Re: [PATCH] x86/vmware: Fix hypercall clobbers

[Alexey Makhalov <alexey.makhalov@broadcom.com>]

On 1/23/26 1:47 AM, Thorsten Leemhuis wrote:
> Well, fixing bugs right where they are obviously is a good thing.
> 
> But well, the problem according to the description quoted below was
> exposed by a change that went into 6.19-rc1 -- which makes it a kernel
> regression that must be fixed in the kernel (ideally before 6.19 is out).

The Linux change 34bf25e820ae ("x86/vmware: Introduce VMware hypercall 
API") that revealed QEMU issue was introduced not now, but back in June 
2024 in v6.11-rc1. Two LTS versions 6.12 and 6.18 have it.

This issue can be triggered only for specific hypercall (usually 
statically inlined) and _only_ if compiler keeps rsi/rdi in use during 
the hypercall.

The proposed kernel fix forces compiler to use memory instead of rsi/rdi 
registers for _every_ VMware hypercall. Every VMware hypercall caller 
should pay a penalty for QEMU issue, even if it runs on non qemu or 
fixed/newer qemu platform.

IMHO, user who faced this issue should be given one choice to fix it - 
update QEMU as the system may not be operable with this combination 
QEMU+Kernel.

If kernel colleagues still believe it should be also solved from the 
kernel side, then my proposal is to perform precise fix by using wider 
hypercall just for VMWARE_CMD_ABSPOINTER_DATA:

--- a/drivers/input/mouse/vmmouse.c
+++ b/drivers/input/mouse/vmmouse.c
@@ -145,7 +145,7 @@ static psmouse_ret_t vmmouse_report_events(struct 
psmouse *psmouse)
                 }

                 /* Now get it */
-               status = vmware_hypercall4(VMWARE_CMD_ABSPOINTER_DATA, 4,
+               status = vmware_hypercall7(VMWARE_CMD_ABSPOINTER_DATA, 
4, 0, 0, 0,
                                            &x, &y, &z);

                 /*

Regards,
--Alexey