2021-04-15 20:45:45

by Steve Wahl

[permalink] [raw]
Subject: [PATCH] perf/x86/intel/uncore: Avoid null dereferences (uncore_extra_pci_dev)

If an uncore has no pci_init routine, or that routine fails,
uncore_pci_init is not called, and memory is not allocated for
uncore_extra_pci_dev.

So check to make sure uncore_extra_pci_dev is not NULL before use.

And fix the case that led us to discover the null derefs; don't fail
snbep_pci2phy_map_init if BIOS doesn't supply pcibus_to_node
information.

Fixes: 9a7832ce3d92 ("perf/x86/intel/uncore: With > 8 nodes, get pci bus die id from NUMA info")

Signed-off-by: Steve Wahl <[email protected]>
---
arch/x86/events/intel/uncore_snbep.c | 10 +++++++---
1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/arch/x86/events/intel/uncore_snbep.c b/arch/x86/events/intel/uncore_snbep.c
index b79951d0707c..14c24356a2fa 100644
--- a/arch/x86/events/intel/uncore_snbep.c
+++ b/arch/x86/events/intel/uncore_snbep.c
@@ -1373,11 +1373,11 @@ static int snbep_pci2phy_map_init(int devid, int nodeid_loc, int idmap_loc, bool
/*
* The nodeid and idmap registers only contain enough
* information to handle 8 nodes. On systems with more
- * than 8 nodes, we need to rely on NUMA information,
+ * than 8 nodes, if available we rely on NUMA information,
* filled in from BIOS supplied information, to determine
* the topology.
*/
- if (nr_node_ids <= 8) {
+ if ((nr_node_ids <= 8) || (pcibus_to_node(ubox_dev->bus) == -1)) {
/* get the Node ID of the local register */
err = pci_read_config_dword(ubox_dev, nodeid_loc, &config);
if (err)
@@ -2865,7 +2865,9 @@ void hswep_uncore_cpu_init(void)
hswep_uncore_cbox.num_boxes = boot_cpu_data.x86_max_cores;

/* Detect 6-8 core systems with only two SBOXes */
- if (uncore_extra_pci_dev[pkg].dev[HSWEP_PCI_PCU_3]) {
+ if (!uncore_extra_pci_dev)
+ hswep_uncore_sbox.num_boxes = 2;
+ else if (uncore_extra_pci_dev[pkg].dev[HSWEP_PCI_PCU_3]) {
u32 capid4;

pci_read_config_dword(uncore_extra_pci_dev[pkg].dev[HSWEP_PCI_PCU_3],
@@ -3243,6 +3245,8 @@ void bdx_uncore_cpu_init(void)
if (boot_cpu_data.x86_model == 86) {
uncore_msr_uncores[BDX_MSR_UNCORE_SBOX] = NULL;
/* Detect systems with no SBOXes */
+ } else if (!uncore_extra_pci_dev) {
+ bdx_msr_uncores[BDX_MSR_UNCORE_SBOX] = NULL;
} else if (uncore_extra_pci_dev[pkg].dev[HSWEP_PCI_PCU_3]) {
struct pci_dev *pdev;
u32 capid4;
--
2.26.2


2021-04-15 21:22:49

by Liang, Kan

[permalink] [raw]
Subject: Re: [PATCH] perf/x86/intel/uncore: Avoid null dereferences (uncore_extra_pci_dev)

Hi Steve,

On 4/15/2021 4:37 PM, Steve Wahl wrote:
> If an uncore has no pci_init routine, or that routine fails,
> uncore_pci_init is not called, and memory is not allocated for
> uncore_extra_pci_dev.
>
> So check to make sure uncore_extra_pci_dev is not NULL before use.
>

I think more after yesterday's discussion. There may be a better
solution than this.

Actually, we don't have to probe all the PCU devices and stores them
into the uncore_extra_pci_dev for the cpu_init().
We just need to pick up the first PCU device and check the existence of
the SBOX once.

I will send out a patch shortly.

> And fix the case that led us to discover the null derefs; don't fail
> snbep_pci2phy_map_init if BIOS doesn't supply pcibus_to_node
> information.
>
> Fixes: 9a7832ce3d92 ("perf/x86/intel/uncore: With > 8 nodes, get pci bus die id from NUMA info")

In theory, the pci_read_config_dword() may fails as well. It has
possible that the issue can still be observed before the "> 8 nodes" patch.

I think the fixes should be 5306c31c5733 ("perf/x86/uncore/hsw-ep:
Handle systems with only two SBOXes")

Thanks,
Kan

> Signed-off-by: Steve Wahl <[email protected]>
> ---
> arch/x86/events/intel/uncore_snbep.c | 10 +++++++---
> 1 file changed, 7 insertions(+), 3 deletions(-)
>
> diff --git a/arch/x86/events/intel/uncore_snbep.c b/arch/x86/events/intel/uncore_snbep.c
> index b79951d0707c..14c24356a2fa 100644
> --- a/arch/x86/events/intel/uncore_snbep.c
> +++ b/arch/x86/events/intel/uncore_snbep.c
> @@ -1373,11 +1373,11 @@ static int snbep_pci2phy_map_init(int devid, int nodeid_loc, int idmap_loc, bool
> /*
> * The nodeid and idmap registers only contain enough
> * information to handle 8 nodes. On systems with more
> - * than 8 nodes, we need to rely on NUMA information,
> + * than 8 nodes, if available we rely on NUMA information,
> * filled in from BIOS supplied information, to determine
> * the topology.
> */
> - if (nr_node_ids <= 8) {
> + if ((nr_node_ids <= 8) || (pcibus_to_node(ubox_dev->bus) == -1)) {
> /* get the Node ID of the local register */
> err = pci_read_config_dword(ubox_dev, nodeid_loc, &config);
> if (err)
> @@ -2865,7 +2865,9 @@ void hswep_uncore_cpu_init(void)
> hswep_uncore_cbox.num_boxes = boot_cpu_data.x86_max_cores;
>
> /* Detect 6-8 core systems with only two SBOXes */
> - if (uncore_extra_pci_dev[pkg].dev[HSWEP_PCI_PCU_3]) {
> + if (!uncore_extra_pci_dev)
> + hswep_uncore_sbox.num_boxes = 2;
> + else if (uncore_extra_pci_dev[pkg].dev[HSWEP_PCI_PCU_3]) {
> u32 capid4;
>
> pci_read_config_dword(uncore_extra_pci_dev[pkg].dev[HSWEP_PCI_PCU_3],
> @@ -3243,6 +3245,8 @@ void bdx_uncore_cpu_init(void)
> if (boot_cpu_data.x86_model == 86) {
> uncore_msr_uncores[BDX_MSR_UNCORE_SBOX] = NULL;
> /* Detect systems with no SBOXes */
> + } else if (!uncore_extra_pci_dev) {
> + bdx_msr_uncores[BDX_MSR_UNCORE_SBOX] = NULL;
> } else if (uncore_extra_pci_dev[pkg].dev[HSWEP_PCI_PCU_3]) {
> struct pci_dev *pdev;
> u32 capid4;
>

2021-04-15 22:16:49

by Liang, Kan

[permalink] [raw]
Subject: Re: [PATCH] perf/x86/intel/uncore: Avoid null dereferences (uncore_extra_pci_dev)



On 4/15/2021 5:19 PM, Liang, Kan wrote:
> Hi Steve,
>
> On 4/15/2021 4:37 PM, Steve Wahl wrote:
>> If an uncore has no pci_init routine, or that routine fails,
>> uncore_pci_init is not called, and memory is not allocated for
>> uncore_extra_pci_dev.
>>
>> So check to make sure uncore_extra_pci_dev is not NULL before use.
>>
>
> I think more after yesterday's discussion. There may be a better
> solution than this.
>
> Actually, we don't have to probe all the PCU devices and stores them
> into the uncore_extra_pci_dev for the cpu_init().
> We just need to pick up the first PCU device and check the existence of
> the SBOX once.
>
> I will send out a patch shortly.

Here is the patch I mentioned.

https://lore.kernel.org/lkml/[email protected]

Thanks,
Kan

>
>> And fix the case that led us to discover the null derefs; don't fail
>> snbep_pci2phy_map_init if BIOS doesn't supply pcibus_to_node
>> information.
>>
>> Fixes: 9a7832ce3d92 ("perf/x86/intel/uncore: With > 8 nodes, get pci
>> bus die id from NUMA info")
>
> In theory, the pci_read_config_dword() may fails as well. It has
> possible that the issue can still be observed before the "> 8 nodes" patch.
>
> I think the fixes should be 5306c31c5733 ("perf/x86/uncore/hsw-ep:
> Handle systems with only two SBOXes")
>
> Thanks,
> Kan
>
>> Signed-off-by: Steve Wahl <[email protected]>
>> ---
>>   arch/x86/events/intel/uncore_snbep.c | 10 +++++++---
>>   1 file changed, 7 insertions(+), 3 deletions(-)
>>
>> diff --git a/arch/x86/events/intel/uncore_snbep.c
>> b/arch/x86/events/intel/uncore_snbep.c
>> index b79951d0707c..14c24356a2fa 100644
>> --- a/arch/x86/events/intel/uncore_snbep.c
>> +++ b/arch/x86/events/intel/uncore_snbep.c
>> @@ -1373,11 +1373,11 @@ static int snbep_pci2phy_map_init(int devid,
>> int nodeid_loc, int idmap_loc, bool
>>           /*
>>            * The nodeid and idmap registers only contain enough
>>            * information to handle 8 nodes.  On systems with more
>> -         * than 8 nodes, we need to rely on NUMA information,
>> +         * than 8 nodes, if available we rely on NUMA information,
>>            * filled in from BIOS supplied information, to determine
>>            * the topology.
>>            */
>> -        if (nr_node_ids <= 8) {
>> +        if ((nr_node_ids <= 8) || (pcibus_to_node(ubox_dev->bus) ==
>> -1)) {
>>               /* get the Node ID of the local register */
>>               err = pci_read_config_dword(ubox_dev, nodeid_loc, &config);
>>               if (err)
>> @@ -2865,7 +2865,9 @@ void hswep_uncore_cpu_init(void)
>>           hswep_uncore_cbox.num_boxes = boot_cpu_data.x86_max_cores;
>>       /* Detect 6-8 core systems with only two SBOXes */
>> -    if (uncore_extra_pci_dev[pkg].dev[HSWEP_PCI_PCU_3]) {
>> +    if (!uncore_extra_pci_dev)
>> +        hswep_uncore_sbox.num_boxes = 2;
>> +    else if (uncore_extra_pci_dev[pkg].dev[HSWEP_PCI_PCU_3]) {
>>           u32 capid4;
>>
>> pci_read_config_dword(uncore_extra_pci_dev[pkg].dev[HSWEP_PCI_PCU_3],
>> @@ -3243,6 +3245,8 @@ void bdx_uncore_cpu_init(void)
>>       if (boot_cpu_data.x86_model == 86) {
>>           uncore_msr_uncores[BDX_MSR_UNCORE_SBOX] = NULL;
>>       /* Detect systems with no SBOXes */
>> +    } else if (!uncore_extra_pci_dev) {
>> +        bdx_msr_uncores[BDX_MSR_UNCORE_SBOX] = NULL;
>>       } else if (uncore_extra_pci_dev[pkg].dev[HSWEP_PCI_PCU_3]) {
>>           struct pci_dev *pdev;
>>           u32 capid4;
>>