Hi Mukesh, On Fri, 24 Jul 2026 12:14:08 -0700 Mukesh R <[email protected]> wrote:
> On 7/23/26 15:21, Jacob Pan wrote: > > Hi Mukesh, > > Hey Jacob, pl see inline.. > > > On Fri, 17 Jul 2026 19:19:41 -0700 > > Mukesh R <[email protected]> wrote: > > ... snip... > > >> static int > >> add_partition(struct mshv_partition *partition) > >> { > >> @@ -2073,6 +2094,7 @@ mshv_ioctl_create_partition(void __user > >> *user_arg, struct device *module_dev) goto cleanup_irq_srcu; > >> > >> partition->pt_id = pt_id; > >> + partition->pt_vmm_tgid = current->tgid; > > I wonder how robust this mechanism is to identify target partition > > via tgid. > > 1) what prevents a VMM process create more than one partition? in > > that case each partition would have the same tgid. > > Currently, none of the VMMs we support do that, and doesn't look like > there is much of a demand for it. > My point is that from kernel UAPI pov, we cannot count on user behavior nor current VMM's implementation. > > 2) IIUC, the lifetime of the partition is tied to FD, which is > > different than the lifetime of a PID. The partition FD can be > > inherited or passed to another process. The VMM tg can exit while > > the FDs can be alive. Then the tgid can be reused by another > > unrelated process, right? > > yeah, AI keeps telling me that, but not super accurate imo. > > we are using tgid and not pid. tgid is process group id, and that will > stay around as long as there is at least one process in it. if we used > pid, then that would be the case. > I understand it is tgid but don't think tgid makes difference in terms of FD lifetime. e.g. process A creates a partition, then passes the partition fd to process B over a Unix socket. Process A can then exit while process B still holds a reference to the partition file. At that point the tgid that was stored in pt_vmm_tgid can be reused by an unrelated process C, while the partition's file object remains alive. > > Would it be more robust to based this on the partition FD instead of > > tgid? > > it might be, but problem with that is we need pt-id in other cases > where that is not available: for example in > hv_iommu_domain_alloc_paging and in irq remapping paths for direct > attached devices. if we can sort that out somehow, then we can do > that. but i suspect, it would take some time to figure that out, so i > hope we can make that a future enhancement. > > For now, i've been thinking of just putting a check and returning > ENOTSUPP if a vmm tries to create another partition. > IMHO, that only solves the uniqueness issue of partition ID but not the lifetime issue. maybe EEXIST instead of ENOTSUPP? > Thanks, > -Mukesh > > >> ret = add_partition(partition); > >> if (ret) > >> diff --git a/include/asm-generic/mshyperv.h > >> b/include/asm-generic/mshyperv.h index bf601d67cecb..e8cbc4e3f7ad > >> 100644 --- a/include/asm-generic/mshyperv.h > >> +++ b/include/asm-generic/mshyperv.h > >> @@ -350,6 +350,7 @@ int hv_call_add_logical_proc(int node, u32 > >> lp_index, u32 acpi_id); int > >> hv_call_notify_all_processors_started(void); bool hv_lp_exists(u32 > >> lp_index); int hv_call_create_vp(int node, u64 partition_id, u32 > >> vp_index, u32 flags); +u64 mshv_current_partid(void); > >> > >> #else /* CONFIG_MSHV_ROOT */ > >> static inline bool hv_root_partition(void) { return false; } > >> @@ -380,6 +381,10 @@ static inline int hv_call_create_vp(int node, > >> u64 partition_id, u32 vp_index, u3 { > >> return -EOPNOTSUPP; > >> } > >> +static inline u64 mshv_current_partid(void) > >> +{ > >> + return HV_PARTITION_ID_INVALID; > >> +} > >> #endif /* CONFIG_MSHV_ROOT */ > >> > >> static inline int hv_deposit_memory(u64 partition_id, u64 > >> status)
