Skip to contentSkip to conversation

MMCONFIG: Fix x86_64 ioremap base_address

PATCH In review OGAWA Hirofumi 4 participants 7 messages Dec 23, 2006 – Dec 25, 2006

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 MMCONFIG: Fix x86_64 ioremap base_address

Current -mm's mmconfig has some problems of remapped range.

a) In the case of broken MCFG tables on Asus etc., we need to remap 256M range, but currently only remap 1M.

b) The base address always corresponds to bus number 0, but currently we are assuming it corresponds to start bus number.

This patch fixes the above problems.

---

 arch/x86_64/pci/mmconfig.c |   48 ++++++++++++++++++++++++++++++++-------------
 1 file changed, 35 insertions(+), 13 deletions(-)
arch/x86_64/pci/mmconfig.c renamed from arch/x86_64/pci/mmconfig.c~pci-mmconfig-ioremap-range-fix +35 -13
diff -puN arch/x86_64/pci/mmconfig.c~pci-mmconfig-ioremap-range-fix arch/x86_64/pci/mmconfig.c
@@ -24,6 +24,39 @@ struct mmcfg_virt {
 };
 static struct mmcfg_virt *pci_mmcfg_virt;
 
+static inline int mcfg_broken(void)
+{
+	struct acpi_table_mcfg_config *cfg = &pci_mmcfg_config[0];
+
+	/* Handle more broken MCFG tables on Asus etc.
+	   They only contain a single entry for bus 0-0. Assume
+ 	   this applies to all busses. */
+	if (pci_mmcfg_config_num == 1 &&
+	    cfg->pci_segment_group_number == 0 &&
+	    (cfg->start_bus_number | cfg->end_bus_number) == 0)
+		return 1;
+	return 0;
+}
+
+static void __iomem *mcfg_ioremap(struct acpi_table_mcfg_config *cfg)
+{
+	void __iomem *addr;
+	u32 size;
+
+	if (mcfg_broken())
+		size = 256 << 20;
+	else
+		size = (cfg->end_bus_number + 1) << 20;
+
+	addr = ioremap_nocache(cfg->base_address, size);
+	if (addr) {
+		printk(KERN_INFO "PCI: Using MMCONFIG at %x - %x\n",
+		       cfg->base_address,
+		       cfg->base_address + size - 1);
+	}
+	return addr;
+}
+
 static char __iomem *get_virt(unsigned int seg, unsigned bus)
 {
 	int cfg_num = -1;
@@ -41,13 +74,7 @@ static char __iomem *get_virt(unsigned i
 			return pci_mmcfg_virt[cfg_num].virt;
 	}
 
-	/* Handle more broken MCFG tables on Asus etc.
-	   They only contain a single entry for bus 0-0. Assume
- 	   this applies to all busses. */
-	cfg = &pci_mmcfg_config[0];
-	if (pci_mmcfg_config_num == 1 &&
-		cfg->pci_segment_group_number == 0 &&
-		(cfg->start_bus_number | cfg->end_bus_number) == 0)
+	if (mcfg_broken())
 		return pci_mmcfg_virt[0].virt;
 
 	/* Fall back to type 0 */
@@ -139,19 +166,14 @@ int __init pci_mmcfg_arch_init(void)
 	}
 
 	for (i = 0; i < pci_mmcfg_config_num; ++i) {
-		u32 size = (pci_mmcfg_config[0].end_bus_number - pci_mmcfg_config[0].start_bus_number + 1) << 20;
 		pci_mmcfg_virt[i].cfg = &pci_mmcfg_config[i];
-		pci_mmcfg_virt[i].virt = ioremap_nocache(pci_mmcfg_config[i].base_address,
-							 size);
+		pci_mmcfg_virt[i].virt = mcfg_ioremap(&pci_mmcfg_config[i]);
 		if (!pci_mmcfg_virt[i].virt) {
 			printk(KERN_ERR "PCI: Cannot map mmconfig aperture for "
 					"segment %d\n",
 			       pci_mmcfg_config[i].pci_segment_group_number);
 			return 0;
 		}
-		printk(KERN_INFO "PCI: Using MMCONFIG at %x-%x\n",
-		       pci_mmcfg_config[i].base_address,
-		       pci_mmcfg_config[i].base_address + size - 1);
 	}
 
 	raw_pci_ops = &pci_mmcfg;

_

Signature · 1 line
OGAWA Hirofumi <[email protected]>
#1 lore raw
Quoted 4 lines · OGAWA Hirofumi

Current -mm's mmconfig has some problems of remapped range.

a) In the case of broken MCFG tables on Asus etc., we need to remap 256M range, but currently only remap 1M.

Hi,

I respectfully disagree. If the MCFG table is broken, we should not use it AT ALL. It's either a correct table, and we can use it, or it's so broken that we know it never has been tested etc etc, we're just better of to not trust it in that case.

Greetings,
Arjan van de Ven

#2 lore raw

Hi,

Quoted 10 lines · Arjan van de Ven
Quoted 4 lines · OGAWA Hirofumi

Current -mm's mmconfig has some problems of remapped range.

a) In the case of broken MCFG tables on Asus etc., we need to remap 256M range, but currently only remap 1M.

I respectfully disagree. If the MCFG table is broken, we should not use it AT ALL. It's either a correct table, and we can use it, or it's so broken that we know it never has been tested etc etc, we're just better of to not trust it in that case.

Hm.. I see. Sounds sane to me. Well, my patch didn't add the new workaround of broken table, it just fixes the oops of existence workaround.

Current workaround is the following (both of linus and -mm),

	if (pci_mmcfg_config_num == 1 &&
		cfg->pci_segment_group_number == 0 &&
		(cfg->start_bus_number | cfg->end_bus_number) == 0)
                /* Assume it can access 256M range */

But, if the system has bus number 0 only, it's a correct table. We may need to detect the broken system. blacklist?

Signature · 1 line
OGAWA Hirofumi <[email protected]>
#3 lore raw
Quoted 10 lines

Current workaround is the following (both of linus and -mm),

	if (pci_mmcfg_config_num == 1 &&
		cfg->pci_segment_group_number == 0 &&
		(cfg->start_bus_number | cfg->end_bus_number) == 0)
                /* Assume it can access 256M range */

But, if the system has bus number 0 only, it's a correct table. We may need to detect the broken system. blacklist?

or just... not assume 256Mb but assume broken?

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

Sorry I missed the original email, but what is the chipset (name, pci ID) of the board(s) with the problematic bios?

OG.

#5 lore raw
Quoted 2 lines · Olivier Galibert

Sorry I missed the original email, but what is the chipset (name, pci ID) of the board(s) with the problematic bios?

I don't know, here is a comment of current code.

/* Handle more broken MCFG tables on Asus etc.
They only contain a single entry for bus 0-0. Assume this applies to all busses. */

Signature · 1 line
OGAWA Hirofumi <[email protected]>
#6 lore raw
In reply to Arjan van de Ven
Quoted 11 lines · Arjan van de Ven
Quoted 9 lines

Current workaround is the following (both of linus and -mm),

	if (pci_mmcfg_config_num == 1 &&
		cfg->pci_segment_group_number == 0 &&
		(cfg->start_bus_number | cfg->end_bus_number) == 0)
                /* Assume it can access 256M range */

But, if the system has bus number 0 only, it's a correct table. We may need to detect the broken system. blacklist?

or just... not assume 256Mb but assume broken?

I see. And instead, add force enable option?

Signature · 1 line
OGAWA Hirofumi <[email protected]>