Skip to contentSkip to conversation

pcmcia: add struct pcmcia_device

PATCH In review Dominik Brodowski 2 participants 2 messages Mar 29, 2003 – Mar 30, 2003 Pcmcia

lore
Share & tools
lore

Dominik Brodowski

author
#0 lore raw

PATCH 1/1 pcmcia: add struct pcmcia_device

This patch adds a struct pcmcia_device, and fills it with content for every function of every pcmcia card inserted into the socket.

Unfortunately, information about the card can only be detected when "cardmgr"[*] has told the pcmcia rsrc_mgr on what IO-ports and what IO-mem is available for usage by the pcmcia core or drivers. So the cards need to be re-scanned on completion of the pcmcia initialization.

Re-starting cardmgr will not re-enable the resources.

Dominik

 drivers/pcmcia/cistpl.c |    3
 drivers/pcmcia/ds.c     |  207 ++++++++++++++++++++++++++++++++++++++++++++++--
 include/pcmcia/ds.h     |   24 +++++
 3 files changed, 227 insertions(+), 7 deletions(-)
drivers/pcmcia/cistpl.c +2 -1
diff -ruN linux-original/drivers/pcmcia/cistpl.c linux/drivers/pcmcia/cistpl.c
@@ -1392,6 +1392,7 @@
     ret = pcmcia_parse_tuple(handle, &tuple, parse);
     return ret;
 }
+EXPORT_SYMBOL(read_tuple);
 
 /*======================================================================
 

@@ -1451,4 +1452,4 @@
 
     return CS_SUCCESS;
 }
-
+EXPORT_SYMBOL(pcmcia_validate_cis);
drivers/pcmcia/ds.c +201 -6

291 lines not on this page · raw message

include/pcmcia/ds.h +24 -0
diff -ruN linux-original/include/pcmcia/ds.h linux/include/pcmcia/ds.h
@@ -165,5 +165,29 @@
 /* error reporting */
 void cs_error(client_handle_t handle, int func, int ret);
 
+struct pcmcia_bus_socket;
+
+struct pcmcia_device {
+	unsigned int			socket_no; /* deprecated */
+	struct pcmcia_bus_socket	*bus_sock;
+	struct list_head		socket_device_list;
+	unsigned int			func;
+
+	/* cis information */
+	unsigned int			has_cis;
+	cistpl_vers_1_t			vers1;
+
+	unsigned int			has_manfid;
+	cistpl_manfid_t			manfid;
+
+	unsigned int			has_funcid;
+	cistpl_funcid_t			funcid;
+
+	struct device			dev;
+};
+
+#define to_pcmcia_dev(n) container_of(n, struct pcmcia_device, dev)
+#define to_pcmcia_drv(n) container_of(n, struct pcmcia_driver, drv)
+
 #endif /* __KERNEL__ */
 #endif /* _LINUX_DS_H */
#1 lore raw

Ok, some comments on this patch.

Quoted 38 lines · Dominik Brodowski

diff -ruN linux-original/drivers/pcmcia/cistpl.c linux/drivers/pcmcia/cistpl.c

drivers/pcmcia/cistpl.c +2 -1
@@ -1392,6 +1392,7 @@
     ret = pcmcia_parse_tuple(handle, &tuple, parse);
     return ret;
 }
+EXPORT_SYMBOL(read_tuple);
 
 /*======================================================================
 

@@ -1451,4 +1452,4 @@
 
     return CS_SUCCESS;
 }
-
+EXPORT_SYMBOL(pcmcia_validate_cis);

diff -ruN linux-original/drivers/pcmcia/ds.c linux/drivers/pcmcia/ds.c

drivers/pcmcia/ds.c +5 -0
@@ -108,11 +108,16 @@
 	struct device		*socket_dev;
 	struct list_head	socket_list;
 	unsigned int		socket_no; /* deprecated */
+	/* the PCMCIA devices connected to this socket (normally one, more
+	 * for multifunction devices: */
+	struct list_head	devices_list;
+	struct semaphore	devices_list_sem;
 };
 
 #define SOCKET_PRESENT		0x01
 #define SOCKET_BUSY		0x02
 #define SOCKET_REMOVAL_PENDING	0x10
+#define SOCKET_ADDING_PENDING	0x20
 
 /*====================================================================*/

drivers/pcmcia/ds.c line 134

Quoted 9 lines
drivers/pcmcia/ds.c +6 -0
@@ -121,6 +126,12 @@
 
 static int major_dev = -1;
 
+/* this is set to 1 when the iomem/ioport configuration of rsrc_mgr.c
+ * is completed -- only then it is possible to decode the name, manfid,
+ * prodid, etc. of the pcmcia card.
+ */
+static int resources_available = 0;
+

This isn't actually such a clear-cut situation. Statically mapped sockets like SA11x0-based stuff are able to access the card with no resources loaded - in fact, there are no resources to be loaded.

One of my patches (which was the troublesome pcmcia-10 patch from akpm's patch set) changes the way we deal with the CIS. For windowed sockets, when a card detected, we immediately try to claim a resource for the CIS window. If this fails, we whinge about "PCMCIA: insert failed" but continue anyway. We could fail at this point.

When resources are changed, I simulate a card removal + insertion request so that the CIS gets re-setup. The longer term solution, imho, is to change the socket setup/reset/unreset code in cs.c to be more of a state machine which could be re-started when resources become available.

Quoted 4 lines
drivers/pcmcia/ds.c +4 -0
+	/* first case: no resources are available, so we can't decode
+	 * any device information */
+	if (!resources_available) {
+		sprintf (p_dev->dev.bus_id, "pcmcia%d:%d.?",

drivers/pcmcia/ds.c line 304

Quoted 10 lines
drivers/pcmcia/ds.c +10 -0
+			 s->socket_dev->class_num, s->socket_no);
+		sprintf (p_dev->dev.name, "unknown");
+		ret = device_register(&p_dev->dev);
+		if (ret) {
+			kfree(p_dev);
+			goto out;
+		}
+		list_add(&p_dev->socket_device_list, &s->devices_list);
+		goto out;
+	}

I'm not completely sure we want to add a device in this case - if we don't have any resources, there isn't much point in making the device available to drivers.

drivers/pcmcia/ds.c line 312

Quoted 5 lines
drivers/pcmcia/ds.c +5 -0
+	if (!func) {
+		cisinfo_t cisinfo;
+		ret = pcmcia_validate_cis(s->handle, &cisinfo);
+		if (ret || !cisinfo.Chains) {
+			printk(KERN_INFO "pcmcia: inserted card has invalid CIS.\n");

I have a memory-type pcmcia card here, and I believe it doesn't have any CIS what so ever, so this would fail.

Signature · 2 lines
Russell King ([email protected])                The developer of ARM Linux
             http://www.arm.linux.org.uk/personal/aboutme.html