eichelbe@nadc.arpa (03/21/86)
---
Here's a problem that asks wizards to test out their long-term memory.
I have a VAX 11/780 under 4.1 BSD UNIX and I have a version of a supposedly
4.1 BSD UNIX uda.c (UDA50/RA81) driver that compiles but for one error.
In the routine udprobe (which is shown below) there is a reference to
a variable uh_lastiv in the uba_hd structure. This structure is in the header
file /sys/h/ubavar.h, but I don't find a uh_lastiv in the structure. Here come
the questions:
(1) What is uh_lastiv supposed to do?
(2) Can I (and how if I can) change udprobe to get rid of the uh_lastiv
reference safely?
(3) If I can't get rid of the reference, how should I change the uba_hd
structure in /sys/h/ubavar.h so I too can use uh_lastiv, and if I do this
will I have to change any code anywhere else in the kernel, etc. ?
(This last part, about changes to other code, is important since I am
really at a loss here. If someone could just do a grep for uh_lastiv
through the system code if you have uh_lastiv in your /sys/h/ubavar.h.)
---
udprobe(reg, ctlr)
caddr_t reg;
int ctlr;
{
register int br, cvec;
register struct uda_softc *sc = &uda_softc[ctlr];
#ifdef lint
br = 0; cvec = br; br = cvec;
#endif
/* SHOULD CHECK THAT IT REALLY IS A UDA */
br = 0x15;
cvec = sc->sc_ivec = (uba_hd[numuba].uh_lastiv -= 4);
return(1);
}
---
Oh, and here's uba_hd, extracted from /sys/h/ubavar.h.
---
struct uba_hd {
struct uba_regs *uh_uba; /* virt addr of uba */
struct uba_regs *uh_physuba; /* phys addr of uba */
int (**uh_vec)(); /* interrupt vector */
struct uba_device *uh_actf; /* head of queue to transfer */
struct uba_device *uh_actl; /* tail of queue to transfer */
short uh_mrwant; /* someone is waiting for map reg */
short uh_bdpwant; /* someone awaits bdp's */
int uh_bdpfree; /* free bdp's */
int uh_hangcnt; /* number of ticks hung */
int uh_zvcnt; /* number of 0 vectors */
short uh_users; /* transient bdp use count */
short uh_xclu; /* an rk07 is using this uba! */
#define UAMSIZ 25
struct map *uh_map; /* buffered data path regs free */
};
---
Thanks.
Jon Eichelberger
eichelbe@NADC.ARPAchris@umcp-cs.UUCP (Chris Torek) (03/24/86)
In article <1986@brl-smoke.ARPA> eichelbe@nadc.arpa writes: >I have a VAX 11/780 under 4.1 BSD UNIX and [a] (UDA50/RA81) driver >that compiles but for one error. In the routine udprobe ... there >is a reference to a variable uh_lastiv in the uba_hd structure. ... >I don't find a uh_lastiv in the structure. Ah, at last, I can show off my wizardly skills again! :-) Actually, I had the exact same thing happen when I installed a uda.c on a 4.1 system, long ago. >(1) What is uh_lastiv supposed to do? It is the `last interrupt vector' address, for devices with programmable interrupt vectors. These did not appear until well after the 4.1 release, so only `late 4.1s' have the code. The variable is set to 0x200 (01000) in autoconf.c, and predecremented by the appropriate amount by any driver needing a programmable interrupt vector. >(2) Can I (and how if I can) change udprobe to get rid of the uh_lastiv > reference safely? If you have exactly one UDA50, no DMF32s, no DMZ32s, and so forth, you can use br = 0x15; cvec = sc->sc_ivec = 0774; but this creates a bug just waiting until the time is ripe to leap out and strangle you. >(3) If I can't get rid of the reference, how should I change the uba_hd > structure in /sys/h/ubavar.h so I too can use uh_lastiv .... 1) Add `uh_lastiv' to `struct uba_hd' in ubavar.h (at the end; that may be unnecessary, but is definitely safe). 2) In autoconf.c, find the routine named `unifind' or `UNIfind' or some variation thereof. In there will be code resembling the following: if (numuba == 0) uhp->uh_vec = UNIvec; else uhp->uh_vec = (int(**)())calloc(512); for (i = 0; i < 128; i++) uhp->uh_vec[i] = scbentry(&catcher[i*2], SCB_ISTACK); After this, add: /* * Set last free interrupt vector for devices with * programmable interrupt vectors. Use is to decrement * this number and use result as interrupt vector. */ uhp->uh_lastiv = 0x200; Unfortunately, that uda.c is fairly buggy. If I had to run RA81s on a 4.1 system, I would reconvert my new driver to 4.1 code. But if you must run that driver on a 780, use it on a Unibus all by itself, or at least with nothing else that uses BDPs. -- In-Real-Life: Chris Torek, Univ of MD Comp Sci Dept (+1 301 454 1415) UUCP: seismo!umcp-cs!chris CSNet: chris@umcp-cs ARPA: chris@mimsy.umd.edu
ed@mtxinu.UUCP (Ed Gould) (03/25/86)
In article <1986@brl-smoke.ARPA> eichelbe@nadc.arpa writes: >--- >Here's a problem that asks wizards to test out their long-term memory. >I have a VAX 11/780 under 4.1 BSD UNIX and I have a version of a supposedly >4.1 BSD UNIX uda.c (UDA50/RA81) driver that compiles but for one error. >In the routine udprobe (which is shown below) there is a reference to >a variable uh_lastiv in the uba_hd structure. This structure is in the header >file /sys/h/ubavar.h, but I don't find a uh_lastiv in the structure. Here come >the questions: >(1) What is uh_lastiv supposed to do? >(2) Can I (and how if I can) change udprobe to get rid of the uh_lastiv > reference safely? >(3) If I can't get rid of the reference, how should I change the uba_hd > structure in /sys/h/ubavar.h so I too can use uh_lastiv, and if I do this > will I have to change any code anywhere else in the kernel, etc. ? > (This last part, about changes to other code, is important since I am > really at a loss here. If someone could just do a grep for uh_lastiv > through the system code if you have uh_lastiv in your /sys/h/ubavar.h.) It sounds like you really have a 4.2 driver. 4.1 didn't have uh_lastiv, nor did it support UDA50's or DMF32's or any other devices with software- programmable interrupt vectors. Uh_lastiv (last interrupt vector) was added to support these kind of devices. It is initialized to 01000 before autoconfiguration starts. When it's needed, as in the UDA probe routine, it is decremented by the size of the vector space (4 for the UDA), and the result is handed to the device to use as its interrupt vector: cvec = sc->sc_ivec = (uba_hd[numuba].uh_lastiv -= 4); (sc_ivec is later handed to the device; cvec is the return-in-register from probe). -- Ed Gould mt Xinu, 2910 Seventh St., Berkeley, CA 94710 USA {ucbvax,decvax}!mtxinu!ed +1 415 644 0146 "A man of quality is not threatened by a woman of equality."