Drivers: hv: hv_common: use meaningful errnos for hypercall statuses by hargar19 · Pull Request #146 · microsoft/OHCL-Linux-Kernel

hargar19 · 2026-06-22T23:25:01Z

Several HV_STATUS_* codes were collapsed to -EIO in commit 3817854 ("hyperv: Log hypercall status codes as strings") when hv_result_to_errno() was converted from a switch statement to a table-driven lookup. The change was a side-effect of consolidating string/errno mappings into one place; the commit message focused on adding hv_status_printk() for debugging.

The previous mapping in hv_status_to_errno() (used in OHCL v6.12 and earlier) preserved more meaningful errnos for several codes that have natural POSIX counterparts. The collapse to -EIO loses information at call sites that key behavior on the converted errno (for example mshv_vtl_configure_reg_page() whitelists -EINVAL/-EACCES to fall back gracefully when the host refuses to install the register-page overlay; with -EIO that call site now hits a BUG()).

Restore the semantically meaningful mappings for the affected statuses:

Hyper-V Status	Old errno	New errno
HV_STATUS_ACCESS_DENIED	-EIO	-EACCES
HV_STATUS_OPERATION_DENIED	-EIO	-EACCES
HV_STATUS_UNKNOWN_PROPERTY	-EIO	-EINVAL
HV_STATUS_PROPERTY_VALUE_OUT_OF_RANGE	-EIO	-EINVAL
HV_STATUS_PROCESSOR_FEATURE_NOT_SUPPORTED	-EIO	-EOPNOTSUPP

EACCES is the canonical errno for permission denied, EINVAL for invalid argument, and EOPNOTSUPP for unsupported operations. These mappings match both their <errno.h> definitions and the established meanings in the v6.12 mapping.

Also drop two duplicate _STATUS_INFO() rows for HV_STATUS_INVALID_LP_INDEX and HV_STATUS_INVALID_REGISTER_VALUE that were leftover from the original table conversion.

Signed-off-by: Hardik Garg hargar@microsoft.com

Several HV_STATUS_* codes were collapsed to -EIO in commit 3817854 ("hyperv: Log hypercall status codes as strings") when hv_result_to_errno() was converted from a switch statement to a table-driven lookup. The change was a side-effect of consolidating string/errno mappings into one place; the commit message focused on adding hv_status_printk() for debugging. The previous OOT helper hv_status_to_errno() (in v6.12) preserved more meaningful errnos for several codes that have natural POSIX counterparts. The collapse-to-EIO loses information at call sites that key behavior on the converted errno (for example mshv_vtl_configure_reg_page() whitelists -EINVAL/-EACCES to fall back gracefully when the host refuses to install the register-page overlay; with -EIO that call site now hits a BUG()). Restore the semantically meaningful mappings for the affected statuses: HV_STATUS_ACCESS_DENIED -EIO -> -EACCES HV_STATUS_OPERATION_DENIED -EIO -> -EACCES HV_STATUS_UNKNOWN_PROPERTY -EIO -> -EINVAL HV_STATUS_PROPERTY_VALUE_OUT_OF_RANGE -EIO -> -EINVAL HV_STATUS_PROCESSOR_FEATURE_NOT_SUPPORTED -EIO -> -EOPNOTSUPP EACCES is the canonical errno for permission denied, EINVAL for invalid argument, and EOPNOTSUPP for unsupported operations. These mappings match both their <errno.h> definitions and the established meanings in the OOT v6.12 helper. Also drop two duplicate _STATUS_INFO() rows for HV_STATUS_INVALID_LP_INDEX and HV_STATUS_INVALID_REGISTER_VALUE that were leftover from the original table conversion. find_hv_status_info() is first-match-wins so the dead rows were not reachable, but they are confusing on review. Impact analysis --------------- The change only matters at sites that distinguish between specific errnos (e.g. `ret == -EIO` vs `ret == -EACCES`). To confirm safety, the entire kernel tree was audited for equality comparisons against any of the affected errnos in Hyper-V-touching code: $ git grep -nE '(ret|status|err|rc|r)[[:space:]]*==[[:space:]]*\ -(EACCES|EBADFD|ENOTRECOVERABLE|EOPNOTSUPP|EIO|EINVAL|ENOMEM)' \ -- drivers/hv/ arch/x86/hyperv/ arch/arm64/hyperv/ \ drivers/iommu/hyperv-iommu.c Results in Hyper-V code: drivers/hv/hv_balloon.c: ret == -EEXIST | ret == -EAGAIN (not in table) drivers/hv/mshv_eventfd.c: ret == -EAGAIN (not in table) drivers/hv/connection.c: ret == -ETIMEDOUT (not in table) drivers/hv/mshv_vtl_main.c: ret == -EINVAL || ret == -EACCES (the consumer this patch fixes) No other in-tree code path inspects -EOPNOTSUPP, -EACCES, -EINVAL or -EIO from a hypercall result to make a behavioral decision. The remaining HV_STATUS_* codes that 3817854 mapped to -EIO are therefore left as -EIO; nothing keys behavior off the distinction today. Signed-off-by: Hardik Garg <hargar@microsoft.com>

Copilot

Pull request overview

This PR restores more semantically meaningful Linux errno mappings for specific Hyper-V hypercall status codes in hv_result_to_errno() by adjusting the hv_status_infos[] lookup table, and removes unreachable duplicate table rows introduced during the earlier table conversion.

Changes:

Map HV_STATUS_ACCESS_DENIED / HV_STATUS_OPERATION_DENIED to -EACCES instead of -EIO.
Map HV_STATUS_UNKNOWN_PROPERTY / HV_STATUS_PROPERTY_VALUE_OUT_OF_RANGE to -EINVAL instead of -EIO, and HV_STATUS_PROCESSOR_FEATURE_NOT_SUPPORTED to -EOPNOTSUPP.
Remove duplicate (dead) _STATUS_INFO() entries for HV_STATUS_INVALID_LP_INDEX and HV_STATUS_INVALID_REGISTER_VALUE.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

hargar19 · 2026-06-22T23:46:03Z

Why arm VM fails on v6.18 but not v6.12:

The bug surface: same hypercall, same response, same hardware
On the failing Cobalt 100 ARM VM, both v6.12 and v6.18 issue the identical hypercall and receive the identical response from Hyper-V:

Field	Value
Hypercall	HvCallSetVpRegisters (0x0051) with count=1
VP	HV_VP_INDEX_SELF, VTL=0
Register	HV_REGISTER_REG_PAGE (0x9001c)
Value	overlay.enabled=1, overlay.pfn=<allocated>
Host advertises	intercept_page_available = 1 in HvRegisterVsmCapabilities
Host returns	HV_STATUS_ACCESS_DENIED (0x6)

This is a benign per-VP policy denial — the host supports the register-page feature in principle, but refuses the install on this configuration. The register page is an optional performance optimization (lets user-mode skip a hypercall for some VP register accesses); kernels are expected to fall back gracefully.

v6.12: failure handled gracefully
OHCL v6.12 used the in-tree helper hv_status_to_errno() (include/asm-generic/mshyperv.h) which mapped:

case HV_STATUS_ACCESS_DENIED:
case HV_STATUS_OPERATION_DENIED:
    return -EACCES;

The caller mshv_vtl_configure_reg_page() (drivers/hv/mshv_vtl_main.c) whitelists exactly this:

if (ret == -EINVAL || ret == -EACCES) {
    pr_info("not using the register page");
} else {
    pr_emerg("error when setting up the register page: %d\n", ret);
    BUG();
}

So -EACCES lands in the silent-fallback path. v6.12 dmesg on Cobalt 100:
[0.013509] not using the register page (×8, once per CPU)

Kernel boots normally. The host's denial is logged and ignored, hypercall-based path takes over.

v6.18: same response, but mapped to -EIO
Upstream commit 3817854ba892 ("hyperv: Log hypercall status codes as strings") refactored hv_result_to_errno() from a switch statement to a table-driven lookup. The author's stated intent was to add string lookup support for hv_status_printk(); the errno mapping table was a side-effect. The new table preserves the existing -EIO fallback for most codes, including HV_STATUS_ACCESS_DENIED:
_STATUS_INFO(HV_STATUS_ACCESS_DENIED, -EIO),

Result: HV_STATUS_ACCESS_DENIED is now mapped to -EIO, the whitelist misses it, and the code hits BUG():

[0.020022] error when setting up the register page: -5
[0.020031] Kernel BUG at mshv_vtl_alloc_context+0x178/0x210
[0.020036] Internal error: Oops - BUG: 00000000f2000800 [#1]  SMP
[0.020080] Kernel panic - not syncing: Oops - BUG: Fatal exception

-5 is -EIO. BUG() in a CPU hotplug callback during boot → fatal.

Diagnostic instrumentation confirming the diagnosis
Captured on both kernels with debug pr_warns added to mshv_vtl_configure_reg_page() and hv_call_set_vp_registers():

v6.12 (boots):

mshv_vtl_configure_reg_page: cpu=0 ... reg_name=0x9001c overlay.enabled=1 overlay.pfn=0x42205f
                              vsm_caps=0xf8020002 intercept_page_available=1
hv_call_set_vp_registers: ... status=0x6 (HV_STATUS_ACCESS_DENIED) errno=-13
not using the register page

v6.18 (crashes pre-patch):

mshv_vtl_configure_reg_page: cpu=0 ... reg_name=0x9001c overlay.enabled=1 overlay.pfn=0x4221fc
                              vsm_caps=0xf8020002 intercept_page_available=1
hv_call_set_vp_registers: ... status=0x6 (HV_STATUS_ACCESS_DENIED) errno=-5
Kernel BUG at mshv_vtl_alloc_context+0x178/0x210

Identical hypercall, identical raw hypervisor response (0x6), different errno → different kernel reaction.

Impact analysis for this change:

The change only matters at sites that distinguish between specific errnos (e.g. ret == -EIO vs ret == -EACCES). To confirm safety, the entire kernel tree was audited for equality comparisons against any of the affected errnos in Hyper-V-touching code:

$ git grep -nE '(ret|status|err|rc|r)[[:space:]]*==[[:space:]]*\
      -(EACCES|EBADFD|ENOTRECOVERABLE|EOPNOTSUPP|EIO|EINVAL|ENOMEM)' \
  -- drivers/hv/ arch/x86/hyperv/ arch/arm64/hyperv/ \
     drivers/iommu/hyperv-iommu.c

Results in Hyper-V code:

Consumer	Expected errno(s)	Status
drivers/hv/hv_balloon.c	-EEXIST, -EAGAIN	Not in table
drivers/hv/mshv_eventfd.c	-EAGAIN	Not in table
drivers/hv/connection.c	-ETIMEDOUT	Not in table
drivers/hv/mshv_vtl_main.c	-EINVAL, -EACCES	Covered by this patch

No other in-tree code path inspects -EOPNOTSUPP, -EACCES, -EINVAL or -EIO from a hypercall result to make a behavioral decision. The remaining HV_STATUS_* codes that 3817854 mapped to -EIO are therefore left as -EIO; nothing keys behavior off the distinction today.

namancse · 2026-06-23T04:36:44Z

Change LGTM, have we run Sashiko or AI agent on this?

Nit, subject: s/"Drivers: hv: hv_common:"/Drivers: hv:"

namancse

No issues with Sashiko, Opus 4.7.

saurabh-sengar

LGTM, please make sure we do proper testing

hargar19 requested a review from Copilot June 22, 2026 23:25

Copilot started reviewing on behalf of hargar19 June 22, 2026 23:26 View session

Copilot AI reviewed Jun 22, 2026

View reviewed changes

hargar19 requested review from namancse and saurabh-sengar June 22, 2026 23:46

hargar19 marked this pull request as ready for review June 22, 2026 23:46

namancse approved these changes Jun 23, 2026

View reviewed changes

saurabh-sengar approved these changes Jun 23, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Drivers: hv: hv_common: use meaningful errnos for hypercall statuses#146

Drivers: hv: hv_common: use meaningful errnos for hypercall statuses#146
hargar19 wants to merge 1 commit into
product/hcl-main/6.18from
user/hargar/6.18-hv-status-errno-fix

hargar19 commented Jun 22, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

hargar19 commented Jun 22, 2026 •

edited

Loading

Uh oh!

namancse commented Jun 23, 2026 •

edited

Loading

Uh oh!

namancse left a comment

Uh oh!

saurabh-sengar left a comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

hargar19 commented Jun 22, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

hargar19 commented Jun 22, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Impact analysis for this change:

Uh oh!

namancse commented Jun 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

namancse left a comment

Choose a reason for hiding this comment

Uh oh!

saurabh-sengar left a comment

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

hargar19 commented Jun 22, 2026 •

edited

Loading

hargar19 commented Jun 22, 2026 •

edited

Loading

namancse commented Jun 23, 2026 •

edited

Loading