NetBSD Problem Report #32938
From sverre@abbor.fesk.com Sun Feb 26 20:58:18 2006
Return-Path: <sverre@abbor.fesk.com>
Received: from abbor.fesk.com (c-24-8-234-68.hsd1.co.comcast.net [24.8.234.68])
by narn.netbsd.org (Postfix) with ESMTP id 1DD0763B863
for <gnats-bugs@gnats.NetBSD.org>; Sun, 26 Feb 2006 20:58:18 +0000 (UTC)
Message-Id: <200602262058.k1QKwG9O000747@abbor.fesk.com>
Date: Sun, 26 Feb 2006 13:58:16 -0700 (MST)
From: sverre@viewmark.com
Reply-To: sverre@viewmark.com
To: gnats-bugs@netbsd.org
Subject: Remapping of PCMCIA IO addresses by the controller
X-Send-Pr-Version: 3.95
>Number: 32938
>Category: kern
>Synopsis: PCMCIA cards with fixed IO addresses are not working.
>Confidential: no
>Severity: serious
>Priority: high
>Responsible: kern-bug-people
>State: open
>Class: sw-bug
>Submitter-Id: net
>Arrival-Date: Sun Feb 26 21:00:00 +0000 2006
>Last-Modified: Sun Oct 29 19:20:01 +0000 2006
>Originator: Sverre Froyen
>Release: NetBSD 3.99.15 (2006-02-21)
>Organization:
Viewmark
>Environment:
System: NetBSD abbor.fesk.com 3.99.15 NetBSD 3.99.15 (ABBOR) #40: Sun Feb 26 13:01:08 MST 2006 toor@abbor.fesk.com:/usr/src/sys/arch/i386/compile/obj.i386/ABBOR i386
Architecture: i386
Machine: i386
>Description:
Some PCMCIA cards do not provide IO addresses that can be assigned using
the card's IOBASE registers. (In fact, the PCMCIA standard appears to not
even require IOBASE registers for single function cards.) Such cards do
not work under NetBSD (at least when "rbus" is defined) because the kernel
will try to access the card using an address assigned by the rbus system
(usually an address above 0x4000).
The attached patch modifies the code for cbb (in sys/dev/pci/pccbb.c) to
remap the rbus address to the card's IO address. The patch also modifies
the pcmcia code (in sys/dev/pcmcia/pcmcia.c) to set the IOBASE registers
to correct values when remapping is being done.
I note that because the ExCA standard does not require the ability to
remap IO windows, not all PCMCIA controllers can be made to work with
these cards.
Comments on the patch:
Passing the IO window offset from pccbb_pcmcia_io_alloc to
pccbb_pcmcia_io_map to pccbb_pcmcia_do_io_map was painful. The
It is done first by using the new opaque ihandle member in the
pcmcia_io_handle structure. Then, in order to make it accessible
to pccbb_pcmcia_do_io_map (which is being called from two separate places),
I added an offset to the pcic_handle.io structure (in analogy to
the offset in the pcic_handle.mem structure). I also added the offset to
the cbb_pcic_handle.io but I am not certain that that was necessary.
The use of variable pf->pf_mfc_iobase to modify the IOBASE register handling
is a bit of a kludge and does not work when a function has multiple IO
spaces. I added a test for this.
>How-To-Repeat:
Insert any number of PCMCIA modem cards and attempt to use them.
>Fix:
Apply the attached patch.
Index: src/sys/dev/pcmcia/pcmcia.c
===================================================================
RCS file: /cvsroot/src/sys/dev/pcmcia/pcmcia.c,v
retrieving revision 1.78
diff -u -r1.78 pcmcia.c
--- src/sys/dev/pcmcia/pcmcia.c 11 Dec 2005 12:23:23 -0000 1.78
+++ src/sys/dev/pcmcia/pcmcia.c 26 Feb 2006 20:14:46 -0000
@@ -548,6 +548,9 @@
}
}
+ DPRINTF(("%s: pf_mfc_iobase=0x%lx pf_mfc_iomax=0x%lx\n",
+ sc->dev.dv_xname, pf->pf_mfc_iobase, pf->pf_mfc_iomax));
+
if (pcmcia_mfc(sc) || 1) {
pcmcia_ccr_write(pf, PCMCIA_CCR_IOBASE0,
(pf->pf_mfc_iobase >> 0) & 0xff);
@@ -780,6 +783,8 @@
int error = 0;
int n, m;
+ pf->pf_mfc_iobase = 0;
+
for (n = 0; n < cfe->num_iospace; n++) {
bus_addr_t start = cfe->iospace[n].start;
bus_size_t length = cfe->iospace[n].length;
@@ -800,6 +805,18 @@
&cfe->iospace[n].handle);
if (error)
break;
+ /*
+ * Set pf_mfc_iobase equal to start. This flags the case
+ * where the card specifies a fixed IO base address and the
+ * controller remaps that address.
+ * XXX Need to rethink this in the case of multiple IO
+ * spaces per function. Fail for now.
+ */
+ if (pf->pf_mfc_iobase != 0) {
+ DPRINTF(("pcmcia_config_alloc: cannot (yet) remap multiple IO spaces\n"));
+ break;
+ }
+ pf->pf_mfc_iobase = start;
}
if (n < cfe->num_iospace) {
for (m = 0; m < n; m++)
Index: src/sys/dev/pci/pccbb.c
===================================================================
RCS file: /cvsroot/src/sys/dev/pci/pccbb.c,v
retrieving revision 1.127
diff -u -r1.127 pccbb.c
--- src/sys/dev/pci/pccbb.c 18 Dec 2005 11:04:00 -0000 1.127
+++ src/sys/dev/pci/pccbb.c 26 Feb 2006 20:14:48 -0000
@@ -2073,6 +2073,7 @@
bus_space_tag_t iot;
bus_space_handle_t ioh;
bus_addr_t mask;
+ long *remap_offset;
#if rbus
rbus_tag_t rb;
#endif
@@ -2141,12 +2142,20 @@
(u_long) ioaddr, (u_long) size));
}
#endif
+ remap_offset = (long *) malloc(sizeof(long), M_DEVBUF, M_WAITOK);
+ if (! remap_offset)
+ return 1;
+ *remap_offset = 0;
+#if rbus
+ *remap_offset = start - ioaddr;
+#endif
pcihp->iot = iot;
pcihp->ioh = ioh;
pcihp->addr = ioaddr;
pcihp->size = size;
pcihp->flags = flags;
+ pcihp->ihandle = (void *) remap_offset;
return 0;
}
@@ -2170,6 +2179,9 @@
bus_space_handle_t ioh = pcihp->ioh;
bus_size_t size = pcihp->size;
+ if (pcihp->ihandle)
+ free(pcihp->ihandle, M_DEVBUF);
+
#if rbus
struct pccbb_softc *sc =
(struct pccbb_softc *)((struct pcic_handle *)pch)->ph_parent;
@@ -2209,6 +2221,7 @@
struct pcic_handle *ph = (struct pcic_handle *)pch;
bus_addr_t ioaddr = pcihp->addr + offset;
int i, win;
+ long *remap_offset = (long *) pcihp->ihandle;
#if defined CBB_DEBUG
static const char *width_names[] = { "dynamic", "io8", "io16" };
#endif
@@ -2253,6 +2266,7 @@
ph->io[win].addr = ioaddr;
ph->io[win].size = size;
ph->io[win].width = width;
+ ph->io[win].offset = *remap_offset;
/* actual dirty register-value changing in the function below. */
pccbb_pcmcia_do_io_map(ph, win);
@@ -2282,6 +2296,8 @@
#define PCIC_SIA_START_HIGH 1
#define PCIC_SIA_STOP_LOW 2
#define PCIC_SIA_STOP_HIGH 3
+#define PCIC_SIA_OFFSET_LOW 0x2e
+#define PCIC_SIA_OFFSET_HIGH 0x2f
int regbase_win = 0x8 + win * 0x04;
u_int8_t ioctl, enable;
@@ -2300,6 +2316,11 @@
Pcic_write(ph, regbase_win + PCIC_SIA_STOP_HIGH,
((ph->io[win].addr + ph->io[win].size - 1) >> 8) & 0xff);
+ Pcic_write(ph, regbase_win + PCIC_SIA_OFFSET_LOW,
+ ph->io[win].offset & 0xff);
+ Pcic_write(ph, regbase_win + PCIC_SIA_OFFSET_HIGH,
+ (ph->io[win].offset >> 8) & 0xff);
+
ioctl = Pcic_read(ph, PCIC_IOCTL);
enable = Pcic_read(ph, PCIC_ADDRWIN_ENABLE);
switch (win) {
@@ -2333,6 +2354,7 @@
printf
(" start %02x %02x, stop %02x %02x, ioctl %02x enable %02x\n",
start_low, start_high, stop_low, stop_high, ioctl, enable);
+ printf("pccbb_pcmcia_do_io_map: PCIC_SIA_OFFSET = %lx\n", ph->io[win].offset);
}
#endif
}
Index: src/sys/dev/pci/pccbbvar.h
===================================================================
RCS file: /cvsroot/src/sys/dev/pci/pccbbvar.h,v
retrieving revision 1.25
diff -u -r1.25 pccbbvar.h
--- src/sys/dev/pci/pccbbvar.h 11 Dec 2005 12:22:50 -0000 1.25
+++ src/sys/dev/pci/pccbbvar.h 26 Feb 2006 20:14:48 -0000
@@ -85,6 +85,7 @@
struct {
bus_addr_t addr;
bus_size_t size;
+ long offset;
int width;
} io[PCIC_IO_WINS];
int ih_irq;
Index: src/sys/dev/ic/i82365var.h
===================================================================
RCS file: /cvsroot/src/sys/dev/ic/i82365var.h,v
retrieving revision 1.26
diff -u -r1.26 i82365var.h
--- src/sys/dev/ic/i82365var.h 16 Feb 2006 20:17:16 -0000 1.26
+++ src/sys/dev/ic/i82365var.h 26 Feb 2006 20:14:49 -0000
@@ -73,6 +73,7 @@
struct {
bus_addr_t addr;
bus_size_t size;
+ long offset;
int width;
} io[PCIC_IO_WINS];
int ih_irq;
>Audit-Trail:
From: Sverre Froyen <sverre@viewmark.com>
To: gnats-bugs@netbsd.org
Cc:
Subject: Re: kern/32938: Remapping of PCMCIA IO addresses by the controller
Date: Thu, 11 May 2006 12:47:21 -0600
Updated patch for src/sys/dev/pcmcia/pvmcia.c. The new patch contains
comments from "The PCMCIA Developer's Guide" about the correct interpretation
of the iomask, start, and length CIS parameters, new code matching the
comments, and a couple of bug fixes. Please verify that the removal of the
tests around line 715 does not cause compatibility problems (these tests were
removed in order to make the debug printout, just below the tests, appear
correct, it had no effect on the working of my modem card).
Index: src/sys/dev/pcmcia/pcmcia.c
===================================================================
RCS file: /cvsroot/src/sys/dev/pcmcia/pcmcia.c,v
retrieving revision 1.79
diff -u -r1.79 pcmcia.c
--- src/sys/dev/pcmcia/pcmcia.c 29 Mar 2006 06:00:46 -0000 1.79
+++ src/sys/dev/pcmcia/pcmcia.c 11 May 2006 17:53:24 -0000
@@ -548,6 +548,9 @@
}
}
+ DPRINTF(("%s: pf_mfc_iobase=0x%lx pf_mfc_iomax=0x%lx\n",
+ sc->dev.dv_xname, pf->pf_mfc_iobase, pf->pf_mfc_iomax));
+
if (pcmcia_mfc(sc) || 1) {
pcmcia_ccr_write(pf, PCMCIA_CCR_IOBASE0,
(pf->pf_mfc_iobase >> 0) & 0xff);
@@ -711,12 +714,8 @@
if (pf->pf_mfc_iobase == 0) {
pf->pf_mfc_iobase = iobase;
pf->pf_mfc_iomax = iomax;
- } else {
- if (iobase < pf->pf_mfc_iobase)
- pf->pf_mfc_iobase = iobase;
- if (iomax > pf->pf_mfc_iomax)
- pf->pf_mfc_iomax = iomax;
}
+
DPRINTF(("function iobase %lx iomax %lx\n",
(long)pf->pf_mfc_iobase, (long)pf->pf_mfc_iomax));
}
@@ -780,26 +779,87 @@
int error = 0;
int n, m;
+ pf->pf_mfc_iobase = 0;
+ pf->pf_mfc_iomax = 0;
+
+ /*
+ * From Michael T. Mori & W. Dean Welder, The PCMCIA Developer's
+ * Guide, 3rd Edition, Sycard Tech. (1999), ISBN 0-9640342-2-0,
+ * p. 111. (With iomask the same as IOAddrLines.)
+ *
+ * case iomask start length description
+ * A 0 0 0 invalid
+ * B 0 0 d length d at any d boundary +
+ * C 0 c 0 invalid
+ * D 0 c d length d at c
+ * E b 0 0 length 2^b at any 2^b boundary
+ * F b 0 d length d at any 2^b boundary
+ * G b c 0 length 2^b at c (makes no sense ++)
+ * H b c d length d at c
+ *
+ * This table ignores the book's distinction between
+ * "Not Specified" and 0.
+ * + rounded up to nearest power of two.
+ * ++ as the upper end of the address range would not fit
+ * within iomask.
+ */
+
for (n = 0; n < cfe->num_iospace; n++) {
bus_addr_t start = cfe->iospace[n].start;
bus_size_t length = cfe->iospace[n].length;
- bus_size_t align = cfe->iomask ? (1 << cfe->iomask) :
- length;
- bus_size_t skew = start & (align - 1);
-
- if ((start - skew) == 0 && align < 0x400) {
- if (skew)
- printf("Drats! I need a skew!\n");
- start = 0;
+ bus_size_t align = 1;
+
+ /* D and H need no processing */
+
+ if (start == 0) {
+ if (cfe->iomask > 0) {
+ /* E or F */
+ align = (1 << cfe->iomask);
+ if (length == 0)
+ length = align; /* E */
+ }
+ else if (length > 0) {
+ /* B */
+ /* align equals 1 at this point */
+ /* XXX fix possible infinite loop */
+ while (align < length)
+ align <<= 1;
+ }
+ }
+
+ DPRINTF(("pcmcia_config_alloc: io %d start=%lx length=%lx align=%lx\n",
+ n, (long)start, (long)length, (long)align));
+
+ if (length == 0) {
+ /* A, C or G -- invalid */
+ error = 1;
+ break;
}
- DPRINTF(("pcmcia_config_alloc: io %d start=%lx length=%lx align=%lx
skew=%lx\n",
- n, (long)start, (long)length, (long)align, (long)skew));
+ /* Sanity check: start + length - 1 should fit within iomask */
+ if (cfe->iomask > 0 && start + length > 1 << cfe->iomask) {
+ error = 1;
+ break;
+ }
error = pcmcia_io_alloc(pf, start, length, align,
&cfe->iospace[n].handle);
if (error)
break;
+ /*
+ * Set pf_mfc_iobase equal to start. This flags the case
+ * where the card specifies a fixed IO base address and the
+ * controller remaps that address.
+ * XXX Need to rethink this in the case of multiple IO
+ * spaces per function. Fail for now.
+ */
+ if (pf->pf_mfc_iobase != 0) {
+ DPRINTF(("pcmcia_config_alloc: cannot (yet) remap multiple IO spaces\n"));
+ error = 1;
+ break;
+ }
+ pf->pf_mfc_iobase = start;
+ pf->pf_mfc_iomax = start + length - 1;
}
if (n < cfe->num_iospace) {
for (m = 0; m < n; m++)
From: "Charles M. Hannum" <mycroft@MIT.EDU>
To: gnats-bugs@NetBSD.org
Cc: kern-bug-people@NetBSD.org, gnats-admin@NetBSD.org,
netbsd-bugs@NetBSD.org, sverre@viewmark.com
Subject: Re: kern/32938: Remapping of PCMCIA IO addresses by the controller
Date: Thu, 11 May 2006 15:36:54 -0400
On Thu, May 11, 2006 at 06:50:02PM +0000, Sverre Froyen wrote:
> Updated patch for src/sys/dev/pcmcia/pvmcia.c. The new patch contains
> comments from "The PCMCIA Developer's Guide" about the correct interpretation
> of the iomask, start, and length CIS parameters, new code matching the
> comments, and a couple of bug fixes. Please verify that the removal of the
> tests around line 715 does not cause compatibility problems (these tests were
> removed in order to make the debug printout, just below the tests, appear
> correct, it had no effect on the working of my modem card).
I am not even vaguely convinced that the book is correct or complete.
The definitive reference is, of course, the PCMCIA spec, and it is not
written this way at all. Also, implementing what happens to work with
your modem card is totally inadequate -- I have at least 50 unique
devices that I tested when writing the current code.
There are three potential problems with the existing code:
1) It doesn't round "length" up to a power of two when it uses it for
alignment. This is a bug. Most cards specify the length as a power
of 2, but I've seen one counterexample to this.
2) It doesn't deal well with a length of 0. I don't recall this being
allowed by the spec, nor did it occur in the CIS of any of the 50+
cards that I tested, but who knows? At the very least it should be
robust enough to not crash in bus_space_alloc() or something.
3) The handling of "skew". This occurs because IOAddrLines actually
tells you how many address lines the device watches, and if it's less
than 16, there may be multiple valid aliases -- e.g. a device with
IOAddrLines=10 and a base address specified as 0x3f8 could be mapped
at 0x3f8, 0x7f8, 0xbf8, 0xff8, 0x13f8, etc. Unfortunately, there is
currently no way to pass this information on to bus_space_alloc(). I
inquired about this and suggested adding a "skew" parameter to the
interface (probably on tech-kern) at the time, but nobody cared, so
it was left open.
It's possible that the only values of IOAddrLines that exist in the
wild are 16, 10, 0 and log2(length), all of which are easy to special
case, but special cases seem like a definitely wrong answer.
Also:
> + * Set pf_mfc_iobase equal to start. This flags the case
> + * where the card specifies a fixed IO base address and the
> + * controller remaps that address.
> + * XXX Need to rethink this in the case of multiple IO
> + * spaces per function. Fail for now.
This is not acceptable. There are many such cards, including many
CF/IDE/ATA/flash devices. You cannot ignore them.
From: Sverre Froyen <sverre@viewmark.com>
To: gnats-bugs@netbsd.org
Cc:
Subject: Re: kern/32938: Remapping of PCMCIA IO addresses by the controller
Date: Sun, 29 Oct 2006 12:14:21 -0700
Hi,
Here's my current set of patches. New since last time is
1) allowing more than one IO address space per function and
2) passing back the new flag PCMCIA_IO_REMAPPED to indicate that the IO space
has been remapped. The flag is then used in pcmcia.c when setting addresses.
The flag provides a foolproof way for the various controllers to tell
pcmcia.c whether or not an address space has been remapped.
I believe this patch might help with PR32328 as well.
Sverre
Index: src/sys/dev/ic/i82365var.h
===================================================================
RCS file: /cvsroot/src/sys/dev/ic/i82365var.h,v
retrieving revision 1.26
diff -u -r1.26 i82365var.h
--- src/sys/dev/ic/i82365var.h 16 Feb 2006 20:17:16 -0000 1.26
+++ src/sys/dev/ic/i82365var.h 29 Oct 2006 18:55:43 -0000
@@ -73,6 +73,7 @@
struct {
bus_addr_t addr;
bus_size_t size;
+ long offset;
int width;
} io[PCIC_IO_WINS];
int ih_irq;
Index: src/sys/dev/pci/pccbb.c
===================================================================
RCS file: /cvsroot/src/sys/dev/pci/pccbb.c,v
retrieving revision 1.136
diff -u -r1.136 pccbb.c
--- src/sys/dev/pci/pccbb.c 24 Oct 2006 14:16:39 -0000 1.136
+++ src/sys/dev/pci/pccbb.c 29 Oct 2006 18:55:48 -0000
@@ -2067,40 +2067,26 @@
bus_space_tag_t iot;
bus_space_handle_t ioh;
bus_addr_t mask;
+ long *remap_offset;
#if rbus
rbus_tag_t rb;
-#endif
- if (align == 0) {
- align = size; /* XXX: funny??? */
- }
- if (start != 0) {
- /* XXX: assume all card decode lower 10 bits by its hardware */
- mask = 0x3ff;
- /* enforce to use only masked address */
- start &= mask;
- } else {
- /*
- * calculate mask:
- * 1. get the most significant bit of size (call it msb).
- * 2. compare msb with the value of size.
- * 3. if size is larger, shift msb left once.
- * 4. obtain mask value to decrement msb.
- */
- bus_size_t size_tmp = size;
- int shifts = 0;
+ /*
+ * Set minimum alignment and mask (OK since IO
+ * space will be remapped by the cardbus bridge).
+ * XXX I think that align = 1 would be OK too except for:
+ * XXX size == 1 => mask = 0 which, in rbus, results
+ * in a full decode. Sounds incorrect!
+ */
- mask = 1;
- while (size_tmp) {
- ++shifts;
- size_tmp >>= 1;
- }
- mask = (1 << shifts);
- if (mask < size) {
- mask <<= 1;
- }
- --mask;
- }
+ align = 1;
+ while (align < size)
+ align <<= 1;
+ mask = align - 1;
+#else
+ if (align == 0)
+ align = size; /* XXX: funny???, still needed? */
+#endif
/*
* Allocate some arbitrary I/O space.
@@ -2116,6 +2102,7 @@
DPRINTF(("pccbb_pcmcia_io_alloc alloc port 0x%lx+0x%lx\n",
(u_long) ioaddr, (u_long) size));
#else
+
if (start) {
ioaddr = start;
if (bus_space_map(iot, start, size, 0, &ioh)) {
@@ -2135,12 +2122,21 @@
(u_long) ioaddr, (u_long) size));
}
#endif
+ remap_offset = (long *) malloc(sizeof(long), M_DEVBUF, M_WAITOK);
+ if (! remap_offset)
+ return 1;
+ *remap_offset = 0;
+#if rbus
+ *remap_offset = start - ioaddr;
+ flags |= PCMCIA_IO_REMAPPED;
+#endif
pcihp->iot = iot;
pcihp->ioh = ioh;
pcihp->addr = ioaddr;
pcihp->size = size;
pcihp->flags = flags;
+ pcihp->ihandle = (void *) remap_offset;
return 0;
}
@@ -2164,6 +2160,9 @@
bus_space_handle_t ioh = pcihp->ioh;
bus_size_t size = pcihp->size;
+ if (pcihp->ihandle)
+ free(pcihp->ihandle, M_DEVBUF);
+
#if rbus
struct pccbb_softc *sc =
(struct pccbb_softc *)((struct pcic_handle *)pch)->ph_parent;
@@ -2203,6 +2202,7 @@
struct pcic_handle *ph = (struct pcic_handle *)pch;
bus_addr_t ioaddr = pcihp->addr + offset;
int i, win;
+ long *remap_offset = (long *) pcihp->ihandle;
#if defined CBB_DEBUG
static const char *width_names[] = { "dynamic", "io8", "io16" };
#endif
@@ -2247,6 +2247,7 @@
ph->io[win].addr = ioaddr;
ph->io[win].size = size;
ph->io[win].width = width;
+ ph->io[win].offset = *remap_offset;
/* actual dirty register-value changing in the function below. */
pccbb_pcmcia_do_io_map(ph, win);
@@ -2276,6 +2277,8 @@
#define PCIC_SIA_START_HIGH 1
#define PCIC_SIA_STOP_LOW 2
#define PCIC_SIA_STOP_HIGH 3
+#define PCIC_SIA_OFFSET_LOW 0x2e
+#define PCIC_SIA_OFFSET_HIGH 0x2f
int regbase_win = 0x8 + win * 0x04;
u_int8_t ioctl, enable;
@@ -2294,6 +2297,11 @@
Pcic_write(ph, regbase_win + PCIC_SIA_STOP_HIGH,
((ph->io[win].addr + ph->io[win].size - 1) >> 8) & 0xff);
+ Pcic_write(ph, regbase_win + PCIC_SIA_OFFSET_LOW,
+ ph->io[win].offset & 0xff);
+ Pcic_write(ph, regbase_win + PCIC_SIA_OFFSET_HIGH,
+ (ph->io[win].offset >> 8) & 0xff);
+
ioctl = Pcic_read(ph, PCIC_IOCTL);
enable = Pcic_read(ph, PCIC_ADDRWIN_ENABLE);
switch (win) {
@@ -2327,6 +2335,7 @@
printf("pccbb_pcmcia_do_io_map start %02x %02x, "
"stop %02x %02x, ioctl %02x enable %02x\n",
start_low, start_high, stop_low, stop_high, ioctl, enable);
+ printf("pccbb_pcmcia_do_io_map: PCIC_SIA_OFFSET = %lx\n",
ph->io[win].offset);
}
#endif
}
Index: src/sys/dev/pci/pccbbvar.h
===================================================================
RCS file: /cvsroot/src/sys/dev/pci/pccbbvar.h,v
retrieving revision 1.26
diff -u -r1.26 pccbbvar.h
--- src/sys/dev/pci/pccbbvar.h 17 Jun 2006 17:06:52 -0000 1.26
+++ src/sys/dev/pci/pccbbvar.h 29 Oct 2006 18:55:49 -0000
@@ -85,6 +85,7 @@
struct {
bus_addr_t addr;
bus_size_t size;
+ long offset;
int width;
} io[PCIC_IO_WINS];
int ih_irq;
Index: src/sys/dev/pcmcia/pcmcia.c
===================================================================
RCS file: /cvsroot/src/sys/dev/pcmcia/pcmcia.c,v
retrieving revision 1.81
diff -u -r1.81 pcmcia.c
--- src/sys/dev/pcmcia/pcmcia.c 24 Oct 2006 20:54:10 -0000 1.81
+++ src/sys/dev/pcmcia/pcmcia.c 29 Oct 2006 18:55:49 -0000
@@ -543,6 +543,9 @@
}
}
+ DPRINTF(("%s: pf_mfc_iobase=0x%lx pf_mfc_iomax=0x%lx\n",
+ sc->dev.dv_xname, pf->pf_mfc_iobase, pf->pf_mfc_iomax));
+
if (pcmcia_mfc(sc) || 1) {
pcmcia_ccr_write(pf, PCMCIA_CCR_IOBASE0,
(pf->pf_mfc_iobase >> 0) & 0xff);
@@ -691,27 +694,11 @@
if (error)
return (error);
- /*
- * XXX in the multifunction multi-iospace-per-function case, this
- * needs to cooperate with io_alloc to make sure that the spaces
- * don't overlap, and that the ccr's are set correctly
- */
-
if (pcmcia_mfc(sc) || 1) {
- bus_addr_t iobase = pcihp->addr;
- bus_addr_t iomax = pcihp->addr + pcihp->size - 1;
- DPRINTF(("window iobase %lx iomax %lx\n", (long)iobase,
- (long)iomax));
- if (pf->pf_mfc_iobase == 0) {
- pf->pf_mfc_iobase = iobase;
- pf->pf_mfc_iomax = iomax;
- } else {
- if (iobase < pf->pf_mfc_iobase)
- pf->pf_mfc_iobase = iobase;
- if (iomax > pf->pf_mfc_iomax)
- pf->pf_mfc_iomax = iomax;
- }
+ DPRINTF(("window iobase %lx iomax %lx\n", (long)pcihp->addr,
+ (long)(pcihp->addr + pcihp->size - 1)));
+
DPRINTF(("function iobase %lx iomax %lx\n",
(long)pf->pf_mfc_iobase, (long)pf->pf_mfc_iomax));
}
@@ -774,28 +761,118 @@
{
int error = 0;
int n, m;
+ bus_addr_t iobase;
+ bus_addr_t iomax;
+ bus_size_t align;
+
+ pf->pf_mfc_iobase = 1 << 26; /* Largest possible value(?) */
+ pf->pf_mfc_iomax = 0;
+
+ /*
+ * From Michael T. Mori & W. Dean Welder, The PCMCIA Developer's
+ * Guide, 3rd Edition, Sycard Tech. (1999), ISBN 0-9640342-2-0,
+ * p. 111. (With iomask the same as IOAddrLines.)
+ *
+ * case iomask start length description
+ * A 0 0 0 invalid
+ * B 0 0 d length d at any d boundary +
+ * C 0 c 0 invalid
+ * D 0 c d length d at c ++
+ * E b 0 0 length 2^b at any 2^b boundary +++
+ * F b 0 d length d at any 2^b boundary
+ * G b c 0 length 2^b at c (makes no sense ++++)
+ * H b c d length d at c ++
+ *
+ * This table ignores the book's distinction between
+ * "Not Specified" and 0.
+ *
+ * + rounded up to nearest power of two.
+ *
+ * ++ We will set align to d (rounded up to nearest power of two) [D]
+ * or 2^b [H] in order to be compatible with existing controller
+ * code. These cases call for adding a skew/phase parameter to the
+ * pcmcia_io_alloc interface [old code had skew = start & (align - 1)].
+ *
+ * +++ E cannot be valid for multiple IO spaces.
+ *
+ * ++++ as the upper end of the address range would not fit
+ * within iomask.
+ */
+
+ /* Determine alignment */
+
+ align = 1;
+ if (cfe->iomask > 0) {
+ /* E, F, G or H */
+ align <<= cfe->iomask;
+ }
+ else {
+ /* Determine the maximum alignment requirement */
+ for (n = 0; n < cfe->num_iospace; n++) {
+ bus_size_t length = cfe->iospace[n].length;
+ if (length > 0) {
+ /* B or D */
+ /* XXX fix possible infinite loop */
+ while (align < length)
+ align <<= 1;
+ }
+ }
+ }
for (n = 0; n < cfe->num_iospace; n++) {
bus_addr_t start = cfe->iospace[n].start;
bus_size_t length = cfe->iospace[n].length;
- bus_size_t align = cfe->iomask ? (1 << cfe->iomask) :
- length;
- bus_size_t skew = start & (align - 1);
-
- if ((start - skew) == 0 && align < 0x400) {
- if (skew)
- printf("Drats! I need a skew!\n");
- start = 0;
+
+ if (length == 0 && start == 0 && cfe->num_iospace == 1)
+ length = align; /* E (single IO space only) */
+
+ DPRINTF(("pcmcia_config_alloc: io %d start=%lx length=%lx align=%lx\n",
+ n, (long)start, (long)length, (long)align));
+
+ if (length == 0) {
+ /* A, C or G -- invalid */
+ error = 1;
+ break;
}
- DPRINTF(("pcmcia_config_alloc: io %d start=%lx length=%lx align=%lx
skew=%lx\n",
- n, (long)start, (long)length, (long)align, (long)skew));
+ /* Sanity check: start + length - 1 should fit within iomask */
+ if (cfe->iomask > 0 && start + length > 1 << cfe->iomask) {
+ error = 1;
+ break;
+ }
error = pcmcia_io_alloc(pf, start, length, align,
&cfe->iospace[n].handle);
if (error)
break;
+
+ /*
+ * Adjust pf->pf_mfc_iobase and pf->pf_mfc_iomax as needed.
+ * If the controller has set the remapped flag, use the
+ * card addresses, otherwise, use the addresses returned by
+ * the controller.
+ *
+ * XXX in the multifunction multi-iospace-per-function case,
+ * this needs to cooperate with io_alloc to make sure that
+ * the spaces don't overlap, and that the ccr's are set
+ * correctly.
+ */
+
+ if (cfe->iospace[n].handle.flags & PCMCIA_IO_REMAPPED) {
+ iobase = start;
+ iomax = start + length - 1;
+ }
+ else {
+ iobase = cfe->iospace[n].handle.addr;
+ iomax = cfe->iospace[n].handle.addr +
+ cfe->iospace[n].handle.size - 1;
+ }
+ if (pf->pf_mfc_iobase > iobase)
+ pf->pf_mfc_iobase = iobase;
+ if (pf->pf_mfc_iomax < iomax)
+ pf->pf_mfc_iomax = iomax;
}
+
if (n < cfe->num_iospace) {
for (m = 0; m < n; m++)
pcmcia_io_free(pf, &cfe->iospace[m].handle);
Index: src/sys/dev/pcmcia/pcmciavar.h
===================================================================
RCS file: /cvsroot/src/sys/dev/pcmcia/pcmciavar.h,v
retrieving revision 1.32
diff -u -r1.32 pcmciavar.h
--- src/sys/dev/pcmcia/pcmciavar.h 23 Feb 2006 03:18:42 -0000 1.32
+++ src/sys/dev/pcmcia/pcmciavar.h 29 Oct 2006 18:55:50 -0000
@@ -49,6 +49,8 @@
};
#define PCMCIA_IO_ALLOCATED 0x01 /* i/o space was allocated */
+#define PCMCIA_IO_REMAPPED 0x02 /* i/o space has been remapped, */
+ /* card uses its own base address */
/*
* Contains information about allocated memory space.
(Contact us)
$NetBSD: query-full-pr,v 1.39 2013/11/01 18:47:49 spz Exp $
$NetBSD: gnats_config.sh,v 1.8 2006/05/07 09:23:38 tsutsui Exp $
Copyright © 1994-2007
The NetBSD Foundation, Inc. ALL RIGHTS RESERVED.
Home