OpenWrt – Rev 1
?pathlinks?
From d56622a9c76ab913da66bb22c3158b10f9fb0543 Mon Sep 17 00:00:00 2001
From: Siarhei Siamashka <siarhei.siamashka@gmail.com>
Date: Mon, 17 Jun 2013 13:32:11 +0300
Subject: [PATCH 056/454] fbdev: add FBIOCOPYAREA ioctl
Based on the patch authored by Ali Gholami Rudi at
https://lkml.org/lkml/2009/7/13/153
Provide an ioctl for userspace applications, but only if this operation
is hardware accelerated (otherwide it does not make any sense).
Signed-off-by: Siarhei Siamashka <siarhei.siamashka@gmail.com>
bcm2708_fb: Add ioctl for reading gpu memory through dma
---
drivers/video/fbdev/bcm2708_fb.c | 111 +++++++++++++++++++++++++++++++
drivers/video/fbdev/core/fbmem.c | 36 ++++++++++
include/uapi/linux/fb.h | 12 ++++
3 files changed, 159 insertions(+)
--- a/drivers/video/fbdev/bcm2708_fb.c
+++ b/drivers/video/fbdev/bcm2708_fb.c
@@ -31,8 +31,10 @@
#include <linux/console.h>
#include <linux/debugfs.h>
#include <asm/sizes.h>
+#include <linux/uaccess.h>
#include <linux/io.h>
#include <linux/dma-mapping.h>
+#include <linux/cred.h>
#include <soc/bcm2835/raspberrypi-firmware.h>
//#define BCM2708_FB_DEBUG
@@ -94,6 +96,7 @@ struct bcm2708_fb {
wait_queue_head_t dma_waitq;
struct bcm2708_fb_stats stats;
unsigned long fb_bus_address;
+ struct { u32 base, length; } gpu;
};
#define to_bcm2708(info) container_of(info, struct bcm2708_fb, fb)
@@ -426,6 +429,106 @@ static int bcm2708_fb_pan_display(struct
return result;
}
+static void dma_memcpy(struct bcm2708_fb *fb, dma_addr_t dst, dma_addr_t src, int size)
+{
+ int burst_size = (fb->dma_chan == 0) ? 8 : 2;
+ struct bcm2708_dma_cb *cb = fb->cb_base;
+
+ cb->info = BCM2708_DMA_BURST(burst_size) | BCM2708_DMA_S_WIDTH |
+ BCM2708_DMA_S_INC | BCM2708_DMA_D_WIDTH |
+ BCM2708_DMA_D_INC;
+ cb->dst = dst;
+ cb->src = src;
+ cb->length = size;
+ cb->stride = 0;
+ cb->pad[0] = 0;
+ cb->pad[1] = 0;
+ cb->next = 0;
+
+ if (size < dma_busy_wait_threshold) {
+ bcm_dma_start(fb->dma_chan_base, fb->cb_handle);
+ bcm_dma_wait_idle(fb->dma_chan_base);
+ } else {
+ void __iomem *dma_chan = fb->dma_chan_base;
+ cb->info |= BCM2708_DMA_INT_EN;
+ bcm_dma_start(fb->dma_chan_base, fb->cb_handle);
+ while (bcm_dma_is_busy(dma_chan)) {
+ wait_event_interruptible(
+ fb->dma_waitq,
+ !bcm_dma_is_busy(dma_chan));
+ }
+ fb->stats.dma_irqs++;
+ }
+ fb->stats.dma_copies++;
+}
+
+#define INTALIAS_NORMAL(x) ((x)&~0xc0000000) // address with no aliases
+#define INTALIAS_L1L2_NONALLOCATING(x) (((x)&~0xc0000000)|0x80000000) // cache coherent but non-allocating in L1 and L2
+
+static long vc_mem_copy(struct bcm2708_fb *fb, unsigned long arg)
+{
+ struct fb_dmacopy ioparam;
+ size_t size = PAGE_SIZE;
+ u32 *buf = NULL;
+ dma_addr_t bus_addr;
+ long rc = 0;
+ size_t offset;
+
+ /* restrict this to root user */
+ if (!uid_eq(current_euid(), GLOBAL_ROOT_UID))
+ {
+ rc = -EFAULT;
+ goto out;
+ }
+
+ /* Get the parameter data.
+ */
+ if (copy_from_user
+ (&ioparam, (void *)arg, sizeof(ioparam)) != 0) {
+ pr_err("[%s]: failed to copy-from-user\n",
+ __func__);
+ rc = -EFAULT;
+ goto out;
+ }
+
+ if (fb->gpu.base == 0 || fb->gpu.length == 0) {
+ pr_err("[%s]: Unable to determine gpu memory (%x,%x)\n", __func__, fb->gpu.base, fb->gpu.length);
+ return -EFAULT;
+ }
+
+ if (INTALIAS_NORMAL(ioparam.src) < fb->gpu.base || INTALIAS_NORMAL(ioparam.src) >= fb->gpu.base + fb->gpu.length) {
+ pr_err("[%s]: Invalid memory access %x (%x-%x)", __func__, INTALIAS_NORMAL(ioparam.src), fb->gpu.base, fb->gpu.base + fb->gpu.length);
+ return -EFAULT;
+ }
+
+ buf = dma_alloc_coherent(fb->fb.device, PAGE_ALIGN(size), &bus_addr,
+ GFP_ATOMIC);
+ if (!buf) {
+ pr_err("[%s]: failed to dma_alloc_coherent(%d)\n",
+ __func__, size);
+ rc = -ENOMEM;
+ goto out;
+ }
+
+ for (offset = 0; offset < ioparam.length; offset += size) {
+ size_t remaining = ioparam.length - offset;
+ size_t s = min(size, remaining);
+ unsigned char *p = (unsigned char *)ioparam.src + offset;
+ unsigned char *q = (unsigned char *)ioparam.dst + offset;
+ dma_memcpy(fb, bus_addr, INTALIAS_L1L2_NONALLOCATING((dma_addr_t)p), size);
+ if (copy_to_user(q, buf, s) != 0) {
+ pr_err("[%s]: failed to copy-to-user\n",
+ __func__);
+ rc = -EFAULT;
+ goto out;
+ }
+ }
+out:
+ if (buf)
+ dma_free_coherent(fb->fb.device, PAGE_ALIGN(size), buf, bus_addr);
+ return rc;
+}
+
static int bcm2708_ioctl(struct fb_info *info, unsigned int cmd, unsigned long arg)
{
struct bcm2708_fb *fb = to_bcm2708(info);
@@ -438,6 +541,9 @@ static int bcm2708_ioctl(struct fb_info
RPI_FIRMWARE_FRAMEBUFFER_SET_VSYNC,
&dummy, sizeof(dummy));
break;
+ case FBIODMACOPY:
+ ret = vc_mem_copy(fb, arg);
+ break;
default:
dev_dbg(info->device, "Unknown ioctl 0x%x\n", cmd);
return -ENOTTY;
@@ -760,6 +866,11 @@ static int bcm2708_fb_probe(struct platf
fb->dev = dev;
fb->fb.device = &dev->dev;
+ // failure here isn't fatal, but we'll fail in vc_mem_copy if fb->gpu is not valid
+ rpi_firmware_property(fb->fw,
+ RPI_FIRMWARE_GET_VC_MEMORY,
+ &fb->gpu, sizeof(fb->gpu));
+
ret = bcm2708_fb_register(fb);
if (ret == 0) {
platform_set_drvdata(dev, fb);
--- a/drivers/video/fbdev/core/fbmem.c
+++ b/drivers/video/fbdev/core/fbmem.c
@@ -1090,6 +1090,31 @@ fb_blank(struct fb_info *info, int blank
}
EXPORT_SYMBOL(fb_blank);
+static int fb_copyarea_user(struct fb_info *info,
+ struct fb_copyarea *copy)
+{
+ int ret = 0;
+ if (!lock_fb_info(info))
+ return -ENODEV;
+ if (copy->dx >= info->var.xres ||
+ copy->sx >= info->var.xres ||
+ copy->width > info->var.xres ||
+ copy->dy >= info->var.yres ||
+ copy->sy >= info->var.yres ||
+ copy->height > info->var.yres ||
+ copy->dx + copy->width > info->var.xres ||
+ copy->sx + copy->width > info->var.xres ||
+ copy->dy + copy->height > info->var.yres ||
+ copy->sy + copy->height > info->var.yres) {
+ ret = -EINVAL;
+ goto out;
+ }
+ info->fbops->fb_copyarea(info, copy);
+out:
+ unlock_fb_info(info);
+ return ret;
+}
+
static long do_fb_ioctl(struct fb_info *info, unsigned int cmd,
unsigned long arg)
{
@@ -1100,6 +1125,7 @@ static long do_fb_ioctl(struct fb_info *
struct fb_cmap cmap_from;
struct fb_cmap_user cmap;
struct fb_event event;
+ struct fb_copyarea copy;
void __user *argp = (void __user *)arg;
long ret = 0;
@@ -1217,6 +1243,15 @@ static long do_fb_ioctl(struct fb_info *
unlock_fb_info(info);
console_unlock();
break;
+ case FBIOCOPYAREA:
+ if (info->flags & FBINFO_HWACCEL_COPYAREA) {
+ /* only provide this ioctl if it is accelerated */
+ if (copy_from_user(©, argp, sizeof(copy)))
+ return -EFAULT;
+ ret = fb_copyarea_user(info, ©);
+ break;
+ }
+ /* fall through */
default:
if (!lock_fb_info(info))
return -ENODEV;
@@ -1362,6 +1397,7 @@ static long fb_compat_ioctl(struct file
case FBIOPAN_DISPLAY:
case FBIOGET_CON2FBMAP:
case FBIOPUT_CON2FBMAP:
+ case FBIOCOPYAREA:
arg = (unsigned long) compat_ptr(arg);
case FBIOBLANK:
ret = do_fb_ioctl(info, cmd, arg);
--- a/include/uapi/linux/fb.h
+++ b/include/uapi/linux/fb.h
@@ -35,6 +35,12 @@
#define FBIOPUT_MODEINFO 0x4617
#define FBIOGET_DISPINFO 0x4618
#define FBIO_WAITFORVSYNC _IOW('F', 0x20, __u32)
+/*
+ * HACK: use 'z' in order not to clash with any other ioctl numbers which might
+ * be concurrently added to the mainline kernel
+ */
+#define FBIOCOPYAREA _IOW('z', 0x21, struct fb_copyarea)
+#define FBIODMACOPY _IOW('z', 0x22, struct fb_dmacopy)
#define FB_TYPE_PACKED_PIXELS 0 /* Packed Pixels */
#define FB_TYPE_PLANES 1 /* Non interleaved planes */
@@ -347,6 +353,12 @@ struct fb_copyarea {
__u32 sy;
};
+struct fb_dmacopy {
+ void *dst;
+ __u32 src;
+ __u32 length;
+};
+
struct fb_fillrect {
__u32 dx; /* screen-relative */
__u32 dy;