linux,tce-size property is 32 bits
PATCH 1/1 linux,tce-size property is 32 bits
The "linux,tce-size" property is only 32 bits (see prom_initialize_tce_table() in arch/powerpc/kernel/prom_init.c). Treating it as an unsigned long in iommu_table_setparms() leads to access beyond the end of the property's buffer, so we pass garbage to the memset() in that function.
[boot]0020 XICS Init
i8259 legacy interrupt controller initialized [boot]0021 XICS Done
PID hash table entries: 4096 (order: 12, 32768 bytes) cpu 0x0: Vector: 300 (Data Access) at [c0000000fe783850]
pc: c000000000035e90: .memset+0x60/0xfc
lr: c000000000044fa4: .iommu_table_setparms+0xb0/0x158
sp: c0000000fe783ad0 msr: 9000000000009032
dar: c000000100000000
dsisr: 42010000
current = 0xc00000000450e810
paca = 0xc000000000411580
pid = 1, comm = swapper
enter ? for help
[link register ] c000000000044fa4 .iommu_table_setparms+0xb0/0x158
[c0000000fe783ad0] c000000000044f4c .iommu_table_setparms+0x58/0x158 (unreliable)
[c0000000fe783b70] c00000000004529c
.iommu_bus_setup_pSeries+0x1c4/0x254 [c0000000fe783c00] c00000000002b8ac .do_bus_setup+0x3c/0xe4 [c0000000fe783c80] c00000000002c924 .pcibios_fixup_bus+0x64/0xd8 [c0000000fe783d00] c0000000001a2d5c .pci_scan_child_bus+0x6c/0x10c [c0000000fe783da0] c00000000002be28 .scan_phb+0x17c/0x1b4 [c0000000fe783e40] c0000000003cfa00 .pcibios_init+0x58/0x19c [c0000000fe783ec0] c0000000000094b4 .init+0x1e8/0x3d8 [c0000000fe783f90] c000000000026e54 .kernel_thread+0x4c/0x68
---
I believe this is a regression since 2.6.18, so please push for inclusion in 2.6.19.
arch/powerpc/platforms/pseries/iommu.c +2 -1
@@ -267,7 +267,8 @@ static void iommu_table_setparms(struct struct iommu_table *tbl) {struct device_node *node; -const unsigned long *basep, *sizep; +const unsigned long *basep; +const u32 *sizep; node = (struct device_node *)phb->arch_data;
Quoted 5 lines · Nathan Lynch
The "linux,tce-size" property is only 32 bits (see
prom_initialize_tce_table()inarch/powerpc/kernel/prom_init.c). Treating it as an unsigned long iniommu_table_setparms()leads to access beyond the end of the property's buffer, so we pass garbage to thememset()in that function.
Probably needs to go into stable as well. Do you know if RHEL5 is affected too ?
Cheers
Ben.
Quoted 52 lines
[boot]0020 XICS Init
i8259 legacy interrupt controller initialized [boot]0021 XICS Done
PID hash table entries: 4096 (order: 12, 32768 bytes) cpu0x0: Vector: 300 (Data Access) at [c0000000fe783850]pc: c000000000035e90: .memset+0x60/0xfc lr: c000000000044fa4: .iommu_table_setparms+0xb0/0x158 sp: c0000000fe783ad0msr: 9000000000009032
dar:c000000100000000
dsisr: 42010000
current =0xc00000000450e810
paca =0xc000000000411580
pid = 1, comm = swapper
enter ? for help
[link register ]c000000000044fa4.iommu_table_setparms+0xb0/0x158
[c0000000fe783ad0]c000000000044f4c.iommu_table_setparms+0x58/0x158 (unreliable)
[c0000000fe783b70]c00000000004529c.iommu_bus_setup_pSeries+0x1c4/0x254 [c0000000fe783c00] c00000000002b8ac .do_bus_setup+0x3c/0xe4 [c0000000fe783c80] c00000000002c924 .pcibios_fixup_bus+0x64/0xd8 [c0000000fe783d00] c0000000001a2d5c .pci_scan_child_bus+0x6c/0x10c [c0000000fe783da0] c00000000002be28 .scan_phb+0x17c/0x1b4 [c0000000fe783e40] c0000000003cfa00 .pcibios_init+0x58/0x19c [c0000000fe783ec0] c0000000000094b4 .init+0x1e8/0x3d8 [c0000000fe783f90] c000000000026e54 .kernel_thread+0x4c/0x68Signed-off-by Nathan Lynch <[email protected]>---
I believe this is a regression since 2.6.18, so please push for inclusion in 2.6.19.
arch/powerpc/platforms/pseries/iommu.c +2
-1@@ -267,7 +267,8 @@ static void iommu_table_setparms(struct struct iommu_table *tbl) {struct device_node *node; -const unsigned long *basep, *sizep; +const unsigned long *basep; +const u32 *sizep; node = (struct device_node *)phb->arch_data;_______________________________________________
Linuxppc-dev mailing list
[email protected]
https://ozlabs.org/mailman/listinfo/linuxppc-dev
Quoted 9 lines · Benjamin Herrenschmidt
Quoted 5 lines · Nathan Lynch
The "linux,tce-size" property is only 32 bits (see
prom_initialize_tce_table()inarch/powerpc/kernel/prom_init.c). Treating it as an unsigned long iniommu_table_setparms()leads to access beyond the end of the property's buffer, so we pass garbage to thememset()in that function.Probably needs to go into stable as well. Do you know if RHEL5 is affected too ?
AFAICT this was introduced by the "Constify & voidify get_property()" series which went in post-2.6.18 so no backport for -stable is needed. Not sure about RHEL5.
Quoted 4 lines
AFAICT this was introduced by the "Constify & voidify
get_property()" series which went in post-2.6.18 so no backport for -stable is needed. Not sure about RHEL5.
That patch (954a46e2d5aec6f59976ddeb1d232b486e59b54a) did touch the declaration of sizep, but only changed it from an unsigned long * to an const unsigned long *.
Looking at the git log, it's been an unsigned long from the beginning (ie, being moved from arch/ppc64). You may need to check the stable and RHEL trees explicitly.
Regards,
Jeremy
Quoted 8 lines · On Fri, 6 Oct 2006 12:16:30 +1000 Jeremy Kerr <[email protected]> wrote:
Quoted 4 lines
AFAICT this was introduced by the "Constify & voidify
get_property()" series which went in post-2.6.18 so no backport for -stable is needed. Not sure about RHEL5.That patch (
954a46e2d5aec6f59976ddeb1d232b486e59b54a) did touch the declaration of sizep, but only changed it from an unsigned long * to an const unsigned long *.
No it didn't:
file:d67af2c657544ef5b0a6386a9c45a31de2fb4681 -> file:bbf2e34dc3582442a9e65c2a404c4e7ddd31619e
arch/powerpc/platforms/pseries/iommu.c +1 -3
@@ -267,13 +267,12 @@ static void iommu_table_setparms(struct struct iommu_table *tbl) {struct device_node *node; -unsigned long *basep; -unsigned int *sizep; +const unsigned long *basep, *sizep; -Olof
Jeremy Kerr
Acked-byQuoted 2 lines
No it didn't:
Wow, looks like I can't read :)
Jeremy
Quoted 34 lines · On Wed, 4 Oct 2006 22:28:00 -0500 Nathan Lynch <[email protected]> wrote:
The "linux,tce-size" property is only 32 bits (see
prom_initialize_tce_table()inarch/powerpc/kernel/prom_init.c). Treating it as an unsigned long iniommu_table_setparms()leads to access beyond the end of the property's buffer, so we pass garbage to thememset()in that function.[boot]0020 XICS Init
i8259 legacy interrupt controller initialized [boot]0021 XICS Done
PID hash table entries: 4096 (order: 12, 32768 bytes) cpu0x0: Vector: 300 (Data Access) at [c0000000fe783850]pc: c000000000035e90: .memset+0x60/0xfc lr: c000000000044fa4: .iommu_table_setparms+0xb0/0x158 sp: c0000000fe783ad0msr: 9000000000009032
dar:c000000100000000
dsisr: 42010000
current =0xc00000000450e810
paca =0xc000000000411580
pid = 1, comm = swapper
enter ? for help
[link register ]c000000000044fa4.iommu_table_setparms+0xb0/0x158
[c0000000fe783ad0]c000000000044f4c.iommu_table_setparms+0x58/0x158 (unreliable)
[c0000000fe783b70]c00000000004529c.iommu_bus_setup_pSeries+0x1c4/0x254 [c0000000fe783c00] c00000000002b8ac .do_bus_setup+0x3c/0xe4 [c0000000fe783c80] c00000000002c924 .pcibios_fixup_bus+0x64/0xd8 [c0000000fe783d00] c0000000001a2d5c .pci_scan_child_bus+0x6c/0x10c [c0000000fe783da0] c00000000002be28 .scan_phb+0x17c/0x1b4 [c0000000fe783e40] c0000000003cfa00 .pcibios_init+0x58/0x19c [c0000000fe783ec0] c0000000000094b4 .init+0x1e8/0x3d8 [c0000000fe783f90] c000000000026e54 .kernel_thread+0x4c/0x68
Quoted 1 line
Signed-off-by Nathan Lynch <[email protected]>
It does not reproduce on the OpenPower box I have, but it's clearly a bug. Good catch.