Re: [PATCH v2] xen/console: introduce domain_console struct

[Jan Beulich <jbeulich@xxxxxxxx>]

On 17.06.2025 03:27, dmkhn@xxxxxxxxx wrote:
> From: Denis Mukhin <dmukhin@xxxxxxxx>
> 
> Introduce domain_console for grouping data structures used for integrating
> domain's diagnostic console with Xen's console driver.
> 
> Group all pbuf-related data structures under domain_console. Rename the moved
> fields to plain .buf, .idx and .lock names, since all uses of the fields are
> touched.
> 
> Bump the domain console buffer size to the closest power of 2 (256) and
> rename the symbol to DOMAIN_CONSOLE_BUF_SIZE.
> 
> Move d->console->buf management under CONFIG_VERBOSE_DEBUG when the
> HYPERCALL_console_io handler is enabled.

This, if at all, needs to be a separate change (with its own justification).
I for one don't think VERBOSE_DEBUG is intended to control any kind of guest
output.

> --- a/xen/arch/x86/hvm/hvm.c
> +++ b/xen/arch/x86/hvm/hvm.c
> @@ -559,7 +559,6 @@ void hvm_do_resume(struct vcpu *v)
>  static int cf_check hvm_print_line(
>      int dir, unsigned int port, unsigned int bytes, uint32_t *val)
>  {
> -    struct domain *cd = current->domain;
>      char c = *val;
>  
>      ASSERT(bytes == 1 && port == XEN_HVM_DEBUGCONS_IOPORT);
> @@ -570,17 +569,24 @@ static int cf_check hvm_print_line(
>  
>      if ( !is_console_printable(c) )
>          return X86EMUL_OKAY;

After this "return" ...

> -
> -    spin_lock(&cd->pbuf_lock);
> -    if ( c != '\n' )
> -        cd->pbuf[cd->pbuf_idx++] = c;
> -    if ( (cd->pbuf_idx == (DOMAIN_PBUF_SIZE - 1)) || (c == '\n') )
> +#ifdef CONFIG_VERBOSE_DEBUG
> +    else

... there's no point to have "else" here.

>      {
> -        cd->pbuf[cd->pbuf_idx] = '\0';
> -        guest_printk(cd, XENLOG_G_DEBUG "%s\n", cd->pbuf);
> -        cd->pbuf_idx = 0;
> +        struct domain *cd = current->domain;

We normally name such a variable d. It also looks as if this could be pointer-
to-const.

> +        struct domain_console *cons = cd->console;
> +
> +        spin_lock(&cons->lock);
> +        if ( c != '\n' )
> +            cons->buf[cons->idx++] = c;
> +        if ( (cons->idx == (DOMAIN_CONSOLE_BUF_SIZE - 1)) || (c == '\n') )
> +        {
> +            cons->buf[cons->idx] = '\0';
> +            guest_printk(cd, XENLOG_G_DEBUG "%s\n", cons->buf);
> +            cons->idx = 0;
> +        }
> +        spin_unlock(&cons->lock);
>      }
> -    spin_unlock(&cd->pbuf_lock);
> +#endif

None of the re-indentation is really warranted here (and will likely go away
anyway once the #ifdef is dropped).

> --- a/xen/common/domain.c
> +++ b/xen/common/domain.c
> @@ -669,7 +669,7 @@ static void _domain_destroy(struct domain *d)
>      BUG_ON(!d->is_dying);
>      BUG_ON(atomic_read(&d->refcnt) != DOMAIN_DESTROYED);
>  
> -    xfree(d->pbuf);
> +    xfree(d->console);
>  
>      argo_destroy(d);
>  
> @@ -800,6 +800,11 @@ struct domain *domain_create(domid_t domid,
>      if ( (d = alloc_domain_struct()) == NULL )
>          return ERR_PTR(-ENOMEM);
>  
> +    err = -ENOMEM;
> +    d->console = xvzalloc(typeof(*d->console));
> +    if ( !d->console )
> +        goto fail;

This definitely need to move down some, at least ...

>      /* Sort out our idea of is_system_domain(). */
>      d->domain_id = domid;
>      d->unique_id = get_unique_id();

... past here. There absolutely must not be struct domain instances be
passed around (see e.g. the call to sched_destroy_domain()) without the
domain ID set. It's hard to see ...

> @@ -862,7 +867,9 @@ struct domain *domain_create(domid_t domid,
>      spin_lock_init(&d->shutdown_lock);
>      d->shutdown_code = SHUTDOWN_CODE_INVALID;
>  
> -    spin_lock_init(&d->pbuf_lock);
> +#ifdef CONFIG_VERBOSE_DEBUG
> +    spin_lock_init(&d->console->lock);
> +#endif

.. why here or ...

> @@ -955,11 +962,6 @@ struct domain *domain_create(domid_t domid,
>      if ( (err = argo_init(d)) != 0 )
>          goto fail;
>  
> -    err = -ENOMEM;
> -    d->pbuf = xzalloc_array(char, DOMAIN_PBUF_SIZE);
> -    if ( !d->pbuf )
> -        goto fail;

... even here wouldn't be early enough anyway.

And btw - where did this buffer allocation move? I don't see anywhere
that d->console->buf would now be initialized. (However, see below.)

> --- a/xen/include/xen/sched.h
> +++ b/xen/include/xen/sched.h
> @@ -371,6 +371,22 @@ struct evtchn_port_ops;
>  
>  #define MAX_NR_IOREQ_SERVERS 8
>  
> +/* Arbitrary value; must be a multiple of the cacheline size. */
> +#define DOMAIN_CONSOLE_BUF_SIZE 256

Where does the relationship with cache line size come from? What if
Xen was to run on hardware (whichever arch) with 512-byte cache lines?

> +/* Domain console settings. */
> +struct domain_console {
> +#ifdef CONFIG_VERBOSE_DEBUG
> +    /* hvm_print_line() and guest_console_write() logging. */
> +    char *buf;

To avoid the need to do yet another separate allocation, how about ...

> +    unsigned int idx;
> +    spinlock_t lock;
> +#endif /* CONFIG_VERBOSE_DEBUG */
> +
> +    /* Permission to take ownership of the physical console input. */
> +    bool input_allowed;

    char buf[DOMAIN_CONSOLE_BUF_SIZE];

Ultimately this would allow the buffer size to be e.g. command line
controlled (if so desired), by then simply converting to a flexible
array member.

Jan