[Cbe-oss-dev] [RFC] More cell interrupt updates
Benjamin Herrenschmidt
benh at kernel.crashing.org
Mon Aug 21 18:12:39 EST 2006
This is a work-in-progress patch. This changes (again) the numbering of
iic interrupts so that:
- NodeID is back in the interrupt number (only one IRQ host, or domain,
is created). This
allows interrupts from sources on another node to be routed non-locally
and things will still work (that wouldn't work with my previous code).
This will allow possibly one day to fix maxcpus=1 or 2 and still get
interrupts from devices on BE 1. (A bit more fixing is needed for that)
- Added handling of the IO exceptions interrupts (badly named, but I
re-used the name used by STI in their initial drop, could be renamed to
the "internal interrupts" maybe ?). Those are the interrupts exposed by
IIC_ISR and IIC_IRR, such as the IOC translation exception, livelock,
performance monitor, etc... Those get their special numbers in the IRQ
number space and are internally implemented as a cascade on unit 0xe,
class 1 of each node. This is untested at this point, it would be good
to adapt & test the oprofile code against that.
We still need to cleanup assumptions here or there about HW CPU == linux
CPU number, or node <-> CPU mapping, etc... I suspect we really need to
put together some global layer to handle these generically, at least for
the powerpc architecture, but I didn't have time to work on that. For
now, the hacks in the iic code should work for us.
Ben.
Index: linux-work/arch/powerpc/platforms/cell/interrupt.c
===================================================================
--- linux-work.orig/arch/powerpc/platforms/cell/interrupt.c 2006-08-17 09:26:19.000000000 +1000
+++ linux-work/arch/powerpc/platforms/cell/interrupt.c 2006-08-21 18:03:57.000000000 +1000
@@ -21,6 +21,12 @@
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ *
+ * TODO:
+ * - Fix various assumptions related to HW CPU numbers vs. linux CPU numbers
+ * vs node numbers in the setup code
+ * - Implement proper handling of maxcpus=1/2 (that is, routing of irqs from
+ * a non-active node to the active node)
*/
#include <linux/interrupt.h>
@@ -44,24 +50,25 @@ struct iic {
u8 target_id;
u8 eoi_stack[16];
int eoi_ptr;
- struct irq_host *host;
+ struct device_node *node;
};
static DEFINE_PER_CPU(struct iic, iic);
#define IIC_NODE_COUNT 2
-static struct irq_host *iic_hosts[IIC_NODE_COUNT];
+static struct irq_host *iic_host;
/* Convert between "pending" bits and hw irq number */
static irq_hw_number_t iic_pending_to_hwnum(struct cbe_iic_pending_bits bits)
{
unsigned char unit = bits.source & 0xf;
+ unsigned char node = bits.source >> 4;
+ unsigned char class = bits.class & 3;
+ /* Decode IPIs */
if (bits.flags & CBE_IIC_IRQ_IPI)
- return IIC_IRQ_IPI0 | (bits.prio >> 4);
- else if (bits.class <= 3)
- return (bits.class << 4) | unit;
+ return IIC_IRQ_TYPE_IPI | (bits.prio >> 4);
else
- return IIC_IRQ_INVALID;
+ return (node << IIC_IRQ_NODE_SHIFT) | (class << 4) | unit;
}
static void iic_mask(unsigned int irq)
@@ -86,6 +93,44 @@ static struct irq_chip iic_chip = {
.eoi = iic_eoi,
};
+
+static void iic_ioexc_eoi(unsigned int irq)
+{
+}
+
+static void iic_ioexc_cascade(unsigned int irq, struct irq_desc *desc,
+ struct pt_regs *regs)
+{
+ struct cbe_iic_regs *node_iic = desc->handler_data;
+ unsigned int base = (irq & 0xffffff00) | IIC_IRQ_TYPE_IOEXC;
+ unsigned int bits;
+ int cascade;
+
+ for (;;) {
+ bits = in_be64(&node_iic->iic_is);
+ if (bits == 0)
+ break;
+ for (cascade = 63; cascade >= 0; cascade--)
+ if (bits & (0x8000000000000000UL >> cascade)) {
+ unsigned int cirq =
+ irq_linear_revmap(iic_host,
+ base | cascade);
+ if (cirq != NO_IRQ)
+ generic_handle_irq(cirq, regs);
+ }
+ out_be64(&node_iic->iic_is, bits);
+ }
+ desc->chip->eoi(irq);
+}
+
+
+static struct irq_chip iic_ioexc_chip = {
+ .typename = " CELL-IOEX",
+ .mask = iic_mask,
+ .unmask = iic_unmask,
+ .eoi = iic_ioexc_eoi,
+};
+
/* Get an IRQ number from the pending state register of the IIC */
static unsigned int iic_get_irq(struct pt_regs *regs)
{
@@ -98,7 +143,7 @@ static unsigned int iic_get_irq(struct p
iic->eoi_stack[++iic->eoi_ptr] = pending.prio;
BUG_ON(iic->eoi_ptr > 15);
if (pending.flags & CBE_IIC_IRQ_VALID)
- return irq_linear_revmap(iic->host,
+ return irq_linear_revmap(iic_host,
iic_pending_to_hwnum(pending));
return NO_IRQ;
}
@@ -108,12 +153,7 @@ static unsigned int iic_get_irq(struct p
/* Use the highest interrupt priorities for IPI */
static inline int iic_ipi_to_irq(int ipi)
{
- return IIC_IRQ_IPI0 + IIC_NUM_IPIS - 1 - ipi;
-}
-
-static inline int iic_irq_to_ipi(int irq)
-{
- return IIC_NUM_IPIS - 1 - (irq - IIC_IRQ_IPI0);
+ return IIC_IRQ_TYPE_IPI + 0xf - ipi;
}
void iic_setup_cpu(void)
@@ -123,7 +163,7 @@ void iic_setup_cpu(void)
void iic_cause_IPI(int cpu, int mesg)
{
- out_be64(&per_cpu(iic, cpu).regs->generate, (IIC_NUM_IPIS - 1 - mesg) << 4);
+ out_be64(&per_cpu(iic, cpu).regs->generate, (0xf - mesg) << 4);
}
u8 iic_get_target_id(int cpu)
@@ -134,9 +174,7 @@ EXPORT_SYMBOL_GPL(iic_get_target_id);
struct irq_host *iic_get_irq_host(int node)
{
- if (node < 0 || node >= IIC_NODE_COUNT)
- return NULL;
- return iic_hosts[node];
+ return iic_host;
}
EXPORT_SYMBOL_GPL(iic_get_irq_host);
@@ -149,34 +187,20 @@ static irqreturn_t iic_ipi_action(int ir
return IRQ_HANDLED;
}
-
static void iic_request_ipi(int ipi, const char *name)
{
- int node, virq;
+ int virq;
- for (node = 0; node < IIC_NODE_COUNT; node++) {
- char *rname;
- if (iic_hosts[node] == NULL)
- continue;
- virq = irq_create_mapping(iic_hosts[node],
- iic_ipi_to_irq(ipi));
- if (virq == NO_IRQ) {
- printk(KERN_ERR
- "iic: failed to map IPI %s on node %d\n",
- name, node);
- continue;
- }
- rname = kzalloc(strlen(name) + 16, GFP_KERNEL);
- if (rname)
- sprintf(rname, "%s node %d", name, node);
- else
- rname = (char *)name;
- if (request_irq(virq, iic_ipi_action, IRQF_DISABLED,
- rname, (void *)(long)ipi))
- printk(KERN_ERR
- "iic: failed to request IPI %s on node %d\n",
- name, node);
+ virq = irq_create_mapping(iic_host, iic_ipi_to_irq(ipi));
+ if (virq == NO_IRQ) {
+ printk(KERN_ERR
+ "iic: failed to map IPI %s\n", name);
+ return;
}
+ if (request_irq(virq, iic_ipi_action, IRQF_DISABLED, name,
+ (void *)(long)ipi))
+ printk(KERN_ERR
+ "iic: failed to request IPI %s\n", name);
}
void iic_request_IPIs(void)
@@ -193,16 +217,23 @@ void iic_request_IPIs(void)
static int iic_host_match(struct irq_host *h, struct device_node *node)
{
- return h->host_data != NULL && node == h->host_data;
+ return device_is_compatible(node,
+ "IBM,CBEA-Internal-Interrupt-Controller");
}
static int iic_host_map(struct irq_host *h, unsigned int virq,
irq_hw_number_t hw)
{
- if (hw < IIC_IRQ_IPI0)
- set_irq_chip_and_handler(virq, &iic_chip, handle_fasteoi_irq);
- else
+ switch (hw & IIC_IRQ_TYPE_MASK) {
+ case IIC_IRQ_TYPE_IPI:
set_irq_chip_and_handler(virq, &iic_chip, handle_percpu_irq);
+ break;
+ case IIC_IRQ_TYPE_IOEXC:
+ set_irq_chip_and_handler(virq, &iic_ioexc_chip, handle_fasteoi_irq);
+ break;
+ default:
+ set_irq_chip_and_handler(virq, &iic_chip, handle_fasteoi_irq);
+ }
return 0;
}
@@ -225,7 +256,7 @@ static struct irq_host_ops iic_host_ops
};
static void __init init_one_iic(unsigned int hw_cpu, unsigned long addr,
- struct irq_host *host)
+ struct device_node *node)
{
/* XXX FIXME: should locate the linux CPU number from the HW cpu
* number properly. We are lucky for now
@@ -237,19 +268,19 @@ static void __init init_one_iic(unsigned
iic->target_id = ((hw_cpu & 2) << 3) | ((hw_cpu & 1) ? 0xf : 0xe);
iic->eoi_stack[0] = 0xff;
- iic->host = host;
+ iic->node = of_node_get(node);
out_be64(&iic->regs->prio, 0);
- printk(KERN_INFO "IIC for CPU %d at %lx mapped to %p, target id 0x%x\n",
- hw_cpu, addr, iic->regs, iic->target_id);
+ printk(KERN_INFO "IIC for CPU %d at %lx mapped to %p, target id 0x%x : %s\n",
+ hw_cpu, addr, iic->regs, iic->target_id, node->full_name);
}
static int __init setup_iic(void)
{
struct device_node *dn;
struct resource r0, r1;
- struct irq_host *host;
- int found = 0;
+ unsigned int node, cascade, found = 0;
+ struct cbe_iic_regs *node_iic;
u32 *np;
for (dn = NULL;
@@ -270,19 +301,27 @@ static int __init setup_iic(void)
of_node_put(dn);
return -ENODEV;
}
- host = NULL;
- if (found < IIC_NODE_COUNT) {
- host = irq_alloc_host(IRQ_HOST_MAP_LINEAR,
- IIC_SOURCE_COUNT,
- &iic_host_ops,
- IIC_IRQ_INVALID);
- iic_hosts[found] = host;
- BUG_ON(iic_hosts[found] == NULL);
- iic_hosts[found]->host_data = of_node_get(dn);
- found++;
- }
- init_one_iic(np[0], r0.start, host);
- init_one_iic(np[1], r1.start, host);
+ found++;
+ init_one_iic(np[0], r0.start, dn);
+ init_one_iic(np[1], r1.start, dn);
+
+ /* Setup cascade for IO exceptions. XXX cleanup tricks to get node
+ * vs CPU etc... Note that we configure the IIC_IRR here with a hard
+ * coded priority of 1. We might want to improve that later.
+ */
+ node = np[0] >> 1;
+ node_iic = cbe_get_cpu_iic_regs(np[0]);
+ cascade = node << IIC_IRQ_NODE_SHIFT;
+ cascade |= 1 << IIC_IRQ_CLASS_SHIFT;
+ cascade |= IIC_UNIT_IIC;
+ set_irq_data(cascade, node_iic);
+ set_irq_chained_handler(cascade , iic_ioexc_cascade);
+ out_be64(&node_iic->iic_ir,
+ (1 << 12) /* priority */ |
+ (node << 4) /* dest node */ |
+ IIC_UNIT_THREAD_0 /* route them to thread 0 */);
+ /* Flush pending (make sure it triggers if there is any pending */
+ out_be64(&node_iic->iic_is, 0xfffffffffffffffful);
}
if (found)
@@ -293,6 +332,12 @@ static int __init setup_iic(void)
void __init iic_init_IRQ(void)
{
+ /* Setup an irq host data structure */
+ iic_host = irq_alloc_host(IRQ_HOST_MAP_LINEAR, IIC_SOURCE_COUNT,
+ &iic_host_ops, IIC_IRQ_INVALID);
+ BUG_ON(iic_host == NULL);
+ irq_set_default_host(iic_host);
+
/* Discover and initialize iics */
if (setup_iic() < 0)
panic("IIC: Failed to initialize !\n");
Index: linux-work/arch/powerpc/platforms/cell/interrupt.h
===================================================================
--- linux-work.orig/arch/powerpc/platforms/cell/interrupt.h 2006-08-17 09:26:19.000000000 +1000
+++ linux-work/arch/powerpc/platforms/cell/interrupt.h 2006-08-21 18:04:08.000000000 +1000
@@ -2,48 +2,60 @@
#define ASM_CELL_PIC_H
#ifdef __KERNEL__
/*
- * Mapping of IIC pending bits into per-node
- * interrupt numbers.
+ * Mapping of IIC pending bits into per-node interrupt numbers.
*
- * IRQ FF CC SS PP FF CC SS PP Description
+ * Interrupt numbers are in the range 0...0x1ff where the top bit
+ * (0x100) represent the source node. Only 2 nodes are supported with
+ * the current code though it's trivial to extend that if necessary using
+ * higher level bits
*
- * 00-3f 80 02 +0 00 - 80 02 +0 3f South Bridge
- * 00-3f 80 02 +b 00 - 80 02 +b 3f South Bridge
- * 41-4a 80 00 +1 ** - 80 00 +a ** SPU Class 0
- * 51-5a 80 01 +1 ** - 80 01 +a ** SPU Class 1
- * 61-6a 80 02 +1 ** - 80 02 +a ** SPU Class 2
- * 70-7f C0 ** ** 00 - C0 ** ** 0f IPI
+ * The bottom 8 bits are split into 2 type bits and 6 data bits that
+ * depend on the type:
*
- * F flags
- * C class
- * S source
- * P Priority
- * + node number
- * * don't care
+ * 00 (0x00 | data) : normal interrupt. data is (class << 4) | source
+ * 01 (0x40 | data) : IO exception. data is the exception number as
+ * defined by bit numbers in IIC_SR
+ * 10 (0x80 | data) : IPI. data is the IPI number (obtained from the priority)
+ * and node is always 0 (IPIs are per-cpu, their source is
+ * not relevant)
+ * 11 (0xc0 | data) : reserved
*
- * A node consists of a Cell Broadband Engine and an optional
- * south bridge device providing a maximum of 64 IRQs.
- * The south bridge may be connected to either IOIF0
- * or IOIF1.
- * Each SPE is represented as three IRQ lines, one per
- * interrupt class.
- * 16 IRQ numbers are reserved for inter processor
- * interruptions, although these are only used in the
- * range of the first node.
+ * In addition, interrupt number 0x80000000 is defined as always invalid
+ * (that is the node field is expected to never extend to move than 23 bits)
*
- * This scheme needs 128 IRQ numbers per BIF node ID,
- * which means that with the total of 512 lines
- * available, we can have a maximum of four nodes.
*/
enum {
- IIC_IRQ_INVALID = 0xff,
- IIC_IRQ_MAX = 0x3f,
- IIC_IRQ_EXT_IOIF0 = 0x20,
- IIC_IRQ_EXT_IOIF1 = 0x2b,
- IIC_IRQ_IPI0 = 0x40,
- IIC_NUM_IPIS = 0x10, /* IRQs reserved for IPI */
- IIC_SOURCE_COUNT = 0x50,
+ IIC_IRQ_INVALID = 0x8000,
+ IIC_IRQ_NODE_MASK = 0x100,
+ IIC_IRQ_NODE_SHIFT = 8,
+ IIC_IRQ_MAX = 0x1ff,
+ IIC_IRQ_TYPE_MASK = 0xc0,
+ IIC_IRQ_TYPE_NORMAL = 0x00,
+ IIC_IRQ_TYPE_IOEXC = 0x40,
+ IIC_IRQ_TYPE_IPI = 0x80,
+ IIC_IRQ_CLASS_SHIFT = 4,
+ IIC_IRQ_EXT_IOIF0 = 0x20, /* Node 0 IOIF 0 */
+ IIC_IRQ_EXT_IOIF1 = 0x2b, /* Node 0 IOIF 1 */
+ IIC_SOURCE_COUNT = 0x200,
+
+ /* Here are defined the various source/dest units. Avoid using those
+ * definitions if you can, they are mostly here for reference
+ */
+ IIC_UNIT_SPU_0 = 0x4,
+ IIC_UNIT_SPU_1 = 0x7,
+ IIC_UNIT_SPU_2 = 0x3,
+ IIC_UNIT_SPU_3 = 0x8,
+ IIC_UNIT_SPU_4 = 0x2,
+ IIC_UNIT_SPU_5 = 0x9,
+ IIC_UNIT_SPU_6 = 0x1,
+ IIC_UNIT_SPU_7 = 0xa,
+ IIC_UNIT_IOC_0 = 0x0,
+ IIC_UNIT_IOC_1 = 0xb,
+ IIC_UNIT_THREAD_0 = 0xe,
+ IIC_UNIT_THREAD_1 = 0xf,
+ IIC_UNIT_IIC = 0xe /* special case for iic originated
+ * IO exceptions */
};
extern void iic_init_IRQ(void);
@@ -52,7 +64,6 @@ extern void iic_request_IPIs(void);
extern void iic_setup_cpu(void);
extern u8 iic_get_target_id(int cpu);
-extern struct irq_host *iic_get_irq_host(int node);
extern void spider_init_IRQ(void);
Index: linux-work/arch/powerpc/platforms/cell/spider-pic.c
===================================================================
--- linux-work.orig/arch/powerpc/platforms/cell/spider-pic.c 2006-08-17 09:26:19.000000000 +1000
+++ linux-work/arch/powerpc/platforms/cell/spider-pic.c 2006-08-21 17:42:32.000000000 +1000
@@ -243,7 +243,6 @@ static unsigned int __init spider_find_c
u32 *imap, *tmp;
int imaplen, intsize, unit;
struct device_node *iic;
- struct irq_host *iic_host;
#if 0 /* Enable that when we have a way to retreive the node as well */
/* First, we check wether we have a real "interrupts" in the device
@@ -289,11 +288,11 @@ static unsigned int __init spider_find_c
* the iic host from the iic OF node, but that way I'm still compatible
* with really really old old firmwares for which we don't have a node
*/
- iic_host = iic_get_irq_host(pic->node_id);
- if (iic_host == NULL)
- return NO_IRQ;
/* Manufacture an IIC interrupt number of class 2 */
- virq = irq_create_mapping(iic_host, 0x20 | unit);
+ virq = irq_create_mapping(NULL,
+ (pic->node_id << IIC_IRQ_NODE_SHIFT) |
+ (2 << IIC_IRQ_CLASS_SHIFT) |
+ unit);
if (virq == NO_IRQ)
printk(KERN_ERR "spider_pic: failed to map cascade !");
return virq;
Index: linux-work/arch/powerpc/platforms/cell/spu_base.c
===================================================================
--- linux-work.orig/arch/powerpc/platforms/cell/spu_base.c 2006-08-17 09:26:19.000000000 +1000
+++ linux-work/arch/powerpc/platforms/cell/spu_base.c 2006-08-21 17:38:37.000000000 +1000
@@ -568,24 +568,21 @@ static void spu_unmap(struct spu *spu)
/* This function shall be abstracted for HV platforms */
static int __init spu_map_interrupts(struct spu *spu, struct device_node *np)
{
- struct irq_host *host;
unsigned int isrc;
u32 *tmp;
- host = iic_get_irq_host(spu->node);
- if (host == NULL)
- return -ENODEV;
-
/* Get the interrupt source from the device-tree */
tmp = (u32 *)get_property(np, "isrc", NULL);
if (!tmp)
return -ENODEV;
- spu->isrc = isrc = tmp[0];
+ isrc = tmp[0];
+ isrc |= spu->node << IIC_IRQ_NODE_SHIFT;
+ spu->isrc = isrc;
/* Now map interrupts of all 3 classes */
- spu->irqs[0] = irq_create_mapping(host, 0x00 | isrc);
- spu->irqs[1] = irq_create_mapping(host, 0x10 | isrc);
- spu->irqs[2] = irq_create_mapping(host, 0x20 | isrc);
+ spu->irqs[0] = irq_create_mapping(NULL, 0x00 | isrc);
+ spu->irqs[1] = irq_create_mapping(NULL, 0x10 | isrc);
+ spu->irqs[2] = irq_create_mapping(NULL, 0x20 | isrc);
/* Right now, we only fail if class 2 failed */
return spu->irqs[2] == NO_IRQ ? -EINVAL : 0;
More information about the cbe-oss-dev
mailing list