> static int i830_map_buffer(drm_buf_t * buf, struct file *filp) > { > drm_file_t *priv = filp->private_data; > drm_device_t *dev = priv->head->dev; > drm_i830_buf_priv_t *buf_priv = buf->dev_private; > drm_i830_private_t *dev_priv = dev->dev_private; > const struct file_operations *old_fops; > unsigned long virtual; > int retcode = 0; > > if (buf_priv->currently_mapped == I830_BUF_MAPPED) > return -EINVAL; > > down_write(¤t->mm->mmap_sem); > old_fops = filp->f_op; > filp->f_op = &i830_buffer_fops; > dev_priv->mmap_buffer = buf; > virtual = do_mmap(filp, 0, buf->total, PROT_READ | PROT_WRITE, > MAP_SHARED, buf->bus_address); > dev_priv->mmap_buffer = NULL; > filp->f_op = old_fops; > if (IS_ERR((void *)virtual)) { /* ugh */ > /* Real error */ > DRM_ERROR("mmap error\n"); > retcode = PTR_ERR((void *)virtual); > buf_priv->virtual = NULL; > } else { > buf_priv->virtual = (void __user *)virtual; > } > up_write(¤t->mm->mmap_sem); > > return retcode; > } > > (and same crap in i810_dma.c aswell) > > Overriding the file operations just for mmap is for one thing racy > as hell and for another very fragile as the mmap and nopage routines > have to agree closely on what to do. > > Even further why in hell do you call do_mmap from a driver? Mapping > memory into userspace from anything but syscall dedicated to it is > surely a desaster waiting to happen. > > Is there any chance we can get rid of this crap (and similar stuff in > drm_bufs.c) as part of the memory manager overhaul? Long-term I'd > like to get rid of the do_mmap(_pgoff) export to avoid that people > introduce similar braindamage again. >
We can't change it without breaking userspace apps, so although the code is ugly it has been working for years is only used on i810 systems, most i830 use i915 driver now, and I don't think you can implement it any other way that will remain compatible with userspace... the drm doesn't use syscalls it uses ioctls, and the ioctl in this case is dedicated to mapping buffers into userspace... The code is mainly concerned with mapping userspace-generic buffers whether they are in AGP/SG/framebuffer space to the userspace driver via a single ioctl call, the userspace doesn't need to know what type of buffers they are in theory, if you look at drm_mmap in drm_vm.c you'll see the other half of this great fun, I'm not sure reimplementing this stuff is worth it, to just avoid do_mmap_pgoff export, and if it can be implemented without busting userspace all over the place.. The new memory manager system can't fix old systems, it isn't a fix old brain damage, it is a whole new interface, it just provides the new interfaces and the drivers have to be ported to it, but the old interfaces have to remain as far as I can see for ever.. the old memory manager interfaces are driver specific in a lot of cases and it would be quite impossible to overhaul and remain userspace compatible.. Dave. -- David Airlie, Software Engineer http://www.skynet.ie/~airlied / airlied at skynet.ie Linux kernel - DRI, VAX / pam_smb / ILUG ------------------------------------------------------------------------- Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys - and earn cash http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV -- _______________________________________________ Dri-devel mailing list Dri-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dri-devel