RE: [BK PATCH] PCI Express patches for 2.4.27-pre3

RE: [BK PATCH] PCI Express patches for 2.4.27-pre3

[Durairaj, Sundarapandian]

Hi Andi,

That's a very good point. I agree for x86_64 we can get rid of dynamic
fixmap and can use ioremap like in  2.6. ( I think it's a fall over from
the i386 version patch)

Please find a updated patch, this closely follows 2.6 version. This
patch is for 2.4.27-pre3 kernel on both x86 and x86-64 architecture and
tested in PCI Express platforms of both x86 and x86_64 architecture.

	 
I think its important that we have this patch for 2.4 kernel as well, as
it will enable the PCI express devices to access extended config space
(above 256 bytes), where all Advance feature of PCI Express config
registers resides. Also this patch is backward compatible in any case,
meaning the user can still fall back to direct config (Existing) access
method, by selecting proper kernel option.

One other option, is we can keep the existing direct access as default
and provide this (MCFG)access method as an kernel configurable option.
Please let me know your opinion on this?


Thanks,
Sundar

-----Original Message-----
From: owner-linux-pci@atrey.karlin.mff.cuni.cz
[mailto:owner-linux-pci@atrey.karlin.mff.cuni.cz] On Behalf Of Andi
Kleen
Sent: Tuesday, May 25, 2004 3:52 AM
To: Greg KH
Cc: linux-kernel@vger.kernel.org; linux-pci@atrey.karlin.mff.cuni.cz;
marcelo.tosatti@cyclades.com
Subject: Re: [BK PATCH] PCI Express patches for 2.4.27-pre3

Greg KH <greg@kroah.com> writes:
>  obj-y			+= pci-pc.o pci-irq.o
> diff -Nru a/arch/x86_64/kernel/mmconfig.c
b/arch/x86_64/kernel/mmconfig.c
> --- /dev/null	Wed Dec 31 16:00:00 1969
> +++ b/arch/x86_64/kernel/mmconfig.c	Mon May 24 13:52:10 2004


> +static inline void pci_exp_set_dev_base(int bus, int devfn)
> +{
> +	u32 dev_base = pci_mmcfg_base_addr | (bus << 20) | (devfn <<
12);
> +	if (dev_base != mmcfg_last_accessed_device) {
> +		mmcfg_last_accessed_device = dev_base;
> +		set_fixmap(FIX_PCIE_MCFG, dev_base);
> +	}

Please no dynamic fixmap crap on x86-64. Do it like 2.6 does  -
ioremap()
the complete mmconfig aperture once and just just reference it directly.

Then you can also get rid of the spinlocks in the actual access
functions,
since everything will be stateless.

-Andi


#################### Patch Start #############################

diff -Naur linux-2.4.27-pre3/arch/i386/config.in
linux-2.4.27-pre3-patched/arch/i386/config.in
--- linux-2.4.27-pre3/arch/i386/config.in	2004-05-25
16:44:16.000000000 +0530
+++ linux-2.4.27-pre3-patched/arch/i386/config.in	2004-05-25
16:49:40.000000000 +0530
@@ -286,6 +286,7 @@
       choice '  PCI access mode' \
 	"BIOS		CONFIG_PCI_GOBIOS	\
 	 Direct		CONFIG_PCI_GODIRECT	\
+	 MMConfig 	CONFIG_PCI_GOMMCONFIG	\
 	 Any		CONFIG_PCI_GOANY"	Any
       if [ "$CONFIG_PCI_GOBIOS" = "y" -o "$CONFIG_PCI_GOANY" = "y" ];
then
          define_bool CONFIG_PCI_BIOS y
@@ -293,6 +294,9 @@
       if [ "$CONFIG_PCI_GODIRECT" = "y" -o "$CONFIG_PCI_GOANY" = "y" ];
then
          define_bool CONFIG_PCI_DIRECT y
       fi
+      if [ "$CONFIG_PCI_GOMMCONFIG" = "y" -o "$CONFIG_PCI_GOANY" = "y"
]; then
+         define_bool CONFIG_PCI_MMCONFIG y
+      fi
    fi
    bool 'ISA bus support' CONFIG_ISA
 fi
diff -Naur linux-2.4.27-pre3/arch/i386/kernel/Makefile
linux-2.4.27-pre3-patched/arch/i386/kernel/Makefile
--- linux-2.4.27-pre3/arch/i386/kernel/Makefile	2004-05-25
16:44:16.000000000 +0530
+++ linux-2.4.27-pre3-patched/arch/i386/kernel/Makefile	2004-05-25
16:49:40.000000000 +0530
@@ -30,6 +30,10 @@
 endif
 endif
 
+ifdef CONFIG_PCI_MMCONFIG
+obj-y			+= mmconfig.o
+endif
+
 obj-$(CONFIG_MCA)		+= mca.o
 obj-$(CONFIG_MTRR)		+= mtrr.o
 obj-$(CONFIG_X86_MSR)		+= msr.o
diff -Naur linux-2.4.27-pre3/arch/i386/kernel/mmconfig.c
linux-2.4.27-pre3-patched/arch/i386/kernel/mmconfig.c
--- linux-2.4.27-pre3/arch/i386/kernel/mmconfig.c	1970-01-01
05:30:00.000000000 +0530
+++ linux-2.4.27-pre3-patched/arch/i386/kernel/mmconfig.c
2004-05-25 16:49:40.000000000 +0530
@@ -0,0 +1,178 @@
+/*
+ * mmconfig.c - Low-level direct PCI config space access via MMCONFIG
+ */
+
+#include <linux/pci.h>
+#include <linux/init.h>
+#include "pci-i386.h"
+
+/* The physical address of the MMCONFIG aperture.  Set from ACPI
tables. */
+extern u32 pci_mmcfg_base_addr;
+
+#define mmcfg_virt_addr (fix_to_virt(FIX_PCIE_MCFG))
+
+/* The base address of the last MMCONFIG device accessed */
+static u32 mmcfg_last_accessed_device;
+
+/*
+ * Functions for accessing PCI configuration space with MMCONFIG
accesses
+ */
+
+static inline void pci_exp_set_dev_base(int bus, int devfn)
+{
+	u32 dev_base = pci_mmcfg_base_addr | (bus << 20) | (devfn <<
12);
+	if (dev_base != mmcfg_last_accessed_device) {
+		mmcfg_last_accessed_device = dev_base;
+		set_fixmap(FIX_PCIE_MCFG, dev_base);
+	}
+}
+
+static int pci_mmcfg_read(int seg, int bus, int devfn, int reg, int
len, u32 *value)
+{
+	unsigned long flags;
+
+	if (!value || (bus > 255) || (devfn > 255) || (reg > 4095))
+		return -EINVAL;
+
+	spin_lock_irqsave(&pci_config_lock, flags);
+
+	pci_exp_set_dev_base(bus, devfn);
+
+	switch (len) {
+	case 1:
+		*value = readb(mmcfg_virt_addr + reg);
+		break;
+	case 2:
+		*value = readw(mmcfg_virt_addr + reg);
+		break;
+	case 4:
+		*value = readl(mmcfg_virt_addr + reg);
+		break;
+	}
+
+	spin_unlock_irqrestore(&pci_config_lock, flags);
+
+	return 0;
+}
+
+static int pci_mmcfg_write(int seg, int bus, int devfn, int reg, int
len, u32 value)
+{
+	unsigned long flags;
+
+	if ((bus > 255) || (devfn > 255) || (reg > 4095)) 
+		return -EINVAL;
+
+	spin_lock_irqsave(&pci_config_lock, flags);
+
+	pci_exp_set_dev_base(bus, devfn);
+
+	switch (len) {
+	case 1:
+		writeb(value, mmcfg_virt_addr + reg);
+		break;
+	case 2:
+		writew(value, mmcfg_virt_addr + reg);
+		break;
+	case 4:
+		writel(value, mmcfg_virt_addr + reg);
+		break;
+	}
+
+	/* Dummy read to flush PCI write */
+	readl(mmcfg_virt_addr);
+
+	spin_unlock_irqrestore(&pci_config_lock, flags);
+
+	return 0;
+}
+
+static int pci_mmcfg_read_config_byte(struct pci_dev *dev, int where,
u8 *value)
+{
+	int result; 
+	u32 data;
+
+	result = pci_mmcfg_read(0, dev->bus->number, dev->devfn, 
+				where, 1, &data);
+
+	*value = (u8)data;
+
+	return result;
+}
+
+static int pci_mmcfg_read_config_word(struct pci_dev *dev, int where,
u16 *value)
+{
+	int result; 
+	u32 data;
+
+	result = pci_mmcfg_read(0, dev->bus->number, dev->devfn, 
+ 				where, 2, &data);
+
+	*value = (u16)data;
+
+	return result;
+}
+
+static int pci_mmcfg_read_config_dword(struct pci_dev *dev, int where,
u32 *value)
+{
+	return pci_mmcfg_read(0, dev->bus->number, dev->devfn, 
+				where, 4, value);
+}
+
+static int pci_mmcfg_write_config_byte(struct pci_dev *dev, int where,
u8 value)
+{
+	return pci_mmcfg_write(0, dev->bus->number, dev->devfn, 
+				where, 1, value);
+}
+
+static int pci_mmcfg_write_config_word(struct pci_dev *dev, int where,
u16 value)
+{
+	return pci_mmcfg_write(0, dev->bus->number, dev->devfn, 
+				where, 2, value);
+}
+
+static int pci_mmcfg_write_config_dword(struct pci_dev *dev, int where,
u32 value)
+{
+	return pci_mmcfg_write(0, dev->bus->number, dev->devfn, 
+				where, 4, value);
+}
+
+static struct pci_ops pci_mmcfg = {
+	pci_mmcfg_read_config_byte,
+	pci_mmcfg_read_config_word,
+	pci_mmcfg_read_config_dword,
+	pci_mmcfg_write_config_byte,
+	pci_mmcfg_write_config_word,
+	pci_mmcfg_write_config_dword
+};
+
+struct pci_ops * __devinit pci_mmcfg_init(void)
+{
+	if ((pci_probe & PCI_PROBE_MMCONF) == 0)
+	{
+		goto out;
+	}
+	if (!pci_mmcfg_base_addr)
+	{
+		goto out;
+	}
+
+	printk(KERN_INFO "PCI: Using MMCONFIG\n");
+	pci_probe = (pci_probe & ~PCI_PROBE_MASK) | PCI_PROBE_MMCONF;
+
+	return &pci_mmcfg;
+
+out:
+	return NULL;
+}
+
+int pci_mcfg_read(int seg, int bus, int dev, int fn, int reg, int len, 
+				u32 *value)
+{
+	pci_mmcfg_read(seg, bus, PCI_DEVFN(dev, fn), reg, len, value);
+}
+
+int pci_mcfg_write(int seg, int bus, int dev, int fn, int reg, int len,

+				u32 *value)
+{
+	pci_mmcfg_write(seg, bus, PCI_DEVFN(dev, fn), reg, len, value);
+}
diff -Naur linux-2.4.27-pre3/arch/i386/kernel/pci-i386.h
linux-2.4.27-pre3-patched/arch/i386/kernel/pci-i386.h
--- linux-2.4.27-pre3/arch/i386/kernel/pci-i386.h	2004-05-25
16:44:16.000000000 +0530
+++ linux-2.4.27-pre3-patched/arch/i386/kernel/pci-i386.h
2004-05-25 16:49:40.000000000 +0530
@@ -15,6 +15,8 @@
 #define PCI_PROBE_BIOS		0x0001
 #define PCI_PROBE_CONF1		0x0002
 #define PCI_PROBE_CONF2		0x0004
+#define PCI_PROBE_MMCONF	0x0008
+#define PCI_PROBE_MASK		0x000f
 #define PCI_NO_SORT		0x0100
 #define PCI_BIOS_SORT		0x0200
 #define PCI_NO_CHECKS		0x0400
@@ -23,6 +25,7 @@
 #define PCI_ASSIGN_ALL_BUSSES	0x4000
 
 extern unsigned int pci_probe;
+extern spinlock_t pci_config_lock;
 
 /* pci-i386.c */
 
@@ -69,3 +72,10 @@
 void pcibios_irq_init(void);
 void pcibios_fixup_irqs(void);
 void pcibios_enable_irq(struct pci_dev *dev);
+
+/* mmconfig.c */
+int pci_mcfg_read(int seg, int bus, int dev, int fn, int reg, int len, 
+				u32 *value);
+int pci_mcfg_write(int seg, int bus, int dev, int fn, int reg, int len,

+				u32 *value);
+struct pci_ops * __devinit pci_mmcfg_init(void);
diff -Naur linux-2.4.27-pre3/arch/i386/kernel/pci-pc.c
linux-2.4.27-pre3-patched/arch/i386/kernel/pci-pc.c
--- linux-2.4.27-pre3/arch/i386/kernel/pci-pc.c	2004-05-25
16:48:45.000000000 +0530
+++ linux-2.4.27-pre3-patched/arch/i386/kernel/pci-pc.c	2004-05-25
16:49:40.000000000 +0530
@@ -20,7 +20,8 @@
 
 #include "pci-i386.h"
 
-unsigned int pci_probe = PCI_PROBE_BIOS | PCI_PROBE_CONF1 |
PCI_PROBE_CONF2;
+unsigned int pci_probe = PCI_PROBE_BIOS | PCI_PROBE_CONF1 |
PCI_PROBE_CONF2 |
+				PCI_PROBE_MMCONF;
 
 int pcibios_last_bus = -1;
 struct pci_bus *pci_root_bus = NULL;
@@ -45,7 +46,7 @@
  * This interrupt-safe spinlock protects all accesses to PCI
  * configuration space.
  */
-static spinlock_t pci_config_lock = SPIN_LOCK_UNLOCKED;
+spinlock_t pci_config_lock = SPIN_LOCK_UNLOCKED;
 
 
 /*
@@ -1446,6 +1447,33 @@
 	}
 #endif
 
+#ifdef CONFIG_PCI_MMCONFIG
+	pci_root_ops = pci_mmcfg_init();
+	if (pci_root_ops){
+		pci_config_read = pci_mcfg_read;
+		pci_config_write = pci_mcfg_write;
+		return;
+	}
+#ifdef CONFIG_PCI_DIRECT
+	/*
+	 * PCI Express option is enabled in NON-PCI Express
+	 * platforms. Reset the PCI root ops
+	 */
+	if ((pci_probe & (PCI_PROBE_CONF1 | PCI_PROBE_CONF2)) 
+		&& (tmp = pci_check_direct())) {
+		pci_root_ops = tmp;
+		if (pci_root_ops == &pci_direct_conf1) {
+			pci_config_read = pci_conf1_read;
+			pci_config_write = pci_conf1_write;
+		}
+		else {
+			pci_config_read = pci_conf2_read;
+			pci_config_write = pci_conf2_write;
+		}
+	}
+#endif
+#endif
+
 	return;
 }
 
@@ -1523,6 +1551,12 @@
 		return NULL;
 	}
 #endif
+#ifdef CONFIG_PCI_MMCONFIG
+	else if (!strcmp(str, "nommconf")) {
+		pci_probe &= ~PCI_PROBE_MMCONF;
+		return NULL;
+	}
+#endif
 	else if (!strcmp(str, "rom")) {
 		pci_probe |= PCI_ASSIGN_ROMS;
 		return NULL;
diff -Naur linux-2.4.27-pre3/arch/x86_64/config.in
linux-2.4.27-pre3-patched/arch/x86_64/config.in
--- linux-2.4.27-pre3/arch/x86_64/config.in	2004-05-25
16:44:15.000000000 +0530
+++ linux-2.4.27-pre3-patched/arch/x86_64/config.in	2004-05-25
16:49:40.000000000 +0530
@@ -95,7 +95,16 @@
 bool 'Networking support' CONFIG_NET
 bool 'PCI support' CONFIG_PCI
 if [ "$CONFIG_PCI" = "y" ]; then
-   define_bool CONFIG_PCI_DIRECT y 
+      choice '  PCI access mode' \
+	"Direct		CONFIG_PCI_GODIRECT 	\
+	 MMConfig 	CONFIG_PCI_GOMMCONFIG 	\
+	 Any		CONFIG_PCI_GOANY"	Any
+      if [ "$CONFIG_PCI_GODIRECT" = "y" -o "$CONFIG_PCI_GOANY" = "y" ];
then
+         define_bool CONFIG_PCI_DIRECT y
+      fi
+      if [ "$CONFIG_PCI_GOMMCONFIG" = "y" -o "$CONFIG_PCI_GOANY" = "y"
]; then
+         define_bool CONFIG_PCI_MMCONFIG y
+      fi
 fi
 
 source drivers/pci/Config.in
diff -Naur linux-2.4.27-pre3/arch/x86_64/kernel/Makefile
linux-2.4.27-pre3-patched/arch/x86_64/kernel/Makefile
--- linux-2.4.27-pre3/arch/x86_64/kernel/Makefile	2004-05-25
16:44:15.000000000 +0530
+++ linux-2.4.27-pre3-patched/arch/x86_64/kernel/Makefile
2004-05-25 16:49:40.000000000 +0530
@@ -21,6 +21,10 @@
 		pci-dma.o x8664_ksyms.o i387.o syscall.o vsyscall.o \
 		setup64.o e820.o warmreboot.o
 
+ifdef CONFIG_PCI_MMCONFIG
+obj-y			+= mmconfig.o
+endif
+
 ifdef CONFIG_PCI
 obj-y			+= pci-x86_64.o
 obj-y			+= pci-pc.o pci-irq.o
diff -Naur linux-2.4.27-pre3/arch/x86_64/kernel/mmconfig.c
linux-2.4.27-pre3-patched/arch/x86_64/kernel/mmconfig.c
--- linux-2.4.27-pre3/arch/x86_64/kernel/mmconfig.c	1970-01-01
05:30:00.000000000 +0530
+++ linux-2.4.27-pre3-patched/arch/x86_64/kernel/mmconfig.c
2004-05-25 17:06:59.000000000 +0530
@@ -0,0 +1,173 @@
+/*
+ * mmconfig.c - Low-level direct PCI config space access via MMCONFIG
+ */
+
+#include <linux/pci.h>
+#include <linux/init.h>
+#include "pci-x86_64.h"
+
+#define MMCONFIG_APER_SIZE (256*1024*1024)
+
+/* The physical address of the MMCONFIG aperture.  Set from ACPI
tables. */
+extern u32 pci_mmcfg_base_addr;
+
+/* Static virtual mapping of the MMCONFIG aperture */
+char *pci_mmcfg_virt;
+
+/*
+ * Functions for accessing PCI configuration space with MMCONFIG
accesses
+ */
+
+static inline char* pci_dev_base(int bus, int devfn)
+{
+	return pci_mmcfg_virt + ((bus << 20) | (devfn << 12));
+}
+
+static int pci_mmcfg_read(int seg, int bus, int devfn, int reg, int
len, 
+				u32 *value)
+{
+	char *addr = pci_dev_base(bus, devfn); 
+	
+	if (!value || (bus > 255) || (devfn > 255) || (reg > 4095))
+		return -EINVAL;
+
+	switch (len) {
+	case 1:
+		*value = readb(addr + reg);
+		break;
+	case 2:
+		*value = readw(addr + reg);
+		break;
+	case 4:
+		*value = readl(addr + reg);
+		break;
+	}
+
+	return 0;
+}
+
+static int pci_mmcfg_write(int seg, int bus, int devfn, int reg, int
len, 
+				u32 value)
+{
+	char *addr = pci_dev_base(bus, devfn); 
+	
+	if ((bus > 255) || (devfn > 255) || (reg > 4095)) 
+		return -EINVAL;
+
+	switch (len) {
+	case 1:
+		writeb(value, addr + reg);
+		break;
+	case 2:
+		writew(value, addr + reg);
+		break;
+	case 4:
+		writel(value, addr + reg);
+		break;
+	}
+
+	/* Dummy read to flush PCI write */
+	readl(addr);
+
+	return 0;
+}
+
+static int pci_mmcfg_read_config_byte(struct pci_dev *dev, int where, 
+					u8 *value)
+{
+	int result; 
+	u32 data;
+
+	result = pci_mmcfg_read(0, dev->bus->number, dev->devfn, 
+				where, 1, &data);
+
+	*value = (u8)data;
+
+	return result;
+}
+
+static int pci_mmcfg_read_config_word(struct pci_dev *dev, int where, 
+					u16 *value)
+{
+	int result; 
+	u32 data;
+
+	result = pci_mmcfg_read(0, dev->bus->number, dev->devfn, 
+ 				where, 2, &data);
+
+	*value = (u16)data;
+
+	return result;
+}
+
+static int pci_mmcfg_read_config_dword(struct pci_dev *dev, int where, 
+					u32 *value)
+{
+	return pci_mmcfg_read(0, dev->bus->number, dev->devfn, 
+				where, 4, value);
+}
+
+static int pci_mmcfg_write_config_byte(struct pci_dev *dev, int where, 
+					u8 value)
+{
+	return pci_mmcfg_write(0, dev->bus->number, dev->devfn, 
+				where, 1, value);
+}
+
+static int pci_mmcfg_write_config_word(struct pci_dev *dev, int where, 
+					u16 value)
+{
+	return pci_mmcfg_write(0, dev->bus->number, dev->devfn, 
+				where, 2, value);
+}
+
+static int pci_mmcfg_write_config_dword(struct pci_dev *dev, int where,

+					u32 value)
+{
+	return pci_mmcfg_write(0, dev->bus->number, dev->devfn, 
+				where, 4, value);
+}
+
+static struct pci_ops pci_mmcfg = {
+	pci_mmcfg_read_config_byte,
+	pci_mmcfg_read_config_word,
+	pci_mmcfg_read_config_dword,
+	pci_mmcfg_write_config_byte,
+	pci_mmcfg_write_config_word,
+	pci_mmcfg_write_config_dword
+};
+
+struct pci_ops * __devinit pci_mmcfg_init(void)
+{
+	if ((pci_probe & PCI_PROBE_MMCONF) == 0)
+		goto out;
+	if (!pci_mmcfg_base_addr)
+		goto out;
+
+	/* i386 doesn't do _nocache right now */
+	pci_mmcfg_virt = ioremap_nocache(pci_mmcfg_base_addr, 
+						MMCONFIG_APER_SIZE);
+	if (!pci_mmcfg_virt) { 
+		printk("PCI: Cannot map mmconfig aperture\n");
+		goto out;
+	}	
+
+	printk(KERN_INFO "PCI: Using MMCONFIG at %x\n",
pci_mmcfg_base_addr);
+
+	return &pci_mmcfg;
+
+out:
+	return NULL;
+}
+
+int pci_mcfg_read(int seg, int bus, int dev, int fn, int reg, int len, 
+				u32 *value)
+{
+	pci_mmcfg_read(seg, bus, PCI_DEVFN(dev, fn), reg, len, value);
+}
+
+int pci_mcfg_write(int seg, int bus, int dev, int fn, int reg, int len,

+				u32 *value)
+{
+	pci_mmcfg_write(seg, bus, PCI_DEVFN(dev, fn), reg, len, value);
+}
diff -Naur linux-2.4.27-pre3/arch/x86_64/kernel/pci-pc.c
linux-2.4.27-pre3-patched/arch/x86_64/kernel/pci-pc.c
--- linux-2.4.27-pre3/arch/x86_64/kernel/pci-pc.c	2004-05-25
16:48:47.000000000 +0530
+++ linux-2.4.27-pre3-patched/arch/x86_64/kernel/pci-pc.c
2004-05-25 16:49:40.000000000 +0530
@@ -27,7 +27,7 @@
 
 #include "pci-x86_64.h"
 
-unsigned int pci_probe = PCI_PROBE_CONF1 | PCI_PROBE_CONF2;
+unsigned int pci_probe = PCI_PROBE_CONF1 | PCI_PROBE_CONF2 |
PCI_PROBE_MMCONF;
 
 int pcibios_last_bus = -1;
 struct pci_bus *pci_root_bus;
@@ -566,6 +566,31 @@
 		printk("??? no pci access\n"); 
 #endif
 
+#ifdef CONFIG_PCI_MMCONFIG
+	pci_root_ops = pci_mmcfg_init();
+	if (pci_root_ops) {
+		pci_config_read = pci_mcfg_read;
+		pci_config_write = pci_mcfg_write;
+	}
+#ifdef CONFIG_PCI_DIRECT
+	else {
+	/*
+	 * PCI Express option is enabled in NON-PCI Express
+	 * platforms. Reset the PCI root ops
+	 */
+		pci_root_ops = pci_check_direct();
+		if (pci_root_ops == &pci_direct_conf1) {
+			pci_config_read = pci_conf1_read;
+			pci_config_write = pci_conf1_write;
+		}
+		else {
+			pci_config_read = pci_conf2_read;
+			pci_config_write = pci_conf2_write;
+		}
+	}
+#endif
+#endif
+
 	return;
 }
 
@@ -580,6 +605,19 @@
 	if (pci_probe & (PCI_PROBE_CONF1 | PCI_PROBE_CONF2))
 		dir = pci_check_direct();
 #endif
+
+#ifdef CONFIG_PCI_MMCONFIG
+	dir = pci_mmcfg_init();
+#ifdef CONFIG_PCI_DIRECT
+	/*
+	 * PCI Express option is enabled in NON-PCI Express
+	 * platforms. Reset the PCI root ops
+	 */
+	if (!dir) {
+		dir = pci_check_direct();
+	}
+#endif
+#endif
 	if (dir)
 		pci_root_ops = dir;
 	else {
@@ -633,6 +671,12 @@
 		return NULL;
 	}
 #endif
+#ifdef CONFIG_PCI_MMCONFIG
+	else if (!strcmp(str, "nommconf")) {
+		pci_probe &= ~PCI_PROBE_MMCONF;
+		return NULL;
+	}
+#endif
 	else if (!strcmp(str, "rom")) {
 		pci_probe |= PCI_ASSIGN_ROMS;
 		return NULL;
diff -Naur linux-2.4.27-pre3/arch/x86_64/kernel/pci-x86_64.h
linux-2.4.27-pre3-patched/arch/x86_64/kernel/pci-x86_64.h
--- linux-2.4.27-pre3/arch/x86_64/kernel/pci-x86_64.h	2004-05-25
16:44:15.000000000 +0530
+++ linux-2.4.27-pre3-patched/arch/x86_64/kernel/pci-x86_64.h
2004-05-25 16:49:40.000000000 +0530
@@ -15,6 +15,8 @@
 #define PCI_PROBE_BIOS		0x0001
 #define PCI_PROBE_CONF1		0x0002
 #define PCI_PROBE_CONF2		0x0004
+#define PCI_PROBE_MMCONF	0x0008
+#define PCI_PROBE_MASK		0x000f
 #define PCI_NO_SORT		0x0100
 #define PCI_BIOS_SORT		0x0200
 #define PCI_NO_CHECKS		0x0400
@@ -74,3 +76,10 @@
 void pcibios_enable_irq(struct pci_dev *dev);
 
 void pci_iommu_init(void);
+
+/* mmconfig.c */
+int pci_mcfg_read(int seg, int bus, int dev, int fn, int reg, int len, 
+				u32 *value);
+int pci_mcfg_write(int seg, int bus, int dev, int fn, int reg, int len,

+				u32 *value);
+struct pci_ops * __devinit pci_mmcfg_init(void);
diff -Naur linux-2.4.27-pre3/drivers/pci/pci.c
linux-2.4.27-pre3-patched/drivers/pci/pci.c
--- linux-2.4.27-pre3/drivers/pci/pci.c	2004-05-25 16:48:51.000000000
+0530
+++ linux-2.4.27-pre3-patched/drivers/pci/pci.c	2004-05-25
16:49:40.000000000 +0530
@@ -36,6 +36,9 @@
 #define DBG(x...)
 #endif
 
+#define PCI_CFG_SPACE_SIZE	256
+#define PCI_CFG_SPACE_EXP_SIZE	4096
+
 LIST_HEAD(pci_root_buses);
 LIST_HEAD(pci_devices);
 
@@ -165,6 +168,8 @@
  *  %PCI_CAP_ID_CHSWP        CompactPCI HotSwap 
  *
  *  %PCI_CAP_ID_PCIX         PCI-X
+ *
+ *  %PCI_CAP_ID_EXP          PCI Express
  */
 int
 pci_find_capability(struct pci_dev *dev, int cap)
@@ -199,6 +204,21 @@
 	return 0;
 }
 
+/**
+ * pci_cfg_space_size - get the configuration space size of the PCI
device.
+ * Regular PCI devices have 256 bytes, but PCI-X 2 and PCI Express
devices
+ * have 4096 bytes.
+ */
+static int pci_cfg_space_size(struct pci_dev *dev)
+{
+	int pos;
+
+	pos = pci_find_capability(dev, PCI_CAP_ID_EXP);
+	if (pos)
+		return PCI_CFG_SPACE_EXP_SIZE;
+	
+	return PCI_CFG_SPACE_SIZE;
+}
 
 /**
  * pci_find_parent_resource - return resource region of parent bus of
given region
@@ -1427,6 +1447,9 @@
 		dev->class = PCI_CLASS_NOT_DEFINED;
 	}
 
+	/* Get the config space size */
+	dev->cfg_size = pci_cfg_space_size(dev);
+
 	/* We found a fine healthy device, go go go... */
 	return 0;
 }
diff -Naur linux-2.4.27-pre3/drivers/pci/proc.c
linux-2.4.27-pre3-patched/drivers/pci/proc.c
--- linux-2.4.27-pre3/drivers/pci/proc.c	2004-05-25
16:44:07.000000000 +0530
+++ linux-2.4.27-pre3-patched/drivers/pci/proc.c	2004-05-25
16:49:41.000000000 +0530
@@ -16,12 +16,13 @@
 #include <asm/uaccess.h>
 #include <asm/byteorder.h>
 
-#define PCI_CFG_SPACE_SIZE 256
-
 static loff_t
 proc_bus_pci_lseek(struct file *file, loff_t off, int whence)
 {
 	loff_t new;
+	const struct inode *ino = file->f_dentry->d_inode;
+	const struct proc_dir_entry *dp = ino->u.generic_ip;
+	struct pci_dev *dev = dp->data;
 
 	switch (whence) {
 	case 0:
@@ -31,12 +32,12 @@
 		new = file->f_pos + off;
 		break;
 	case 2:
-		new = PCI_CFG_SPACE_SIZE + off;
+		new = dev->cfg_size + off;
 		break;
 	default:
 		return -EINVAL;
 	}
-	if (new < 0 || new > PCI_CFG_SPACE_SIZE)
+	if (new < 0 || new > dev->cfg_size)
 		return -EINVAL;
 	return (file->f_pos = new);
 }
@@ -57,7 +58,7 @@
 	 */
 
 	if (capable(CAP_SYS_ADMIN))
-		size = PCI_CFG_SPACE_SIZE;
+		size = dev->cfg_size;
 	else if (dev->hdr_type == PCI_HEADER_TYPE_CARDBUS)
 		size = 128;
 	else
@@ -132,12 +133,12 @@
 	int pos = *ppos;
 	int cnt;
 
-	if (pos >= PCI_CFG_SPACE_SIZE)
+	if (pos >= dev->cfg_size)
 		return 0;
-	if (nbytes >= PCI_CFG_SPACE_SIZE)
-		nbytes = PCI_CFG_SPACE_SIZE;
-	if (pos + nbytes > PCI_CFG_SPACE_SIZE)
-		nbytes = PCI_CFG_SPACE_SIZE - pos;
+	if (nbytes >= dev->cfg_size)
+		nbytes = dev->cfg_size;
+	if (pos + nbytes > dev->cfg_size)
+		nbytes = dev->cfg_size - pos;
 	cnt = nbytes;
 
 	if (!access_ok(VERIFY_READ, buf, cnt))
@@ -389,7 +390,7 @@
 		return -ENOMEM;
 	e->proc_fops = &proc_bus_pci_operations;
 	e->data = dev;
-	e->size = PCI_CFG_SPACE_SIZE;
+	e->size = dev->cfg_size;
 	return 0;
 }
 
diff -Naur linux-2.4.27-pre3/include/asm-i386/fixmap.h
linux-2.4.27-pre3-patched/include/asm-i386/fixmap.h
--- linux-2.4.27-pre3/include/asm-i386/fixmap.h	2004-05-25
16:43:56.000000000 +0530
+++ linux-2.4.27-pre3-patched/include/asm-i386/fixmap.h	2004-05-25
16:49:41.000000000 +0530
@@ -76,6 +76,9 @@
 	FIX_ACPI_BEGIN,
 	FIX_ACPI_END = FIX_ACPI_BEGIN + FIX_ACPI_PAGES - 1,
 #endif
+#ifdef CONFIG_PCI_MMCONFIG
+	FIX_PCIE_MCFG,
+#endif
 	__end_of_permanent_fixed_addresses,
 	/* temporary boot-time mappings, used before ioremap() is
functional */
 #define NR_FIX_BTMAPS	16
diff -Naur linux-2.4.27-pre3/include/linux/pci.h
linux-2.4.27-pre3-patched/include/linux/pci.h
--- linux-2.4.27-pre3/include/linux/pci.h	2004-05-25
16:48:53.000000000 +0530
+++ linux-2.4.27-pre3-patched/include/linux/pci.h	2004-05-25
16:49:41.000000000 +0530
@@ -401,6 +401,7 @@
 	/* device is compatible with these IDs */
 	unsigned short vendor_compatible[DEVICE_COUNT_COMPATIBLE];
 	unsigned short device_compatible[DEVICE_COUNT_COMPATIBLE];
+	int		cfg_size;	/* Size of configuration space
*/
 
 	/*
 	 * Instead of touching interrupt line and base address registers

#################### Patch END #############################

Re: [BK PATCH] PCI Express patches for 2.4.27-pre3

[Matthew Wilcox]

On Wed, May 26, 2004 at 11:59:43AM +0530, Durairaj, Sundarapandian wrote:
> I think its important that we have this patch for 2.4 kernel as well, as
> it will enable the PCI express devices to access extended config space
> (above 256 bytes), where all Advance feature of PCI Express config
> registers resides.

That's not a good reason to add it to 2.4.  If people want access to
that advanced stuff, they can upgrade to 2.6.

By the way, are you working on using any of these advanced features yet?
I'm waiting until I actually have some hardware in my hands before I
do anything.

-- 
"Next the statesmen will invent cheap lies, putting the blame upon 
the nation that is attacked, and every man will be glad of those
conscience-soothing falsities, and will diligently study them, and refuse
to examine any refutations of them; and thus he will by and by convince 
himself that the war is just, and will thank God for the better sleep 
he enjoys after this process of grotesque self-deception." -- Mark Twain

Re: [BK PATCH] PCI Express patches for 2.4.27-pre3

[Greg KH]

On Wed, May 26, 2004 at 11:59:43AM +0530, Durairaj, Sundarapandian wrote:
> I think its important that we have this patch for 2.4 kernel as well, as
> it will enable the PCI express devices to access extended config space
> (above 256 bytes), where all Advance feature of PCI Express config
> registers resides.

Are there any drivers or devices on the market today that need access to
this extended config space?

I agree that if customers want this kind of new features, they should
use the 2.6 kernel.

thanks,

greg k-h

Re: [BK PATCH] PCI Express patches for 2.4.27-pre3

[Jeff Garzik]

Greg KH wrote:
> On Wed, May 26, 2004 at 11:59:43AM +0530, Durairaj, Sundarapandian wrote:
> 
>>I think its important that we have this patch for 2.4 kernel as well, as
>>it will enable the PCI express devices to access extended config space
>>(above 256 bytes), where all Advance feature of PCI Express config
>>registers resides.
> 
> 
> Are there any drivers or devices on the market today that need access to
> this extended config space?


"need"?  Not AFAIK.  Not yet, anyway.

Most PCI-Ex devices I've seen are designed such that they work under 
OS's and drivers that do not support the extended configuration area.

	Jeff

Re: [BK PATCH] PCI Express patches for 2.4.27-pre3

[Andi Kleen]

Greg KH <greg@kroah.com> writes:
>  obj-y			+= pci-pc.o pci-irq.o
> diff -Nru a/arch/x86_64/kernel/mmconfig.c b/arch/x86_64/kernel/mmconfig.c
> --- /dev/null	Wed Dec 31 16:00:00 1969
> +++ b/arch/x86_64/kernel/mmconfig.c	Mon May 24 13:52:10 2004


> +static inline void pci_exp_set_dev_base(int bus, int devfn)
> +{
> +	u32 dev_base = pci_mmcfg_base_addr | (bus << 20) | (devfn << 12);
> +	if (dev_base != mmcfg_last_accessed_device) {
> +		mmcfg_last_accessed_device = dev_base;
> +		set_fixmap(FIX_PCIE_MCFG, dev_base);
> +	}

Please no dynamic fixmap crap on x86-64. Do it like 2.6 does  - ioremap()
the complete mmconfig aperture once and just just reference it directly.

Then you can also get rid of the spinlocks in the actual access functions,
since everything will be stateless.

-Andi

Re: [BK PATCH] PCI Express patches for 2.4.27-pre3

[Greg KH]

I'll let the original authors of this patch answer the question as to
why they did not implement it in this manner.

Dely?

thanks,

greg k-h

On Tue, May 25, 2004 at 12:21:59AM +0200, Andi Kleen wrote:
> Greg KH <greg@kroah.com> writes:
> >  obj-y			+= pci-pc.o pci-irq.o
> > diff -Nru a/arch/x86_64/kernel/mmconfig.c b/arch/x86_64/kernel/mmconfig.c
> > --- /dev/null	Wed Dec 31 16:00:00 1969
> > +++ b/arch/x86_64/kernel/mmconfig.c	Mon May 24 13:52:10 2004
> 
> 
> > +static inline void pci_exp_set_dev_base(int bus, int devfn)
> > +{
> > +	u32 dev_base = pci_mmcfg_base_addr | (bus << 20) | (devfn << 12);
> > +	if (dev_base != mmcfg_last_accessed_device) {
> > +		mmcfg_last_accessed_device = dev_base;
> > +		set_fixmap(FIX_PCIE_MCFG, dev_base);
> > +	}
> 
> Please no dynamic fixmap crap on x86-64. Do it like 2.6 does  - ioremap()
> the complete mmconfig aperture once and just just reference it directly.
> 
> Then you can also get rid of the spinlocks in the actual access functions,
> since everything will be stateless.
> 
> -Andi
> 
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

Re: [BK PATCH] PCI Express patches for 2.4.27-pre3

[Andi Kleen]

Marcelo Tosatti <marcelo.tosatti@cyclades.com> writes:

>
> major? the MMConfig support is minimal as I can see? 
>
>> > Marcelo, feel free to tell me otherwise if you do not want
>> > this in the 2.4 tree. 
>
> Is this code necessary for PCI-Express devices/busses to work properly?

No, it is completely optional right now.

Also we found that it sometimes triggers bugs that are not happening
with old style config space accesses. 

-Andi

Re: [BK PATCH] PCI Express patches for 2.4.27-pre3

[Andi Kleen]

Marcelo Tosatti <marcelo.tosatti@cyclades.com> writes:
>
> Yeap, I would prefer not to apply them at this time. For one, Arjan told
> me privately it can break XFree86 which accesses the PCI config space directly.
> Right?

Erm,no. In fact it could even fix XFree86 for that, because the way
they do it currently in user space is racy and the kernel doing it 
differently would fix that race. But I would still recommend to 
not apply it, since it is somewhat risky e.g. for triggering 
new hardware bugs and it is not really needed and would dilute the
message of freeze %) XFree86 should be fixed in user space to use the proper
access methods provided by the kernel (/proc/sys etc.) 

-Andi

[BK PATCH] PCI Express patches for 2.4.27-pre3

[Greg KH]

Hi,

Now that the ACPI portion of PCI Express support is in the 2.4 kernel
tree, I can send you the remaining portion to actually enable this
feature to work properly.  Here is the PCI Express support for i386 and
x86_64 platforms backported from 2.6 to the 2.4 kernel.

Please pull from:  bk://kernel.bkbits.net/gregkh/linux/pci-2.4

The raw patches will follow, for those who want to see them.

thanks,

greg k-h


 Documentation/Configure.help    |   13 +-
 arch/i386/config.in             |    4 
 arch/i386/kernel/Makefile       |    4 
 arch/i386/kernel/mmconfig.c     |  178 ++++++++++++++++++++++++++++++++++++++++
 arch/i386/kernel/pci-i386.h     |   10 ++
 arch/i386/kernel/pci-pc.c       |   38 ++++++++
 arch/x86_64/config.in           |   11 ++
 arch/x86_64/kernel/Makefile     |    4 
 arch/x86_64/kernel/mmconfig.c   |  170 ++++++++++++++++++++++++++++++++++++++
 arch/x86_64/kernel/pci-pc.c     |   46 ++++++++++
 arch/x86_64/kernel/pci-x86_64.h |    9 ++
 drivers/acpi/Config.in          |    2 
 drivers/pci/pci.c               |   23 +++++
 drivers/pci/proc.c              |   23 ++---
 include/asm-i386/fixmap.h       |    3 
 include/asm-x86_64/fixmap.h     |    3 
 include/linux/pci.h             |    1 
 17 files changed, 521 insertions(+), 21 deletions(-)
-----

<blujuice:us.ibm.com>:
  o PCI: PCI Express help in 2.4

Dely Sy:
  o PCI: PCI Express for 2.4

Re: [BK PATCH] PCI Express patches for 2.4.27-pre3

[Greg KH]

ChangeSet 1.1427, 2004/05/24 13:32:48-07:00, dlsy@snoqualmie.dp.intel.com

[PATCH] PCI: PCI Express for 2.4

Here is an updated patch for PCI Express support based off
of the 2.6 code.  Now that the ACPI support for this is in
the 2.4 tree, this patch can be accepted.


 arch/i386/config.in             |    4 
 arch/i386/kernel/Makefile       |    4 
 arch/i386/kernel/mmconfig.c     |  178 ++++++++++++++++++++++++++++++++++++++++
 arch/i386/kernel/pci-i386.h     |   10 ++
 arch/i386/kernel/pci-pc.c       |   38 ++++++++
 arch/x86_64/config.in           |   11 ++
 arch/x86_64/kernel/Makefile     |    4 
 arch/x86_64/kernel/mmconfig.c   |  170 ++++++++++++++++++++++++++++++++++++++
 arch/x86_64/kernel/pci-pc.c     |   46 ++++++++++
 arch/x86_64/kernel/pci-x86_64.h |    9 ++
 drivers/acpi/Config.in          |    2 
 drivers/pci/pci.c               |   23 +++++
 drivers/pci/proc.c              |   23 ++---
 include/asm-i386/fixmap.h       |    3 
 include/asm-x86_64/fixmap.h     |    3 
 include/linux/pci.h             |    1 
 16 files changed, 514 insertions(+), 15 deletions(-)


diff -Nru a/arch/i386/config.in b/arch/i386/config.in
--- a/arch/i386/config.in	Mon May 24 13:52:10 2004
+++ b/arch/i386/config.in	Mon May 24 13:52:10 2004
@@ -286,12 +286,16 @@
       choice '  PCI access mode' \
 	"BIOS		CONFIG_PCI_GOBIOS	\
 	 Direct		CONFIG_PCI_GODIRECT	\
+	 MMConfig 	CONFIG_PCI_GOMMCONFIG	\
 	 Any		CONFIG_PCI_GOANY"	Any
       if [ "$CONFIG_PCI_GOBIOS" = "y" -o "$CONFIG_PCI_GOANY" = "y" ]; then
          define_bool CONFIG_PCI_BIOS y
       fi
       if [ "$CONFIG_PCI_GODIRECT" = "y" -o "$CONFIG_PCI_GOANY" = "y" ]; then
          define_bool CONFIG_PCI_DIRECT y
+      fi
+      if [ "$CONFIG_PCI_GOMMCONFIG" = "y" -o "$CONFIG_PCI_GOANY" = "y" ]; then
+         define_bool CONFIG_PCI_MMCONFIG y
       fi
    fi
    bool 'ISA bus support' CONFIG_ISA
diff -Nru a/arch/i386/kernel/Makefile b/arch/i386/kernel/Makefile
--- a/arch/i386/kernel/Makefile	Mon May 24 13:52:10 2004
+++ b/arch/i386/kernel/Makefile	Mon May 24 13:52:10 2004
@@ -30,6 +30,10 @@
 endif
 endif
 
+ifdef CONFIG_PCI_MMCONFIG
+obj-y			+= mmconfig.o
+endif
+
 obj-$(CONFIG_MCA)		+= mca.o
 obj-$(CONFIG_MTRR)		+= mtrr.o
 obj-$(CONFIG_X86_MSR)		+= msr.o
diff -Nru a/arch/i386/kernel/mmconfig.c b/arch/i386/kernel/mmconfig.c
--- /dev/null	Wed Dec 31 16:00:00 1969
+++ b/arch/i386/kernel/mmconfig.c	Mon May 24 13:52:10 2004
@@ -0,0 +1,178 @@
+/*
+ * mmconfig.c - Low-level direct PCI config space access via MMCONFIG
+ */
+
+#include <linux/pci.h>
+#include <linux/init.h>
+#include "pci-i386.h"
+
+/* The physical address of the MMCONFIG aperture.  Set from ACPI tables. */
+extern u32 pci_mmcfg_base_addr;
+
+#define mmcfg_virt_addr (fix_to_virt(FIX_PCIE_MCFG))
+
+/* The base address of the last MMCONFIG device accessed */
+static u32 mmcfg_last_accessed_device;
+
+/*
+ * Functions for accessing PCI configuration space with MMCONFIG accesses
+ */
+
+static inline void pci_exp_set_dev_base(int bus, int devfn)
+{
+	u32 dev_base = pci_mmcfg_base_addr | (bus << 20) | (devfn << 12);
+	if (dev_base != mmcfg_last_accessed_device) {
+		mmcfg_last_accessed_device = dev_base;
+		set_fixmap(FIX_PCIE_MCFG, dev_base);
+	}
+}
+
+static int pci_mmcfg_read(int seg, int bus, int devfn, int reg, int len, u32 *value)
+{
+	unsigned long flags;
+
+	if (!value || (bus > 255) || (devfn > 255) || (reg > 4095))
+		return -EINVAL;
+
+	spin_lock_irqsave(&pci_config_lock, flags);
+
+	pci_exp_set_dev_base(bus, devfn);
+
+	switch (len) {
+	case 1:
+		*value = readb(mmcfg_virt_addr + reg);
+		break;
+	case 2:
+		*value = readw(mmcfg_virt_addr + reg);
+		break;
+	case 4:
+		*value = readl(mmcfg_virt_addr + reg);
+		break;
+	}
+
+	spin_unlock_irqrestore(&pci_config_lock, flags);
+
+	return 0;
+}
+
+static int pci_mmcfg_write(int seg, int bus, int devfn, int reg, int len, u32 value)
+{
+	unsigned long flags;
+
+	if ((bus > 255) || (devfn > 255) || (reg > 4095)) 
+		return -EINVAL;
+
+	spin_lock_irqsave(&pci_config_lock, flags);
+
+	pci_exp_set_dev_base(bus, devfn);
+
+	switch (len) {
+	case 1:
+		writeb(value, mmcfg_virt_addr + reg);
+		break;
+	case 2:
+		writew(value, mmcfg_virt_addr + reg);
+		break;
+	case 4:
+		writel(value, mmcfg_virt_addr + reg);
+		break;
+	}
+
+	/* Dummy read to flush PCI write */
+	readl(mmcfg_virt_addr);
+
+	spin_unlock_irqrestore(&pci_config_lock, flags);
+
+	return 0;
+}
+
+static int pci_mmcfg_read_config_byte(struct pci_dev *dev, int where, u8 *value)
+{
+	int result; 
+	u32 data;
+
+	result = pci_mmcfg_read(0, dev->bus->number, dev->devfn, 
+				where, 1, &data);
+
+	*value = (u8)data;
+
+	return result;
+}
+
+static int pci_mmcfg_read_config_word(struct pci_dev *dev, int where, u16 *value)
+{
+	int result; 
+	u32 data;
+
+	result = pci_mmcfg_read(0, dev->bus->number, dev->devfn, 
+ 				where, 2, &data);
+
+	*value = (u16)data;
+
+	return result;
+}
+
+static int pci_mmcfg_read_config_dword(struct pci_dev *dev, int where, u32 *value)
+{
+	return pci_mmcfg_read(0, dev->bus->number, dev->devfn, 
+				where, 4, value);
+}
+
+static int pci_mmcfg_write_config_byte(struct pci_dev *dev, int where, u8 value)
+{
+	return pci_mmcfg_write(0, dev->bus->number, dev->devfn, 
+				where, 1, value);
+}
+
+static int pci_mmcfg_write_config_word(struct pci_dev *dev, int where, u16 value)
+{
+	return pci_mmcfg_write(0, dev->bus->number, dev->devfn, 
+				where, 2, value);
+}
+
+static int pci_mmcfg_write_config_dword(struct pci_dev *dev, int where, u32 value)
+{
+	return pci_mmcfg_write(0, dev->bus->number, dev->devfn, 
+				where, 4, value);
+}
+
+static struct pci_ops pci_mmcfg = {
+	pci_mmcfg_read_config_byte,
+	pci_mmcfg_read_config_word,
+	pci_mmcfg_read_config_dword,
+	pci_mmcfg_write_config_byte,
+	pci_mmcfg_write_config_word,
+	pci_mmcfg_write_config_dword
+};
+
+struct pci_ops * __devinit pci_mmcfg_init(void)
+{
+	if ((pci_probe & PCI_PROBE_MMCONF) == 0)
+	{
+		goto out;
+	}
+	if (!pci_mmcfg_base_addr)
+	{
+		goto out;
+	}
+
+	printk(KERN_INFO "PCI: Using MMCONFIG\n");
+	pci_probe = (pci_probe & ~PCI_PROBE_MASK) | PCI_PROBE_MMCONF;
+
+	return &pci_mmcfg;
+
+out:
+	return NULL;
+}
+
+int pci_mcfg_read(int seg, int bus, int dev, int fn, int reg, int len, 
+				u32 *value)
+{
+	pci_mmcfg_read(seg, bus, PCI_DEVFN(dev, fn), reg, len, value);
+}
+
+int pci_mcfg_write(int seg, int bus, int dev, int fn, int reg, int len, 
+				u32 *value)
+{
+	pci_mmcfg_write(seg, bus, PCI_DEVFN(dev, fn), reg, len, value);
+}
diff -Nru a/arch/i386/kernel/pci-i386.h b/arch/i386/kernel/pci-i386.h
--- a/arch/i386/kernel/pci-i386.h	Mon May 24 13:52:10 2004
+++ b/arch/i386/kernel/pci-i386.h	Mon May 24 13:52:10 2004
@@ -15,6 +15,8 @@
 #define PCI_PROBE_BIOS		0x0001
 #define PCI_PROBE_CONF1		0x0002
 #define PCI_PROBE_CONF2		0x0004
+#define PCI_PROBE_MMCONF	0x0008
+#define PCI_PROBE_MASK		0x000f
 #define PCI_NO_SORT		0x0100
 #define PCI_BIOS_SORT		0x0200
 #define PCI_NO_CHECKS		0x0400
@@ -23,6 +25,7 @@
 #define PCI_ASSIGN_ALL_BUSSES	0x4000
 
 extern unsigned int pci_probe;
+extern spinlock_t pci_config_lock;
 
 /* pci-i386.c */
 
@@ -69,3 +72,10 @@
 void pcibios_irq_init(void);
 void pcibios_fixup_irqs(void);
 void pcibios_enable_irq(struct pci_dev *dev);
+
+/* mmconfig.c */
+int pci_mcfg_read(int seg, int bus, int dev, int fn, int reg, int len, 
+				u32 *value);
+int pci_mcfg_write(int seg, int bus, int dev, int fn, int reg, int len, 
+				u32 *value);
+struct pci_ops * __devinit pci_mmcfg_init(void);
diff -Nru a/arch/i386/kernel/pci-pc.c b/arch/i386/kernel/pci-pc.c
--- a/arch/i386/kernel/pci-pc.c	Mon May 24 13:52:10 2004
+++ b/arch/i386/kernel/pci-pc.c	Mon May 24 13:52:10 2004
@@ -20,7 +20,8 @@
 
 #include "pci-i386.h"
 
-unsigned int pci_probe = PCI_PROBE_BIOS | PCI_PROBE_CONF1 | PCI_PROBE_CONF2;
+unsigned int pci_probe = PCI_PROBE_BIOS | PCI_PROBE_CONF1 | PCI_PROBE_CONF2 |
+				PCI_PROBE_MMCONF;
 
 int pcibios_last_bus = -1;
 struct pci_bus *pci_root_bus = NULL;
@@ -45,7 +46,7 @@
  * This interrupt-safe spinlock protects all accesses to PCI
  * configuration space.
  */
-static spinlock_t pci_config_lock = SPIN_LOCK_UNLOCKED;
+spinlock_t pci_config_lock = SPIN_LOCK_UNLOCKED;
 
 
 /*
@@ -1446,6 +1447,33 @@
 	}
 #endif
 
+#ifdef CONFIG_PCI_MMCONFIG
+	pci_root_ops = pci_mmcfg_init();
+	if (pci_root_ops){
+		pci_config_read = pci_mcfg_read;
+		pci_config_write = pci_mcfg_write;
+		return;
+	}
+#ifdef CONFIG_PCI_DIRECT
+	/*
+	 * PCI Express option is enabled in NON-PCI Express
+	 * platforms. Reset the PCI root ops
+	 */
+	if ((pci_probe & (PCI_PROBE_CONF1 | PCI_PROBE_CONF2)) 
+		&& (tmp = pci_check_direct())) {
+		pci_root_ops = tmp;
+		if (pci_root_ops == &pci_direct_conf1) {
+			pci_config_read = pci_conf1_read;
+			pci_config_write = pci_conf1_write;
+		}
+		else {
+			pci_config_read = pci_conf2_read;
+			pci_config_write = pci_conf2_write;
+		}
+	}
+#endif
+#endif
+
 	return;
 }
 
@@ -1520,6 +1548,12 @@
 	}
 	else if (!strcmp(str, "conf2")) {
 		pci_probe = PCI_PROBE_CONF2 | PCI_NO_CHECKS;
+		return NULL;
+	}
+#endif
+#ifdef CONFIG_PCI_MMCONFIG
+	else if (!strcmp(str, "nommconf")) {
+		pci_probe &= ~PCI_PROBE_MMCONF;
 		return NULL;
 	}
 #endif
diff -Nru a/arch/x86_64/config.in b/arch/x86_64/config.in
--- a/arch/x86_64/config.in	Mon May 24 13:52:10 2004
+++ b/arch/x86_64/config.in	Mon May 24 13:52:10 2004
@@ -95,7 +95,16 @@
 bool 'Networking support' CONFIG_NET
 bool 'PCI support' CONFIG_PCI
 if [ "$CONFIG_PCI" = "y" ]; then
-   define_bool CONFIG_PCI_DIRECT y 
+      choice '  PCI access mode' \
+	"Direct		CONFIG_PCI_GODIRECT 	\
+	 MMConfig 	CONFIG_PCI_GOMMCONFIG 	\
+	 Any		CONFIG_PCI_GOANY"	Any
+      if [ "$CONFIG_PCI_GODIRECT" = "y" -o "$CONFIG_PCI_GOANY" = "y" ]; then
+         define_bool CONFIG_PCI_DIRECT y
+      fi
+      if [ "$CONFIG_PCI_GOMMCONFIG" = "y" -o "$CONFIG_PCI_GOANY" = "y" ]; then
+         define_bool CONFIG_PCI_MMCONFIG y
+      fi
 fi
 
 source drivers/pci/Config.in
diff -Nru a/arch/x86_64/kernel/Makefile b/arch/x86_64/kernel/Makefile
--- a/arch/x86_64/kernel/Makefile	Mon May 24 13:52:10 2004
+++ b/arch/x86_64/kernel/Makefile	Mon May 24 13:52:10 2004
@@ -21,6 +21,10 @@
 		pci-dma.o x8664_ksyms.o i387.o syscall.o vsyscall.o \
 		setup64.o e820.o warmreboot.o
 
+ifdef CONFIG_PCI_MMCONFIG
+obj-y			+= mmconfig.o
+endif
+
 ifdef CONFIG_PCI
 obj-y			+= pci-x86_64.o
 obj-y			+= pci-pc.o pci-irq.o
diff -Nru a/arch/x86_64/kernel/mmconfig.c b/arch/x86_64/kernel/mmconfig.c
--- /dev/null	Wed Dec 31 16:00:00 1969
+++ b/arch/x86_64/kernel/mmconfig.c	Mon May 24 13:52:10 2004
@@ -0,0 +1,170 @@
+/*
+ * mmconfig.c - Low-level direct PCI config space access via MMCONFIG
+ */
+
+#include <linux/pci.h>
+#include <linux/init.h>
+#include "pci-x86_64.h"
+
+/* The physical address of the MMCONFIG aperture.  Set from ACPI tables. */
+extern u32 pci_mmcfg_base_addr;
+
+#define mmcfg_virt_addr (fix_to_virt(FIX_PCIE_MCFG))
+
+/* The base address of the last MMCONFIG device accessed */
+static u32 mmcfg_last_accessed_device;
+
+/*
+ * Functions for accessing PCI configuration space with MMCONFIG accesses
+ */
+
+static inline void pci_exp_set_dev_base(int bus, int devfn)
+{
+	u32 dev_base = pci_mmcfg_base_addr | (bus << 20) | (devfn << 12);
+	if (dev_base != mmcfg_last_accessed_device) {
+		mmcfg_last_accessed_device = dev_base;
+		set_fixmap(FIX_PCIE_MCFG, dev_base);
+	}
+}
+
+static int pci_mmcfg_read(int seg, int bus, int devfn, int reg, int len, 
+				u32 *value)
+{
+	if (!value || (bus > 255) || (devfn > 255) || (reg > 4095))
+		return -EINVAL;
+
+	pci_exp_set_dev_base(bus, devfn);
+
+	switch (len) {
+	case 1:
+		*value = readb(mmcfg_virt_addr + reg);
+		break;
+	case 2:
+		*value = readw(mmcfg_virt_addr + reg);
+		break;
+	case 4:
+		*value = readl(mmcfg_virt_addr + reg);
+		break;
+	}
+
+	return 0;
+}
+
+static int pci_mmcfg_write(int seg, int bus, int devfn, int reg, int len, 
+				u32 value)
+{
+	if ((bus > 255) || (devfn > 255) || (reg > 4095)) 
+		return -EINVAL;
+
+	pci_exp_set_dev_base(bus, devfn);
+
+	switch (len) {
+	case 1:
+		writeb(value, mmcfg_virt_addr + reg);
+		break;
+	case 2:
+		writew(value, mmcfg_virt_addr + reg);
+		break;
+	case 4:
+		writel(value, mmcfg_virt_addr + reg);
+		break;
+	}
+
+	/* Dummy read to flush PCI write */
+	readl(mmcfg_virt_addr);
+
+	return 0;
+}
+
+static int pci_mmcfg_read_config_byte(struct pci_dev *dev, int where, 
+					u8 *value)
+{
+	int result; 
+	u32 data;
+
+	result = pci_mmcfg_read(0, dev->bus->number, dev->devfn, 
+				where, 1, &data);
+
+	*value = (u8)data;
+
+	return result;
+}
+
+static int pci_mmcfg_read_config_word(struct pci_dev *dev, int where, 
+					u16 *value)
+{
+	int result; 
+	u32 data;
+
+	result = pci_mmcfg_read(0, dev->bus->number, dev->devfn, 
+ 				where, 2, &data);
+
+	*value = (u16)data;
+
+	return result;
+}
+
+static int pci_mmcfg_read_config_dword(struct pci_dev *dev, int where, 
+					u32 *value)
+{
+	return pci_mmcfg_read(0, dev->bus->number, dev->devfn, 
+				where, 4, value);
+}
+
+static int pci_mmcfg_write_config_byte(struct pci_dev *dev, int where, 
+					u8 value)
+{
+	return pci_mmcfg_write(0, dev->bus->number, dev->devfn, 
+				where, 1, value);
+}
+
+static int pci_mmcfg_write_config_word(struct pci_dev *dev, int where, 
+					u16 value)
+{
+	return pci_mmcfg_write(0, dev->bus->number, dev->devfn, 
+				where, 2, value);
+}
+
+static int pci_mmcfg_write_config_dword(struct pci_dev *dev, int where, 
+					u32 value)
+{
+	return pci_mmcfg_write(0, dev->bus->number, dev->devfn, 
+				where, 4, value);
+}
+
+static struct pci_ops pci_mmcfg = {
+	pci_mmcfg_read_config_byte,
+	pci_mmcfg_read_config_word,
+	pci_mmcfg_read_config_dword,
+	pci_mmcfg_write_config_byte,
+	pci_mmcfg_write_config_word,
+	pci_mmcfg_write_config_dword
+};
+
+struct pci_ops * __devinit pci_mmcfg_init(void)
+{
+	if ((pci_probe & PCI_PROBE_MMCONF) == 0)
+		goto out;
+	if (!pci_mmcfg_base_addr)
+		goto out;
+
+	printk(KERN_INFO "PCI: Using MMCONFIG\n");
+	pci_probe = (pci_probe & ~PCI_PROBE_MASK) | PCI_PROBE_MMCONF;
+
+	return &pci_mmcfg;
+
+out:
+	return NULL;
+}
+
+int pci_mcfg_read(int seg, int bus, int dev, int fn, int reg, int len, 
+				u32 *value)
+{
+	pci_mmcfg_read(seg, bus, PCI_DEVFN(dev, fn), reg, len, value);
+}
+
+int pci_mcfg_write(int seg, int bus, int dev, int fn, int reg, int len, 
+				u32 *value)
+{
+	pci_mmcfg_write(seg, bus, PCI_DEVFN(dev, fn), reg, len, value);
+}
diff -Nru a/arch/x86_64/kernel/pci-pc.c b/arch/x86_64/kernel/pci-pc.c
--- a/arch/x86_64/kernel/pci-pc.c	Mon May 24 13:52:10 2004
+++ b/arch/x86_64/kernel/pci-pc.c	Mon May 24 13:52:10 2004
@@ -27,7 +27,7 @@
 
 #include "pci-x86_64.h"
 
-unsigned int pci_probe = PCI_PROBE_CONF1 | PCI_PROBE_CONF2;
+unsigned int pci_probe = PCI_PROBE_CONF1 | PCI_PROBE_CONF2 | PCI_PROBE_MMCONF;
 
 int pcibios_last_bus = -1;
 struct pci_bus *pci_root_bus;
@@ -566,6 +566,31 @@
 		printk("??? no pci access\n"); 
 #endif
 
+#ifdef CONFIG_PCI_MMCONFIG
+	pci_root_ops = pci_mmcfg_init();
+	if (pci_root_ops) {
+		pci_config_read = pci_mcfg_read;
+		pci_config_write = pci_mcfg_write;
+	}
+#ifdef CONFIG_PCI_DIRECT
+	else {
+	/*
+	 * PCI Express option is enabled in NON-PCI Express
+	 * platforms. Reset the PCI root ops
+	 */
+		pci_root_ops = pci_check_direct();
+		if (pci_root_ops == &pci_direct_conf1) {
+			pci_config_read = pci_conf1_read;
+			pci_config_write = pci_conf1_write;
+		}
+		else {
+			pci_config_read = pci_conf2_read;
+			pci_config_write = pci_conf2_write;
+		}
+	}
+#endif
+#endif
+
 	return;
 }
 
@@ -580,6 +605,19 @@
 	if (pci_probe & (PCI_PROBE_CONF1 | PCI_PROBE_CONF2))
 		dir = pci_check_direct();
 #endif
+
+#ifdef CONFIG_PCI_MMCONFIG
+	dir = pci_mmcfg_init();
+#ifdef CONFIG_PCI_DIRECT
+	/*
+	 * PCI Express option is enabled in NON-PCI Express
+	 * platforms. Reset the PCI root ops
+	 */
+	if (!dir) {
+		dir = pci_check_direct();
+	}
+#endif
+#endif
 	if (dir)
 		pci_root_ops = dir;
 	else {
@@ -630,6 +668,12 @@
 	}
 	else if (!strcmp(str, "conf2")) {
 		pci_probe = PCI_PROBE_CONF2 | PCI_NO_CHECKS;
+		return NULL;
+	}
+#endif
+#ifdef CONFIG_PCI_MMCONFIG
+	else if (!strcmp(str, "nommconf")) {
+		pci_probe &= ~PCI_PROBE_MMCONF;
 		return NULL;
 	}
 #endif
diff -Nru a/arch/x86_64/kernel/pci-x86_64.h b/arch/x86_64/kernel/pci-x86_64.h
--- a/arch/x86_64/kernel/pci-x86_64.h	Mon May 24 13:52:10 2004
+++ b/arch/x86_64/kernel/pci-x86_64.h	Mon May 24 13:52:10 2004
@@ -15,6 +15,8 @@
 #define PCI_PROBE_BIOS		0x0001
 #define PCI_PROBE_CONF1		0x0002
 #define PCI_PROBE_CONF2		0x0004
+#define PCI_PROBE_MMCONF	0x0008
+#define PCI_PROBE_MASK		0x000f
 #define PCI_NO_SORT		0x0100
 #define PCI_BIOS_SORT		0x0200
 #define PCI_NO_CHECKS		0x0400
@@ -74,3 +76,10 @@
 void pcibios_enable_irq(struct pci_dev *dev);
 
 void pci_iommu_init(void);
+
+/* mmconfig.c */
+int pci_mcfg_read(int seg, int bus, int dev, int fn, int reg, int len, 
+				u32 *value);
+int pci_mcfg_write(int seg, int bus, int dev, int fn, int reg, int len, 
+				u32 *value);
+struct pci_ops * __devinit pci_mmcfg_init(void);
diff -Nru a/drivers/acpi/Config.in b/drivers/acpi/Config.in
--- a/drivers/acpi/Config.in	Mon May 24 13:52:10 2004
+++ b/drivers/acpi/Config.in	Mon May 24 13:52:10 2004
@@ -14,6 +14,7 @@
     define_bool CONFIG_ACPI_INTERPRETER		y
     define_bool CONFIG_ACPI_EC			y
     define_bool CONFIG_ACPI_POWER		y
+    define_bool CONFIG_ACPI_MMCONFIG		y
     if [ "$CONFIG_PCI" = "y" ]; then
       define_bool CONFIG_ACPI_PCI		y
       define_bool CONFIG_ACPI_MMCONFIG		y
@@ -73,6 +74,7 @@
     define_bool CONFIG_ACPI_BUS		y
     define_bool CONFIG_ACPI_INTERPRETER y
     define_bool CONFIG_ACPI_POWER 	y
+    define_bool CONFIG_ACPI_MMCONFIG	y
     define_bool CONFIG_ACPI_SYSTEM 	y
     tristate     '  Button'		CONFIG_ACPI_BUTTON
     tristate     '  Fan'		CONFIG_ACPI_FAN
diff -Nru a/drivers/pci/pci.c b/drivers/pci/pci.c
--- a/drivers/pci/pci.c	Mon May 24 13:52:10 2004
+++ b/drivers/pci/pci.c	Mon May 24 13:52:10 2004
@@ -36,6 +36,9 @@
 #define DBG(x...)
 #endif
 
+#define PCI_CFG_SPACE_SIZE	256
+#define PCI_CFG_SPACE_EXP_SIZE	4096
+
 LIST_HEAD(pci_root_buses);
 LIST_HEAD(pci_devices);
 
@@ -165,6 +168,8 @@
  *  %PCI_CAP_ID_CHSWP        CompactPCI HotSwap 
  *
  *  %PCI_CAP_ID_PCIX         PCI-X
+ *
+ *  %PCI_CAP_ID_EXP          PCI Express
  */
 int
 pci_find_capability(struct pci_dev *dev, int cap)
@@ -199,6 +204,21 @@
 	return 0;
 }
 
+/**
+ * pci_cfg_space_size - get the configuration space size of the PCI device.
+ * Regular PCI devices have 256 bytes, but PCI-X 2 and PCI Express devices
+ * have 4096 bytes.
+ */
+static int pci_cfg_space_size(struct pci_dev *dev)
+{
+	int pos;
+
+	pos = pci_find_capability(dev, PCI_CAP_ID_EXP);
+	if (pos)
+		return PCI_CFG_SPACE_EXP_SIZE;
+	
+	return PCI_CFG_SPACE_SIZE;
+}
 
 /**
  * pci_find_parent_resource - return resource region of parent bus of given region
@@ -1426,6 +1446,9 @@
 		       dev->slot_name, class, dev->hdr_type);
 		dev->class = PCI_CLASS_NOT_DEFINED;
 	}
+
+	/* Get the config space size */
+	dev->cfg_size = pci_cfg_space_size(dev);
 
 	/* We found a fine healthy device, go go go... */
 	return 0;
diff -Nru a/drivers/pci/proc.c b/drivers/pci/proc.c
--- a/drivers/pci/proc.c	Mon May 24 13:52:10 2004
+++ b/drivers/pci/proc.c	Mon May 24 13:52:10 2004
@@ -16,12 +16,13 @@
 #include <asm/uaccess.h>
 #include <asm/byteorder.h>
 
-#define PCI_CFG_SPACE_SIZE 256
-
 static loff_t
 proc_bus_pci_lseek(struct file *file, loff_t off, int whence)
 {
 	loff_t new;
+	const struct inode *ino = file->f_dentry->d_inode;
+	const struct proc_dir_entry *dp = ino->u.generic_ip;
+	struct pci_dev *dev = dp->data;
 
 	switch (whence) {
 	case 0:
@@ -31,12 +32,12 @@
 		new = file->f_pos + off;
 		break;
 	case 2:
-		new = PCI_CFG_SPACE_SIZE + off;
+		new = dev->cfg_size + off;
 		break;
 	default:
 		return -EINVAL;
 	}
-	if (new < 0 || new > PCI_CFG_SPACE_SIZE)
+	if (new < 0 || new > dev->cfg_size)
 		return -EINVAL;
 	return (file->f_pos = new);
 }
@@ -57,7 +58,7 @@
 	 */
 
 	if (capable(CAP_SYS_ADMIN))
-		size = PCI_CFG_SPACE_SIZE;
+		size = dev->cfg_size;
 	else if (dev->hdr_type == PCI_HEADER_TYPE_CARDBUS)
 		size = 128;
 	else
@@ -132,12 +133,12 @@
 	int pos = *ppos;
 	int cnt;
 
-	if (pos >= PCI_CFG_SPACE_SIZE)
+	if (pos >= dev->cfg_size)
 		return 0;
-	if (nbytes >= PCI_CFG_SPACE_SIZE)
-		nbytes = PCI_CFG_SPACE_SIZE;
-	if (pos + nbytes > PCI_CFG_SPACE_SIZE)
-		nbytes = PCI_CFG_SPACE_SIZE - pos;
+	if (nbytes >= dev->cfg_size)
+		nbytes = dev->cfg_size;
+	if (pos + nbytes > dev->cfg_size)
+		nbytes = dev->cfg_size - pos;
 	cnt = nbytes;
 
 	if (!access_ok(VERIFY_READ, buf, cnt))
@@ -389,7 +390,7 @@
 		return -ENOMEM;
 	e->proc_fops = &proc_bus_pci_operations;
 	e->data = dev;
-	e->size = PCI_CFG_SPACE_SIZE;
+	e->size = dev->cfg_size;
 	return 0;
 }
 
diff -Nru a/include/asm-i386/fixmap.h b/include/asm-i386/fixmap.h
--- a/include/asm-i386/fixmap.h	Mon May 24 13:52:10 2004
+++ b/include/asm-i386/fixmap.h	Mon May 24 13:52:10 2004
@@ -76,6 +76,9 @@
 	FIX_ACPI_BEGIN,
 	FIX_ACPI_END = FIX_ACPI_BEGIN + FIX_ACPI_PAGES - 1,
 #endif
+#ifdef CONFIG_PCI_MMCONFIG
+	FIX_PCIE_MCFG,
+#endif
 	__end_of_permanent_fixed_addresses,
 	/* temporary boot-time mappings, used before ioremap() is functional */
 #define NR_FIX_BTMAPS	16
diff -Nru a/include/asm-x86_64/fixmap.h b/include/asm-x86_64/fixmap.h
--- a/include/asm-x86_64/fixmap.h	Mon May 24 13:52:10 2004
+++ b/include/asm-x86_64/fixmap.h	Mon May 24 13:52:10 2004
@@ -46,6 +46,9 @@
 	FIX_IO_APIC_BASE_0,
 	FIX_IO_APIC_BASE_END = FIX_IO_APIC_BASE_0 + MAX_IO_APICS-1,
 #endif
+#ifdef CONFIG_PCI_MMCONFIG
+	FIX_PCIE_MCFG,
+#endif
 	__end_of_fixed_addresses
 };
 
diff -Nru a/include/linux/pci.h b/include/linux/pci.h
--- a/include/linux/pci.h	Mon May 24 13:52:10 2004
+++ b/include/linux/pci.h	Mon May 24 13:52:10 2004
@@ -401,6 +401,7 @@
 	/* device is compatible with these IDs */
 	unsigned short vendor_compatible[DEVICE_COUNT_COMPATIBLE];
 	unsigned short device_compatible[DEVICE_COUNT_COMPATIBLE];
+	int		cfg_size;	/* Size of configuration space */
 
 	/*
 	 * Instead of touching interrupt line and base address registers

Re: [BK PATCH] PCI Express patches for 2.4.27-pre3

[Greg KH]

ChangeSet 1.1428, 2004/05/24 13:33:53-07:00, blujuice@us.ibm.com

[PATCH] PCI: PCI Express help in 2.4

Here is a patch for describing PCI access mode - help in 2.4.
This aligns the help with what is in 2.6.


 Documentation/Configure.help |   13 +++++++------
 1 files changed, 7 insertions(+), 6 deletions(-)


diff -Nru a/Documentation/Configure.help b/Documentation/Configure.help
--- a/Documentation/Configure.help	Mon May 24 13:52:08 2004
+++ b/Documentation/Configure.help	Mon May 24 13:52:08 2004
@@ -4236,12 +4236,13 @@
   PCI-based systems don't have any BIOS at all. Linux can also try to
   detect the PCI hardware directly without using the BIOS.
 
-  With this option, you can specify how Linux should detect the PCI
-  devices. If you choose "BIOS", the BIOS will be used, if you choose
-  "Direct", the BIOS won't be used, and if you choose "Any", the
-  kernel will try the direct access method and falls back to the BIOS
-  if that doesn't work. If unsure, go with the default, which is
-  "Any".
+  With this option, you can specify how Linux should detect the
+  PCI devices. If you choose "BIOS", the BIOS will be used,
+  if you choose "Direct", the BIOS won't be used, and if you
+  choose "MMConfig", then PCI Express MMCONFIG will be used.
+  If you choose "Any", the kernel will try MMCONFIG, then the
+  direct access method and falls back to the BIOS if that doesn't
+  work. If unsure, go with the default, which is "Any".
 
 PCI device name database
 CONFIG_PCI_NAMES

Re: [BK PATCH] PCI Express patches for 2.4.27-pre3

[Arjan van de Ven]

[-- Attachment #1: Type: text/plain, Size: 527 bytes --]

On Mon, 2004-05-24 at 23:01, Greg KH wrote:
> Hi,
> 
> Now that the ACPI portion of PCI Express support is in the 2.4 kernel
> tree, I can send you the remaining portion to actually enable this
> feature to work properly.  Here is the PCI Express support for i386 and
> x86_64 platforms backported from 2.6 to the 2.4 kernel.


how does this mesh with the "2.4 is now feature frozen"?
Especially since you don't NEED this patch to run on pci express
hardware, it's just that you can use a few performance tweaks.


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: [BK PATCH] PCI Express patches for 2.4.27-pre3

[Greg KH]

On Tue, May 25, 2004 at 08:53:28AM +0200, Arjan van de Ven wrote:
> On Mon, 2004-05-24 at 23:01, Greg KH wrote:
> > Hi,
> > 
> > Now that the ACPI portion of PCI Express support is in the 2.4 kernel
> > tree, I can send you the remaining portion to actually enable this
> > feature to work properly.  Here is the PCI Express support for i386 and
> > x86_64 platforms backported from 2.6 to the 2.4 kernel.
> 
> 
> how does this mesh with the "2.4 is now feature frozen"?

As the major chunk of ACPI support just got added to the tree, and the
only reason that went in was for this patch, I assumed that it was
acceptable.  Marcelo, feel free to tell me otherwise if you do not want
this in the 2.4 tree.

> Especially since you don't NEED this patch to run on pci express
> hardware, it's just that you can use a few performance tweaks.

I agree.

thanks,

greg k-h

Re: [BK PATCH] PCI Express patches for 2.4.27-pre3

[Matthew Wilcox]

On Tue, May 25, 2004 at 01:00:06AM -0700, Greg KH wrote:
> > how does this mesh with the "2.4 is now feature frozen"?
> 
> As the major chunk of ACPI support just got added to the tree, and the
> only reason that went in was for this patch, I assumed that it was
> acceptable.  Marcelo, feel free to tell me otherwise if you do not want
> this in the 2.4 tree.

I assume it was added because Len tries to keep ACPI in 2.4 and 2.6 as
close to identical as possible.  It certainly doesn't hurt anyone to add
the ACPI functionality without the MMConfig support.

-- 
"Next the statesmen will invent cheap lies, putting the blame upon 
the nation that is attacked, and every man will be glad of those
conscience-soothing falsities, and will diligently study them, and refuse
to examine any refutations of them; and thus he will by and by convince 
himself that the war is just, and will thank God for the better sleep 
he enjoys after this process of grotesque self-deception." -- Mark Twain

Re: [BK PATCH] PCI Express patches for 2.4.27-pre3

[Marcelo Tosatti]

Hi kernel fellows,

On Tue, May 25, 2004 at 12:32:31PM +0100, Matthew Wilcox wrote:
> On Tue, May 25, 2004 at 01:00:06AM -0700, Greg KH wrote:
> > > how does this mesh with the "2.4 is now feature frozen"?
> > 
> > As the major chunk of ACPI support just got added to the tree, and the
> > only reason that went in was for this patch, I assumed that it was
> > acceptable.

major? the MMConfig support is minimal as I can see? 

> > Marcelo, feel free to tell me otherwise if you do not want
> > this in the 2.4 tree. 

Is this code necessary for PCI-Express devices/busses to work properly?

> I assume it was added because Len tries to keep ACPI in 2.4 and 2.6 as
> close to identical as possible.  It certainly doesn't hurt anyone to add
> the ACPI functionality without the MMConfig support.

I've humbly asked Len to stop doing big updates whenever possible on the 
v2.4 ACPI code, and do bugfixes only instead. Is that a pain in the ass for you, Len?    

I asked that because it is common to see new bugs introduced by an ACPI update, 
and you know that more than I do.

Thanks!

PS: Greg, about the PCI-Express hotplug drivers, I assume they are independant 
on any of this?

Re: [BK PATCH] PCI Express patches for 2.4.27-pre3

[Arjan van de Ven]

[-- Attachment #1: Type: text/plain, Size: 369 bytes --]


On Tue, May 25, 2004 at 09:54:53AM -0300, Marcelo Tosatti wrote:
> > > Marcelo, feel free to tell me otherwise if you do not want
> > > this in the 2.4 tree. 
> 
> Is this code necessary for PCI-Express devices/busses to work properly?

afaik not. It's an enhancement to make config space access to them somewhat
faster, but they just work using the existing method.


[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [BK PATCH] PCI Express patches for 2.4.27-pre3

[Matthew Wilcox]

On Tue, May 25, 2004 at 03:01:16PM +0200, Arjan van de Ven wrote:
> 
> On Tue, May 25, 2004 at 09:54:53AM -0300, Marcelo Tosatti wrote:
> > > > Marcelo, feel free to tell me otherwise if you do not want
> > > > this in the 2.4 tree. 
> > 
> > Is this code necessary for PCI-Express devices/busses to work properly?
> 
> afaik not. It's an enhancement to make config space access to them somewhat
> faster, but they just work using the existing method.

It also allows access to the top 3840 bytes of config space.  The *spec*
says you can't require access to that area for correct functioning of
the device, but we all know how much people love to follow specs.

-- 
"Next the statesmen will invent cheap lies, putting the blame upon 
the nation that is attacked, and every man will be glad of those
conscience-soothing falsities, and will diligently study them, and refuse
to examine any refutations of them; and thus he will by and by convince 
himself that the war is just, and will thank God for the better sleep 
he enjoys after this process of grotesque self-deception." -- Mark Twain

Re: [BK PATCH] PCI Express patches for 2.4.27-pre3

[Greg KH]

On Tue, May 25, 2004 at 02:24:13PM +0100, Matthew Wilcox wrote:
> On Tue, May 25, 2004 at 03:01:16PM +0200, Arjan van de Ven wrote:
> > 
> > On Tue, May 25, 2004 at 09:54:53AM -0300, Marcelo Tosatti wrote:
> > > > > Marcelo, feel free to tell me otherwise if you do not want
> > > > > this in the 2.4 tree. 
> > > 
> > > Is this code necessary for PCI-Express devices/busses to work properly?
> > 
> > afaik not. It's an enhancement to make config space access to them somewhat
> > faster, but they just work using the existing method.
> 
> It also allows access to the top 3840 bytes of config space.  The *spec*
> says you can't require access to that area for correct functioning of
> the device, but we all know how much people love to follow specs.

As no one really has PCI Express devices at this time, who knows how
badly the vendors will mess them up :)

thanks,

greg k-h

Re: [BK PATCH] PCI Express patches for 2.4.27-pre3

[Jeff Garzik]

Greg KH wrote:
> As no one really has PCI Express devices at this time, who knows how

This is not really true...

Re: [BK PATCH] PCI Express patches for 2.4.27-pre3

[Greg KH]

On Tue, May 25, 2004 at 10:49:57PM -0400, Jeff Garzik wrote:
> Greg KH wrote:
> >As no one really has PCI Express devices at this time, who knows how
> 
> This is not really true...

Ok, I have seen a few of them in labs, but I'm having a _real_ hard time
finding one at the local Fry's to purchase :)

greg k-h

Re: [BK PATCH] PCI Express patches for 2.4.27-pre3

[Greg KH]

On Tue, May 25, 2004 at 09:54:53AM -0300, Marcelo Tosatti wrote:
> 
> Hi kernel fellows,
> 
> On Tue, May 25, 2004 at 12:32:31PM +0100, Matthew Wilcox wrote:
> > On Tue, May 25, 2004 at 01:00:06AM -0700, Greg KH wrote:
> > > > how does this mesh with the "2.4 is now feature frozen"?
> > > 
> > > As the major chunk of ACPI support just got added to the tree, and the
> > > only reason that went in was for this patch, I assumed that it was
> > > acceptable.
> 
> major? the MMConfig support is minimal as I can see? 

It isn't that big of a patch, but it is make to core PCI code.

> > > Marcelo, feel free to tell me otherwise if you do not want
> > > this in the 2.4 tree. 
> 
> Is this code necessary for PCI-Express devices/busses to work properly?

Not that I can tell, the main point is accessing the extended config
space, and speeding up the access to the device to its natural speed.

> > I assume it was added because Len tries to keep ACPI in 2.4 and 2.6 as
> > close to identical as possible.  It certainly doesn't hurt anyone to add
> > the ACPI functionality without the MMConfig support.
> 
> I've humbly asked Len to stop doing big updates whenever possible on the 
> v2.4 ACPI code, and do bugfixes only instead. Is that a pain in the ass for you, Len?    
> 
> I asked that because it is common to see new bugs introduced by an ACPI update, 
> and you know that more than I do.

Yes, I know that quite well :)

So, because of this, you are saying that we should not apply these
patches at this time?

If so, that's fine with me, as now any distro that wants to add this to
their 2.4 kernel can, as the patches are public.

> PS: Greg, about the PCI-Express hotplug drivers, I assume they are independant 
> on any of this?

Yes, that is independant of these changes (that is required as there is
no other way to control the pci hotplug controller of those systems,
except with the driver I had submitted in the past to you.).

thanks,

greg k-h

Re: [BK PATCH] PCI Express patches for 2.4.27-pre3

[Marcelo Tosatti]

On Tue, May 25, 2004 at 07:40:56AM -0700, Greg KH wrote:
> On Tue, May 25, 2004 at 09:54:53AM -0300, Marcelo Tosatti wrote:
> > 
> > Hi kernel fellows,
> > 
> > On Tue, May 25, 2004 at 12:32:31PM +0100, Matthew Wilcox wrote:
> > > On Tue, May 25, 2004 at 01:00:06AM -0700, Greg KH wrote:
> > > > > how does this mesh with the "2.4 is now feature frozen"?
> > > > 
> > > > As the major chunk of ACPI support just got added to the tree, and the
> > > > only reason that went in was for this patch, I assumed that it was
> > > > acceptable.
> > 
> > major? the MMConfig support is minimal as I can see? 
> 
> It isn't that big of a patch, but it is make to core PCI code.
> 
> > > > Marcelo, feel free to tell me otherwise if you do not want
> > > > this in the 2.4 tree. 
> > 
> > Is this code necessary for PCI-Express devices/busses to work properly?
> 
> Not that I can tell, the main point is accessing the extended config
> space, and speeding up the access to the device to its natural speed.
> 
> > > I assume it was added because Len tries to keep ACPI in 2.4 and 2.6 as
> > > close to identical as possible.  It certainly doesn't hurt anyone to add
> > > the ACPI functionality without the MMConfig support.
> > 
> > I've humbly asked Len to stop doing big updates whenever possible on the 
> > v2.4 ACPI code, and do bugfixes only instead. Is that a pain in the ass for you, Len?    
> > 
> > I asked that because it is common to see new bugs introduced by an ACPI update, 
> > and you know that more than I do.
> 
> Yes, I know that quite well :)
> 
> So, because of this, you are saying that we should not apply these
> patches at this time? 

Yeap, I would prefer not to apply them at this time. For one, Arjan told
me privately it can break XFree86 which accesses the PCI config space directly.
Right?

> If so, that's fine with me, as now any distro that wants to add this to
> their 2.4 kernel can, as the patches are public.

Yeap. 

> > PS: Greg, about the PCI-Express hotplug drivers, I assume they are independant 
> > on any of this?
> 
> Yes, that is independant of these changes (that is required as there is
> no other way to control the pci hotplug controller of those systems,
> except with the driver I had submitted in the past to you.).
> 
> thanks,
> 
> greg k-h

Re: [BK PATCH] PCI Express patches for 2.4.27-pre3

[Greg KH]

On Tue, May 25, 2004 at 01:59:04PM -0300, Marcelo Tosatti wrote:
> On Tue, May 25, 2004 at 07:40:56AM -0700, Greg KH wrote:
> > On Tue, May 25, 2004 at 09:54:53AM -0300, Marcelo Tosatti wrote:
> > > 
> > > Hi kernel fellows,
> > > 
> > > On Tue, May 25, 2004 at 12:32:31PM +0100, Matthew Wilcox wrote:
> > > > On Tue, May 25, 2004 at 01:00:06AM -0700, Greg KH wrote:
> > > > > > how does this mesh with the "2.4 is now feature frozen"?
> > > > > 
> > > > > As the major chunk of ACPI support just got added to the tree, and the
> > > > > only reason that went in was for this patch, I assumed that it was
> > > > > acceptable.
> > > 
> > > major? the MMConfig support is minimal as I can see? 
> > 
> > It isn't that big of a patch, but it is make to core PCI code.
> > 
> > > > > Marcelo, feel free to tell me otherwise if you do not want
> > > > > this in the 2.4 tree. 
> > > 
> > > Is this code necessary for PCI-Express devices/busses to work properly?
> > 
> > Not that I can tell, the main point is accessing the extended config
> > space, and speeding up the access to the device to its natural speed.
> > 
> > > > I assume it was added because Len tries to keep ACPI in 2.4 and 2.6 as
> > > > close to identical as possible.  It certainly doesn't hurt anyone to add
> > > > the ACPI functionality without the MMConfig support.
> > > 
> > > I've humbly asked Len to stop doing big updates whenever possible on the 
> > > v2.4 ACPI code, and do bugfixes only instead. Is that a pain in the ass for you, Len?    
> > > 
> > > I asked that because it is common to see new bugs introduced by an ACPI update, 
> > > and you know that more than I do.
> > 
> > Yes, I know that quite well :)
> > 
> > So, because of this, you are saying that we should not apply these
> > patches at this time? 
> 
> Yeap, I would prefer not to apply them at this time. For one, Arjan told
> me privately it can break XFree86 which accesses the PCI config space directly.
> Right?

Um, not that I know of.  2.6 has had this code for a while now with no
reported problems.

Arjan, is there something I need to know about?  :)

> > If so, that's fine with me, as now any distro that wants to add this to
> > their 2.4 kernel can, as the patches are public.
> 
> Yeap. 

Ok, no problem, thanks for looking at it.

greg k-h