iommu/dmar: Fix crash on boot when DMAR is disabled
PATCH 1/6 iommu/dmar: Fix crash on boot when DMAR is disabled
By default CONFIG_INTEL_IOMMU_DEFAULT_ON is not set and thus dmar_disabled variable is set.
Intel IOMMU driver based on above doesn't set intel_iommu_enabled variable.
The commit b0119e870837 ("iommu: Introduce new 'struct iommu_device'") mistakenly assumes it never happens and tries to unregister not ever registered resources, which crashes the kernel at boot time:
BUG: unable to handle kernel NULL pointer dereference at 0000000000000008 IP: iommu_device_unregister+0x31/0x60
Make unregister procedure conditional in free_iommu().
--- drivers/iommu/dmar.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
drivers/iommu/dmar.c +4 -2
@@ -1108,8 +1108,10 @@ static int alloc_iommu(struct dmar_drhd_unit *drhd)static void free_iommu (struct intel_iommu *iommu) { -iommu_device_sysfs_remove (&iommu->iommu); -iommu_device_unregister (&iommu->iommu); +if (intel_iommu_enabled) { +iommu_device_unregister (&iommu->iommu); +iommu_device_sysfs_remove (&iommu->iommu); + }if (iommu->irq) {if (iommu->pr_irq) {
Signature · 1 line
2.11.0
A message that is not in the archive (sent off-list or not yet received): 20170215144226.34519-1-andriy.shevchenko-VuQAYsv1563Yd54FQh9/[email protected]
PATCH 2/6 iommu/dmar: Rectify return code handling in detect_intel_iommu()
There is inconsistency in return codes across the functions called from detect_intel_iommu().
Make it consistent and propagate return code to the caller.
--- drivers/iommu/dmar.c | 15 +++++++-------- 1 file changed, 7 insertions(+), 8 deletions(-)
drivers/iommu/dmar.c +7 -8
@@ -551,7 +551,7 @@ static int __init dmar_table_detect(void) status =AE_NOT_FOUND ; } -return (ACPI_SUCCESS (status) ?1 :0 ); +return (ACPI_SUCCESS (status) ?0 : -ENOENT ); } static int dmar_walk_remapping_entries(struct acpi_dmar_header *start,
@@ -891,17 +891,17 @@ int __init detect_intel_iommu(void)down_write (&dmar_global_lock); ret =dmar_table_detect (); -if (ret) - ret = !dmar_walk_dmar_table ((struct acpi_table_dmar *)dmar_tbl, - &validate_drhd_cb); -if (ret && !no_iommu && !iommu_detected && !dmar_disabled) { +if (!ret) + ret =dmar_walk_dmar_table ((struct acpi_table_dmar *)dmar_tbl, + &validate_drhd_cb); +if (!ret && !no_iommu && !iommu_detected && !dmar_disabled) { iommu_detected =1 ;/* Make sure ACS will be enabled */ pci_request_acs (); }#ifdef CONFIG_X86 -if (ret) +if (!ret) x86_init.iommu.iommu_init = intel_iommu_init;#endif
@@ -911,10 +911,9 @@ int __init detect_intel_iommu(void) }up_write (&dmar_global_lock); -return ret ?1 : -ENODEV ; +return ret; } -static void unmap_iommu (struct intel_iommu *iommu) {iounmap (iommu->reg);
Signature · 1 line
2.11.0
A message that is not in the archive (sent off-list or not yet received): 20170215144226.34519-2-andriy.shevchenko-VuQAYsv1563Yd54FQh9/[email protected]
Re: [PATCH v1 2/6] iommu/dmar: Rectify return code handling in detect_intel_iommu()
Quoted 5 lines · Andy Shevchenko
There is inconsistency in return codes across the functions called from
detect_intel_iommu().Make it consistent and propagate return code to the caller.
drivers/iommu/dmar.c int __init detect_intel_iommu(void) · line 914
Quoted 6 lines
drivers/iommu/dmar.c +0
-0@@ -911,10 +911,9 @@ int __init detect_intel_iommu(void) } up_write(&dmar_global_lock);drivers/iommu/dmar.c +1
-1-return ret ?1 : -ENODEV ; +return ret;
Accidentally it works since we have no late_init() defined.
I will fix this in v2, but will wait for the comments regarding the rest of the series.
Signature · 6 lines
Andy Shevchenko <[email protected]> Intel Finland Oy _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
PATCH 3/6 iommu/dmar: Return directly from a loop in dmar_dev_scope_status()
In reply to a message not in the archiveThere is no need to have a temporary variable.
--- drivers/iommu/dmar.c | 14 ++++++++------ 1 file changed, 8 insertions(+), 6 deletions(-)
drivers/iommu/dmar.c +8 -6
@@ -557,11 +557,10 @@ static int __init dmar_table_detect(void)static int dmar_walk_remapping_entries (struct acpi_dmar_header *start,size_t len,struct dmar_res_callback *cb) { -int ret =0 ;struct acpi_dmar_header *iter, *next;struct acpi_dmar_header *end = ((void *)start) + len; -for (iter = start; iter < end && ret ==0 ; iter = next) { +for (iter = start; iter < end; iter = next) { next = (void *)iter + iter->length;if (iter->length ==0 ) {/* Avoid looping forever on bad ACPI tables */
@@ -570,8 +569,7 @@ static int dmar_walk_remapping_entries(struct acpi_dmar_header *start, }else if (next > end) {/* Avoid passing table end */ pr_warn (FW_BUG "Record passes table end\n" ); - ret = -EINVAL ; -break ; +return -EINVAL ; }if (cb->print_entry)
@@ -582,15 +580,19 @@ static int dmar_walk_remapping_entries(struct acpi_dmar_header *start,pr_debug ("Unknown DMAR structure type %d\n" , iter->type); }else if (cb->cb[iter->type]) { +int ret; + ret = cb->cb[iter->type](iter, cb->arg[iter->type]); +if (ret) +return ret; }else if (!cb->ignore_unhandled) {pr_warn ("No handler for DMAR structure type %d\n" , iter->type); - ret = -EINVAL ; +return -EINVAL ; } } -return ret; +return 0 ; } static inline int dmar_walk_dmar_table(struct acpi_table_dmar *dmar,
Signature · 1 line
2.11.0
PATCH 4/6 iommu/dmar: Remove redundant assignment of ret
In reply to a message not in the archiveThere is no need to assign ret to 0 in some cases. Moreover it might shadow some errors in the future.
Remove such assignments.
--- drivers/iommu/dmar.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/iommu/dmar.c +2 -2
@@ -391,7 +391,7 @@ static int dmar_parse_one_drhd(struct acpi_dmar_header *header, void *arg) {struct acpi_dmar_hardware_unit *drhd;struct dmar_drhd_unit *dmaru; -int ret =0 ; +int ret; drhd = (struct acpi_dmar_hardware_unit *)header; dmaru =dmar_find_dmaru (drhd);
@@ -609,8 +609,8 @@ static int __initparse_dmar_table (void) {struct acpi_table_dmar *dmar; -int ret =0 ;int drhd_count =0 ; +int ret;struct dmar_res_callback cb = { .print_entry = true, .ignore_unhandled = true,
Signature · 1 line
2.11.0
PATCH 5/6 iommu/dmar: Remove redundant ' != 0' when check return code
In reply to a message not in the archiveUsual patter when we check for return code, which might be negative errno, is either ret or !ret.
Remove extra ' != 0' from condition.
--- drivers/iommu/dmar.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/iommu/dmar.c +1 -1
@@ -311,7 +311,7 @@ static int dmar_pci_bus_add_dev(struct dmar_pci_notify_info *info) ((void *)drhd) + drhd->header.length, dmaru->segment, dmaru->devices, dmaru->devices_cnt); -if (ret !=0 ) +if (ret)break ; }if (ret >=0 )
Signature · 1 line
2.11.0
PATCH 6/6 iommu/vt-d: Use lo_hi_readq() / lo_hi_writeq()
In reply to a message not in the archiveThere is already helper functions to do 64-bit I/O on 32-bit machines or buses, thus we don't need to reinvent the wheel.
--- include/linux/intel-iommu.h | 18 ++---------------- 1 file changed, 2 insertions(+), 16 deletions(-)
include/linux/intel-iommu.h +2 -16
@@ -30,6 +30,8 @@#include <linux/mmu_notifier.h> #include <linux/list.h> #include <linux/iommu.h> +#include <linux/io-64-nonatomic-lo-hi.h> +#include <asm/cacheflush.h> #include <asm/iommu.h>
@@ -72,24 +74,8 @@#define OFFSET_STRIDE (9) -#ifdef CONFIG_64BIT #define dmar_readq (a) readq(a)#define dmar_writeq (a,v) writeq(v,a) -#else -static inline u64 dmar_readq (void __iomem *addr) -{ -u32 lo, hi; - lo =readl (addr); - hi =readl (addr +4 ); -return (((u64 ) hi) <<32 ) + lo; -} - -static inline void dmar_writeq (void __iomem *addr,u64 val) -{ -writel ((u32 )val, addr); -writel ((u32)(val >>32 ), addr +4 ); -} -#endif #define DMAR_VER_MAJOR (v) (((v) & 0xf0) >> 4)#define DMAR_VER_MINOR (v) ((v) & 0x0f)
Signature · 1 line
2.11.0
Quoted 10 lines · Andy Shevchenko
By default
CONFIG_INTEL_IOMMU_DEFAULT_ONis not set and thusdmar_disabledvariable is set.Intel IOMMU driver based on above doesn't set
intel_iommu_enabledvariable.The commit
b0119e870837("iommu: Introduce new 'structiommu_device'") mistakenly assumes it never happens and tries to unregister not ever registered resources, which crashes the kernel at boot time:
Quoted 5 lines
BUG: unable to handle kernel NULL pointer dereference at 0000000000000008 IP: iommu_device_unregister+0x31/0x60Make unregister procedure conditional in
free_iommu().
Joerg, it looks like bug is still present. Can we get the fix merged? Or do you have better ideas?
Quoted 26 lines
Fixes b0119e870837 ("iommu: Introduce new 'struct iommu_device'")Cc Joerg Roedel <[email protected]>Signed-off-by Andy Shevchenko <[email protected]>---
drivers/iommu/dmar.c | 6 ++++--1 file changed, 4 insertions(+), 2 deletions(-)drivers/iommu/dmar.c +0
-0@@ -1108,8 +1108,10 @@ static int alloc_iommu(struct dmar_drhd_unit *drhd)
static voidfree_iommu(struct intel_iommu *iommu){drivers/iommu/dmar.c +4
-2-iommu_device_sysfs_remove (&iommu->iommu); -iommu_device_unregister (&iommu->iommu); +if (intel_iommu_enabled) { +iommu_device_unregister (&iommu->iommu); +iommu_device_sysfs_remove (&iommu->iommu); + }
if (iommu->irq) {
if (iommu->pr_irq) {
Signature · 6 lines
Andy Shevchenko <[email protected]> Intel Finland Oy _______________________________________________ iommu mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/iommu
Hi Andy,
Quoted 10 lines · Andy Shevchenko
By default
CONFIG_INTEL_IOMMU_DEFAULT_ONis not set and thusdmar_disabledvariable is set.Intel IOMMU driver based on above doesn't set
intel_iommu_enabledvariable.The commit
b0119e870837("iommu: Introduce new 'structiommu_device'") mistakenly assumes it never happens and tries to unregister not ever registered resources, which crashes the kernel at boot time:
Quoted 8 lines
BUG: unable to handle kernel NULL pointer dereference at 0000000000000008 IP: iommu_device_unregister+0x31/0x60Make unregister procedure conditional in
free_iommu().Fixes b0119e870837 ("iommu: Introduce new 'struct iommu_device'")Cc Joerg Roedel <[email protected]>
Thanks for the patch, I applied it and will send it upstream asap.
Joerg
A message that is not in the archive (sent off-list or not yet received): [email protected]
Quoted 18 lines · Joerg Roedel
Hi Andy,
Quoted 15 lines · Andy Shevchenko
By default
CONFIG_INTEL_IOMMU_DEFAULT_ONis not set and thusdmar_disabledvariable is set.Intel IOMMU driver based on above doesn't set
intel_iommu_enabledvariable.The commit
b0119e870837("iommu: Introduce new 'structiommu_device'")
mistakenly assumes it never happens and tries to unregister not ever registered resources, which crashes the kernel at boot time:BUG: unable to handle kernel NULL pointer dereference at 0000000000000008 IP: iommu_device_unregister+0x31/0x60
Quoted 7 lines
Quoted 5 lines
Make unregister procedure conditional in
free_iommu().Fixes b0119e870837 ("iommu: Introduce new 'struct iommu_device'")Cc Joerg Roedel <[email protected]>Thanks for the patch, I applied it and will send it upstream asap.
Thanks! Any comments on the rest? I would like to send v2.
Signature · 2 lines
Andy Shevchenko <andriy.shevchenko-VuQAYsv1563Yd54FQh9/[email protected]> Intel Finland Oy