[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <3d257fce5c37e6bd73648614bfaabc26de95737f.camel@intel.com>
Date: Tue, 27 May 2025 11:07:38 +0000
From: "Huang, Kai" <kai.huang@...el.com>
To: "kvm@...r.kernel.org" <kvm@...r.kernel.org>, "pbonzini@...hat.com"
<pbonzini@...hat.com>, "linux-kernel@...r.kernel.org"
<linux-kernel@...r.kernel.org>
CC: "Li, Xiaoyao" <xiaoyao.li@...el.com>, lkp <lkp@...el.com>
Subject: Re: [PATCH] x86/tdx: mark tdh_vp_enter() as __flatten
On Mon, 2025-05-26 at 23:10 +0000, Huang, Kai wrote:
> On Mon, 2025-05-26 at 16:45 -0400, Paolo Bonzini wrote:
> > In some cases tdx_tdvpr_pa() is not fully inlined into tdh_vp_enter(), which
> > causes the following warning:
> >
> > vmlinux.o: warning: objtool: tdh_vp_enter+0x8: call to tdx_tdvpr_pa() leaves .noinstr.text section
> >
> > This happens if the compiler considers tdx_tdvpr_pa() to be "large", for example
> > because CONFIG_SPARSEMEM adds two function calls to page_to_section() and
> > __section_mem_map_addr():
> >
> > ({ const struct page *__pg = (pg); \
> > int __sec = page_to_section(__pg); \
> > (unsigned long)(__pg - __section_mem_map_addr(__nr_to_section(__sec)));
> > \
> > })
>
> Just FYI the above warning can also be triggered when CONFIG_SPARSEMEM_VMEMMAP=y
> (and CONFIG_SPARSEMEM=y as well), in which case the __page_to_pfn() is simply:
>
> #define __page_to_pfn(page) (unsigned long)((page) - vmemmap)
>
> The function call to page_to_section() and __section_mem_map_addr() only happens
> when CONFIG_SPARSEMEM_VMEMMAP=n while CONFIG_SPARSEMEM=y.
>
[...]
>
> I did some test and can confirm this patch can silence the warning mentioned in
> the changelog.
>
> However, with this patch applied, I also tested the case that
> CONFIG_SPARSEMEM_VMEMMAP=n and CONFIG_SPARSEMEM=y [1], but I still got:
>
> vmlinux.o: warning: objtool: tdh_vp_enter+0x10: call to page_to_section() leaves
> .noinstr.text section
>
> Not sure why, but it seems __flatten failed to work as expected, as least
> recursively.
>
I found a recently merged patch from Kirill has made CONFIG_SPARSEMEM_VMEMAP
always true on x86_64:
https://lore.kernel.org/lkml/174748683804.406.11521945321481191771.tip-bot2@tip-bot2/
So we don't need to worry about the case that CONFIG_SPARSEMEM_VMEMMAP=n and
CONFIG_SPARSEMEM=y.
The changelog could be simplified/updated though I think (since seems there's no
need to mention the "large" case now).
So:
Tested-by: Kai Huang <kai.huang@...el.com>
Acked-by: Kai Huang <kai.huang@...el.com>
Powered by blists - more mailing lists