linux/arch/x86_64/kernel/pci-calgary.c

1424 lines
35 KiB
C
Raw Normal View History

/*
* Derived from arch/powerpc/kernel/iommu.c
*
* Copyright IBM Corporation, 2006-2007
* Copyright (C) 2006 Jon Mason <jdmason@kudzu.us>
*
* Author: Jon Mason <jdmason@kudzu.us>
* Author: Muli Ben-Yehuda <muli@il.ibm.com>
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* 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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
#include <linux/kernel.h>
#include <linux/init.h>
#include <linux/types.h>
#include <linux/slab.h>
#include <linux/mm.h>
#include <linux/spinlock.h>
#include <linux/string.h>
#include <linux/dma-mapping.h>
#include <linux/init.h>
#include <linux/bitops.h>
#include <linux/pci_ids.h>
#include <linux/pci.h>
#include <linux/delay.h>
#include <asm/proto.h>
#include <asm/calgary.h>
#include <asm/tce.h>
#include <asm/pci-direct.h>
#include <asm/system.h>
#include <asm/dma.h>
#include <asm/rio.h>
#ifdef CONFIG_CALGARY_IOMMU_ENABLED_BY_DEFAULT
int use_calgary __read_mostly = 1;
#else
int use_calgary __read_mostly = 0;
#endif /* CONFIG_CALGARY_DEFAULT_ENABLED */
#define PCI_DEVICE_ID_IBM_CALGARY 0x02a1
#define PCI_DEVICE_ID_IBM_CALIOC2 0x0308
/* we need these for register space address calculation */
#define START_ADDRESS 0xfe000000
#define CHASSIS_BASE 0
#define ONE_BASED_CHASSIS_NUM 1
/* register offsets inside the host bridge space */
#define CALGARY_CONFIG_REG 0x0108
#define PHB_CSR_OFFSET 0x0110 /* Channel Status */
#define PHB_PLSSR_OFFSET 0x0120
#define PHB_CONFIG_RW_OFFSET 0x0160
#define PHB_IOBASE_BAR_LOW 0x0170
#define PHB_IOBASE_BAR_HIGH 0x0180
#define PHB_MEM_1_LOW 0x0190
#define PHB_MEM_1_HIGH 0x01A0
#define PHB_IO_ADDR_SIZE 0x01B0
#define PHB_MEM_1_SIZE 0x01C0
#define PHB_MEM_ST_OFFSET 0x01D0
#define PHB_AER_OFFSET 0x0200
#define PHB_CONFIG_0_HIGH 0x0220
#define PHB_CONFIG_0_LOW 0x0230
#define PHB_CONFIG_0_END 0x0240
#define PHB_MEM_2_LOW 0x02B0
#define PHB_MEM_2_HIGH 0x02C0
#define PHB_MEM_2_SIZE_HIGH 0x02D0
#define PHB_MEM_2_SIZE_LOW 0x02E0
#define PHB_DOSHOLE_OFFSET 0x08E0
/* CalIOC2 specific */
#define PHB_SAVIOR_L2 0x0DB0
/* PHB_CONFIG_RW */
#define PHB_TCE_ENABLE 0x20000000
#define PHB_SLOT_DISABLE 0x1C000000
#define PHB_DAC_DISABLE 0x01000000
#define PHB_MEM2_ENABLE 0x00400000
#define PHB_MCSR_ENABLE 0x00100000
/* TAR (Table Address Register) */
#define TAR_SW_BITS 0x0000ffffffff800fUL
#define TAR_VALID 0x0000000000000008UL
/* CSR (Channel/DMA Status Register) */
#define CSR_AGENT_MASK 0xffe0ffff
/* CCR (Calgary Configuration Register) */
#define CCR_2SEC_TIMEOUT 0x000000000000000EUL
#define MAX_NUM_OF_PHBS 8 /* how many PHBs in total? */
[PATCH] x86_64: Calgary IOMMU - Multi-Node NULL pointer dereference fix Calgary hits a NULL pointer dereference when booting in a multi-chassis NUMA system. See Redhat bugzilla number 198498, found by Konrad Rzeszutek (konradr@redhat.com). There are many issues that had to be resolved to fix this problem. Firstly when I originally wrote the code to handle NUMA systems, I had a large misunderstanding that was not corrected until now. That was that I thought the "number of nodes online" referred to number of physical systems connected. So that if NUMA was disabled, there would only be 1 node and it would only show that node's PCI bus. In reality if NUMA is disabled, the system displays all of the connected chassis as one node but is only ignorant of the delays in accessing main memory. Therefore, references to num_online_nodes() and MAX_NUMNODES are incorrect and need to be set to the maximum number of nodes that can be accessed (which are 8). I created a variable, MAX_NUM_CHASSIS, and set it to 8 to fix this. Secondly, when walking the PCI in detect_calgary, the code only checked the first "slot" when looking to see if a device is present. This will work for most cases, but unfortunately it isn't always the case. In the NUMA MXE drawers, there are USB devices present on the 3rd slot (with slot 1 being empty). So, to work around this, all slots (up to 8) are scanned to see if there are any devices present. Lastly, the bus is being enumerated on large systems in a different way the we originally thought. This throws the ugly logic we had out the window. To more elegantly handle this, I reorganized the kva array to be sparse (which removed the need to have any bus number to kva slot logic in tce.c) and created a secondary space array to contain the bus number to phb mapping. With these changes Calgary boots on an x460 with 4 nodes with and without NUMA enabled. Signed-off-by: Jon Mason <jdmason@us.ibm.com> Signed-off-by: Muli Ben-Yehuda <muli@il.ibm.com> Signed-off-by: Andi Kleen <ak@suse.de> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-07-29 21:42:43 +02:00
#define MAX_NUM_CHASSIS 8 /* max number of chassis */
/* MAX_PHB_BUS_NUM is the maximal possible dev->bus->number */
#define MAX_PHB_BUS_NUM (MAX_NUM_OF_PHBS * MAX_NUM_CHASSIS * 2)
#define PHBS_PER_CALGARY 4
/* register offsets in Calgary's internal register space */
static const unsigned long tar_offsets[] = {
0x0580 /* TAR0 */,
0x0588 /* TAR1 */,
0x0590 /* TAR2 */,
0x0598 /* TAR3 */
};
static const unsigned long split_queue_offsets[] = {
0x4870 /* SPLIT QUEUE 0 */,
0x5870 /* SPLIT QUEUE 1 */,
0x6870 /* SPLIT QUEUE 2 */,
0x7870 /* SPLIT QUEUE 3 */
};
static const unsigned long phb_offsets[] = {
0x8000 /* PHB0 */,
0x9000 /* PHB1 */,
0xA000 /* PHB2 */,
0xB000 /* PHB3 */
};
/* PHB debug registers */
static const unsigned long phb_debug_offsets[] = {
0x4000 /* PHB 0 DEBUG */,
0x5000 /* PHB 1 DEBUG */,
0x6000 /* PHB 2 DEBUG */,
0x7000 /* PHB 3 DEBUG */
};
/*
* STUFF register for each debug PHB,
* byte 1 = start bus number, byte 2 = end bus number
*/
#define PHB_DEBUG_STUFF_OFFSET 0x0020
#define EMERGENCY_PAGES 32 /* = 128KB */
unsigned int specified_table_size = TCE_TABLE_SIZE_UNSPECIFIED;
static int translate_empty_slots __read_mostly = 0;
static int calgary_detected __read_mostly = 0;
static struct rio_table_hdr *rio_table_hdr __initdata;
static struct scal_detail *scal_devs[MAX_NUMNODES] __initdata;
static struct rio_detail *rio_devs[MAX_NUMNODES * 4] __initdata;
struct calgary_bus_info {
void *tce_space;
unsigned char translation_disabled;
signed char phbid;
void __iomem *bbar;
};
static void calgary_handle_quirks(struct iommu_table *tbl, struct pci_dev *dev);
static void calgary_tce_cache_blast(struct iommu_table *tbl);
static void calioc2_handle_quirks(struct iommu_table *tbl, struct pci_dev *dev);
static struct cal_chipset_ops calgary_chip_ops = {
.handle_quirks = calgary_handle_quirks,
.tce_cache_blast = calgary_tce_cache_blast
};
static struct cal_chipset_ops calioc2_chip_ops = {
.handle_quirks = calioc2_handle_quirks,
.tce_cache_blast = NULL
};
static struct calgary_bus_info bus_info[MAX_PHB_BUS_NUM] = { { NULL, 0, 0 }, };
/* enable this to stress test the chip's TCE cache */
#ifdef CONFIG_IOMMU_DEBUG
int debugging __read_mostly = 1;
static inline unsigned long verify_bit_range(unsigned long* bitmap,
int expected, unsigned long start, unsigned long end)
{
unsigned long idx = start;
BUG_ON(start >= end);
while (idx < end) {
if (!!test_bit(idx, bitmap) != expected)
return idx;
++idx;
}
/* all bits have the expected value */
return ~0UL;
}
#else /* debugging is disabled */
int debugging __read_mostly = 0;
static inline unsigned long verify_bit_range(unsigned long* bitmap,
int expected, unsigned long start, unsigned long end)
{
return ~0UL;
}
#endif /* CONFIG_IOMMU_DEBUG */
static inline unsigned int num_dma_pages(unsigned long dma, unsigned int dmalen)
{
unsigned int npages;
npages = PAGE_ALIGN(dma + dmalen) - (dma & PAGE_MASK);
npages >>= PAGE_SHIFT;
return npages;
}
static inline int translate_phb(struct pci_dev* dev)
{
int disabled = bus_info[dev->bus->number].translation_disabled;
return !disabled;
}
static void iommu_range_reserve(struct iommu_table *tbl,
unsigned long start_addr, unsigned int npages)
{
unsigned long index;
unsigned long end;
unsigned long badbit;
index = start_addr >> PAGE_SHIFT;
/* bail out if we're asked to reserve a region we don't cover */
if (index >= tbl->it_size)
return;
end = index + npages;
if (end > tbl->it_size) /* don't go off the table */
end = tbl->it_size;
badbit = verify_bit_range(tbl->it_map, 0, index, end);
if (badbit != ~0UL) {
if (printk_ratelimit())
printk(KERN_ERR "Calgary: entry already allocated at "
"0x%lx tbl %p dma 0x%lx npages %u\n",
badbit, tbl, start_addr, npages);
}
set_bit_string(tbl->it_map, index, npages);
}
static unsigned long iommu_range_alloc(struct iommu_table *tbl,
unsigned int npages)
{
unsigned long offset;
BUG_ON(npages == 0);
offset = find_next_zero_string(tbl->it_map, tbl->it_hint,
tbl->it_size, npages);
if (offset == ~0UL) {
tbl->chip_ops->tce_cache_blast(tbl);
offset = find_next_zero_string(tbl->it_map, 0,
tbl->it_size, npages);
if (offset == ~0UL) {
printk(KERN_WARNING "Calgary: IOMMU full.\n");
if (panic_on_overflow)
panic("Calgary: fix the allocator.\n");
else
return bad_dma_address;
}
}
set_bit_string(tbl->it_map, offset, npages);
tbl->it_hint = offset + npages;
BUG_ON(tbl->it_hint > tbl->it_size);
return offset;
}
static dma_addr_t iommu_alloc(struct iommu_table *tbl, void *vaddr,
unsigned int npages, int direction)
{
unsigned long entry, flags;
dma_addr_t ret = bad_dma_address;
spin_lock_irqsave(&tbl->it_lock, flags);
entry = iommu_range_alloc(tbl, npages);
if (unlikely(entry == bad_dma_address))
goto error;
/* set the return dma address */
ret = (entry << PAGE_SHIFT) | ((unsigned long)vaddr & ~PAGE_MASK);
/* put the TCEs in the HW table */
tce_build(tbl, entry, npages, (unsigned long)vaddr & PAGE_MASK,
direction);
spin_unlock_irqrestore(&tbl->it_lock, flags);
return ret;
error:
spin_unlock_irqrestore(&tbl->it_lock, flags);
printk(KERN_WARNING "Calgary: failed to allocate %u pages in "
"iommu %p\n", npages, tbl);
return bad_dma_address;
}
static void __iommu_free(struct iommu_table *tbl, dma_addr_t dma_addr,
unsigned int npages)
{
unsigned long entry;
unsigned long badbit;
unsigned long badend;
/* were we called with bad_dma_address? */
badend = bad_dma_address + (EMERGENCY_PAGES * PAGE_SIZE);
if (unlikely((dma_addr >= bad_dma_address) && (dma_addr < badend))) {
printk(KERN_ERR "Calgary: driver tried unmapping bad DMA "
"address 0x%Lx\n", dma_addr);
WARN_ON(1);
return;
}
entry = dma_addr >> PAGE_SHIFT;
BUG_ON(entry + npages > tbl->it_size);
tce_free(tbl, entry, npages);
badbit = verify_bit_range(tbl->it_map, 1, entry, entry + npages);
if (badbit != ~0UL) {
if (printk_ratelimit())
printk(KERN_ERR "Calgary: bit is off at 0x%lx "
"tbl %p dma 0x%Lx entry 0x%lx npages %u\n",
badbit, tbl, dma_addr, entry, npages);
}
__clear_bit_string(tbl->it_map, entry, npages);
}
static void iommu_free(struct iommu_table *tbl, dma_addr_t dma_addr,
unsigned int npages)
{
unsigned long flags;
spin_lock_irqsave(&tbl->it_lock, flags);
__iommu_free(tbl, dma_addr, npages);
spin_unlock_irqrestore(&tbl->it_lock, flags);
}
static inline struct iommu_table *find_iommu_table(struct device *dev)
{
struct pci_dev *pdev;
struct pci_bus *pbus;
struct iommu_table *tbl;
pdev = to_pci_dev(dev);
/* is the device behind a bridge? */
if (unlikely(pdev->bus->parent))
pbus = pdev->bus->parent;
else
pbus = pdev->bus;
tbl = pbus->self->sysdata;
BUG_ON(pdev->bus->parent && (tbl->it_busno != pdev->bus->parent->number));
return tbl;
}
static void __calgary_unmap_sg(struct iommu_table *tbl,
struct scatterlist *sglist, int nelems, int direction)
{
while (nelems--) {
unsigned int npages;
dma_addr_t dma = sglist->dma_address;
unsigned int dmalen = sglist->dma_length;
if (dmalen == 0)
break;
npages = num_dma_pages(dma, dmalen);
__iommu_free(tbl, dma, npages);
sglist++;
}
}
void calgary_unmap_sg(struct device *dev, struct scatterlist *sglist,
int nelems, int direction)
{
unsigned long flags;
struct iommu_table *tbl = find_iommu_table(dev);
if (!translate_phb(to_pci_dev(dev)))
return;
spin_lock_irqsave(&tbl->it_lock, flags);
__calgary_unmap_sg(tbl, sglist, nelems, direction);
spin_unlock_irqrestore(&tbl->it_lock, flags);
}
static int calgary_nontranslate_map_sg(struct device* dev,
struct scatterlist *sg, int nelems, int direction)
{
int i;
for (i = 0; i < nelems; i++ ) {
struct scatterlist *s = &sg[i];
BUG_ON(!s->page);
s->dma_address = virt_to_bus(page_address(s->page) +s->offset);
s->dma_length = s->length;
}
return nelems;
}
int calgary_map_sg(struct device *dev, struct scatterlist *sg,
int nelems, int direction)
{
struct iommu_table *tbl = find_iommu_table(dev);
unsigned long flags;
unsigned long vaddr;
unsigned int npages;
unsigned long entry;
int i;
if (!translate_phb(to_pci_dev(dev)))
return calgary_nontranslate_map_sg(dev, sg, nelems, direction);
spin_lock_irqsave(&tbl->it_lock, flags);
for (i = 0; i < nelems; i++ ) {
struct scatterlist *s = &sg[i];
BUG_ON(!s->page);
vaddr = (unsigned long)page_address(s->page) + s->offset;
npages = num_dma_pages(vaddr, s->length);
entry = iommu_range_alloc(tbl, npages);
if (entry == bad_dma_address) {
/* makes sure unmap knows to stop */
s->dma_length = 0;
goto error;
}
s->dma_address = (entry << PAGE_SHIFT) | s->offset;
/* insert into HW table */
tce_build(tbl, entry, npages, vaddr & PAGE_MASK,
direction);
s->dma_length = s->length;
}
spin_unlock_irqrestore(&tbl->it_lock, flags);
return nelems;
error:
__calgary_unmap_sg(tbl, sg, nelems, direction);
for (i = 0; i < nelems; i++) {
sg[i].dma_address = bad_dma_address;
sg[i].dma_length = 0;
}
spin_unlock_irqrestore(&tbl->it_lock, flags);
return 0;
}
dma_addr_t calgary_map_single(struct device *dev, void *vaddr,
size_t size, int direction)
{
dma_addr_t dma_handle = bad_dma_address;
unsigned long uaddr;
unsigned int npages;
struct iommu_table *tbl = find_iommu_table(dev);
uaddr = (unsigned long)vaddr;
npages = num_dma_pages(uaddr, size);
if (translate_phb(to_pci_dev(dev)))
dma_handle = iommu_alloc(tbl, vaddr, npages, direction);
else
dma_handle = virt_to_bus(vaddr);
return dma_handle;
}
void calgary_unmap_single(struct device *dev, dma_addr_t dma_handle,
size_t size, int direction)
{
struct iommu_table *tbl = find_iommu_table(dev);
unsigned int npages;
if (!translate_phb(to_pci_dev(dev)))
return;
npages = num_dma_pages(dma_handle, size);
iommu_free(tbl, dma_handle, npages);
}
void* calgary_alloc_coherent(struct device *dev, size_t size,
dma_addr_t *dma_handle, gfp_t flag)
{
void *ret = NULL;
dma_addr_t mapping;
unsigned int npages, order;
struct iommu_table *tbl = find_iommu_table(dev);
size = PAGE_ALIGN(size); /* size rounded up to full pages */
npages = size >> PAGE_SHIFT;
order = get_order(size);
/* alloc enough pages (and possibly more) */
ret = (void *)__get_free_pages(flag, order);
if (!ret)
goto error;
memset(ret, 0, size);
if (translate_phb(to_pci_dev(dev))) {
/* set up tces to cover the allocated range */
mapping = iommu_alloc(tbl, ret, npages, DMA_BIDIRECTIONAL);
if (mapping == bad_dma_address)
goto free;
*dma_handle = mapping;
} else /* non translated slot */
*dma_handle = virt_to_bus(ret);
return ret;
free:
free_pages((unsigned long)ret, get_order(size));
ret = NULL;
error:
return ret;
}
static const struct dma_mapping_ops calgary_dma_ops = {
.alloc_coherent = calgary_alloc_coherent,
.map_single = calgary_map_single,
.unmap_single = calgary_unmap_single,
.map_sg = calgary_map_sg,
.unmap_sg = calgary_unmap_sg,
};
static inline void __iomem * busno_to_bbar(unsigned char num)
{
return bus_info[num].bbar;
}
static inline int busno_to_phbid(unsigned char num)
{
return bus_info[num].phbid;
}
static inline unsigned long split_queue_offset(unsigned char num)
{
size_t idx = busno_to_phbid(num);
return split_queue_offsets[idx];
}
static inline unsigned long tar_offset(unsigned char num)
{
size_t idx = busno_to_phbid(num);
return tar_offsets[idx];
}
static inline unsigned long phb_offset(unsigned char num)
{
size_t idx = busno_to_phbid(num);
return phb_offsets[idx];
}
static inline void __iomem* calgary_reg(void __iomem *bar, unsigned long offset)
{
unsigned long target = ((unsigned long)bar) | offset;
return (void __iomem*)target;
}
static inline int is_calioc2(unsigned short device)
{
return (device == PCI_DEVICE_ID_IBM_CALIOC2);
}
static inline int is_calgary(unsigned short device)
{
return (device == PCI_DEVICE_ID_IBM_CALGARY);
}
static inline int is_cal_pci_dev(unsigned short device)
{
return (is_calgary(device) || is_calioc2(device));
}
static void calgary_tce_cache_blast(struct iommu_table *tbl)
{
u64 val;
u32 aer;
int i = 0;
void __iomem *bbar = tbl->bbar;
void __iomem *target;
/* disable arbitration on the bus */
target = calgary_reg(bbar, phb_offset(tbl->it_busno) | PHB_AER_OFFSET);
aer = readl(target);
writel(0, target);
/* read plssr to ensure it got there */
target = calgary_reg(bbar, phb_offset(tbl->it_busno) | PHB_PLSSR_OFFSET);
val = readl(target);
/* poll split queues until all DMA activity is done */
target = calgary_reg(bbar, split_queue_offset(tbl->it_busno));
do {
val = readq(target);
i++;
} while ((val & 0xff) != 0xff && i < 100);
if (i == 100)
printk(KERN_WARNING "Calgary: PCI bus not quiesced, "
"continuing anyway\n");
/* invalidate TCE cache */
target = calgary_reg(bbar, tar_offset(tbl->it_busno));
writeq(tbl->tar_val, target);
/* enable arbitration */
target = calgary_reg(bbar, phb_offset(tbl->it_busno) | PHB_AER_OFFSET);
writel(aer, target);
(void)readl(target); /* flush */
}
static void __init calgary_reserve_mem_region(struct pci_dev *dev, u64 start,
u64 limit)
{
unsigned int numpages;
limit = limit | 0xfffff;
limit++;
numpages = ((limit - start) >> PAGE_SHIFT);
iommu_range_reserve(dev->sysdata, start, numpages);
}
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;
unsigned char busnum = dev->bus->number;
void __iomem *bbar = tbl->bbar;
/* peripheral MEM_1 region */
target = calgary_reg(bbar, phb_offset(busnum) | PHB_MEM_1_LOW);
low = be32_to_cpu(readl(target));
target = calgary_reg(bbar, phb_offset(busnum) | PHB_MEM_1_HIGH);
high = be32_to_cpu(readl(target));
target = calgary_reg(bbar, phb_offset(busnum) | PHB_MEM_1_SIZE);
sizelow = be32_to_cpu(readl(target));
start = (high << 32) | low;
limit = sizelow;
calgary_reserve_mem_region(dev, start, limit);
}
static void __init calgary_reserve_peripheral_mem_2(struct pci_dev *dev)
{
void __iomem *target;
u32 val32;
u64 low, high, sizelow, sizehigh;
u64 start, limit;
struct iommu_table *tbl = dev->sysdata;
unsigned char busnum = dev->bus->number;
void __iomem *bbar = tbl->bbar;
/* is it enabled? */
target = calgary_reg(bbar, phb_offset(busnum) | PHB_CONFIG_RW_OFFSET);
val32 = be32_to_cpu(readl(target));
if (!(val32 & PHB_MEM2_ENABLE))
return;
target = calgary_reg(bbar, phb_offset(busnum) | PHB_MEM_2_LOW);
low = be32_to_cpu(readl(target));
target = calgary_reg(bbar, phb_offset(busnum) | PHB_MEM_2_HIGH);
high = be32_to_cpu(readl(target));
target = calgary_reg(bbar, phb_offset(busnum) | PHB_MEM_2_SIZE_LOW);
sizelow = be32_to_cpu(readl(target));
target = calgary_reg(bbar, phb_offset(busnum) | PHB_MEM_2_SIZE_HIGH);
sizehigh = be32_to_cpu(readl(target));
start = (high << 32) | low;
limit = (sizehigh << 32) | sizelow;
calgary_reserve_mem_region(dev, start, limit);
}
/*
* some regions of the IO address space do not get translated, so we
* must not give devices IO addresses in those regions. The regions
* are the 640KB-1MB region and the two PCI peripheral memory holes.
* Reserve all of them in the IOMMU bitmap to avoid giving them out
* later.
*/
static void __init calgary_reserve_regions(struct pci_dev *dev)
{
unsigned int npages;
u64 start;
struct iommu_table *tbl = dev->sysdata;
/* reserve EMERGENCY_PAGES from bad_dma_address and up */
iommu_range_reserve(tbl, bad_dma_address, EMERGENCY_PAGES);
/* avoid the BIOS/VGA first 640KB-1MB region */
/* for CalIOC2 - avoid the entire first 2MB */
if (is_calgary(dev->device)) {
start = (640 * 1024);
npages = ((1024 - 640) * 1024) >> PAGE_SHIFT;
} else { /* calioc2 */
start = 0;
npages = (2 * 1024 * 1024) >> PAGE_SHIFT;
}
iommu_range_reserve(tbl, start, npages);
/* reserve the two PCI peripheral memory regions in IO space */
calgary_reserve_peripheral_mem_1(dev);
calgary_reserve_peripheral_mem_2(dev);
}
static int __init calgary_setup_tar(struct pci_dev *dev, void __iomem *bbar)
{
u64 val64;
u64 table_phys;
void __iomem *target;
int ret;
struct iommu_table *tbl;
/* build TCE tables for each PHB */
ret = build_tce_table(dev, bbar);
if (ret)
return ret;
tbl = dev->sysdata;
tbl->it_base = (unsigned long)bus_info[dev->bus->number].tce_space;
tce_free(tbl, 0, tbl->it_size);
if (is_calgary(dev->device))
tbl->chip_ops = &calgary_chip_ops;
else if (is_calioc2(dev->device))
tbl->chip_ops = &calioc2_chip_ops;
else
BUG();
calgary_reserve_regions(dev);
/* set TARs for each PHB */
target = calgary_reg(bbar, tar_offset(dev->bus->number));
val64 = be64_to_cpu(readq(target));
/* zero out all TAR bits under sw control */
val64 &= ~TAR_SW_BITS;
table_phys = (u64)__pa(tbl->it_base);
val64 |= table_phys;
BUG_ON(specified_table_size > TCE_TABLE_SIZE_8M);
val64 |= (u64) specified_table_size;
tbl->tar_val = cpu_to_be64(val64);
writeq(tbl->tar_val, target);
readq(target); /* flush */
return 0;
}
static void __init calgary_free_bus(struct pci_dev *dev)
{
u64 val64;
struct iommu_table *tbl = dev->sysdata;
void __iomem *target;
unsigned int bitmapsz;
target = calgary_reg(tbl->bbar, tar_offset(dev->bus->number));
val64 = be64_to_cpu(readq(target));
val64 &= ~TAR_SW_BITS;
writeq(cpu_to_be64(val64), target);
readq(target); /* flush */
bitmapsz = tbl->it_size / BITS_PER_BYTE;
free_pages((unsigned long)tbl->it_map, get_order(bitmapsz));
tbl->it_map = NULL;
kfree(tbl);
dev->sysdata = NULL;
/* Can't free bootmem allocated memory after system is up :-( */
bus_info[dev->bus->number].tce_space = NULL;
}
static void calgary_dump_error_regs(struct iommu_table *tbl)
{
void __iomem *bbar = tbl->bbar;
u32 csr, csmr, plssr, mck;
void __iomem *target;
unsigned long phboff = phb_offset(tbl->it_busno);
unsigned long erroff;
u32 errregs[7];
int i;
/* dump CSR */
target = calgary_reg(bbar, phboff | PHB_CSR_OFFSET);
csr = be32_to_cpu(readl(target));
/* dump PLSSR */
target = calgary_reg(bbar, phboff | PHB_PLSSR_OFFSET);
plssr = be32_to_cpu(readl(target));
/* dump CSMR */
target = calgary_reg(bbar, phboff | 0x290);
csmr = be32_to_cpu(readl(target));
/* dump mck */
target = calgary_reg(bbar, phboff | 0x800);
mck = be32_to_cpu(readl(target));
printk(KERN_EMERG "Calgary: 0x%08x@CSR 0x%08x@PLSSR 0x%08x@CSMR "
"0x%08x@MCK\n", csr, plssr, csmr, mck);
/* dump rest of error regs */
printk(KERN_EMERG "Calgary: ");
for (i = 0; i < ARRAY_SIZE(errregs); i++) {
erroff = (0x810 + (i * 0x10)); /* err regs are at 0x810 - 0x870 */
target = calgary_reg(bbar, phboff | erroff);
errregs[i] = be32_to_cpu(readl(target));
printk("0x%08x@0x%lx ", errregs[i], erroff);
}
printk("\n");
}
static void calgary_watchdog(unsigned long data)
{
struct pci_dev *dev = (struct pci_dev *)data;
struct iommu_table *tbl = dev->sysdata;
void __iomem *bbar = tbl->bbar;
u32 val32;
void __iomem *target;
target = calgary_reg(bbar, phb_offset(tbl->it_busno) | PHB_CSR_OFFSET);
val32 = be32_to_cpu(readl(target));
/* If no error, the agent ID in the CSR is not valid */
if (val32 & CSR_AGENT_MASK) {
printk(KERN_EMERG "Calgary: DMA error on PHB %#x\n",
dev->bus->number);
calgary_dump_error_regs(tbl);
/* reset error */
writel(0, target);
/* Disable bus that caused the error */
target = calgary_reg(bbar, phb_offset(tbl->it_busno) |
PHB_CONFIG_RW_OFFSET);
val32 = be32_to_cpu(readl(target));
val32 |= PHB_SLOT_DISABLE;
writel(cpu_to_be32(val32), target);
readl(target); /* flush */
} else {
/* Reset the timer */
mod_timer(&tbl->watchdog_timer, jiffies + 2 * HZ);
}
}
static void __init calgary_set_split_completion_timeout(void __iomem *bbar,
unsigned char busnum, unsigned long timeout)
{
u64 val64;
void __iomem *target;
unsigned int phb_shift = ~0; /* silence gcc */
u64 mask;
switch (busno_to_phbid(busnum)) {
case 0: phb_shift = (63 - 19);
break;
case 1: phb_shift = (63 - 23);
break;
case 2: phb_shift = (63 - 27);
break;
case 3: phb_shift = (63 - 35);
break;
default:
BUG_ON(busno_to_phbid(busnum));
}
target = calgary_reg(bbar, CALGARY_CONFIG_REG);
val64 = be64_to_cpu(readq(target));
/* zero out this PHB's timer bits */
mask = ~(0xFUL << phb_shift);
val64 &= mask;
val64 |= (timeout << phb_shift);
writeq(cpu_to_be64(val64), target);
readq(target); /* flush */
}
static void calioc2_handle_quirks(struct iommu_table *tbl, struct pci_dev *dev)
{
unsigned char busnum = dev->bus->number;
void __iomem *bbar = tbl->bbar;
void __iomem *target;
u32 val;
/*
* CalIOC2 designers recommend setting bit 8 in 0xnDB0 to 1
*/
target = calgary_reg(bbar, phb_offset(busnum) | PHB_SAVIOR_L2);
val = cpu_to_be32(readl(target));
val |= 0x00800000;
writel(cpu_to_be32(val), target);
}
static void calgary_handle_quirks(struct iommu_table *tbl, struct pci_dev *dev)
{
unsigned char busnum = dev->bus->number;
/*
* Give split completion a longer timeout on bus 1 for aic94xx
* http://bugzilla.kernel.org/show_bug.cgi?id=7180
*/
if (is_calgary(dev->device) && (busnum == 1))
calgary_set_split_completion_timeout(tbl->bbar, busnum,
CCR_2SEC_TIMEOUT);
}
static void __init calgary_enable_translation(struct pci_dev *dev)
{
u32 val32;
unsigned char busnum;
void __iomem *target;
void __iomem *bbar;
struct iommu_table *tbl;
busnum = dev->bus->number;
tbl = dev->sysdata;
bbar = tbl->bbar;
/* enable TCE in PHB Config Register */
target = calgary_reg(bbar, phb_offset(busnum) | PHB_CONFIG_RW_OFFSET);
val32 = be32_to_cpu(readl(target));
val32 |= PHB_TCE_ENABLE | PHB_DAC_DISABLE | PHB_MCSR_ENABLE;
printk(KERN_INFO "Calgary: enabling translation on %s PHB %#x\n",
(dev->device == PCI_DEVICE_ID_IBM_CALGARY) ?
"Calgary" : "CalIOC2", busnum);
printk(KERN_INFO "Calgary: errant DMAs will now be prevented on this "
"bus.\n");
writel(cpu_to_be32(val32), target);
readl(target); /* flush */
init_timer(&tbl->watchdog_timer);
tbl->watchdog_timer.function = &calgary_watchdog;
tbl->watchdog_timer.data = (unsigned long)dev;
mod_timer(&tbl->watchdog_timer, jiffies);
}
static void __init calgary_disable_translation(struct pci_dev *dev)
{
u32 val32;
unsigned char busnum;
void __iomem *target;
void __iomem *bbar;
struct iommu_table *tbl;
busnum = dev->bus->number;
tbl = dev->sysdata;
bbar = tbl->bbar;
/* disable TCE in PHB Config Register */
target = calgary_reg(bbar, phb_offset(busnum) | PHB_CONFIG_RW_OFFSET);
val32 = be32_to_cpu(readl(target));
val32 &= ~(PHB_TCE_ENABLE | PHB_DAC_DISABLE | PHB_MCSR_ENABLE);
printk(KERN_INFO "Calgary: disabling translation on PHB %#x!\n", busnum);
writel(cpu_to_be32(val32), target);
readl(target); /* flush */
del_timer_sync(&tbl->watchdog_timer);
}
static void __init calgary_init_one_nontraslated(struct pci_dev *dev)
{
pci_dev_get(dev);
dev->sysdata = NULL;
/* is the device behind a bridge? */
if (dev->bus->parent)
dev->bus->parent->self = dev;
else
dev->bus->self = dev;
}
static int __init calgary_init_one(struct pci_dev *dev)
{
void __iomem *bbar;
struct iommu_table *tbl;
int ret;
BUG_ON(dev->bus->number >= MAX_PHB_BUS_NUM);
bbar = busno_to_bbar(dev->bus->number);
ret = calgary_setup_tar(dev, bbar);
if (ret)
goto done;
pci_dev_get(dev);
if (dev->bus->parent) {
if (dev->bus->parent->self)
printk(KERN_WARNING "Calgary: IEEEE, dev %p has "
"bus->parent->self!\n", dev);
dev->bus->parent->self = dev;
} else
dev->bus->self = dev;
tbl = dev->sysdata;
tbl->chip_ops->handle_quirks(tbl, dev);
calgary_enable_translation(dev);
return 0;
done:
return ret;
}
static int __init calgary_locate_bbars(void)
{
int ret;
int rioidx, phb, bus;
void __iomem *bbar;
void __iomem *target;
unsigned long offset;
u8 start_bus, end_bus;
u32 val;
ret = -ENODATA;
for (rioidx = 0; rioidx < rio_table_hdr->num_rio_dev; rioidx++) {
struct rio_detail *rio = rio_devs[rioidx];
if ((rio->type != COMPAT_CALGARY) && (rio->type != ALT_CALGARY))
continue;
/* map entire 1MB of Calgary config space */
bbar = ioremap_nocache(rio->BBAR, 1024 * 1024);
if (!bbar)
goto error;
for (phb = 0; phb < PHBS_PER_CALGARY; phb++) {
offset = phb_debug_offsets[phb] | PHB_DEBUG_STUFF_OFFSET;
target = calgary_reg(bbar, offset);
val = be32_to_cpu(readl(target));
start_bus = (u8)((val & 0x00FF0000) >> 16);
end_bus = (u8)((val & 0x0000FF00) >> 8);
if (end_bus) {
for (bus = start_bus; bus <= end_bus; bus++) {
bus_info[bus].bbar = bbar;
bus_info[bus].phbid = phb;
}
} else {
bus_info[start_bus].bbar = bbar;
bus_info[start_bus].phbid = phb;
}
}
}
return 0;
error:
/* scan bus_info and iounmap any bbars we previously ioremap'd */
for (bus = 0; bus < ARRAY_SIZE(bus_info); bus++)
if (bus_info[bus].bbar)
iounmap(bus_info[bus].bbar);
return ret;
}
static int __init calgary_init(void)
{
int ret;
struct pci_dev *dev = NULL;
void* tce_space;
ret = calgary_locate_bbars();
if (ret)
return ret;
do {
dev = pci_get_device(PCI_VENDOR_ID_IBM, PCI_ANY_ID, dev);
if (!dev)
break;
if (!is_cal_pci_dev(dev->device))
continue;
if (!translate_phb(dev)) {
calgary_init_one_nontraslated(dev);
continue;
}
tce_space = bus_info[dev->bus->number].tce_space;
if (!tce_space && !translate_empty_slots) {
printk("Calg: %p failed tce_space check\n", dev);
continue;
}
ret = calgary_init_one(dev);
if (ret)
goto error;
} while (1);
return ret;
error:
do {
dev = pci_get_device_reverse(PCI_VENDOR_ID_IBM,
PCI_ANY_ID, dev);
if (!dev)
break;
if (!is_cal_pci_dev(dev->device))
continue;
if (!translate_phb(dev)) {
pci_dev_put(dev);
continue;
}
if (!bus_info[dev->bus->number].tce_space && !translate_empty_slots)
continue;
calgary_disable_translation(dev);
calgary_free_bus(dev);
pci_dev_put(dev); /* Undo calgary_init_one()'s pci_dev_get() */
} while (1);
return ret;
}
static inline int __init determine_tce_table_size(u64 ram)
{
int ret;
if (specified_table_size != TCE_TABLE_SIZE_UNSPECIFIED)
return specified_table_size;
/*
* Table sizes are from 0 to 7 (TCE_TABLE_SIZE_64K to
* TCE_TABLE_SIZE_8M). Table size 0 has 8K entries and each
* larger table size has twice as many entries, so shift the
* max ram address by 13 to divide by 8K and then look at the
* order of the result to choose between 0-7.
*/
ret = get_order(ram >> 13);
if (ret > TCE_TABLE_SIZE_8M)
ret = TCE_TABLE_SIZE_8M;
return ret;
}
static int __init build_detail_arrays(void)
{
unsigned long ptr;
int i, scal_detail_size, rio_detail_size;
if (rio_table_hdr->num_scal_dev > MAX_NUMNODES){
printk(KERN_WARNING
"Calgary: MAX_NUMNODES too low! Defined as %d, "
"but system has %d nodes.\n",
MAX_NUMNODES, rio_table_hdr->num_scal_dev);
return -ENODEV;
}
switch (rio_table_hdr->version){
case 2:
scal_detail_size = 11;
rio_detail_size = 13;
break;
case 3:
scal_detail_size = 12;
rio_detail_size = 15;
break;
default:
printk(KERN_WARNING
"Calgary: Invalid Rio Grande Table Version: %d\n",
rio_table_hdr->version);
return -EPROTO;
}
ptr = ((unsigned long)rio_table_hdr) + 3;
for (i = 0; i < rio_table_hdr->num_scal_dev;
i++, ptr += scal_detail_size)
scal_devs[i] = (struct scal_detail *)ptr;
for (i = 0; i < rio_table_hdr->num_rio_dev;
i++, ptr += rio_detail_size)
rio_devs[i] = (struct rio_detail *)ptr;
return 0;
}
static int __init calgary_bus_has_devices(int bus, unsigned short pci_dev)
{
int dev;
u32 val;
if (pci_dev == PCI_DEVICE_ID_IBM_CALIOC2) {
/*
* FIXME: properly scan for devices accross the
* PCI-to-PCI bridge on every CalIOC2 port.
*/
return 1;
}
for (dev = 1; dev < 8; dev++) {
val = read_pci_config(bus, dev, 0, 0);
if (val != 0xffffffff)
break;
}
return (val != 0xffffffff);
}
void __init detect_calgary(void)
{
[PATCH] x86_64: Calgary IOMMU - Multi-Node NULL pointer dereference fix Calgary hits a NULL pointer dereference when booting in a multi-chassis NUMA system. See Redhat bugzilla number 198498, found by Konrad Rzeszutek (konradr@redhat.com). There are many issues that had to be resolved to fix this problem. Firstly when I originally wrote the code to handle NUMA systems, I had a large misunderstanding that was not corrected until now. That was that I thought the "number of nodes online" referred to number of physical systems connected. So that if NUMA was disabled, there would only be 1 node and it would only show that node's PCI bus. In reality if NUMA is disabled, the system displays all of the connected chassis as one node but is only ignorant of the delays in accessing main memory. Therefore, references to num_online_nodes() and MAX_NUMNODES are incorrect and need to be set to the maximum number of nodes that can be accessed (which are 8). I created a variable, MAX_NUM_CHASSIS, and set it to 8 to fix this. Secondly, when walking the PCI in detect_calgary, the code only checked the first "slot" when looking to see if a device is present. This will work for most cases, but unfortunately it isn't always the case. In the NUMA MXE drawers, there are USB devices present on the 3rd slot (with slot 1 being empty). So, to work around this, all slots (up to 8) are scanned to see if there are any devices present. Lastly, the bus is being enumerated on large systems in a different way the we originally thought. This throws the ugly logic we had out the window. To more elegantly handle this, I reorganized the kva array to be sparse (which removed the need to have any bus number to kva slot logic in tce.c) and created a secondary space array to contain the bus number to phb mapping. With these changes Calgary boots on an x460 with 4 nodes with and without NUMA enabled. Signed-off-by: Jon Mason <jdmason@us.ibm.com> Signed-off-by: Muli Ben-Yehuda <muli@il.ibm.com> Signed-off-by: Andi Kleen <ak@suse.de> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-07-29 21:42:43 +02:00
int bus;
void *tbl;
[PATCH] x86_64: Calgary IOMMU - Multi-Node NULL pointer dereference fix Calgary hits a NULL pointer dereference when booting in a multi-chassis NUMA system. See Redhat bugzilla number 198498, found by Konrad Rzeszutek (konradr@redhat.com). There are many issues that had to be resolved to fix this problem. Firstly when I originally wrote the code to handle NUMA systems, I had a large misunderstanding that was not corrected until now. That was that I thought the "number of nodes online" referred to number of physical systems connected. So that if NUMA was disabled, there would only be 1 node and it would only show that node's PCI bus. In reality if NUMA is disabled, the system displays all of the connected chassis as one node but is only ignorant of the delays in accessing main memory. Therefore, references to num_online_nodes() and MAX_NUMNODES are incorrect and need to be set to the maximum number of nodes that can be accessed (which are 8). I created a variable, MAX_NUM_CHASSIS, and set it to 8 to fix this. Secondly, when walking the PCI in detect_calgary, the code only checked the first "slot" when looking to see if a device is present. This will work for most cases, but unfortunately it isn't always the case. In the NUMA MXE drawers, there are USB devices present on the 3rd slot (with slot 1 being empty). So, to work around this, all slots (up to 8) are scanned to see if there are any devices present. Lastly, the bus is being enumerated on large systems in a different way the we originally thought. This throws the ugly logic we had out the window. To more elegantly handle this, I reorganized the kva array to be sparse (which removed the need to have any bus number to kva slot logic in tce.c) and created a secondary space array to contain the bus number to phb mapping. With these changes Calgary boots on an x460 with 4 nodes with and without NUMA enabled. Signed-off-by: Jon Mason <jdmason@us.ibm.com> Signed-off-by: Muli Ben-Yehuda <muli@il.ibm.com> Signed-off-by: Andi Kleen <ak@suse.de> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-07-29 21:42:43 +02:00
int calgary_found = 0;
unsigned long ptr;
[PATCH] x86_64: fix boot time hang in detect_calgary() if CONFIG_CALGARY_IOMMU is built into the kernel via CONFIG_CALGARY_IOMMU_ENABLED_BY_DEFAULT, or is enabled via the iommu=calgary boot option, then the detect_calgary() function runs to detect the presence of a Calgary IOMMU. detect_calgary() first searches the BIOS EBDA area for a "rio_table_hdr" BIOS table. It has this parsing algorithm for the EBDA: while (offset) { ... /* The next offset is stored in the 1st word. 0 means no more */ offset = *((unsigned short *)(ptr + offset)); } got that? Lets repeat it slowly: we've got a BIOS-supplied data structure, plus Linux kernel code that will only break out of an infinite parsing loop once the BIOS gives a zero offset. Ok? Translation: what an excellent opportunity for BIOS writers to lock up the Linux boot process in an utterly hard to debug place! Indeed the BIOS jumped on that opportunity on my box, which has the following EBDA chaining layout: 384, 65282, 65535, 65535, 65535, 65535, 65535, 65535 ... see the pattern? So my, definitely non-Calgary system happily locks up in detect_calgary()! the patch below fixes the boot hang by trusting the BIOS-supplied data structure a bit less: the parser always has to make forward progress, and if it doesnt, we break out of the loop and i get the expected kernel message: Calgary: Unable to locate Rio Grande Table in EBDA - bailing! Signed-off-by: Ingo Molnar <mingo@elte.hu> Acked-by: Muli Ben-Yehuda <muli@il.ibm.com> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-20 11:53:32 +01:00
unsigned int offset, prev_offset;
int ret;
/*
* if the user specified iommu=off or iommu=soft or we found
* another HW IOMMU already, bail out.
*/
if (swiotlb || no_iommu || iommu_detected)
return;
if (!use_calgary)
return;
if (!early_pci_allowed())
return;
printk(KERN_DEBUG "Calgary: detecting Calgary via BIOS EBDA area\n");
ptr = (unsigned long)phys_to_virt(get_bios_ebda());
rio_table_hdr = NULL;
[PATCH] x86_64: fix boot time hang in detect_calgary() if CONFIG_CALGARY_IOMMU is built into the kernel via CONFIG_CALGARY_IOMMU_ENABLED_BY_DEFAULT, or is enabled via the iommu=calgary boot option, then the detect_calgary() function runs to detect the presence of a Calgary IOMMU. detect_calgary() first searches the BIOS EBDA area for a "rio_table_hdr" BIOS table. It has this parsing algorithm for the EBDA: while (offset) { ... /* The next offset is stored in the 1st word. 0 means no more */ offset = *((unsigned short *)(ptr + offset)); } got that? Lets repeat it slowly: we've got a BIOS-supplied data structure, plus Linux kernel code that will only break out of an infinite parsing loop once the BIOS gives a zero offset. Ok? Translation: what an excellent opportunity for BIOS writers to lock up the Linux boot process in an utterly hard to debug place! Indeed the BIOS jumped on that opportunity on my box, which has the following EBDA chaining layout: 384, 65282, 65535, 65535, 65535, 65535, 65535, 65535 ... see the pattern? So my, definitely non-Calgary system happily locks up in detect_calgary()! the patch below fixes the boot hang by trusting the BIOS-supplied data structure a bit less: the parser always has to make forward progress, and if it doesnt, we break out of the loop and i get the expected kernel message: Calgary: Unable to locate Rio Grande Table in EBDA - bailing! Signed-off-by: Ingo Molnar <mingo@elte.hu> Acked-by: Muli Ben-Yehuda <muli@il.ibm.com> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-20 11:53:32 +01:00
prev_offset = 0;
offset = 0x180;
[PATCH] x86_64: fix boot time hang in detect_calgary() if CONFIG_CALGARY_IOMMU is built into the kernel via CONFIG_CALGARY_IOMMU_ENABLED_BY_DEFAULT, or is enabled via the iommu=calgary boot option, then the detect_calgary() function runs to detect the presence of a Calgary IOMMU. detect_calgary() first searches the BIOS EBDA area for a "rio_table_hdr" BIOS table. It has this parsing algorithm for the EBDA: while (offset) { ... /* The next offset is stored in the 1st word. 0 means no more */ offset = *((unsigned short *)(ptr + offset)); } got that? Lets repeat it slowly: we've got a BIOS-supplied data structure, plus Linux kernel code that will only break out of an infinite parsing loop once the BIOS gives a zero offset. Ok? Translation: what an excellent opportunity for BIOS writers to lock up the Linux boot process in an utterly hard to debug place! Indeed the BIOS jumped on that opportunity on my box, which has the following EBDA chaining layout: 384, 65282, 65535, 65535, 65535, 65535, 65535, 65535 ... see the pattern? So my, definitely non-Calgary system happily locks up in detect_calgary()! the patch below fixes the boot hang by trusting the BIOS-supplied data structure a bit less: the parser always has to make forward progress, and if it doesnt, we break out of the loop and i get the expected kernel message: Calgary: Unable to locate Rio Grande Table in EBDA - bailing! Signed-off-by: Ingo Molnar <mingo@elte.hu> Acked-by: Muli Ben-Yehuda <muli@il.ibm.com> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-20 11:53:32 +01:00
/*
* The next offset is stored in the 1st word.
* Only parse up until the offset increases:
*/
while (offset > prev_offset) {
/* The block id is stored in the 2nd word */
if (*((unsigned short *)(ptr + offset + 2)) == 0x4752){
/* set the pointer past the offset & block id */
rio_table_hdr = (struct rio_table_hdr *)(ptr + offset + 4);
break;
}
[PATCH] x86_64: fix boot time hang in detect_calgary() if CONFIG_CALGARY_IOMMU is built into the kernel via CONFIG_CALGARY_IOMMU_ENABLED_BY_DEFAULT, or is enabled via the iommu=calgary boot option, then the detect_calgary() function runs to detect the presence of a Calgary IOMMU. detect_calgary() first searches the BIOS EBDA area for a "rio_table_hdr" BIOS table. It has this parsing algorithm for the EBDA: while (offset) { ... /* The next offset is stored in the 1st word. 0 means no more */ offset = *((unsigned short *)(ptr + offset)); } got that? Lets repeat it slowly: we've got a BIOS-supplied data structure, plus Linux kernel code that will only break out of an infinite parsing loop once the BIOS gives a zero offset. Ok? Translation: what an excellent opportunity for BIOS writers to lock up the Linux boot process in an utterly hard to debug place! Indeed the BIOS jumped on that opportunity on my box, which has the following EBDA chaining layout: 384, 65282, 65535, 65535, 65535, 65535, 65535, 65535 ... see the pattern? So my, definitely non-Calgary system happily locks up in detect_calgary()! the patch below fixes the boot hang by trusting the BIOS-supplied data structure a bit less: the parser always has to make forward progress, and if it doesnt, we break out of the loop and i get the expected kernel message: Calgary: Unable to locate Rio Grande Table in EBDA - bailing! Signed-off-by: Ingo Molnar <mingo@elte.hu> Acked-by: Muli Ben-Yehuda <muli@il.ibm.com> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-20 11:53:32 +01:00
prev_offset = offset;
offset = *((unsigned short *)(ptr + offset));
}
if (!rio_table_hdr) {
printk(KERN_DEBUG "Calgary: Unable to locate Rio Grande table "
"in EBDA - bailing!\n");
return;
}
ret = build_detail_arrays();
if (ret) {
printk(KERN_DEBUG "Calgary: build_detail_arrays ret %d\n", ret);
return;
}
specified_table_size = determine_tce_table_size(end_pfn * PAGE_SIZE);
[PATCH] x86_64: Calgary IOMMU - Multi-Node NULL pointer dereference fix Calgary hits a NULL pointer dereference when booting in a multi-chassis NUMA system. See Redhat bugzilla number 198498, found by Konrad Rzeszutek (konradr@redhat.com). There are many issues that had to be resolved to fix this problem. Firstly when I originally wrote the code to handle NUMA systems, I had a large misunderstanding that was not corrected until now. That was that I thought the "number of nodes online" referred to number of physical systems connected. So that if NUMA was disabled, there would only be 1 node and it would only show that node's PCI bus. In reality if NUMA is disabled, the system displays all of the connected chassis as one node but is only ignorant of the delays in accessing main memory. Therefore, references to num_online_nodes() and MAX_NUMNODES are incorrect and need to be set to the maximum number of nodes that can be accessed (which are 8). I created a variable, MAX_NUM_CHASSIS, and set it to 8 to fix this. Secondly, when walking the PCI in detect_calgary, the code only checked the first "slot" when looking to see if a device is present. This will work for most cases, but unfortunately it isn't always the case. In the NUMA MXE drawers, there are USB devices present on the 3rd slot (with slot 1 being empty). So, to work around this, all slots (up to 8) are scanned to see if there are any devices present. Lastly, the bus is being enumerated on large systems in a different way the we originally thought. This throws the ugly logic we had out the window. To more elegantly handle this, I reorganized the kva array to be sparse (which removed the need to have any bus number to kva slot logic in tce.c) and created a secondary space array to contain the bus number to phb mapping. With these changes Calgary boots on an x460 with 4 nodes with and without NUMA enabled. Signed-off-by: Jon Mason <jdmason@us.ibm.com> Signed-off-by: Muli Ben-Yehuda <muli@il.ibm.com> Signed-off-by: Andi Kleen <ak@suse.de> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-07-29 21:42:43 +02:00
for (bus = 0; bus < MAX_PHB_BUS_NUM; bus++) {
struct calgary_bus_info *info = &bus_info[bus];
unsigned short pci_device;
u32 val;
val = read_pci_config(bus, 0, 0, 0);
pci_device = (val & 0xFFFF0000) >> 16;
[PATCH] x86_64: Calgary IOMMU - Multi-Node NULL pointer dereference fix Calgary hits a NULL pointer dereference when booting in a multi-chassis NUMA system. See Redhat bugzilla number 198498, found by Konrad Rzeszutek (konradr@redhat.com). There are many issues that had to be resolved to fix this problem. Firstly when I originally wrote the code to handle NUMA systems, I had a large misunderstanding that was not corrected until now. That was that I thought the "number of nodes online" referred to number of physical systems connected. So that if NUMA was disabled, there would only be 1 node and it would only show that node's PCI bus. In reality if NUMA is disabled, the system displays all of the connected chassis as one node but is only ignorant of the delays in accessing main memory. Therefore, references to num_online_nodes() and MAX_NUMNODES are incorrect and need to be set to the maximum number of nodes that can be accessed (which are 8). I created a variable, MAX_NUM_CHASSIS, and set it to 8 to fix this. Secondly, when walking the PCI in detect_calgary, the code only checked the first "slot" when looking to see if a device is present. This will work for most cases, but unfortunately it isn't always the case. In the NUMA MXE drawers, there are USB devices present on the 3rd slot (with slot 1 being empty). So, to work around this, all slots (up to 8) are scanned to see if there are any devices present. Lastly, the bus is being enumerated on large systems in a different way the we originally thought. This throws the ugly logic we had out the window. To more elegantly handle this, I reorganized the kva array to be sparse (which removed the need to have any bus number to kva slot logic in tce.c) and created a secondary space array to contain the bus number to phb mapping. With these changes Calgary boots on an x460 with 4 nodes with and without NUMA enabled. Signed-off-by: Jon Mason <jdmason@us.ibm.com> Signed-off-by: Muli Ben-Yehuda <muli@il.ibm.com> Signed-off-by: Andi Kleen <ak@suse.de> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-07-29 21:42:43 +02:00
if (!is_cal_pci_dev(pci_device))
continue;
[PATCH] x86_64: Calgary IOMMU - Multi-Node NULL pointer dereference fix Calgary hits a NULL pointer dereference when booting in a multi-chassis NUMA system. See Redhat bugzilla number 198498, found by Konrad Rzeszutek (konradr@redhat.com). There are many issues that had to be resolved to fix this problem. Firstly when I originally wrote the code to handle NUMA systems, I had a large misunderstanding that was not corrected until now. That was that I thought the "number of nodes online" referred to number of physical systems connected. So that if NUMA was disabled, there would only be 1 node and it would only show that node's PCI bus. In reality if NUMA is disabled, the system displays all of the connected chassis as one node but is only ignorant of the delays in accessing main memory. Therefore, references to num_online_nodes() and MAX_NUMNODES are incorrect and need to be set to the maximum number of nodes that can be accessed (which are 8). I created a variable, MAX_NUM_CHASSIS, and set it to 8 to fix this. Secondly, when walking the PCI in detect_calgary, the code only checked the first "slot" when looking to see if a device is present. This will work for most cases, but unfortunately it isn't always the case. In the NUMA MXE drawers, there are USB devices present on the 3rd slot (with slot 1 being empty). So, to work around this, all slots (up to 8) are scanned to see if there are any devices present. Lastly, the bus is being enumerated on large systems in a different way the we originally thought. This throws the ugly logic we had out the window. To more elegantly handle this, I reorganized the kva array to be sparse (which removed the need to have any bus number to kva slot logic in tce.c) and created a secondary space array to contain the bus number to phb mapping. With these changes Calgary boots on an x460 with 4 nodes with and without NUMA enabled. Signed-off-by: Jon Mason <jdmason@us.ibm.com> Signed-off-by: Muli Ben-Yehuda <muli@il.ibm.com> Signed-off-by: Andi Kleen <ak@suse.de> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-07-29 21:42:43 +02:00
if (info->translation_disabled)
continue;
if (calgary_bus_has_devices(bus, pci_device) ||
translate_empty_slots) {
tbl = alloc_tce_table();
if (!tbl)
goto cleanup;
info->tce_space = tbl;
calgary_found = 1;
printk("Calg: allocated tce_table %p for bus 0x%x\n",
info->tce_space, bus);
[PATCH] x86_64: Calgary IOMMU - Multi-Node NULL pointer dereference fix Calgary hits a NULL pointer dereference when booting in a multi-chassis NUMA system. See Redhat bugzilla number 198498, found by Konrad Rzeszutek (konradr@redhat.com). There are many issues that had to be resolved to fix this problem. Firstly when I originally wrote the code to handle NUMA systems, I had a large misunderstanding that was not corrected until now. That was that I thought the "number of nodes online" referred to number of physical systems connected. So that if NUMA was disabled, there would only be 1 node and it would only show that node's PCI bus. In reality if NUMA is disabled, the system displays all of the connected chassis as one node but is only ignorant of the delays in accessing main memory. Therefore, references to num_online_nodes() and MAX_NUMNODES are incorrect and need to be set to the maximum number of nodes that can be accessed (which are 8). I created a variable, MAX_NUM_CHASSIS, and set it to 8 to fix this. Secondly, when walking the PCI in detect_calgary, the code only checked the first "slot" when looking to see if a device is present. This will work for most cases, but unfortunately it isn't always the case. In the NUMA MXE drawers, there are USB devices present on the 3rd slot (with slot 1 being empty). So, to work around this, all slots (up to 8) are scanned to see if there are any devices present. Lastly, the bus is being enumerated on large systems in a different way the we originally thought. This throws the ugly logic we had out the window. To more elegantly handle this, I reorganized the kva array to be sparse (which removed the need to have any bus number to kva slot logic in tce.c) and created a secondary space array to contain the bus number to phb mapping. With these changes Calgary boots on an x460 with 4 nodes with and without NUMA enabled. Signed-off-by: Jon Mason <jdmason@us.ibm.com> Signed-off-by: Muli Ben-Yehuda <muli@il.ibm.com> Signed-off-by: Andi Kleen <ak@suse.de> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-07-29 21:42:43 +02:00
}
}
printk(KERN_DEBUG "Calgary: finished detection, Calgary %s\n",
calgary_found ? "found" : "not found");
[PATCH] x86_64: Calgary IOMMU - Multi-Node NULL pointer dereference fix Calgary hits a NULL pointer dereference when booting in a multi-chassis NUMA system. See Redhat bugzilla number 198498, found by Konrad Rzeszutek (konradr@redhat.com). There are many issues that had to be resolved to fix this problem. Firstly when I originally wrote the code to handle NUMA systems, I had a large misunderstanding that was not corrected until now. That was that I thought the "number of nodes online" referred to number of physical systems connected. So that if NUMA was disabled, there would only be 1 node and it would only show that node's PCI bus. In reality if NUMA is disabled, the system displays all of the connected chassis as one node but is only ignorant of the delays in accessing main memory. Therefore, references to num_online_nodes() and MAX_NUMNODES are incorrect and need to be set to the maximum number of nodes that can be accessed (which are 8). I created a variable, MAX_NUM_CHASSIS, and set it to 8 to fix this. Secondly, when walking the PCI in detect_calgary, the code only checked the first "slot" when looking to see if a device is present. This will work for most cases, but unfortunately it isn't always the case. In the NUMA MXE drawers, there are USB devices present on the 3rd slot (with slot 1 being empty). So, to work around this, all slots (up to 8) are scanned to see if there are any devices present. Lastly, the bus is being enumerated on large systems in a different way the we originally thought. This throws the ugly logic we had out the window. To more elegantly handle this, I reorganized the kva array to be sparse (which removed the need to have any bus number to kva slot logic in tce.c) and created a secondary space array to contain the bus number to phb mapping. With these changes Calgary boots on an x460 with 4 nodes with and without NUMA enabled. Signed-off-by: Jon Mason <jdmason@us.ibm.com> Signed-off-by: Muli Ben-Yehuda <muli@il.ibm.com> Signed-off-by: Andi Kleen <ak@suse.de> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-07-29 21:42:43 +02:00
if (calgary_found) {
iommu_detected = 1;
calgary_detected = 1;
printk(KERN_INFO "PCI-DMA: Calgary IOMMU detected.\n");
printk(KERN_INFO "PCI-DMA: Calgary TCE table spec is %d, "
"CONFIG_IOMMU_DEBUG is %s.\n", specified_table_size,
debugging ? "enabled" : "disabled");
}
return;
cleanup:
for (--bus; bus >= 0; --bus) {
struct calgary_bus_info *info = &bus_info[bus];
if (info->tce_space)
free_tce_table(info->tce_space);
}
}
int __init calgary_iommu_init(void)
{
int ret;
if (no_iommu || swiotlb)
return -ENODEV;
if (!calgary_detected)
return -ENODEV;
/* ok, we're trying to use Calgary - let's roll */
printk(KERN_INFO "PCI-DMA: Using Calgary IOMMU\n");
ret = calgary_init();
if (ret) {
printk(KERN_ERR "PCI-DMA: Calgary init failed %d, "
"falling back to no_iommu\n", ret);
if (end_pfn > MAX_DMA32_PFN)
printk(KERN_ERR "WARNING more than 4GB of memory, "
"32bit PCI may malfunction.\n");
return ret;
}
force_iommu = 1;
bad_dma_address = 0x0;
dma_ops = &calgary_dma_ops;
return 0;
}
static int __init calgary_parse_options(char *p)
{
unsigned int bridge;
size_t len;
char* endp;
while (*p) {
if (!strncmp(p, "64k", 3))
specified_table_size = TCE_TABLE_SIZE_64K;
else if (!strncmp(p, "128k", 4))
specified_table_size = TCE_TABLE_SIZE_128K;
else if (!strncmp(p, "256k", 4))
specified_table_size = TCE_TABLE_SIZE_256K;
else if (!strncmp(p, "512k", 4))
specified_table_size = TCE_TABLE_SIZE_512K;
else if (!strncmp(p, "1M", 2))
specified_table_size = TCE_TABLE_SIZE_1M;
else if (!strncmp(p, "2M", 2))
specified_table_size = TCE_TABLE_SIZE_2M;
else if (!strncmp(p, "4M", 2))
specified_table_size = TCE_TABLE_SIZE_4M;
else if (!strncmp(p, "8M", 2))
specified_table_size = TCE_TABLE_SIZE_8M;
len = strlen("translate_empty_slots");
if (!strncmp(p, "translate_empty_slots", len))
translate_empty_slots = 1;
len = strlen("disable");
if (!strncmp(p, "disable", len)) {
p += len;
if (*p == '=')
++p;
if (*p == '\0')
break;
bridge = simple_strtol(p, &endp, 0);
if (p == endp)
break;
[PATCH] x86_64: Calgary IOMMU - Multi-Node NULL pointer dereference fix Calgary hits a NULL pointer dereference when booting in a multi-chassis NUMA system. See Redhat bugzilla number 198498, found by Konrad Rzeszutek (konradr@redhat.com). There are many issues that had to be resolved to fix this problem. Firstly when I originally wrote the code to handle NUMA systems, I had a large misunderstanding that was not corrected until now. That was that I thought the "number of nodes online" referred to number of physical systems connected. So that if NUMA was disabled, there would only be 1 node and it would only show that node's PCI bus. In reality if NUMA is disabled, the system displays all of the connected chassis as one node but is only ignorant of the delays in accessing main memory. Therefore, references to num_online_nodes() and MAX_NUMNODES are incorrect and need to be set to the maximum number of nodes that can be accessed (which are 8). I created a variable, MAX_NUM_CHASSIS, and set it to 8 to fix this. Secondly, when walking the PCI in detect_calgary, the code only checked the first "slot" when looking to see if a device is present. This will work for most cases, but unfortunately it isn't always the case. In the NUMA MXE drawers, there are USB devices present on the 3rd slot (with slot 1 being empty). So, to work around this, all slots (up to 8) are scanned to see if there are any devices present. Lastly, the bus is being enumerated on large systems in a different way the we originally thought. This throws the ugly logic we had out the window. To more elegantly handle this, I reorganized the kva array to be sparse (which removed the need to have any bus number to kva slot logic in tce.c) and created a secondary space array to contain the bus number to phb mapping. With these changes Calgary boots on an x460 with 4 nodes with and without NUMA enabled. Signed-off-by: Jon Mason <jdmason@us.ibm.com> Signed-off-by: Muli Ben-Yehuda <muli@il.ibm.com> Signed-off-by: Andi Kleen <ak@suse.de> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-07-29 21:42:43 +02:00
if (bridge < MAX_PHB_BUS_NUM) {
printk(KERN_INFO "Calgary: disabling "
"translation for PHB %#x\n", bridge);
bus_info[bridge].translation_disabled = 1;
}
}
p = strpbrk(p, ",");
if (!p)
break;
p++; /* skip ',' */
}
return 1;
}
__setup("calgary=", calgary_parse_options);