xen-devel.lists.xenproject.org archive mirror
 help / color / mirror / Atom feed
From: Julien Grall <julien.grall@arm.com>
To: Sergej Proskurin <proskurin@sec.in.tum.de>,
	xen-devel@lists.xenproject.org
Cc: Stefano Stabellini <sstabellini@kernel.org>
Subject: Re: [PATCH 04/18] arm/altp2m: Add altp2m init/teardown routines.
Date: Mon, 4 Jul 2016 16:17:17 +0100	[thread overview]
Message-ID: <577A7DFD.2070600@arm.com> (raw)
In-Reply-To: <20160704114605.10086-5-proskurin@sec.in.tum.de>

Hello Sergej,

On 04/07/16 12:45, Sergej Proskurin wrote:
> The p2m intialization now invokes intialization routines responsible for

s/intialization/initialization/

> the allocation and intitialization of altp2m structures. The same

Ditto

> applies to teardown routines. The functionality has been adopted from
> the x86 altp2m implementation.

This patch would benefit to be split in 2:
    1) Moving p2m init/teardown in a separate function
    2) Introducing altp2m init/teardown

It will ease the review.

> Signed-off-by: Sergej Proskurin <proskurin@sec.in.tum.de>
> ---
> Cc: Stefano Stabellini <sstabellini@kernel.org>
> Cc: Julien Grall <julien.grall@arm.com>
> ---
>   xen/arch/arm/p2m.c            | 166 ++++++++++++++++++++++++++++++++++++++++--
>   xen/include/asm-arm/domain.h  |   6 ++
>   xen/include/asm-arm/hvm/hvm.h |  12 +++
>   xen/include/asm-arm/p2m.h     |  20 +++++
>   4 files changed, 198 insertions(+), 6 deletions(-)
>
> diff --git a/xen/arch/arm/p2m.c b/xen/arch/arm/p2m.c
> index aa4e774..e72ca7a 100644
> --- a/xen/arch/arm/p2m.c
> +++ b/xen/arch/arm/p2m.c
> @@ -1400,19 +1400,103 @@ static void p2m_free_vmid(struct domain *d)
>       spin_unlock(&vmid_alloc_lock);
>   }
>
> -void p2m_teardown(struct domain *d)
> +static int p2m_initialise(struct domain *d, struct p2m_domain *p2m)

AFAICT, this function is only used by p2m_initialise_one. So I would 
prefer if you fold the code in the latter.

>   {
> -    struct p2m_domain *p2m = &d->arch.p2m;
> +    int ret = 0;
> +
> +    spin_lock_init(&p2m->lock);
> +    INIT_PAGE_LIST_HEAD(&p2m->pages);
> +
> +    spin_lock(&p2m->lock);
> +
> +    p2m->domain = d;
> +    p2m->access_required = false;
> +    p2m->mem_access_enabled = false;
> +    p2m->default_access = p2m_access_rwx;
> +    p2m->p2m_class = p2m_host;
> +    p2m->root = NULL;
> +
> +    /* Adopt VMID of the associated domain */
> +    p2m->vmid = d->arch.p2m.vmid;

It looks like to me that re-using the same VMID will require more TLB 
flush (such as when a VCPU is migrated to another physical CPU). So 
could you explain why you decided to re-use the same VMID?

> +    p2m->vttbr.vttbr = 0;
> +    p2m->vttbr.vttbr_vmid = p2m->vmid;
> +
> +    p2m->max_mapped_gfn = 0;
> +    p2m->lowest_mapped_gfn = ULONG_MAX;
> +    radix_tree_init(&p2m->mem_access_settings);
> +
> +    spin_unlock(&p2m->lock);
> +
> +    return ret;
> +}
> +
> +static void p2m_free_one(struct p2m_domain *p2m)
> +{
> +    mfn_t mfn;
> +    unsigned int i;
>       struct page_info *pg;
>
>       spin_lock(&p2m->lock);
>
>       while ( (pg = page_list_remove_head(&p2m->pages)) )
> -        free_domheap_page(pg);
> +        if ( pg != p2m->root )
> +            free_domheap_page(pg);
> +
> +    for ( i = 0; i < P2M_ROOT_PAGES; i++ )
> +    {
> +        mfn = _mfn(page_to_mfn(p2m->root) + i);
> +        clear_domain_page(mfn);
> +    }
> +    free_domheap_pages(p2m->root, P2M_ROOT_ORDER);
> +    p2m->root = NULL;
> +
> +    radix_tree_destroy(&p2m->mem_access_settings, NULL);
> +
> +    spin_unlock(&p2m->lock);
> +
> +    xfree(p2m);
> +}
> +
> +static struct p2m_domain *p2m_init_one(struct domain *d)
> +{
> +    struct p2m_domain *p2m = xzalloc(struct p2m_domain);
> +
> +    if ( !p2m )
> +        return NULL;
> +
> +    if ( p2m_initialise(d, p2m) )
> +        goto free_p2m;
> +
> +    return p2m;
> +
> +free_p2m:
> +    xfree(p2m);
> +    return NULL;
> +}
> +
> +static void p2m_teardown_hostp2m(struct domain *d)

Why does p2m_teardown_hostp2m not use p2m_teardown_one to teardown the 
p2m? Assuming xfree(p2m) is move out of the function.

> +{
> +    struct p2m_domain *p2m = p2m_get_hostp2m(d);
> +    struct page_info *pg = NULL;
> +    mfn_t mfn;
> +    unsigned int i;
> +
> +    spin_lock(&p2m->lock);
>
> -    if ( p2m->root )

Why did you remove this check? The p2m->root could be NULL if the an 
error occurred before create the root page table.

> -        free_domheap_pages(p2m->root, P2M_ROOT_ORDER);
> +    while ( (pg = page_list_remove_head(&p2m->pages)) )
> +        if ( pg != p2m->root )

Why this check? p2m->root will never be part of p2m->pages.

> +        {
> +            mfn = _mfn(page_to_mfn(pg));
> +            clear_domain_page(mfn);
> +            free_domheap_page(pg);
> +        }
>
> +    for ( i = 0; i < P2M_ROOT_PAGES; i++ )
> +    {
> +        mfn = _mfn(page_to_mfn(p2m->root) + i);
> +        clear_domain_page(mfn);
> +    }
> +    free_domheap_pages(p2m->root, P2M_ROOT_ORDER);
>       p2m->root = NULL;
>
>       p2m_free_vmid(d);
> @@ -1422,7 +1506,7 @@ void p2m_teardown(struct domain *d)
>       spin_unlock(&p2m->lock);
>   }
>
> -int p2m_init(struct domain *d)
> +static int p2m_init_hostp2m(struct domain *d)

Why does p2m_init_hostp2m not use p2m_init_one to initialize the p2m?

>   {
>       struct p2m_domain *p2m = &d->arch.p2m;
>       int rc = 0;
> @@ -1437,6 +1521,8 @@ int p2m_init(struct domain *d)
>       if ( rc != 0 )
>           goto err;
>
> +    p2m->vttbr.vttbr_vmid = p2m->vmid;
> +
>       d->arch.vttbr = 0;
>
>       p2m->root = NULL;
> @@ -1454,6 +1540,74 @@ err:
>       return rc;
>   }
>
> +static void p2m_teardown_altp2m(struct domain *d)
> +{
> +    unsigned int i;
> +    struct p2m_domain *p2m;
> +
> +    for ( i = 0; i < MAX_ALTP2M; i++ )
> +    {
> +        if ( !d->arch.altp2m_p2m[i] )
> +            continue;
> +
> +        p2m = d->arch.altp2m_p2m[i];
> +        p2m_free_one(p2m);
> +        d->arch.altp2m_vttbr[i] = INVALID_MFN;
> +        d->arch.altp2m_p2m[i] = NULL;

These 2 lines are not necessary because the domain is destroyed and the 
memory associated will be free very soon.

> +    }
> +
> +    d->arch.altp2m_active = false;
> +}
> +
> +static int p2m_init_altp2m(struct domain *d)
> +{
> +    unsigned int i;
> +    struct p2m_domain *p2m;
> +
> +    spin_lock_init(&d->arch.altp2m_lock);
> +    for ( i = 0; i < MAX_ALTP2M; i++ )
> +    {
> +        d->arch.altp2m_vttbr[i] = INVALID_MFN;

The VTTBR will be stored in altp2m_p2m[i].vttbr. So why do you need to 
store in a different way?

Also, please don't mix value that have different meaning. MFN_INVALID 
indicates that a MFN is invalid not the VTTBR.

> +        d->arch.altp2m_p2m[i] = p2m = p2m_init_one(d);
> +        if ( p2m == NULL )
> +        {
> +            p2m_teardown_altp2m(d);

This call is not necessary. p2m_teardown_altp2m will be called by 
p2m_teardown as part of arch_domain_destroy.

> +            return -ENOMEM;
> +        }
> +        p2m->p2m_class = p2m_alternate;
> +        p2m->access_required = 1;
> +        _atomic_set(&p2m->active_vcpus, 0);
> +    }
> +
> +    return 0;
> +}
> +
> +void p2m_teardown(struct domain *d)
> +{
> +    /*
> +     * We must teardown altp2m unconditionally because
> +     * we initialise it unconditionally.

Why do we need to initialize altp2m unconditionally? When altp2m is not 
used we will allocate memory that is never used.

I would prefer to see the allocation of the memory only if the domain 
will make use of altp2m.

> +     */
> +    p2m_teardown_altp2m(d);
> +
> +    p2m_teardown_hostp2m(d);
> +}
> +
> +int p2m_init(struct domain *d)
> +{
> +    int rc = 0;
> +
> +    rc = p2m_init_hostp2m(d);
> +    if ( rc )
> +        return rc;
> +
> +    rc = p2m_init_altp2m(d);
> +    if ( rc )
> +        p2m_teardown_hostp2m(d);

This call is not necessary.

> +
> +    return rc;
> +}
> +
>   int relinquish_p2m_mapping(struct domain *d)
>   {
>       struct p2m_domain *p2m = &d->arch.p2m;
> diff --git a/xen/include/asm-arm/domain.h b/xen/include/asm-arm/domain.h
> index 2039f16..6b9770f 100644
> --- a/xen/include/asm-arm/domain.h
> +++ b/xen/include/asm-arm/domain.h
> @@ -29,6 +29,9 @@ enum domain_type {
>   #define is_64bit_domain(d) (0)
>   #endif
>
> +#define MAX_ALTP2M      10 /* arbitrary */
> +#define INVALID_ALTP2M  0xffff

IMHO, this should either be part of p2m.h or altp2m.h

> +
>   extern int dom0_11_mapping;
>   #define is_domain_direct_mapped(d) ((d) == hardware_domain && dom0_11_mapping)
>
> @@ -130,6 +133,9 @@ struct arch_domain
>
>       /* altp2m: allow multiple copies of host p2m */
>       bool_t altp2m_active;
> +    struct p2m_domain *altp2m_p2m[MAX_ALTP2M];
> +    spinlock_t altp2m_lock;

Please document was the lock is protecting.

> +    uint64_t altp2m_vttbr[MAX_ALTP2M];
>   }  __cacheline_aligned;
>
>   struct arch_vcpu
> diff --git a/xen/include/asm-arm/hvm/hvm.h b/xen/include/asm-arm/hvm/hvm.h
> index 96c455c..28d5298 100644
> --- a/xen/include/asm-arm/hvm/hvm.h
> +++ b/xen/include/asm-arm/hvm/hvm.h
> @@ -19,6 +19,18 @@
>   #ifndef __ASM_ARM_HVM_HVM_H__
>   #define __ASM_ARM_HVM_HVM_H__
>
> +struct vttbr_data {

This structure should not be part of hvm.h but processor.h. Also, I 
would rename it to simply vttbr.

> +    union {
> +        struct {
> +            u64 vttbr_baddr :40, /* variable res0: from 0-(x-1) bit */

Please drop vttbr_. Also, this field is 48 bits for ARMv8 (see ARM 
D7.2.102 in DDI 0487A.j).

> +                res1        :8,
> +                vttbr_vmid  :8,

Please drop vttbr_.

> +                res2        :8;
> +        };
> +        u64 vttbr;
> +    };
> +};
> +
>   struct hvm_function_table {
>       char *name;
>
> diff --git a/xen/include/asm-arm/p2m.h b/xen/include/asm-arm/p2m.h
> index 0d1e61e..a78d547 100644
> --- a/xen/include/asm-arm/p2m.h
> +++ b/xen/include/asm-arm/p2m.h
> @@ -8,6 +8,9 @@
>   #include <xen/p2m-common.h>
>   #include <public/memory.h>
>
> +#include <asm/atomic.h>
> +#include <asm/hvm/hvm.h>

ARM has not concept of HVM nor PV. So I would prefer to see a very 
limited usage of hvm.h

> +
>   #define paddr_bits PADDR_BITS
>
>   /* Holds the bit size of IPAs in p2m tables.  */
> @@ -17,6 +20,11 @@ struct domain;
>
>   extern void memory_type_changed(struct domain *);
>
> +typedef enum {
> +    p2m_host,
> +    p2m_alternate,
> +} p2m_class_t;
> +
>   /* Per-p2m-table state */
>   struct p2m_domain {
>       /* Lock that protects updates to the p2m */
> @@ -66,6 +74,18 @@ struct p2m_domain {
>       /* Radix tree to store the p2m_access_t settings as the pte's don't have
>        * enough available bits to store this information. */
>       struct radix_tree_root mem_access_settings;
> +
> +    /* Alternate p2m: count of vcpu's currently using this p2m. */
> +    atomic_t active_vcpus;
> +
> +    /* Choose between: host/alternate */
> +    p2m_class_t p2m_class;

Is there any reason to have this field? It is set but never used.

> +
> +    /* Back pointer to domain */
> +    struct domain *domain;

AFAICT, the only usage of this field is in p2m_altp2m_lazy_copy where 
you have direct access to the domain. So this could be dropped.

> +
> +    /* VTTBR information */
> +    struct vttbr_data vttbr;
>   };
>
>   /* List of possible type for each page in the p2m entry.
>

Regards,

-- 
Julien Grall

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel

  reply	other threads:[~2016-07-04 15:17 UTC|newest]

Thread overview: 126+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2016-07-04 11:45 [PATCH 00/18] arm/altp2m: Introducing altp2m to ARM Sergej Proskurin
2016-07-04 11:45 ` [PATCH 01/18] arm/altp2m: Add cmd-line support for altp2m on ARM Sergej Proskurin
2016-07-04 12:15   ` Andrew Cooper
2016-07-04 13:02     ` Sergej Proskurin
2016-07-04 13:25   ` Julien Grall
2016-07-04 13:43     ` Sergej Proskurin
2016-07-04 17:42   ` Julien Grall
2016-07-04 17:56     ` Tamas K Lengyel
2016-07-04 21:08       ` Sergej Proskurin
2016-07-04 11:45 ` [PATCH 02/18] arm/altp2m: Add first altp2m HVMOP stubs Sergej Proskurin
2016-07-04 13:36   ` Julien Grall
2016-07-04 13:51     ` Sergej Proskurin
2016-07-05 10:19   ` Julien Grall
2016-07-06  9:14     ` Sergej Proskurin
2016-07-06 13:43       ` Julien Grall
2016-07-06 15:23         ` Tamas K Lengyel
2016-07-06 15:54           ` Julien Grall
2016-07-06 16:05             ` Tamas K Lengyel
2016-07-06 16:29               ` Julien Grall
2016-07-06 16:35                 ` Tamas K Lengyel
2016-07-06 18:35                   ` Julien Grall
2016-07-07  9:14                     ` Sergej Proskurin
2016-07-04 11:45 ` [PATCH 03/18] arm/altp2m: Add HVMOP_altp2m_get_domain_state Sergej Proskurin
2016-07-04 11:45 ` [PATCH 04/18] arm/altp2m: Add altp2m init/teardown routines Sergej Proskurin
2016-07-04 15:17   ` Julien Grall [this message]
2016-07-04 16:40     ` Sergej Proskurin
2016-07-04 16:43       ` Andrew Cooper
2016-07-04 16:56         ` Sergej Proskurin
2016-07-04 17:44           ` Julien Grall
2016-07-04 21:19             ` Sergej Proskurin
2016-07-04 21:35               ` Julien Grall
2016-07-04 21:46               ` Sergej Proskurin
2016-07-04 18:18         ` Julien Grall
2016-07-04 21:37           ` Sergej Proskurin
2016-07-04 18:30       ` Julien Grall
2016-07-04 21:56         ` Sergej Proskurin
2016-07-04 16:15   ` Julien Grall
2016-07-04 16:51     ` Sergej Proskurin
2016-07-04 18:34       ` Julien Grall
2016-07-05  7:45         ` Sergej Proskurin
2016-07-04 11:45 ` [PATCH 05/18] arm/altp2m: Add HVMOP_altp2m_set_domain_state Sergej Proskurin
2016-07-04 15:39   ` Julien Grall
2016-07-05  8:45     ` Sergej Proskurin
2016-07-05 10:11       ` Julien Grall
2016-07-05 12:05         ` Sergej Proskurin
2016-07-04 11:45 ` [PATCH 06/18] arm/altp2m: Add a(p2m) table flushing routines Sergej Proskurin
2016-07-04 12:12   ` Sergej Proskurin
2016-07-04 15:42     ` Julien Grall
2016-07-05  8:52       ` Sergej Proskurin
2016-07-04 15:55   ` Julien Grall
2016-07-05  9:51     ` Sergej Proskurin
2016-07-04 16:20   ` Julien Grall
2016-07-05  9:57     ` Sergej Proskurin
2016-07-04 11:45 ` [PATCH 07/18] arm/altp2m: Add HVMOP_altp2m_create_p2m Sergej Proskurin
2016-07-04 11:45 ` [PATCH 08/18] arm/altp2m: Add HVMOP_altp2m_destroy_p2m Sergej Proskurin
2016-07-04 16:32   ` Julien Grall
2016-07-05 11:37     ` Sergej Proskurin
2016-07-05 11:48       ` Julien Grall
2016-07-05 12:18         ` Sergej Proskurin
2016-07-04 11:45 ` [PATCH 09/18] arm/altp2m: Add HVMOP_altp2m_switch_p2m Sergej Proskurin
2016-07-04 11:45 ` [PATCH 10/18] arm/altp2m: Renamed and extended p2m_alloc_table Sergej Proskurin
2016-07-04 18:43   ` Julien Grall
2016-07-05 13:56     ` Sergej Proskurin
2016-07-04 11:45 ` [PATCH 11/18] arm/altp2m: Make flush_tlb_domain ready for altp2m Sergej Proskurin
2016-07-04 12:30   ` Sergej Proskurin
2016-07-04 20:32   ` Julien Grall
2016-07-05 14:48     ` Sergej Proskurin
2016-07-05 15:37       ` Julien Grall
2016-07-05 20:21         ` Sergej Proskurin
2016-07-06 14:28           ` Julien Grall
2016-07-06 14:39             ` Sergej Proskurin
2016-07-07 17:24           ` Julien Grall
2016-07-04 11:45 ` [PATCH 12/18] arm/altp2m: Cosmetic fixes - function prototypes Sergej Proskurin
2016-07-15 13:45   ` Julien Grall
2016-07-16 15:18     ` Sergej Proskurin
2016-07-04 11:45 ` [PATCH 13/18] arm/altp2m: Make get_page_from_gva ready for altp2m Sergej Proskurin
2016-07-04 20:34   ` Julien Grall
2016-07-05 20:31     ` Sergej Proskurin
2016-07-04 11:45 ` [PATCH 14/18] arm/altp2m: Add HVMOP_altp2m_set_mem_access Sergej Proskurin
2016-07-05 12:49   ` Julien Grall
2016-07-05 21:55     ` Sergej Proskurin
2016-07-06 14:32       ` Julien Grall
2016-07-06 16:12         ` Tamas K Lengyel
2016-07-06 16:59           ` Julien Grall
2016-07-06 17:03           ` Sergej Proskurin
2016-07-06 17:08   ` Julien Grall
2016-07-07  9:16     ` Sergej Proskurin
2016-07-04 11:45 ` [PATCH 15/18] arm/altp2m: Add altp2m paging mechanism Sergej Proskurin
2016-07-04 20:53   ` Julien Grall
2016-07-06  8:33     ` Sergej Proskurin
2016-07-06 14:26       ` Julien Grall
2016-07-04 11:45 ` [PATCH 16/18] arm/altp2m: Extended libxl to activate altp2m on ARM Sergej Proskurin
2016-07-07 16:27   ` Wei Liu
2016-07-24 16:06     ` Sergej Proskurin
2016-07-25  8:32       ` Wei Liu
2016-07-25  9:04         ` Sergej Proskurin
2016-07-25  9:49           ` Julien Grall
2016-07-25 10:08             ` Wei Liu
2016-07-25 11:26               ` Sergej Proskurin
2016-07-25 11:37                 ` Wei Liu
2016-07-04 11:45 ` [PATCH 17/18] arm/altp2m: Adjust debug information to altp2m Sergej Proskurin
2016-07-04 20:58   ` Julien Grall
2016-07-06  8:41     ` Sergej Proskurin
2016-07-04 11:45 ` [PATCH 18/18] arm/altp2m: Extend xen-access for altp2m on ARM Sergej Proskurin
2016-07-04 13:38   ` Razvan Cojocaru
2016-07-06  8:44     ` Sergej Proskurin
2016-07-04 11:45 ` [PATCH 01/18] arm/altp2m: Add cmd-line support " Sergej Proskurin
2016-07-04 11:45 ` [PATCH 02/18] arm/altp2m: Add first altp2m HVMOP stubs Sergej Proskurin
2016-07-04 11:45 ` [PATCH 03/18] arm/altp2m: Add HVMOP_altp2m_get_domain_state Sergej Proskurin
2016-07-04 11:45 ` [PATCH 04/18] arm/altp2m: Add altp2m init/teardown routines Sergej Proskurin
2016-07-04 11:45 ` [PATCH 05/18] arm/altp2m: Add HVMOP_altp2m_set_domain_state Sergej Proskurin
2016-07-04 11:45 ` [PATCH 06/18] arm/altp2m: Add a(p2m) table flushing routines Sergej Proskurin
2016-07-04 11:45 ` [PATCH 07/18] arm/altp2m: Add HVMOP_altp2m_create_p2m Sergej Proskurin
2016-07-04 11:45 ` [PATCH 08/18] arm/altp2m: Add HVMOP_altp2m_destroy_p2m Sergej Proskurin
2016-07-04 11:45 ` [PATCH 09/18] arm/altp2m: Add HVMOP_altp2m_switch_p2m Sergej Proskurin
2016-07-04 11:45 ` [PATCH 10/18] arm/altp2m: Renamed and extended p2m_alloc_table Sergej Proskurin
2016-07-04 11:45 ` [PATCH 11/18] arm/altp2m: Make flush_tlb_domain ready for altp2m Sergej Proskurin
2016-07-04 11:45 ` [PATCH 12/18] arm/altp2m: Cosmetic fixes - function prototypes Sergej Proskurin
2016-07-04 11:46 ` [PATCH 13/18] arm/altp2m: Make get_page_from_gva ready for altp2m Sergej Proskurin
2016-07-04 11:46 ` [PATCH 14/18] arm/altp2m: Add HVMOP_altp2m_set_mem_access Sergej Proskurin
2016-07-04 11:46 ` [PATCH 15/18] arm/altp2m: Add altp2m paging mechanism Sergej Proskurin
2016-07-04 11:46 ` [PATCH 16/18] arm/altp2m: Extended libxl to activate altp2m on ARM Sergej Proskurin
2016-07-04 11:46 ` [PATCH 17/18] arm/altp2m: Adjust debug information to altp2m Sergej Proskurin
2016-07-04 11:46 ` [PATCH 18/18] arm/altp2m: Extend xen-access for altp2m on ARM Sergej Proskurin
2016-07-04 12:52 ` [PATCH 00/18] arm/altp2m: Introducing altp2m to ARM Andrew Cooper
2016-07-04 13:05   ` Sergej Proskurin

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=577A7DFD.2070600@arm.com \
    --to=julien.grall@arm.com \
    --cc=proskurin@sec.in.tum.de \
    --cc=sstabellini@kernel.org \
    --cc=xen-devel@lists.xenproject.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).