off topic a bit but i realy need help c++ and kernel

off topic a bit but i realy need help c++ and kernel

[david]

[-- Attachment #1: Type: text/plain, Size: 2576 bytes --]

hi i am writing a video kernel driver for linux lexos and have got stuck

this is how NVIDIA do their regs

#define NV_PFIFO_RAMHT                                   0x00002210 /*
RW-4R */
#define NV_PFIFO_RAMHT_BASE_ADDRESS                             8:4 /*
RWIUF */
#define NV_PFIFO_RAMHT_BASE_ADDRESS_10000                0x00000010 /*
RWI-V */
#define NV_PFIFO_RAMHT_SIZE                                   17:16 /*
RWIUF */
#define NV_PFIFO_RAMHT_SIZE_4K                           0x00000000 /*
RWI-V */
#define NV_PFIFO_RAMHT_SIZE_8K                           0x00000001 /*
RW--V */
#define NV_PFIFO_RAMHT_SIZE_16K                          0x00000002 /*
RW--V */
#define NV_PFIFO_RAMHT_SIZE_32K                          0x00000003 /*
RW--V */
#define NV_PFIFO_RAMHT_SEARCH                                 25:24 /*
RWIUF */
#define NV_PFIFO_RAMHT_SEARCH_16                         0x00000000 /*
RWI-V */
#define NV_PFIFO_RAMHT_SEARCH_32                         0x00000001 /*
RW--V */
#define NV_PFIFO_RAMHT_SEARCH_64                         0x00000002 /*
RW--V */
#define NV_PFIFO_RAMHT_SEARCH_128                        0x00000003 /*
RW--V */

this can be used like setreg( NV_PFIFO_RAMHT , htbase << ( 0 ?
NV_PFIFO_RAMHT_BASE_ADDRESS)) ;
but i wont to make it better so i tried it this way

#define NV_PFIFO_RAMHT_BASE_ADDRESS 0x00002210:8:4:3 /* address : high
bit : low bit : mode 3 = rw
this seemed much better all on one line so you can do
setreg(NV_PFIFO_RAMHT_BASE_ADDRESS , htbase) ;
but ( 2 ? 1:2:3 ) dose not work
and NV_PFIFO_RAMHT_BASE_ADDRESS 0x00002210,8,4,3 dose not work as
setreg(a,h,l,m) thinks this is
all one agr
the other would be to have it as a big struct

but then compiler can not optimize it ( i think )
if i can use #define 's the it quicker and setreg can be an inline asm
marco

how would it work if i used 64 bit like this

bits 32 - 63 reg address
bits 24 - 31 high bit
bits 16 - 23 low  bit
bits 0  - 7  mode

#define NV_PFIFO_RAMHT_BASE_ADDRESS  0x0000221008040003

so if i sent this to the setreg(unsigned long long , data) asm macro it
would be in EAX:EDX (64 bit)
can the kernel use (long long) and (unsigned long long)  ( 64 bit ints)

that 64 bit way look good but is it ?
                   ( opps i wold need an extr 32 bits for value but can
have 64:32 i think )

has anyone got any help they can give me on this !please email me thank
you!
as i am stuck
also is there a down loadable book on c++ as i am just learning from
others ppl code and really
need a book to do it right

thank you david rundle <sector2@ihug.co.nz>








[-- Attachment #2: Type: text/html, Size: 5210 bytes --]

Nvidia GeForce2 kernel driver - kernel 2.4.0 test-10

[Anthony Chatman]

Speaking of Nvidia, I have a Nvidia GeForce2, and had problems loading 
the NV kernel module with a patched test10 kernel (i was running test9 
before).  I took a look at the test10 patch, and noticed the following 2 
lines were taken out of  <linux_dir>/include/linux/wrapper.h:

#define mem_map_inc_count(p) atomic_inc(&(p->count))
#define mem_map_dec_count(p) atomic_dec(&(p->count))

I added those two defines back into wrapper.h and then was able to load 
the NVdriver successfully, with no problems. This doesn't appear to 
break compiliation of the test10 kernel either. I thought I'd let 
everyone know if anyone was having problems with the NV kernel driver. 
Please note, I am not a C programmer, but more of a C hacker, and this 
worked for me on my Slackware 7.1 machine. I have no idea what this may 
have broken in the kernel or whatnot. I only know that this fixed the 
problem on this particular box with the NV kernel driver, so proceed at 
your own risk ;-)





david wrote:

> hi i am writing a video kernel driver for linux lexos and have got stuck
> 
> this is how NVIDIA do their regs
> 

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
Please read the FAQ at http://www.tux.org/lkml/

Re: Nvidia GeForce2 kernel driver - kernel 2.4.0 test-10

[J . A . Magallon]

On Wed, 08 Nov 2000 10:52:41 Anthony Chatman wrote:
> Speaking of Nvidia, I have a Nvidia GeForce2, and had problems loading 
> the NV kernel module with a patched test10 kernel (i was running test9 
> before).  I took a look at the test10 patch, and noticed the following 2 
> lines were taken out of  <linux_dir>/include/linux/wrapper.h:
> 
> #define mem_map_inc_count(p) atomic_inc(&(p->count))
> #define mem_map_dec_count(p) atomic_dec(&(p->count))
> 
> I added those two defines back into wrapper.h and then was able to load 

I think you should never do that. Those macros are outdated and were
removed from kernel. Patch NVdriver instead. Patch is attached.

BTW: does your board run ok on 2.4 ? I have a TNT2 and have not been able
to get it working on a 2.4-smp. In 2.2-smp works fine. I don't know what
is bad...In a recent strace (with test10), it seemed to be hanged on a
poll() call...

-- 
Juan Antonio Magallon Lacarta                                 #> cd /pub
mailto:jamagallon@able.es                                     #> more beer

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
Please read the FAQ at http://www.tux.org/lkml/

Re: Nvidia GeForce2 kernel driver - kernel 2.4.0 test-10

[J . A . Magallon]

[-- Attachment #1: Type: text/plain, Size: 253 bytes --]


On Thu, 09 Nov 2000 08:54:36 Athanasius wrote:
> 
>    Oh no it wasn't, doh *;-).

One other try...

-- 
Juan Antonio Magallon Lacarta                                 #> cd /pub
mailto:jamagallon@able.es                                     #> more beer

[-- Attachment #2: patch-nvdriver-2.4.0-test11 --]
[-- Type: application/octet-stream, Size: 1579 bytes --]

diff -ru NVIDIA_kernel-0.9-5/nv.c nvdriver_/nv.c
--- NVIDIA_kernel-0.9-5/nv.c	Sat Aug 26 02:48:38 2000
+++ nvdriver_/nv.c	Thu Nov  9 00:23:42 2000
@@ -49,6 +49,13 @@
 #include <linux/modversions.h>
 #endif
 
+#ifndef mem_map_dec_count
+  #define mem_map_dec_count(p) atomic_inc(&((p)->count));
+#endif
+#ifndef mem_map_inc_count
+  #define mem_map_inc_count(p) atomic_dec(&((p)->count));
+#endif
+
 #include <nv.h>			        // needs to precede other headers (SMP)
 
 #include <linux/stddef.h>
diff -ru NVIDIA_kernel-0.9-5/os-interface.c nvdriver_/os-interface.c
--- NVIDIA_kernel-0.9-5/os-interface.c	Fri Sep  1 04:19:17 2000
+++ nvdriver_/os-interface.c	Thu Nov  9 00:22:45 2000
@@ -1331,6 +1331,11 @@
     char *parmp;
     char ch;
 
+    spinlock_t unload_lock = SPIN_LOCK_UNLOCKED;
+    struct module *mp = THIS_MODULE;
+    struct module_symbol *sym;
+    int i;
+
     if ((strlen(regParmStr) + NV_SYM_PREFIX_LENGTH) > NV_MAX_SYM_NAME)
         goto done;
 
@@ -1351,11 +1356,17 @@
 
     *symp = '\0';
 
-    symbol_value = get_module_symbol(NV_MODULE_NAME, symbol_name);
-    
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 0)
-    put_module_symbol(symbol_value);
-#endif
+    spin_lock(&unload_lock);
+    if (MOD_CAN_QUERY(mp) && (mp->nsyms > 0)) {
+        for (i = mp->nsyms, sym = mp->syms;
+             i > 0; --i, ++sym) {
+
+            if (strcmp(sym->name, symbol_name) == 0) {
+                symbol_value = sym->value;
+                break;
+            }
+        }
+    }
+    spin_unlock(&unload_lock);
 
  done:
     return (void *) symbol_value;