| Message ID | 20241003035333.49261-5-mhklinux@outlook.com |
|---|---|
| State | New |
| Headers | show |
| Series | hyper-v: Don't assume cpu_possible_mask is dense | expand |
From: mhkelley58@gmail.com <mhkelley58@gmail.com> Sent: Wednesday, October 2, 2024 8:54 PM > > Current code allocates the stor_chns array with size num_possible_cpus(). > This code assumes cpu_possible_mask is dense, which is not true in > the general case per [1]. If cpu_possible_mask is sparse, the array > might be indexed by a value beyond the size of the array. > > However, the configurations that Hyper-V provides to guest VMs on x86 > and ARM64 hardware, in combination with how architecture specific code > assigns Linux CPU numbers, *does* always produce a dense cpu_possible_mask. > So the dense assumption is not currently causing failures. But for > robustness against future changes in how cpu_possible_mask is populated, > update the code to no longer assume dense. > > The correct approach is to allocate and initialize the array using size > "nr_cpu_ids". While this leaves unused array entries corresponding to > holes in cpu_possible_mask, the holes are assumed to be minimal and hence > the amount of memory wasted by unused entries is minimal. > > [1] https://lore.kernel.org/lkml/SN6PR02MB4157210CC36B2593F8572E5ED4692@SN6PR02MB4157.namprd02.prod.outlook.com/ > > Signed-off-by: Michael Kelley <mhklinux@outlook.com> Martin or James -- This entire series was Acked-by: Peter Zijlstra [1]. Patch 5 of the series was picked up by the net-next tree a few weeks back and is in 6.13-rc1. Do you need anything else to pick up this single patch in the appropriate scsi tree? I'll separately pursue getting Patches 1 thru 3 of the series picked up by the Hyper-V tree. There's no interdependency between the patches in the series, so they can each go separately. Michael [1] https://lore.kernel.org/linux-hyperv/20241004100742.GO18071@noisy.programming.kicks-ass.net/ > --- > drivers/scsi/storvsc_drv.c | 13 ++++++------- > 1 file changed, 6 insertions(+), 7 deletions(-) > > diff --git a/drivers/scsi/storvsc_drv.c b/drivers/scsi/storvsc_drv.c > index 11b3fc3b24c9..f2beb6b23284 100644 > --- a/drivers/scsi/storvsc_drv.c > +++ b/drivers/scsi/storvsc_drv.c > @@ -917,14 +917,13 @@ static int storvsc_channel_init(struct hv_device *device, bool > is_fc) > > /* > * Allocate state to manage the sub-channels. > - * We allocate an array based on the numbers of possible CPUs > - * (Hyper-V does not support cpu online/offline). > - * This Array will be sparseley populated with unique > - * channels - primary + sub-channels. > - * We will however populate all the slots to evenly distribute > - * the load. > + * We allocate an array based on the number of CPU ids. This array > + * is initially sparsely populated for the CPUs assigned to channels: > + * primary + sub-channels. As I/Os are initiated by different CPUs, > + * the slots for all online CPUs are populated to evenly distribute > + * the load across all channels. > */ > - stor_device->stor_chns = kcalloc(num_possible_cpus(), sizeof(void *), > + stor_device->stor_chns = kcalloc(nr_cpu_ids, sizeof(void *), > GFP_KERNEL); > if (stor_device->stor_chns == NULL) > return -ENOMEM; > -- > 2.25.1 >
Michael, >> Current code allocates the stor_chns array with size >> num_possible_cpus(). This code assumes cpu_possible_mask is dense, >> which is not true in the general case per [1]. If cpu_possible_mask >> is sparse, the array might be indexed by a value beyond the size of >> the array. Applied to 6.14/scsi-staging, thanks!
diff --git a/drivers/scsi/storvsc_drv.c b/drivers/scsi/storvsc_drv.c index 11b3fc3b24c9..f2beb6b23284 100644 --- a/drivers/scsi/storvsc_drv.c +++ b/drivers/scsi/storvsc_drv.c @@ -917,14 +917,13 @@ static int storvsc_channel_init(struct hv_device *device, bool is_fc) /* * Allocate state to manage the sub-channels. - * We allocate an array based on the numbers of possible CPUs - * (Hyper-V does not support cpu online/offline). - * This Array will be sparseley populated with unique - * channels - primary + sub-channels. - * We will however populate all the slots to evenly distribute - * the load. + * We allocate an array based on the number of CPU ids. This array + * is initially sparsely populated for the CPUs assigned to channels: + * primary + sub-channels. As I/Os are initiated by different CPUs, + * the slots for all online CPUs are populated to evenly distribute + * the load across all channels. */ - stor_device->stor_chns = kcalloc(num_possible_cpus(), sizeof(void *), + stor_device->stor_chns = kcalloc(nr_cpu_ids, sizeof(void *), GFP_KERNEL); if (stor_device->stor_chns == NULL) return -ENOMEM;