[PATCH] drm: Optimise drm_ioctl() for small user args
Chris Wilson
chris at chris-wilson.co.uk
Mon May 29 20:51:15 UTC 2017
When looking at simple ioctls coupled with conveniently small user
parameters, the overhead of the syscall and drm_ioctl() present large
low hanging fruit. Profiling trivial microbenchmarks around
i915_gem_busy_ioctl, the low hanging fruit comprises of the call to
copy_user(). Those calls are only inlined by the macro where the
constant is known at compile-time, but the ioctl argument size depends
on the ioctl number. To help the compiler, explicitly add switches for
the small sizes that expand to simple moves to/from user. Doing the
multiple inlines does add significant code bloat, so it is very
debatable as to its value. Back to the trivial, but frequently used,
example of i915_gem_busy_ioctl() on a Broadwell avoiding the call gives
us a 15-25% improvement:
before after
single 100.173ns 84.496ns
parallel 204.275ns 152.957ns
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
drivers/gpu/drm/drm_ioctl.c | 103 +++++++++++++++++++++++++++++++++-----------
1 file changed, 77 insertions(+), 26 deletions(-)
diff --git a/drivers/gpu/drm/drm_ioctl.c b/drivers/gpu/drm/drm_ioctl.c
index 865e3ee4d743..3fb61985ba08 100644
--- a/drivers/gpu/drm/drm_ioctl.c
+++ b/drivers/gpu/drm/drm_ioctl.c
@@ -716,8 +716,14 @@ long drm_ioctl(struct file *filp,
drm_ioctl_t *func;
unsigned int nr = DRM_IOCTL_NR(cmd);
int retcode = -EINVAL;
- char stack_kdata[128];
- char *kdata = NULL;
+ union {
+ u8 k8;
+ u16 k16;
+ u32 k32;
+ u64 k64;
+ char kdata[128];
+ } stack;
+ char *kdata = stack.kdata;
unsigned int in_size, out_size, drv_size, ksize;
bool is_driver_ioctl;
@@ -731,12 +737,12 @@ long drm_ioctl(struct file *filp,
if (is_driver_ioctl) {
/* driver ioctl */
if (nr - DRM_COMMAND_BASE >= dev->driver->num_ioctls)
- goto err_i1;
+ goto err_invalid_ioctl;
ioctl = &dev->driver->ioctls[nr - DRM_COMMAND_BASE];
} else {
/* core ioctl */
if (nr >= DRM_CORE_IOCTL_COUNT)
- goto err_i1;
+ goto err_invalid_ioctl;
ioctl = &drm_ioctls[nr];
}
@@ -759,27 +765,45 @@ long drm_ioctl(struct file *filp,
if (unlikely(!func)) {
DRM_DEBUG("no function\n");
retcode = -EINVAL;
- goto err_i1;
+ goto out;
}
retcode = drm_ioctl_permit(ioctl->flags, file_priv);
if (unlikely(retcode))
- goto err_i1;
+ goto out;
- if (ksize <= sizeof(stack_kdata)) {
- kdata = stack_kdata;
- } else {
- kdata = kmalloc(ksize, GFP_KERNEL);
- if (!kdata) {
- retcode = -ENOMEM;
- goto err_i1;
- }
+ if (unlikely(!access_ok(VERIFY_READ, arg, ksize))) {
+ retcode = -EFAULT;
+ goto out;
}
- if (copy_from_user(kdata, (void __user *)arg, in_size) != 0) {
- retcode = -EFAULT;
- goto err_i1;
+ user_access_begin();
+ switch (ksize) {
+ case 1:
+ unsafe_get_user(stack.k8, (u8 *)arg, err_invalid_user);
+ break;
+ case 2:
+ unsafe_get_user(stack.k16, (u16 *)arg, err_invalid_user);
+ break;
+ case 4:
+ unsafe_get_user(stack.k32, (u32 *)arg, err_invalid_user);
+ break;
+ case 8:
+ unsafe_get_user(stack.k64, (u64 *)arg, err_invalid_user);
+ break;
+
+ default:
+ if (ksize > sizeof(stack.kdata))
+ kdata = kmalloc(ksize, GFP_KERNEL);
+ if (!kdata)
+ retcode = -ENOMEM;
+ else if (__copy_from_user(kdata, (void __user *)arg, in_size))
+ goto err_invalid_user;
+ break;
}
+ user_access_end();
+ if (unlikely(retcode))
+ goto out;
if (ksize > in_size)
memset(kdata + in_size, 0, ksize - in_size);
@@ -794,21 +818,48 @@ long drm_ioctl(struct file *filp,
mutex_unlock(&drm_global_mutex);
}
- if (copy_to_user((void __user *)arg, kdata, out_size) != 0)
- retcode = -EFAULT;
+ user_access_begin();
+ switch (out_size) {
+ case 0:
+ break;
+ case 1:
+ unsafe_put_user(stack.k8, (u8 *)arg, err_invalid_user);
+ break;
+ case 2:
+ unsafe_put_user(stack.k16, (u16 *)arg, err_invalid_user);
+ break;
+ case 4:
+ unsafe_put_user(stack.k32, (u32 *)arg, err_invalid_user);
+ break;
+ case 8:
+ unsafe_put_user(stack.k64, (u64 *)arg, err_invalid_user);
+ break;
- err_i1:
- if (!ioctl)
- DRM_DEBUG("invalid ioctl: pid=%d, dev=0x%lx, auth=%d, cmd=0x%02x, nr=0x%02x\n",
- task_pid_nr(current),
- (long)old_encode_dev(file_priv->minor->kdev->devt),
- file_priv->authenticated, cmd, nr);
+ default:
+ if (__copy_to_user((void __user *)arg, kdata, out_size))
+ goto err_invalid_user;
+ }
+ user_access_end();
- if (kdata != stack_kdata)
+out:
+ if (kdata != stack.kdata)
kfree(kdata);
if (retcode)
DRM_DEBUG("ret = %d\n", retcode);
return retcode;
+
+err_invalid_ioctl:
+ DRM_DEBUG("invalid ioctl: pid=%d, dev=0x%lx, auth=%d, cmd=0x%02x, nr=0x%02x\n",
+ task_pid_nr(current),
+ (long)old_encode_dev(file_priv->minor->kdev->devt),
+ file_priv->authenticated, cmd, nr);
+ goto out;
+
+err_invalid_user:
+ user_access_end();
+ retcode = -EFAULT;
+ goto out;
+
}
EXPORT_SYMBOL(drm_ioctl);
--
2.11.0
More information about the Intel-gfx-trybot
mailing list