ocxlflash_mmap_fault(struct vm_fault *vmf)
+{
+       struct vm_area_struct *vma = vmf->vma;
+       struct ocxlflash_context *ctx = vma->vm_file->private_data;
+       u64 mmio_area, offset;
+
+       offset = vmf->pgoff << PAGE_SHIFT;
+       if (offset >= ctx->psn_size)
+               return VM_FAULT_SIGBUS;
+
+       mmio_area = ctx->psn_phys;
+       mmio_area += offset;
+
+       vm_insert_pfn(vma, vmf->address, mmio_area >> PAGE_SHIFT);
+       return VM_FAULT_NOPAGE;
+}
+
+static const struct vm_operations_struct ocxlflash_vmops = {
+       .fault = ocxlflash_mmap_fault,
+};
+
+/**
+ * afu_mmap() - map the fault handler operations
+ * @file:      File associated with the context.
+ * @vma:       VM area associated with mapping.
+ *
+ * Return: 0 on success, -errno on failure
+ */
+static int afu_mmap(struct file *file, struct vm_area_struct *vma)
+{
+       struct ocxlflash_context *ctx = file->private_data;
+
+       if ((vma_pages(vma) + vma->vm_pgoff) >
+           (ctx->psn_size >> PAGE_SHIFT))
+               return -EINVAL;
+
+       vma->vm_flags |= VM_IO | VM_PFNMAP;
+       vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
+       vma->vm_ops = &ocxlflash_vmops;
+       return 0;
+}
+
 static const struct file_operations ocxl_afu_fops = {
        .owner          = THIS_MODULE,
        .poll           = afu_poll,
        .read           = afu_read,
+       .release        = afu_release,
+       .mmap           = afu_mmap,
 };
 
 #define PATCH_FOPS(NAME)                                               \
@@ -1079,6 +1149,8 @@ static struct file *ocxlflash_get_fd(void *ctx_cookie,
        if (fops) {
                PATCH_FOPS(poll);
                PATCH_FOPS(read);
+               PATCH_FOPS(release);
+               PATCH_FOPS(mmap);
        } else /* Use default ops */
                fops = (struct file_operations *)&ocxl_afu_fops;