O2 VICE support
Attached is patch set to add support to kernel for O2 video compression engine (VICE). It should apply cleanly to latest CVS.
Ilya.
Index: arch/mips64/kernel/ioctl32.c
===================================================================
RCS file: /home/cvs/linux/arch/mips64/kernel/ioctl32.c,v retrieving revision 1.26
mips64/kernel/ioctl32.c +10 -0
@@ -55,6 +57,10 @@#include <linux/rtc.h> +#ifdef CONFIG_O2_VICE +#include <linux/vice.h> +#endif +long sys_ioctl (unsigned int fd,unsigned int cmd,unsigned long arg);static int w_long (unsigned int fd,unsigned int cmd,unsigned long arg)
@@ -827,6 +968,12 @@ COMPATIBLE_IOCTL(TIOCGSERIAL) COMPATIBLE_IOCTL(TIOCSSERIAL) COMPATIBLE_IOCTL(TIOCSERGETLSR) +#ifdef CONFIG_O2_VICE +COMPATIBLE_IOCTL(VICE_IOCTL_MAP_DMA) +COMPATIBLE_IOCTL(VICE_IOCTL_MSP_RUN) +COMPATIBLE_IOCTL(VICE_IOCTL_BSP_RUN) +COMPATIBLE_IOCTL(VICE_IOCTL_DO_DMA) +#endif COMPATIBLE_IOCTL(FIOCLEX) COMPATIBLE_IOCTL(FIONCLEX)
Index: drivers/char/Kconfig
===================================================================
RCS file: /home/cvs/linux/drivers/char/Kconfig,v
retrieving revision 1.1
char/Kconfig +2 -0
@@ -1281,6 +1281,8 @@ source "drivers/char/pcmcia/Kconfig" +source drivers/char/o2vice/Kconfig + config MWAVE tristate "ACP Modem (Mwave) support" depends on X86
Index: drivers/char/Makefile
===================================================================
RCS file: /home/cvs/linux/drivers/char/Makefile,v
retrieving revision 1.100
char/Makefile +1 -0
@@ -111,6 +111,7 @@ obj-$(CONFIG_AGP) += agp/ obj-$(CONFIG_DRM) += drm/ obj-$(CONFIG_PCMCIA) += pcmcia/ +obj-$(CONFIG_O2_VICE) += o2vice/ # Files generated that shall be removed upon make clean clean-files := consolemap_deftbl.c defkeymap.c qtronixmap.c
char/o2vice/Kconfig new file +32 -0
@@ -0,0 +1,32 @@ +# +# O2 VICE Engine confiuration +# + +config O2_VICE + tristate "O2 VICE Engine Support" + depends on SGI_IP32 + ---help--- + This option enables O2 VICE Engine support. + VICE stands for Video Image Compression Engine. This is very powerfull + piece of silicon, that can greatly speed up lots of graphics, vide, or + sound related tasks. To be able to use it, you will also need special + library, that can be found at <insert URL here> + +config O2_VICE_DBGG + bool "VICE Debugger Support. READ HELP!" + depends on O2_VICE + ---help--- + This option enables features of VICE driver needed to debug VICE library. + This is probably serious security risk. You don't need it. If you think + you do, you are wrong. Say NO. + +config O2_VICE_DBG + bool "You seem to insist... did you read help? Yes? No? READ HELP!" + depends on O2_VICE_DBGG + ---help--- + You are still here? Didn't I just tell you that it is not needed? + Or do you want to say you *legally* obtained information needed for + programming VICE? That you got all the tools needed? As a matter of + fact, these tools don't even exist yet! + + Sigh... You've been warned...
char/o2vice/Makefile new file +21 -0
@@ -0,0 +1,21 @@ +# +# drivers/char/o2vice/Makefile +# +# Makefile for the O2 VICE Engine driver. +# + +SUB_DIRS := +MOD_SUB_DIRS := $(SUB_DIRS) +ALL_SUB_DIRS := $(SUB_DIRS) + +#O_TARGET := vice.o + +obj-y := +obj-m := +obj-n := +obj- := + +obj-$(CONFIG_O2_VICE) += main.o msp.o bsp.o dma.o +#obj-$(CONFIG_O2_VICE_DBG) += vicedebug.o + +include $(TOPDIR)/Rules.make
linux/vice.h new file +462 -0
462 lines not on this page · raw message
char/o2vice/bsp.c new file +161 -0
161 lines not on this page · raw message
char/o2vice/msp.c new file +157 -0
157 lines not on this page · raw message
char/o2vice/Kconfig new file +32 -0
@@ -0,0 +1,32 @@ +# +# O2 VICE Engine confiuration +# + +config O2_VICE + tristate "O2 VICE Engine Support" + depends on SGI_IP32 + ---help--- + This option enables O2 VICE Engine support. + VICE stands for Video Image Compression Engine. This is very powerfull + piece of silicon, that can greatly speed up lots of graphics, vide, or + sound related tasks. To be able to use it, you will also need special + library, that can be found at <insert URL here> + +config O2_VICE_DBGG + bool "VICE Debugger Support. READ HELP!" + depends on O2_VICE + ---help--- + This option enables features of VICE driver needed to debug VICE library. + This is probably serious security risk. You don't need it. If you think + you do, you are wrong. Say NO. + +config O2_VICE_DBG + bool "You seem to insist... did you read help? Yes? No? READ HELP!" + depends on O2_VICE_DBGG + ---help--- + You are still here? Didn't I just tell you that it is not needed? + Or do you want to say you *legally* obtained information needed for + programming VICE? That you got all the tools needed? As a matter of + fact, these tools don't even exist yet! + + Sigh... You've been warned...
char/o2vice/Makefile new file +21 -0
@@ -0,0 +1,21 @@ +# +# drivers/char/o2vice/Makefile +# +# Makefile for the O2 VICE Engine driver. +# + +SUB_DIRS := +MOD_SUB_DIRS := $(SUB_DIRS) +ALL_SUB_DIRS := $(SUB_DIRS) + +#O_TARGET := vice.o + +obj-y := +obj-m := +obj-n := +obj- := + +obj-$(CONFIG_O2_VICE) += main.o msp.o bsp.o dma.o +#obj-$(CONFIG_O2_VICE_DBG) += vicedebug.o + +include $(TOPDIR)/Rules.make
char/o2vice/main.c new file +438 -0
438 lines not on this page · raw message
char/o2vice/dma.c new file +183 -0
183 lines not on this page · raw message
mips64/kernel/ioctl32.c line 62
Quoted 12 lines · On Tue, Dec 10, 2002 at 11:11:20AM -0800, [email protected] wrote:
RCS file:
/home/cvs/linux/arch/mips64/kernel/ioctl32.c,vretrieving revision 1.26
diff -u -r1.26ioctl32.cmips64/kernel/ioctl32.c +3
-0@@ -55,6 +57,10 @@#include <linux/rtc.h> +#ifdef CONFIG_O2_VICE +#include <linux/vice.h> +#endif
Just include the header unconditionally.
mips64/kernel/ioctl32.c line 976
Quoted 11 lines
mips64/kernel/ioctl32.c +6
-0static int w_long (unsigned int fd,unsigned int cmd,unsigned long arg)@@ -827,6 +968,12 @@ COMPATIBLE_IOCTL(TIOCGSERIAL) COMPATIBLE_IOCTL(TIOCSSERIAL) COMPATIBLE_IOCTL(TIOCSERGETLSR) +#ifdef CONFIG_O2_VICE +COMPATIBLE_IOCTL(VICE_IOCTL_MAP_DMA) +COMPATIBLE_IOCTL(VICE_IOCTL_MSP_RUN) +COMPATIBLE_IOCTL(VICE_IOCTL_BSP_RUN) +COMPATIBLE_IOCTL(VICE_IOCTL_DO_DMA) +#endif
Dito. COMPATIBLE_IOCTL() isn't a big burgden, so it doesn't harm to get rid of the ifdefs.
mips64/kernel/ioctl32.c line 977
Quoted 1 line
mips64/kernel/ioctl32.c +1
-0+source drivers/char/o2vice/Kconfig
I think it should be in drivers/media/vice (i.e. it's one of the media devices, and I don't think you need the o2, especially as you don't use it e.g. in the header name).
Quoted 9 lines
char/o2vice/Makefile new file +6
-0@@ -0,0 +1,21 @@ +# +# drivers/char/o2vice/Makefile +# +# Makefile for the O2 VICE Engine driver. +# +
char/o2vice/Makefile line 16
Quoted 10 lines
char/o2vice/Makefile new file +10
-0+SUB_DIRS := +MOD_SUB_DIRS := $(SUB_DIRS) +ALL_SUB_DIRS := $(SUB_DIRS) + +#O_TARGET := vice.o + +obj-y := +obj-m := +obj-n := +obj- :=
This is 2.2 crap. Just get rid of it..
char/o2vice/Makefile line 21
Quoted 5 lines
char/o2vice/Makefile new file +5
-0+ +obj-$(CONFIG_O2_VICE) += main.o msp.o bsp.o dma.o +#obj-$(CONFIG_O2_VICE_DBG) += vicedebug.o + +include $(TOPDIR)/Rules.make
The include isn't needed anymore either in 2.5.
char/o2vice/Makefile
Quoted 11 lines
char/o2vice/Makefile new file +11
-0+#include <linux/ioctl.h> /* needed for the _IOW etc stuff used later */ +#ifdef __KERNEL__ +#include <asm/pci.h> +#endif +#include <linux/types.h> /* size_t */ + +#ifndef __KERNEL__ +typedef __u64 u64; +typedef __u32 u32; +typedef __u16 u16; +#endif /* __KERNEL */
Urgg. Usually kernel headers aren't supposed to be used in userspace. If you want to use a copy anyway it should be done with much less burden on the kernel code.
char/o2vice/Makefile
Quoted 7 lines
char/o2vice/Makefile new file +7
-0+#ifdef VICE_DEBUG +# define DPRINTK(fmt, args...) printk( KERN_DEBUG "vice: " fmt, ## args) +#else +# define DPRINTK(fmt, args...) /* not debugging: nothing */ +#endif + +#define PDEBUG DPRINTK
Just use pr_debug?
char/o2vice/Makefile
Quoted 4 lines
char/o2vice/Makefile new file +4
-0+ +#ifndef VICE_MAJOR +#define VICE_MAJOR 240 /* dynamic major by default */ +#endif
Comment doesn't match code.
char/o2vice/Makefile
Quoted 10 lines
char/o2vice/Makefile new file +10
-0+/* + * Prototypes for shared functions + */ +ssize_t vice_read (struct file *filp, char *buf, size_t count, + loff_t *f_pos); +ssize_t vice_write (struct file *filp, const char *buf, size_t count, + loff_t *f_pos); +loff_t vice_llseek (struct file *filp, loff_t off, int whence); +int vice_ioctl (struct inode *inode, struct file *filp, + unsigned int cmd, unsigned long arg);
Exported file operations are usually a very bad sign. But it looks like you could just make them static anyway :)
Quoted 5 lines
char/o2vice/Makefile new file +5
-0+/* + * VICE ioctl commands + */ +#ifdef __KERNEL__ +#define VICE_IOCTL_MAGIC 0x96
char/o2vice/Makefile
Quoted 10 lines
char/o2vice/Makefile new file +10
-0+#define VICE_IOCTL_MAP_DMA _IOR(VICE_IOCTL_MAGIC,1,unsigned long) +#define VICE_IOCTL_MSP_RUN _IOR(VICE_IOCTL_MAGIC,2,msp_run) +#define VICE_IOCTL_BSP_RUN _IOR(VICE_IOCTL_MAGIC,3,bsp_run) +#define VICE_IOCTL_DO_DMA _IOWR(VICE_IOCTL_MAGIC,4,dma_run) +#else +#define VICE_IOCTL_MAP_DMA 0x40089601 +#define VICE_IOCTL_MSP_RUN 0x400c9602 +#define VICE_IOCTL_BSP_RUN 0x400c9603 +#define VICE_IOCTL_DO_DMA 0xc00c9604 +#endif
Ummm..
char/o2vice/Makefile
Quoted 10 lines
char/o2vice/Makefile new file +10
-0+static inline int vice_dma_mem_init(vice_dev *vice) +{ + int i; + for (i=0; i<64*2;i++){ + if(!(vice->dma_kmem[i]=pci_alloc_consistent(0,VICE_PAGE_SIZE,&(vice->dma_mem[i])))) { + DPRINTK("failed at page# %i\n",i); + return -ENOMEM; + } + } + return 0;
Please adopt to Documentation/CodingStyle
char/o2vice/Makefile
Quoted 3 lines
char/o2vice/Makefile new file +3
-0+ DPRINTK("Falling asleep\n"); + while(vice->msp_status==VICE_RES_STATUS_INPROGRESS) { + interruptible_sleep_on(&wq);
sleep_on is racy. You should at least use wait_even, but it doesn't look like vice->msp_status is atomically updated anyway.
char/o2vice/Makefile
Quoted 2 lines
char/o2vice/Makefile new file +0
-0
The diff seems to include the same hunk twice..
char/o2vice/Makefile
Quoted 7 lines
char/o2vice/Makefile new file +7
-0+/* In vice_open, the fop_array is used according to TYPE(dev) */ +int vice_open(struct inode *inode, struct file *filp) +{ + vice_dev *vice=vice_device; /* device information */ + int res; + + MOD_INC_USE_COUNT; /* Before we maybe sleep */
Please set .owner in the file_operations instead.
char/o2vice/Makefile
Quoted 3 lines
char/o2vice/Makefile new file +3
-0+ + if(vice->is_open) + return -EBUSY;
This seems pretty racy.
char/o2vice/Makefile
Quoted 4 lines
char/o2vice/Makefile new file +4
-0+ + if (!filp->private_data) { + filp->private_data = vice_device; + }
filp->private_data can't be set.
Quoted 4 lines
char/o2vice/Makefile new file +4
-0+ssize_t vice_read(struct file * filp, char *buf, size_t count, loff_t * f_pos) +{ + printk(KERN_WARNING + "Processing bit streams through reading/writing is not supported yet\n");
char/o2vice/Makefile
Quoted 10 lines
char/o2vice/Makefile new file +10
-0+ return -ENOSYS; +} + +ssize_t vice_write(struct file * filp, const char *buf, size_t count, + loff_t * f_pos) +{ + printk(KERN_WARNING + "Processing bit streams through reading/writing is not supported (yet)\n"); + return -ENOSYS; +}
What about just not implementing the methods instead?
char/o2vice/Makefile
Quoted 5 lines
char/o2vice/Makefile new file +5
-0+static void vice_vma_open(struct vm_area_struct *vma) +{ MOD_INC_USE_COUNT; } + +static void vice_vma_close(struct vm_area_struct *vma) +{ MOD_DEC_USE_COUNT; }
This is silly. You get a reference for vma->vm_file as long as you have any mmaps. That's enough for the refcounting.
char/o2vice/Makefile
Quoted 5 lines
char/o2vice/Makefile new file +5
-0+static struct vm_operations_struct vice_vm_ops = { + open: vice_vma_open, + close: vice_vma_close, + nopage: vice_vma_nopage, +};
Please use C99-initializers (i.e. .foo = bar)
char/o2vice/Makefile
Quoted 6 lines
char/o2vice/Makefile new file +6
-0+void vice_cleanup_module(void) +{ +#ifndef CONFIG_DEVFS_FS + /* cleanup_module is never called if registering failed */ + unregister_chrdev(vice_major, "vice"); +#endif
Umm, just because someone makes the mistake of enabling devfs he doesn't have to use it.. :)
char/o2vice/Makefile
Quoted 3 lines
char/o2vice/Makefile new file +3
-0+#ifndef VICE_DEBUG + EXPORT_NO_SYMBOLS; /* otherwise, leave global symbols visible */ +#endif
EXPORT_NO_SYMBOLS is a noop on 2.5
Quoted 3 lines
Urgg. Usually kernel headers aren't supposed to be used in userspace. If you want to use a copy anyway it should be done with much less burden on the kernel code.
Well, most of manipulation of this device will be done bu user-space code. Kernel's only job is synchronising access and managing memory. Thus we would like to keep all definitions in one place.
Quoted 5 lines
char/o2vice/main.c line 150
Quoted 3 lines
char/o2vice/main.c new file +3
-0+if (!filp->private_data) { + filp->private_data = vice_device; + }
filp->private_datacan't be set.
???
I can live without using it of course, since it isn't currently possible to have more then one instance of VICE in same machine, but theoretical possibility does exist. Where should I pass information about which specific device current call is?
Quoted 17 lines
Quoted 5 lines
char/o2vice/main.c new file +4
-0+ssize_t vice_read (struct file * filp,char *buf,size_t count,loff_t * f_pos) +{ +printk (KERN_WARNING +"Processing bit streams through reading/writing is not supported yet\n" );char/o2vice/main.c line 213
Quoted 10 lines
char/o2vice/main.c new file +10
-0+return -ENOSYS ; +} + +ssize_t vice_write (struct file * filp,const char *buf,size_t count, +loff_t * f_pos) +{ +printk (KERN_WARNING +"Processing bit streams through reading/writing is not supported (yet)\n" ); +return -ENOSYS ; +}What about just not implementing the methods instead?
This is more like reminder to myself that I should do that one day :)
Quoted 8 lines
char/o2vice/main.c line 247
Quoted 5 lines
char/o2vice/main.c new file +5
-0+static void vice_vma_open (struct vm_area_struct *vma) +{MOD_INC_USE_COUNT ; } + +static void vice_vma_close (struct vm_area_struct *vma) +{MOD_DEC_USE_COUNT ; }This is silly. You get a reference for
vma->vm_fileas long as you have any mmaps. That's enough for the refcounting.
I was wondering about that too, but this is the way i was in book :)
Quoted 9 lines
char/o2vice/main.c line 273
Quoted 6 lines
char/o2vice/main.c new file +5
-0+static struct vm_operations_struct vice_vm_ops = { + open: vice_vma_open, + close: vice_vma_close, + nopage: vice_vma_nopage, +};Please use C99-initializers (i.e. .foo = bar)
Quoted 9 lines
char/o2vice/main.c line 352
Quoted 6 lines
char/o2vice/main.c new file +6
-0+void vice_cleanup_module (void ) +{ +#ifndef CONFIG_DEVFS_FS +/* cleanup_module is never called if registering failed */ +unregister_chrdev (vice_major,"vice" ); +#endif Umm, just because someone makes the mistake of enabling devfs he doesn't have to use it.. :)
I'm not buying that one :)
Quoted 6 lines
char/o2vice/main.c line 425
Quoted 4 lines
char/o2vice/main.c new file +3
-0+#ifndef VICE_DEBUG +EXPORT_NO_SYMBOLS ;/* otherwise, leave global symbols visible */ +#endif
EXPORT_NO_SYMBOLSis a noop on 2.5
Driver was written for 2.4 first.
Thanks a lot for constructive criticism.
Ilya.
Quoted 4 lines
Quoted 4 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
"ilya" == ilya <[email protected]> writes:
Quoted 9 lines
Quoted 9 lines
char/o2vice/main.c line 352
Quoted 6 lines
char/o2vice/main.c new file +6
-0+void vice_cleanup_module (void ) +{ +#ifndef CONFIG_DEVFS_FS +/* cleanup_module is never called if registering failed */ +unregister_chrdev (vice_major,"vice" ); +#endif Umm, just because someone makes the mistake of enabling devfs he doesn't have to use it.. :)
ilya> I'm not buying that one :)
Read behind the joke:
- you have devfs compiled in
- you don't have devfs mounted
Conclusion:
there is a bug in your code :)
Later, Juan.
Signature · 2 lines
In theory, practice and theory are the same, but in practice they are different -- Larry McVoy
Quoted 10 lines · On Wed, Dec 11, 2002 at 02:16:29PM -0800, [email protected] wrote:
Quoted 5 lines
char/o2vice/main.c line 150
Quoted 3 lines
char/o2vice/main.c new file +3
-0+if (!filp->private_data) { + filp->private_data = vice_device; + }
filp->private_datacan't be set.???
I can live without using it of course, since it isn't currently possible to have more then one instance of VICE in same machine, but theoretical possibility does exist. Where should I pass information about which specific device current call is?
Sorry if that sentence was confusing. I meant it can't be already set in ->open so the if is superflous.
Quoted 10 lines
Quoted 9 lines
char/o2vice/main.c line 352
Quoted 6 lines
char/o2vice/main.c new file +6
-0+void vice_cleanup_module (void ) +{ +#ifndef CONFIG_DEVFS_FS +/* cleanup_module is never called if registering failed */ +unregister_chrdev (vice_major,"vice" ); +#endif Umm, just because someone makes the mistake of enabling devfs he doesn't have to use it.. :)
I'm not buying that one :)
You can have CONFIG_DEVFS_FS set but devfs not mounted and used. Thus #ifndef CONFIG_DEVFS_FS code is a very bad idea.
char/o2vice/Kconfig line 13
Quoted 12 lines · On Tue, 10 Dec 2002 [email protected] wrote:
Attached is patch set to add support to kernel for O2 video compression engine (VICE). It should apply cleanly to latest CVS.
char/o2vice/Kconfig new file +9
-0+config O2_VICE + tristate "O2 VICE Engine Support" + depends on SGI_IP32 + ---help--- + This option enables O2 VICE Engine support. + VICE stands for Video Image Compression Engine. This is very powerfull + piece of silicon, that can greatly speed up lots of graphics, vide, or + sound related tasks. To be able to use it, you will also need special + library, that can be found at <insert URL here>
^^^^^^^^^^^^^^^^^
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]
In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds