Skip to contentSkip to conversation

x86-64: introduce struct pci_sysdata to facilitate sharing of ->sysdata

PATCH In review Muli Ben-Yehuda 3 participants 7 messages Jul 11, 2007 – Jul 12, 2007

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86-64: introduce struct pci_sysdata to facilitate sharing of ->sysdata

Andi, please consider applying for 2.6.23. Applies on top of the Calgary update I just sent out ("Calgary: more updates for 2.6.23").

This patch introduces struct pci_sysdata to x86 and x86-64, and converts the existing two users (NUMA, Calgary) to use it.

This lays the groundwork for having other users of sysdata, such as the PCI domains work.

The Calgary bits are tested, the NUMA bits just look ok.

---
 arch/i386/pci/acpi.c             |   32 +++++++++++++++++++++++++++-----
 arch/i386/pci/common.c           |   13 ++++++++++++-
 arch/x86_64/kernel/pci-calgary.c |   29 +++++++++++++++--------------
 arch/x86_64/kernel/tce.c         |   12 ++++--------
 arch/x86_64/pci/k8-bus.c         |    6 +++++-
 include/asm-i386/pci.h           |    5 +++++
 include/asm-i386/topology.h      |    2 +-
 include/asm-x86_64/pci.h         |   19 +++++++++++++++++++
 include/asm-x86_64/topology.h    |    2 +-
 9 files changed, 89 insertions(+), 31 deletions(-)
arch/i386/pci/acpi.c +27 -5
index b33aea8..bc8a44b 100644
@@ -8,20 +8,42 @@
 struct pci_bus * __devinit pci_acpi_scan_root(struct acpi_device *device, int domain, int busnum)
 {
 	struct pci_bus *bus;
+	struct pci_sysdata *sd;
+	int pxm;
+
+	/* Allocate per-root-bus (not per bus) arch-specific data.
+	 * TODO: leak; this memory is never freed.
+	 * It's arguable whether it's worth the trouble to care.
+	 */
+	sd = kzalloc(sizeof(*sd), GFP_KERNEL);
+	if (!sd) {
+		printk(KERN_ERR "PCI: OOM, not probing PCI bus %02x\n", busnum);
+		return NULL;
+	}
 
 	if (domain != 0) {
 		printk(KERN_WARNING "PCI: Multiple domains not supported\n");
+		kfree(sd);
 		return NULL;
 	}
 
-	bus = pcibios_scan_root(busnum);
+	sd->node = -1;
+
+	pxm = acpi_get_pxm(device->handle);
+#ifdef CONFIG_ACPI_NUMA
+	if (pxm >= 0)
+		sd->node = pxm_to_node(pxm);
+#endif
+
+	bus = pci_scan_bus_parented(NULL, busnum, &pci_root_ops, sd);
+	if (!bus)
+		kfree(sd);
+
 #ifdef CONFIG_ACPI_NUMA
 	if (bus != NULL) {
-		int pxm = acpi_get_pxm(device->handle);
 		if (pxm >= 0) {
-			bus->sysdata = (void *)(unsigned long)pxm_to_node(pxm);
-			printk("bus %d -> pxm %d -> node %ld\n",
-				busnum, pxm, (long)(bus->sysdata));
+			printk("bus %d -> pxm %d -> node %d\n",
+				busnum, pxm, sd->node);
 		}
 	}
 #endif
arch/i386/pci/common.c +12 -1
index 3f78d4d..85503de 100644
@@ -293,6 +293,7 @@ static struct dmi_system_id __devinitdata pciprobe_dmi_table[] = {
 struct pci_bus * __devinit pcibios_scan_root(int busnum)
 {
 	struct pci_bus *bus = NULL;
+	struct pci_sysdata *sd;
 
 	dmi_check_system(pciprobe_dmi_table);
 
@@ -303,9 +304,19 @@ struct pci_bus * __devinit pcibios_scan_root(int busnum)
 		}
 	}
 
+	/* Allocate per-root-bus (not per bus) arch-specific data.
+	 * TODO: leak; this memory is never freed.
+	 * It's arguable whether it's worth the trouble to care.
+	 */
+	sd = kzalloc(sizeof(*sd), GFP_KERNEL);
+	if (!sd) {
+		printk(KERN_ERR "PCI: OOM, not probing PCI bus %02x\n", busnum);
+		return NULL;
+	}
+
 	printk(KERN_DEBUG "PCI: Probing PCI hardware (bus %02x)\n", busnum);
 
-	return pci_scan_bus_parented(NULL, busnum, &pci_root_ops, NULL);
+	return pci_scan_bus_parented(NULL, busnum, &pci_root_ops, sd);
 }
 
 extern u8 pci_cache_line_size;
arch/x86_64/kernel/pci-calgary.c +15 -14
index aec8f76..a186f82 100644
@@ -375,7 +375,7 @@ static inline struct iommu_table *find_iommu_table(struct device *dev)
 	else
 		pbus = pdev->bus;
 
-	tbl = pbus->self->sysdata;
+	tbl = pci_iommu(pbus);
 
 	BUG_ON(pdev->bus->parent &&
 	       (tbl->it_busno != pdev->bus->parent->number));
@@ -718,7 +718,7 @@ static void __init calgary_reserve_mem_region(struct pci_dev *dev, u64 start,
 	limit++;
 
 	numpages = ((limit - start) >> PAGE_SHIFT);
-	iommu_range_reserve(dev->sysdata, start, numpages);
+	iommu_range_reserve(pci_iommu(dev->bus), start, numpages);
 }
 
 static void __init calgary_reserve_peripheral_mem_1(struct pci_dev *dev)
@@ -726,7 +726,7 @@ static void __init calgary_reserve_peripheral_mem_1(struct pci_dev *dev)
 	void __iomem *target;
 	u64 low, high, sizelow;
 	u64 start, limit;
-	struct iommu_table *tbl = dev->sysdata;
+	struct iommu_table *tbl = pci_iommu(dev->bus);
 	unsigned char busnum = dev->bus->number;
 	void __iomem *bbar = tbl->bbar;
 
@@ -750,7 +750,7 @@ static void __init calgary_reserve_peripheral_mem_2(struct pci_dev *dev)
 	u32 val32;
 	u64 low, high, sizelow, sizehigh;
 	u64 start, limit;
-	struct iommu_table *tbl = dev->sysdata;
+	struct iommu_table *tbl = pci_iommu(dev->bus);
 	unsigned char busnum = dev->bus->number;
 	void __iomem *bbar = tbl->bbar;
 
@@ -786,7 +786,7 @@ static void __init calgary_reserve_regions(struct pci_dev *dev)
 {
 	unsigned int npages;
 	u64 start;
-	struct iommu_table *tbl = dev->sysdata;
+	struct iommu_table *tbl = pci_iommu(dev->bus);
 
 	/* reserve EMERGENCY_PAGES from bad_dma_address and up */
 	iommu_range_reserve(tbl, bad_dma_address, EMERGENCY_PAGES);
@@ -820,7 +820,7 @@ static int __init calgary_setup_tar(struct pci_dev *dev, void __iomem *bbar)
 	if (ret)
 		return ret;
 
-	tbl = dev->sysdata;
+	tbl = pci_iommu(dev->bus);
 	tbl->it_base = (unsigned long)bus_info[dev->bus->number].tce_space;
 	tce_free(tbl, 0, tbl->it_size);
 
@@ -857,7 +857,7 @@ static int __init calgary_setup_tar(struct pci_dev *dev, void __iomem *bbar)
 static void __init calgary_free_bus(struct pci_dev *dev)
 {
 	u64 val64;
-	struct iommu_table *tbl = dev->sysdata;
+	struct iommu_table *tbl = pci_iommu(dev->bus);
 	void __iomem *target;
 	unsigned int bitmapsz;
 
@@ -872,7 +872,8 @@ static void __init calgary_free_bus(struct pci_dev *dev)
 	tbl->it_map = NULL;
 
 	kfree(tbl);
-	dev->sysdata = NULL;
+	
+	set_pci_iommu(dev->bus, NULL);
 
 	/* Can't free bootmem allocated memory after system is up :-( */
 	bus_info[dev->bus->number].tce_space = NULL;
@@ -945,7 +946,7 @@ static void calioc2_dump_error_regs(struct iommu_table *tbl)
 static void calgary_watchdog(unsigned long data)
 {
 	struct pci_dev *dev = (struct pci_dev *)data;
-	struct iommu_table *tbl = dev->sysdata;
+	struct iommu_table *tbl = pci_iommu(dev->bus);
 	void __iomem *bbar = tbl->bbar;
 	u32 val32;
 	void __iomem *target;
@@ -1043,7 +1044,7 @@ static void __init calgary_enable_translation(struct pci_dev *dev)
 	struct iommu_table *tbl;
 
 	busnum = dev->bus->number;
-	tbl = dev->sysdata;
+	tbl = pci_iommu(dev->bus);
 	bbar = tbl->bbar;
 
 	/* enable TCE in PHB Config Register */
@@ -1075,7 +1076,7 @@ static void __init calgary_disable_translation(struct pci_dev *dev)
 	struct iommu_table *tbl;
 
 	busnum = dev->bus->number;
-	tbl = dev->sysdata;
+	tbl = pci_iommu(dev->bus);
 	bbar = tbl->bbar;
 
 	/* disable TCE in PHB Config Register */
@@ -1093,7 +1094,7 @@ static void __init calgary_disable_translation(struct pci_dev *dev)
 static void __init calgary_init_one_nontraslated(struct pci_dev *dev)
 {
 	pci_dev_get(dev);
-	dev->sysdata = NULL;
+	set_pci_iommu(dev->bus, NULL);
 
 	/* is the device behind a bridge? */
 	if (dev->bus->parent)
@@ -1125,7 +1126,7 @@ static int __init calgary_init_one(struct pci_dev *dev)
 	} else
 		dev->bus->self = dev;
 
-	tbl = dev->sysdata;
+	tbl = pci_iommu(dev->bus);
 	tbl->chip_ops->handle_quirks(tbl, dev);
 
 	calgary_enable_translation(dev);
@@ -1522,7 +1523,7 @@ static void __init calgary_fixup_one_tce_space(struct pci_dev *dev)
 	unsigned int npages;
 	int i;
 
-	tbl = dev->sysdata;
+	tbl = pci_iommu(dev->bus);
 
 	for (i = 0; i < 4; i++) {
 		struct resource *r = &dev->resource[PCI_BRIDGE_RESOURCES + i];
arch/x86_64/kernel/tce.c +4 -8
index f61fb8e..3aeae2f 100644
@@ -136,9 +136,9 @@ int build_tce_table(struct pci_dev *dev, void __iomem *bbar)
 	struct iommu_table *tbl;
 	int ret;
 
-	if (dev->sysdata) {
-		printk(KERN_ERR "Calgary: dev %p has sysdata %p\n",
-		       dev, dev->sysdata);
+	if (pci_iommu(dev->bus)) {
+		printk(KERN_ERR "Calgary: dev %p has sysdata->iommu %p\n",
+		       dev, pci_iommu(dev->bus));
 		BUG();
 	}
 
@@ -155,11 +155,7 @@ int build_tce_table(struct pci_dev *dev, void __iomem *bbar)
 
 	tbl->bbar = bbar;
 
-	/*
-	 * NUMA is already using the bus's sysdata pointer, so we use
-	 * the bus's pci_dev's sysdata instead.
-	 */
-	dev->sysdata = tbl;
+	set_pci_iommu(dev->bus, tbl);
 
 	return 0;
 
arch/x86_64/pci/k8-bus.c +5 -1
index 3acf60d..9cc813e 100644
@@ -59,6 +59,8 @@ fill_mp_bus_to_cpumask(void)
 				     j <= SUBORDINATE_LDT_BUS_NUMBER(ldtbus);
 				     j++) { 
 					struct pci_bus *bus;
+					struct pci_sysdata *sd;
+
 					long node = NODE_ID(nid);
 					/* Algorithm a bit dumb, but
  					   it shouldn't matter here */
@@ -67,7 +69,9 @@ fill_mp_bus_to_cpumask(void)
 						continue;
 					if (!node_online(node))
 						node = 0;
-					bus->sysdata = (void *)node;
+
+					sd = bus->sysdata;
+					sd->node = node;
 				}		
 			}
 		}
include/asm-i386/pci.h +5 -0
index 64b6d0b..64c7e94 100644
@@ -3,6 +3,11 @@
 
 
 #ifdef __KERNEL__
+
+struct pci_sysdata {
+	int		node;		/* NUMA node */
+};
+
 #include <linux/mm.h>		/* for struct page */
 
 /* Can be used to override the logic in pci_scan_bus for skipping
include/asm-i386/topology.h +1 -1
index 7fc512d..19b2daf 100644
@@ -67,7 +67,7 @@ static inline int node_to_first_cpu(int node)
 	return first_cpu(mask);
 }
 
-#define pcibus_to_node(bus) ((long) (bus)->sysdata)
+#define pcibus_to_node(bus) ((struct pci_sysdata *)((bus)->sysdata))->node
 #define pcibus_to_cpumask(bus) node_to_cpumask(pcibus_to_node(bus))
 
 /* sched_domains SD_NODE_INIT for NUMAQ machines */
include/asm-x86_64/pci.h +19 -0
index 49c5e92..12ec371 100644
@@ -5,6 +5,25 @@
 
 #ifdef __KERNEL__
 
+struct pci_sysdata {
+	int		node;		/* NUMA node */
+	void*		iommu;		/* IOMMU private data */
+};
+
+#ifdef CONFIG_CALGARY_IOMMU
+static inline void* pci_iommu(struct pci_bus *bus)
+{
+	struct pci_sysdata *sd = bus->sysdata;
+	return sd->iommu;
+}
+
+static inline void set_pci_iommu(struct pci_bus *bus, void *val)
+{
+	struct pci_sysdata *sd = bus->sysdata;
+	sd->iommu = val;
+}
+#endif /* CONFIG_CALGARY_IOMMU */
+
 #include <linux/mm.h> /* for struct page */
 
 /* Can be used to override the logic in pci_scan_bus for skipping
include/asm-x86_64/topology.h +1 -1
index 4fd6fb2..36e52fb 100644
@@ -22,7 +22,7 @@ extern int __node_distance(int, int);
 #define parent_node(node)		(node)
 #define node_to_first_cpu(node) 	(first_cpu(node_to_cpumask[node]))
 #define node_to_cpumask(node)		(node_to_cpumask[node])
-#define pcibus_to_node(bus)		((long)(bus->sysdata))	
+#define pcibus_to_node(bus)	((struct pci_sysdata *)((bus)->sysdata))->node
 #define pcibus_to_cpumask(bus)		node_to_cpumask(pcibus_to_node(bus));
 
 #define numa_node_id()			read_pda(nodenumber)
Signature · 1 line
1.5.2
#1 lore raw
Quoted 6 lines · Muli Ben-Yehuda

Andi, please consider applying for 2.6.23. Applies on top of the Calgary update I just sent out ("Calgary: more updates for 2.6.23").

This patch introduces struct pci_sysdata to x86 and x86-64, and converts the existing two users (NUMA, Calgary) to use it.

Quoted 7 lines

This lays the groundwork for having other users of sysdata, such as the PCI domains work.

The Calgary bits are tested, the NUMA bits just look ok.

how about acpi=off in x86_64 system?

You maybe update this with
http://lkml.org/lkml/2007/6/29/246
Andrew already sent that to Andi for 2.6.23

YH

#2 lore raw
Quoted 10 lines · Yinghai Lu
Quoted 9 lines · Muli Ben-Yehuda

Andi, please consider applying for 2.6.23. Applies on top of the Calgary update I just sent out ("Calgary: more updates for 2.6.23").

This patch introduces struct pci_sysdata to x86 and x86-64, and converts the existing two users (NUMA, Calgary) to use it.

This lays the groundwork for having other users of sysdata, such as the PCI domains work.

Quoted 6 lines
Quoted 4 lines

The Calgary bits are tested, the NUMA bits just look ok.

how about acpi=off in x86_64 system?

Are you asking if it works? my main victim machines don't boot at all with acpi=off regardless of this patch so I can't test it there. I'll try to find a sacrifical Opteron machine to try it on.

Quoted 3 lines

You maybe update this with
http://lkml.org/lkml/2007/6/29/246
Andrew already sent that to Andi for 2.6.23

It's in Andi's "need more review" list, so I guess it depends on Andi's preference which (if any) of these patches will go in first. Andi?

Cheers,
Muli

#3 lore raw
In reply to Muli Ben-Yehuda
Quoted 10 lines · Muli Ben-Yehuda

Andi, please consider applying for 2.6.23. Applies on top of the Calgary update I just sent out ("Calgary: more updates for 2.6.23").

This patch introduces struct pci_sysdata to x86 and x86-64, and converts the existing two users (NUMA, Calgary) to use it.

This lays the groundwork for having other users of sysdata, such as the PCI domains work.

The Calgary bits are tested, the NUMA bits just look ok.

I very much start to dislike the untyped "sysdata"... I much rather have separate fields for the different uses (like a IOMMU field) that aren't going to share ever. Possibly even typed, but for IOMMU that may be tricky....

#4 lore raw
Quoted 10 lines · Arjan van de Ven
Quoted 9 lines · Muli Ben-Yehuda

Andi, please consider applying for 2.6.23. Applies on top of the Calgary update I just sent out ("Calgary: more updates for 2.6.23").

This patch introduces struct pci_sysdata to x86 and x86-64, and converts the existing two users (NUMA, Calgary) to use it.

This lays the groundwork for having other users of sysdata, such as the PCI domains work.

Quoted 6 lines
Quoted 1 line

The Calgary bits are tested, the NUMA bits just look ok.

I very much start to dislike the untyped "sysdata"... I much rather have separate fields for the different uses (like a IOMMU field) that aren't going to share ever. Possibly even typed, but for IOMMU that may be tricky....

Could you elaborate on what you mean here?

If you mean instead of hanging a pci_sysdata off of struct pci_bus, hang a 'void *iommu' and an 'int node' directly off of the pci_bus, that looks like a lot of churn (most architecture use sysdata as an opaque pointer to an arch specific structure) and bloat (you would need to hang off pci_bus everything that is in every arch's use of sysdata!) for very little type safety.

Cheers,
Muli

#5 lore raw
Quoted 18 lines · Muli Ben-Yehuda
Quoted 10 lines · Arjan van de Ven
Quoted 9 lines · Muli Ben-Yehuda

Andi, please consider applying for 2.6.23. Applies on top of the Calgary update I just sent out ("Calgary: more updates for 2.6.23").

This patch introduces struct pci_sysdata to x86 and x86-64, and converts the existing two users (NUMA, Calgary) to use it.

This lays the groundwork for having other users of sysdata, such as the PCI domains work.

Quoted 6 lines
Quoted 1 line

The Calgary bits are tested, the NUMA bits just look ok.

I very much start to dislike the untyped "sysdata"... I much rather have separate fields for the different uses (like a IOMMU field) that aren't going to share ever. Possibly even typed, but for IOMMU that may be tricky....

Quoted 4 lines

Could you elaborate on what you mean here?

If you mean instead of hanging a pci_sysdata off of struct pci_bus, hang a 'void *iommu' and an 'int node' directly off of the pci_bus,

yes that is what I mean.

Quoted 2 lines

that looks like a lot of churn (most architecture use sysdata as an opaque pointer to an arch specific structure) and bloat (you would

it's only a few bytes per pci device/bus. not all that bad for having something which involves splitting a shared pointer into logical users.

(and heck, if you really really are worried we could make it a union for some of them, although that defeats the "split is safer" benefit)

Signature · 2 lines
if you want to mail me at work (you don't), use arjan (at) linux.intel.com
Test the interaction between Linux and your BIOS via http://www.linuxfirmwarekit.org
#6 lore raw
Quoted 17 lines · Arjan van de Ven
Quoted 10 lines
Quoted 4 lines

I very much start to dislike the untyped "sysdata"... I much rather have separate fields for the different uses (like a IOMMU field) that aren't going to share ever. Possibly even typed, but for IOMMU that may be tricky....

Could you elaborate on what you mean here?

If you mean instead of hanging a pci_sysdata off of struct pci_bus, hang a 'void *iommu' and an 'int node' directly off of the pci_bus,

yes that is what I mean.

Quoted 3 lines

that looks like a lot of churn (most architecture use sysdata as an opaque pointer to an arch specific structure) and bloat (you would

Quoted 7 lines

it's only a few bytes per pci device/bus. not all that bad for having something which involves splitting a shared pointer into logical users.

(and heck, if you really really are worried we could make it a union for some of them, although that defeats the "split is safer" benefit)

It's not just bloat and code churn, I also don't like teaching every architecture about every other architecture's use of sysdata, which is inherently architecture specific. I think the benefits of modularity here outweigh the benefits of a little type safety.

Cheers,
Muli