Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Vitaly Kuznetsov <vkuznets@redhat.com>
To:     Tianyu Lan <ltykernel@gmail.com>, kys@microsoft.com,
        haiyangz@microsoft.com, wei.liu@kernel.org, decui@microsoft.com,
        tglx@linutronix.de, mingo@redhat.com, bp@alien8.de,
        dave.hansen@linux.intel.com, x86@kernel.org, hpa@zytor.com,
        daniel.lezcano@linaro.org, arnd@arndb.de,
        michael.h.kelley@microsoft.com
Cc:     Tianyu Lan <tiala@microsoft.com>, linux-arch@vger.kernel.org,
        linux-hyperv@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH 4/9] drivers: hv: Mark shared pages unencrypted in
 SEV-SNP enlightened guest
In-Reply-To: <20230601151624.1757616-5-ltykernel@gmail.com>
References: <20230601151624.1757616-1-ltykernel@gmail.com>
 <20230601151624.1757616-5-ltykernel@gmail.com>
Date:   Mon, 05 Jun 2023 14:54:26 +0200
Message-ID: <87zg5ejchp.fsf@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain
Precedence: bulk

Tianyu Lan <ltykernel@gmail.com> writes:

> From: Tianyu Lan <tiala@microsoft.com>
>
> Hypervisor needs to access iput arg, VMBus synic event and
> message pages. Mask these pages unencrypted in the sev-snp
> guest and free them only if they have been marked encrypted
> successfully.
>
> Signed-off-by: Tianyu Lan <tiala@microsoft.com>
> ---
>  drivers/hv/hv.c        | 57 +++++++++++++++++++++++++++++++++++++++---
>  drivers/hv/hv_common.c | 24 +++++++++++++++++-
>  2 files changed, 77 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/hv/hv.c b/drivers/hv/hv.c
> index de6708dbe0df..94406dbe0df0 100644
> --- a/drivers/hv/hv.c
> +++ b/drivers/hv/hv.c
> @@ -20,6 +20,7 @@
>  #include <linux/interrupt.h>
>  #include <clocksource/hyperv_timer.h>
>  #include <asm/mshyperv.h>
> +#include <linux/set_memory.h>
>  #include "hyperv_vmbus.h"
>  
>  /* The one and only */
> @@ -78,7 +79,7 @@ int hv_post_message(union hv_connection_id connection_id,
>  
>  int hv_synic_alloc(void)
>  {
> -	int cpu;
> +	int cpu, ret = -ENOMEM;
>  	struct hv_per_cpu_context *hv_cpu;
>  
>  	/*
> @@ -123,26 +124,76 @@ int hv_synic_alloc(void)
>  				goto err;
>  			}
>  		}
> +
> +		if (hv_isolation_type_en_snp()) {
> +			ret = set_memory_decrypted((unsigned long)
> +				hv_cpu->synic_message_page, 1);
> +			if (ret) {
> +				pr_err("Failed to decrypt SYNIC msg page: %d\n", ret);
> +				hv_cpu->synic_message_page = NULL;
> +
> +				/*
> +				 * Free the event page here and not encrypt
> +				 * the page in hv_synic_free().
> +				 */
> +				free_page((unsigned long)hv_cpu->synic_event_page);
> +				hv_cpu->synic_event_page = NULL;
> +				goto err;
> +			}
> +
> +			ret = set_memory_decrypted((unsigned long)
> +				hv_cpu->synic_event_page, 1);
> +			if (ret) {
> +				pr_err("Failed to decrypt SYNIC event page: %d\n", ret);
> +				hv_cpu->synic_event_page = NULL;
> +				goto err;
> +			}
> +
> +			memset(hv_cpu->synic_message_page, 0, PAGE_SIZE);
> +			memset(hv_cpu->synic_event_page, 0, PAGE_SIZE);
> +		}
>  	}
>  
>  	return 0;
> +
>  err:
>  	/*
>  	 * Any memory allocations that succeeded will be freed when
>  	 * the caller cleans up by calling hv_synic_free()
>  	 */
> -	return -ENOMEM;
> +	return ret;
>  }
>  
>  
>  void hv_synic_free(void)
>  {
> -	int cpu;
> +	int cpu, ret;
>  
>  	for_each_present_cpu(cpu) {
>  		struct hv_per_cpu_context *hv_cpu
>  			= per_cpu_ptr(hv_context.cpu_context, cpu);
>  
> +		/* It's better to leak the page if the encryption fails. */
> +		if (hv_isolation_type_en_snp()) {
> +			if (hv_cpu->synic_message_page) {
> +				ret = set_memory_encrypted((unsigned long)
> +					hv_cpu->synic_message_page, 1);
> +				if (ret) {
> +					pr_err("Failed to encrypt SYNIC msg page: %d\n", ret);
> +					hv_cpu->synic_message_page = NULL;
> +				}
> +			}
> +
> +			if (hv_cpu->synic_event_page) {
> +				ret = set_memory_encrypted((unsigned long)
> +					hv_cpu->synic_event_page, 1);
> +				if (ret) {
> +					pr_err("Failed to encrypt SYNIC event page: %d\n", ret);
> +					hv_cpu->synic_event_page = NULL;
> +				}
> +			}
> +		}
> +
>  		free_page((unsigned long)hv_cpu->synic_event_page);
>  		free_page((unsigned long)hv_cpu->synic_message_page);
>  	}
> diff --git a/drivers/hv/hv_common.c b/drivers/hv/hv_common.c
> index 179bc5f5bf52..bed9aa6ac19a 100644
> --- a/drivers/hv/hv_common.c
> +++ b/drivers/hv/hv_common.c
> @@ -24,6 +24,7 @@
>  #include <linux/kmsg_dump.h>
>  #include <linux/slab.h>
>  #include <linux/dma-map-ops.h>
> +#include <linux/set_memory.h>
>  #include <asm/hyperv-tlfs.h>
>  #include <asm/mshyperv.h>
>  
> @@ -359,6 +360,7 @@ int hv_common_cpu_init(unsigned int cpu)
>  	u64 msr_vp_index;
>  	gfp_t flags;
>  	int pgcount = hv_root_partition ? 2 : 1;
> +	int ret;
>  
>  	/* hv_cpu_init() can be called with IRQs disabled from hv_resume() */
>  	flags = irqs_disabled() ? GFP_ATOMIC : GFP_KERNEL;
> @@ -368,6 +370,17 @@ int hv_common_cpu_init(unsigned int cpu)
>  	if (!(*inputarg))
>  		return -ENOMEM;
>  
> +	if (hv_isolation_type_en_snp()) {
> +		ret = set_memory_decrypted((unsigned long)*inputarg, pgcount);
> +		if (ret) {
> +			kfree(*inputarg);
> +			*inputarg = NULL;
> +			return ret;
> +		}
> +
> +		memset(*inputarg, 0x00, pgcount * PAGE_SIZE);
> +	}
> +
>  	if (hv_root_partition) {
>  		outputarg = (void **)this_cpu_ptr(hyperv_pcpu_output_arg);
>  		*outputarg = (char *)(*inputarg) + HV_HYP_PAGE_SIZE;
> @@ -387,7 +400,9 @@ int hv_common_cpu_die(unsigned int cpu)
>  {
>  	unsigned long flags;
>  	void **inputarg, **outputarg;
> +	int pgcount = hv_root_partition ? 2 : 1;
>  	void *mem;
> +	int ret;
>  
>  	local_irq_save(flags);
>  
> @@ -402,7 +417,14 @@ int hv_common_cpu_die(unsigned int cpu)
>  
>  	local_irq_restore(flags);
>  
> -	kfree(mem);
> +	if (hv_isolation_type_en_snp()) {
> +		ret = set_memory_encrypted((unsigned long)mem, pgcount);
> +		if (ret)
> +			pr_warn("Hyper-V: Failed to encrypt input arg on cpu%d: %d\n",
> +				cpu, ret);
> +		/* It's unsafe to free 'mem'. */
> +		return 0;

Why is it unsafe to free 'mem' if ret == 0? Also, why don't we want to
proparate non-zero 'ret' from here to fail CPU offlining?


> +	}
>  
>  	return 0;
>  }

-- 
Vitaly