From: Michael Kelley Date: Tue, 18 Mar 2025 21:49:19 +0000 (-0700) Subject: x86/hyperv: Add comments about hv_vpset and var size hypercall input args X-Git-Tag: io_uring-6.15-20250403~115^2~1 X-Git-Url: https://git.kernel.dk/?a=commitdiff_plain;h=999ad14259a0d45cb3b616e2e95a7c8b622a7ecd;p=linux-block.git x86/hyperv: Add comments about hv_vpset and var size hypercall input args Current code varies in how the size of the variable size input header for hypercalls is calculated when the input contains struct hv_vpset. Surprisingly, this variation is correct, as different hypercalls make different choices for what portion of struct hv_vpset is treated as part of the variable size input header. The Hyper-V TLFS is silent on these details, but the behavior has been confirmed with Hyper-V developers. To avoid future confusion about these differences, add comments to struct hv_vpset, and to hypercall call sites with input that contains a struct hv_vpset. The comments describe the overall situation and the calculation that should be used at each particular call site. No functional change as only comments are updated. Signed-off-by: Michael Kelley Link: https://lore.kernel.org/r/20250318214919.958953-1-mhklinux@outlook.com Signed-off-by: Wei Liu Message-ID: <20250318214919.958953-1-mhklinux@outlook.com> --- diff --git a/arch/x86/hyperv/hv_apic.c b/arch/x86/hyperv/hv_apic.c index f022d5f64fb6..6d91ac5f9836 100644 --- a/arch/x86/hyperv/hv_apic.c +++ b/arch/x86/hyperv/hv_apic.c @@ -145,6 +145,11 @@ static bool __send_ipi_mask_ex(const struct cpumask *mask, int vector, ipi_arg->vp_set.format = HV_GENERIC_SET_ALL; } + /* + * For this hypercall, Hyper-V treats the valid_bank_mask field + * of ipi_arg->vp_set as part of the fixed size input header. + * So the variable input header size is equal to nr_bank. + */ status = hv_do_rep_hypercall(HVCALL_SEND_IPI_EX, 0, nr_bank, ipi_arg, NULL); diff --git a/arch/x86/hyperv/mmu.c b/arch/x86/hyperv/mmu.c index cc8c3bd0e7c2..80d9350612d2 100644 --- a/arch/x86/hyperv/mmu.c +++ b/arch/x86/hyperv/mmu.c @@ -205,6 +205,10 @@ static u64 hyperv_flush_tlb_others_ex(const struct cpumask *cpus, /* * We can flush not more than max_gvas with one hypercall. Flush the * whole address space if we were asked to do more. + * + * For these hypercalls, Hyper-V treats the valid_bank_mask field + * of flush->hv_vp_set as part of the fixed size input header. + * So the variable input header size is equal to nr_bank. */ max_gvas = (PAGE_SIZE - sizeof(*flush) - nr_bank * diff --git a/include/hyperv/hvgdk_mini.h b/include/hyperv/hvgdk_mini.h index 735329859f21..abf0bd76e370 100644 --- a/include/hyperv/hvgdk_mini.h +++ b/include/hyperv/hvgdk_mini.h @@ -205,7 +205,14 @@ union hv_reference_tsc_msr { /* The number of vCPUs in one sparse bank */ #define HV_VCPUS_PER_SPARSE_BANK (64) -/* Some of Hyper-V structs do not use hv_vpset where linux uses them */ +/* + * Some of Hyper-V structs do not use hv_vpset where linux uses them. + * + * struct hv_vpset is usually used as part of hypercall input. The portion + * that counts as "fixed size input header" vs. "variable size input header" + * varies per hypercall. See comments at relevant hypercall call sites as to + * how the "valid_bank_mask" field should be accounted. + */ struct hv_vpset { /* HV_VP_SET */ u64 format; u64 valid_bank_mask;