[PATCHv4] x86/sgx: Avoid softlockup from sgx_vepc_release

[PATCHv4] x86/sgx: Avoid softlockup from sgx_vepc_release

[Jack Wang]

We hit softlocup with following call trace:

? asm_sysvec_apic_timer_interrupt+0x16/0x20
xa_erase+0x21/0xb0
? sgx_free_epc_page+0x20/0x50
sgx_vepc_release+0x75/0x220
__fput+0x89/0x250
task_work_run+0x59/0x90
do_exit+0x337/0x9a0

Similar like commit 8795359e35bc ("x86/sgx: Silence softlockup detection
when releasing large enclaves"). The test system has 64GB of enclave memory,
and all assigned to a single VM. Release vepc take longer time and triggers
the softlockup warning.

Add cond_resched() to give other tasks a chance to run and placate
the softlockup detector.

Cc: Jarkko Sakkinen <jarkko@kernel.org>
Cc: Haitao Huang <haitao.huang@linux.intel.com>
Cc: stable@vger.kernel.org
Cc: x86@kernel.org
Fixes: 540745ddbc70 ("x86/sgx: Introduce virtual EPC for use by KVM guests")
Reported-by: Yu Zhang <yu.zhang@ionos.com>
Tested-by: Yu Zhang <yu.zhang@ionos.com>
Acked-by: Haitao Huang <haitao.huang@linux.intel.com>
Reviewed-by: Jarkko Sakkinen <jarkko@kernel.org>
Reviewed-by: Kai Huang <kai.huang@intel.com>
Signed-off-by: Jack Wang <jinpu.wang@ionos.com>
---
v4: add rob from Kai.

 arch/x86/kernel/cpu/sgx/virt.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/arch/x86/kernel/cpu/sgx/virt.c b/arch/x86/kernel/cpu/sgx/virt.c
index c3e37eaec8ec..7aaa3652e31d 100644
--- a/arch/x86/kernel/cpu/sgx/virt.c
+++ b/arch/x86/kernel/cpu/sgx/virt.c
@@ -204,6 +204,7 @@ static int sgx_vepc_release(struct inode *inode, struct file *file)
 			continue;
 
 		xa_erase(&vepc->page_array, index);
+		cond_resched();
 	}
 
 	/*
@@ -222,6 +223,7 @@ static int sgx_vepc_release(struct inode *inode, struct file *file)
 			list_add_tail(&epc_page->list, &secs_pages);
 
 		xa_erase(&vepc->page_array, index);
+		cond_resched();
 	}
 
 	/*
@@ -243,6 +245,7 @@ static int sgx_vepc_release(struct inode *inode, struct file *file)
 
 		if (sgx_vepc_free_page(epc_page))
 			list_add_tail(&epc_page->list, &secs_pages);
+		cond_resched();
 	}
 
 	if (!list_empty(&secs_pages))
-- 
2.34.1

Re: [PATCHv4] x86/sgx: Avoid softlockup from sgx_vepc_release

[Ingo Molnar]

* Jack Wang <jinpu.wang@ionos.com> wrote:

> We hit softlocup with following call trace:
> 
> ? asm_sysvec_apic_timer_interrupt+0x16/0x20
> xa_erase+0x21/0xb0
> ? sgx_free_epc_page+0x20/0x50
> sgx_vepc_release+0x75/0x220
> __fput+0x89/0x250
> task_work_run+0x59/0x90
> do_exit+0x337/0x9a0
> 
> Similar like commit 8795359e35bc ("x86/sgx: Silence softlockup detection
> when releasing large enclaves"). The test system has 64GB of enclave memory,
> and all assigned to a single VM. Release vepc take longer time and triggers
> the softlockup warning.
> 
> Add cond_resched() to give other tasks a chance to run and placate
> the softlockup detector.

I've rewritten the changelog: it's not just a change to 'placate' the 
softlockup detector, the SGX code was causing *real*, very long 
non-preemptible delays in the kernel. That's a real kernel bug, not some 
softlockup whingle that needs to be silenced ...

So what this patch does is to break up those delays & latencies, and that 
softlockup doesn't warn anymore is a side effect fix.

I've changed the description accordingly, and applied it to tip:x86/urgent.

Thanks,

	Ingo