Re: [patch v3 31/36] x86/apic: Provide cpu_primary_thread mask

From: Thomas Gleixner <tglx@linutronix.de>
To: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: LKML <linux-kernel@vger.kernel.org>,
	x86@kernel.org, David Woodhouse <dwmw2@infradead.org>,
	Andrew Cooper <andrew.cooper3@citrix.com>,
	Brian Gerst <brgerst@gmail.com>,
	Arjan van de Veen <arjan@linux.intel.com>,
	Paolo Bonzini <pbonzini@redhat.com>,
	Paul McKenney <paulmck@kernel.org>,
	Tom Lendacky <thomas.lendacky@amd.com>,
	Sean Christopherson <seanjc@google.com>,
	Oleksandr Natalenko <oleksandr@natalenko.name>,
	Paul Menzel <pmenzel@molgen.mpg.de>,
	"Guilherme G. Piccoli" <gpiccoli@igalia.com>,
	Piotr Gorski <lucjan.lucjanov@gmail.com>,
	Usama Arif <usama.arif@bytedance.com>,
	Juergen Gross <jgross@suse.com>,
	Boris Ostrovsky <boris.ostrovsky@oracle.com>,
	xen-devel@lists.xenproject.org,
	Russell King <linux@armlinux.org.uk>,
	Arnd Bergmann <arnd@arndb.de>,
	linux-arm-kernel@lists.infradead.org,
	Catalin Marinas <catalin.marinas@arm.com>,
	Will Deacon <will@kernel.org>, Guo Ren <guoren@kernel.org>,
	linux-csky@vger.kernel.org,
	Thomas Bogendoerfer <tsbogend@alpha.franken.de>,
	linux-mips@vger.kernel.org,
	"James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>,
	Helge Deller <deller@gmx.de>,
	linux-parisc@vger.kernel.org,
	Paul Walmsley <paul.walmsley@sifive.com>,
	Palmer Dabbelt <palmer@dabbelt.com>,
	linux-riscv@lists.infradead.org,
	Mark Rutland <mark.rutland@arm.com>,
	Sabin Rapan <sabrapan@amazon.com>,
	"Michael Kelley (LINUX)" <mikelley@microsoft.com>,
	Dave Hansen <dave.hansen@linux.intel.com>
Subject: Re: [patch v3 31/36] x86/apic: Provide cpu_primary_thread mask
Date: Mon, 29 May 2023 21:27:13 +0200	[thread overview]
Message-ID: <87bki3kkfi.ffs@tglx> (raw)
In-Reply-To: <20230529023939.mc2akptpxcg3eh2f@box.shutemov.name>

On Mon, May 29 2023 at 05:39, Kirill A. Shutemov wrote:
> On Sat, May 27, 2023 at 03:40:02PM +0200, Thomas Gleixner wrote:
> But it gets broken again on "x86/smpboot: Implement a bit spinlock to
> protect the realmode stack" with
>
> [    0.554079] .... node  #0, CPUs:        #1  #2
> [    0.738071] Callback from call_rcu_tasks() invoked.
> [   10.562065] CPU2 failed to report alive state
> [   10.566337]   #3
> [   20.570066] CPU3 failed to report alive state
> [   20.574268]   #4
> ...
>
> Notably CPU1 is missing from "failed to report" list. So CPU1 takes the
> lock fine, but seems never unlocks it.
>
> Maybe trampoline_lock(%rip) in head_64.S somehow is not the same as
> &tr_lock in trampoline_64.S. I donno.

It's definitely the same in the regular startup (16bit mode), but TDX
starts up via:

trampoline_start64
  trampoline_compat
    LOAD_REALMODE_ESP <- lock

That place cannot work with that LOAD_REALMODE_ESP macro. The untested
below should cure it.

Thanks,

        tglx
---

--- a/arch/x86/realmode/rm/trampoline_64.S
+++ b/arch/x86/realmode/rm/trampoline_64.S
@@ -37,12 +37,12 @@
 	.text
 	.code16
 
-.macro LOAD_REALMODE_ESP
+.macro LOAD_REALMODE_ESP lock:req
 	/*
 	 * Make sure only one CPU fiddles with the realmode stack
 	 */
 .Llock_rm\@:
-        lock btsl       $0, tr_lock
+        lock btsl       $0, \lock
         jnc             2f
         pause
         jmp             .Llock_rm\@
@@ -63,7 +63,7 @@ SYM_CODE_START(trampoline_start)
 	mov	%ax, %es
 	mov	%ax, %ss
 
-	LOAD_REALMODE_ESP
+	LOAD_REALMODE_ESP tr_lock
 
 	call	verify_cpu		# Verify the cpu supports long mode
 	testl   %eax, %eax		# Check for return code
@@ -106,7 +106,7 @@ SYM_CODE_START(sev_es_trampoline_start)
 	mov	%ax, %es
 	mov	%ax, %ss
 
-	LOAD_REALMODE_ESP
+	LOAD_REALMODE_ESP tr_lock
 
 	jmp	.Lswitch_to_protected
 SYM_CODE_END(sev_es_trampoline_start)
@@ -189,7 +189,7 @@ SYM_CODE_START(pa_trampoline_compat)
 	 * In compatibility mode.  Prep ESP and DX for startup_32, then disable
 	 * paging and complete the switch to legacy 32-bit mode.
 	 */
-	LOAD_REALMODE_ESP
+	LOAD_REALMODE_ESP pa_tr_lock
 	movw	$__KERNEL_DS, %dx
 
 	movl	$(CR0_STATE & ~X86_CR0_PG), %eax

