Re: [PATCH V5 3/4] perf/x86/intel/uncore: Fix die ID init and look up bugs

[Chen, Zide]

On 3/26/2026 7:03 PM, Mi, Dapeng wrote:
> 
> On 3/27/2026 7:57 AM, Chen, Zide wrote:
>>
>> On 3/25/2026 11:03 PM, Mi, Dapeng wrote:
>>> Zide, Sashiko gave some comments on this patch. Could you please have a
>>> look if they are reasonable? Thanks.
>>>
>>> https://sashiko.dev/#/patchset/20260324214932.10068-1-zide.chen%40intel.com
>> 1. Regarding the concern that this change may replace an offline node's
>> -1 with the die ID of an adjacent online node, I do not think this is an
>> issue.
>>
>> After this fix, the logic is the same for both (nr_node_ids <= 8) and
>> (nr_node_ids > 8): map->pbus_to_dieid[bus] may be written with an
>> invalid die_id (e.g., -1). This is not an error and is expected in some
>> cases. We should continue to populate the map->pbus_to_dieid[] array.
>>
>> Regardless of the traversal order (as determined by the reverse
>> argument), for a given die, the UBOX device is expected to reside on the
>> first valid bus in the die it is affined to.
>>
>> Under the current assignment algorithm, all buses following a UBOX
>> device, up to the next UBOX device or the end of traversal, are assigned
>> the same die ID.
>>
>> For example, on SPR, there are two UBOX devices: one device on bus 0x7e
>> in die 0, and another on bus 0xfe in die 1. With reversed traversal
>> order, buses 0xff–0x7f are assigned die ID 1, while buses 0x7e–0x00 are
>> assigned die ID 0.
>>
>> If all CPUs in die 1 are offline, then buses 0xff–0x7f are assigned -1.
>> This is fine.
>>
>> That being said, the die ID for invalid buses is not consistent, which
>> is not ideal.
> 
> Yes, for the case with 2 sockets and socket 1 is offline, it's correct. But
> assume there are 4 sockets (0/1/2/3), buses 0x0-0x3f are attached to socket
> 0, buses 0x40-0x7f are attached to socket 1, buses 0x80-0xbf are attached
> to socket 2 and buses 0xc0-0xff are attached to socket 3, the socket 2 is
> offline. In reverse order, the die id of  buses 0x80-0xbf would be
> overwritten to 3 instead of -1, right?

Yes, that’s correct theoretically.

> 
> But it seems there is not a good way to fix this issue and the function
> spr_update_device_location() won't really find the ubox device of socket 2
> since socket 2 has been offline. So it won't cause a real issue. 

Yes. In this algorithm, for any "invalid" buses — whether they belong to
an offline die or are not present in the system — the associated
map->pbus_to_dieid[] entry is arbitrary.

PMON units on offline dies won't be enumerated.


>>
>> 2. Regarding the repeated snbep_pci2phy_map_init() calls. I wanted a
>> "simple" fix initially.  I may need to split this patch into two
>> separate patches.
>>
>>> On 3/25/2026 5:49 AM, Zide Chen wrote:
>>>> In snbep_pci2phy_map_init(), in the nr_node_ids > 8 path,
>>>> uncore_device_to_die() may return -1 when all CPUs associated
>>>> with the UBOX device are offline.
>>>>
>>>> Remove the WARN_ON_ONCE(die_id == -1) check for two reasons:
>>>>
>>>> - The current code breaks out of the loop. This is incorrect because
>>>>   pci_get_device() does not guarantee iteration in domain or bus order,
>>>>   so additional UBOX devices may be skipped during the scan.
>>>>
>>>> - Returning -EINVAL is incorrect, since marking offline buses with
>>>>   die_id == -1 is expected and should not be treated as an error.
>>>>
>>>> Separately, when NUMA is disabled on a NUMA-capable platform,
>>>> pcibus_to_node() returns NUMA_NO_NODE, causing uncore_device_to_die()
>>>> to return -1 for all PCI devices.  As a result,
>>>> spr_update_device_location(), used on Intel SPR and EMR, ignores the
>>>> corresponding PMON units and does not add them to the RB tree.
>>>>
>>>> Fix this by using uncore_pcibus_to_dieid(), which retrieves topology
>>>> from the UBOX GIDNIDMAP register and works regardless of whether NUMA
>>>> is enabled in Linux.  This requires snbep_pci2phy_map_init() to be
>>>> added in spr_uncore_pci_init().
>>>>
>>>> Keep uncore_device_to_die() only for the nr_node_ids > 8 case, where
>>>> NUMA is expected to be enabled.
>>>>
>>>> Fixes: 9a7832ce3d92 ("perf/x86/intel/uncore: With > 8 nodes, get pci bus die id from NUMA info")
>>>> Fixes: 65248a9a9ee1 ("perf/x86/uncore: Add a quirk for UPI on SPR")
>>>> Tested-by: Steve Wahl <steve.wahl@xxxxxxx>
>>>> Signed-off-by: Zide Chen <zide.chen@xxxxxxxxx>
>>>> ---
>>>> V2:
>>>> - Fix the commit message to note that spr_update_device_location() is
>>>>   used by EMR, not GNR.
>>>> - Rewrite the commit message for clarity.
>>>> - Add a Tested-by tag.
>>>>
>>>> V5:
>>>> - Remove unused variable die_id (Dapeng).
>>>> ---
>>>>  arch/x86/events/intel/uncore.c       |  1 +
>>>>  arch/x86/events/intel/uncore_snbep.c | 17 ++++++++---------
>>>>  2 files changed, 9 insertions(+), 9 deletions(-)
>>>>
>>>> diff --git a/arch/x86/events/intel/uncore.c b/arch/x86/events/intel/uncore.c
>>>> index 786bd51a0d89..e9cc1ba921c5 100644
>>>> --- a/arch/x86/events/intel/uncore.c
>>>> +++ b/arch/x86/events/intel/uncore.c
>>>> @@ -67,6 +67,7 @@ int uncore_die_to_segment(int die)
>>>>  	return bus ? pci_domain_nr(bus) : -EINVAL;
>>>>  }
>>>>  
>>>> +/* Note: This API can only be used when NUMA information is available. */
>>>>  int uncore_device_to_die(struct pci_dev *dev)
>>>>  {
>>>>  	int node = pcibus_to_node(dev->bus);
>>>> diff --git a/arch/x86/events/intel/uncore_snbep.c b/arch/x86/events/intel/uncore_snbep.c
>>>> index 9b51883fd6fd..5ef205a70559 100644
>>>> --- a/arch/x86/events/intel/uncore_snbep.c
>>>> +++ b/arch/x86/events/intel/uncore_snbep.c
>>>> @@ -1413,7 +1413,7 @@ static int topology_gidnid_map(int nodeid, u32 gidnid)
>>>>  static int snbep_pci2phy_map_init(int devid, int nodeid_loc, int idmap_loc, bool reverse)
>>>>  {
>>>>  	struct pci_dev *ubox_dev = NULL;
>>>> -	int i, bus, nodeid, segment, die_id;
>>>> +	int i, bus, nodeid, segment;
>>>>  	struct pci2phy_map *map;
>>>>  	int err = 0;
>>>>  	u32 config = 0;
>>>> @@ -1458,14 +1458,8 @@ static int snbep_pci2phy_map_init(int devid, int nodeid_loc, int idmap_loc, bool
>>>>  				break;
>>>>  			}
>>>>  
>>>> -			map->pbus_to_dieid[bus] = die_id = uncore_device_to_die(ubox_dev);
>>>> -
>>>> +			map->pbus_to_dieid[bus] = uncore_device_to_die(ubox_dev);
>>>>  			raw_spin_unlock(&pci2phy_map_lock);
>>>> -
>>>> -			if (WARN_ON_ONCE(die_id == -1)) {
>>>> -				err = -EINVAL;
>>>> -				break;
>>>> -			}
>>>>  		}
>>>>  	}
>>>>  
>>>> @@ -6420,7 +6414,7 @@ static void spr_update_device_location(int type_id)
>>>>  
>>>>  	while ((dev = pci_get_device(PCI_VENDOR_ID_INTEL, device, dev)) != NULL) {
>>>>  
>>>> -		die = uncore_device_to_die(dev);
>>>> +		die = uncore_pcibus_to_dieid(dev->bus);
>>>>  		if (die < 0)
>>>>  			continue;
>>>>  
>>>> @@ -6444,6 +6438,11 @@ static void spr_update_device_location(int type_id)
>>>>  
>>>>  int spr_uncore_pci_init(void)
>>>>  {
>>>> +	int ret = snbep_pci2phy_map_init(0x3250, SKX_CPUNODEID, SKX_GIDNIDMAP, true);
>>>> +
>>>> +	if (ret)
>>>> +		return ret;
>>>> +
>>>>  	/*
>>>>  	 * The discovery table of UPI on some SPR variant is broken,
>>>>  	 * which impacts the detection of both UPI and M3UPI uncore PMON.