Re: [PATCH PREVIEW RFC 3/6] crypto: use bootcache to cache fastest algorithm

On Tue, 30 Sept 2025 at 00:48, Bird, Tim <[email protected]> wrote:
>
> > -----Original Message-----
> > From: acampanella-thegoodpenguin <[email protected]>
> > Sent: Tuesday, September 23, 2025 8:24 AM
> > To: [email protected]
> > Subject: [PATCH PREVIEW RFC 3/6] crypto: use bootcache to cache fastest algorithm
> >
> > From: Andrew Murray <amurray@ thegoodpenguin. co. uk> During boot xor_blocks may determine the fastest xor algorithm by using
> > do_xor_speed to perform a speed test on available algorithms. This process can increase the overall boot time. Let's
> > From: Andrew Murray <[email protected]>
> >
> > During boot xor_blocks may determine the fastest xor algorithm
> > by using do_xor_speed to perform a speed test on available
> > algorithms. This process can increase the overall boot time.
>
> It would be good to mention the amount of time we're talking about here.
> It won't be the same value for all platforms, but you could mention the
> amount of time this takes on the platform you're working on (or on a particularly
> slow machine, if you want, just to highlight where it might be a problem
> for some situations but not all.)

Good idea, thanks!

Though, it may be that a whole bunch of patches like this are
accumulated, you boot the kernel with them enabled, blindly cache all
the results (for reuse in subsequent boots), and gain a quicker boot.
I.e. it may not be necessary for users to pick and choose.


> >
> > Let's make use of bootcache to cache the result of the speed
> > test for subsequent boots.
> >
> > Signed-off-by: Andrew Murray <[email protected]>
> > ---
> >  crypto/xor.c | 29 ++++++++++++++++++++++++++++-
> >  1 file changed, 28 insertions(+), 1 deletion(-)
> >
> > diff --git a/crypto/xor.c b/crypto/xor.c
> > index f39621a57bb33c4015c06dff00e03a07716618f6..3457df0414064758a1923752e91642d2237af7b3 100644
> > --- a/crypto/xor.c
> > +++ b/crypto/xor.c
> > @@ -14,6 +14,7 @@
> >  #include <linux/raid/xor.h>
> >  #include <linux/jiffies.h>
> >  #include <linux/preempt.h>
> > +#include <linux/bootcache.h>
> >  #include <asm/xor.h>
> >
> >  #ifndef XOR_SELECT_TEMPLATE
> > @@ -54,13 +55,13 @@ EXPORT_SYMBOL(xor_blocks);
> >  /* Set of all registered templates.  */
> >  static struct xor_block_template *__initdata template_list;
> >
> > -#ifndef MODULE
> >  static void __init do_xor_register(struct xor_block_template *tmpl)
> >  {
> >       tmpl->next = template_list;
> >       template_list = tmpl;
> >  }
> >
> > +#ifndef MODULE
> >  static int __init register_xor_blocks(void)
> >  {
> >       active_template = XOR_SELECT_TEMPLATE(NULL);
> > @@ -79,6 +80,21 @@ static int __init register_xor_blocks(void)
> >  #define BENCH_SIZE   4096
> >  #define REPS         800U
> >
> > +static struct xor_block_template * __init
> > +xor_get_template_by_name(char *fastest_name)
> > +{
> > +     struct xor_block_template *f;
> > +
> > +#define xor_speed    do_xor_register
> > +     // build a list of templates
> > +     XOR_TRY_TEMPLATES;
> > +#undef xor_speed
> > +     for (f = template_list; f; f = f->next)
> > +             if (!strcmp(f->name, fastest_name))
> > +                     return f;
> > +     return NULL;
> > +}
> > +
> >  static void __init
> >  do_xor_speed(struct xor_block_template *tmpl, void *b1, void *b2)
> >  {
> > @@ -117,9 +133,18 @@ calibrate_xor_blocks(void)
> >  {
> >       void *b1, *b2;
> >       struct xor_block_template *f, *fastest;
> > +     char cached_name[32];
> > +     int ret;
> >
> >       fastest = XOR_SELECT_TEMPLATE(NULL);
> >
> > +     if (!fastest) {
> > +             ret = bootcache_get_string("xor_blocks_fastest",
> > +                             cached_name, sizeof(cached_name));
> > +             if (!ret)
> > +                     fastest = xor_get_template_by_name(cached_name);
>
> I presume that if CONFIG_BOOTCACHE is not defined, then ret ends up being -ENOENT
> always, which makes this whole block completely evaporate. Is that right?
> In that case, do you end up with a warning about unused variable cached_name?
> Or does the compiler know not to complain about that?  Do you still end up with
> space reserved on the stack?

Good spot, I'm not sure the answer but i'll find out.

Thanks,

Andrew Murray


>
> > +     }
> > +
> >       if (fastest) {
> >               printk(KERN_INFO "xor: automatically using best "
> >                                "checksumming function   %-10s\n",
> > @@ -149,6 +174,8 @@ calibrate_xor_blocks(void)
> >               if (f->speed > fastest->speed)
> >                       fastest = f;
> >
> > +     bootcache_set_string("xor_blocks_fastest", fastest->name);
> > +
> >       pr_info("xor: using function: %s (%d MB/sec)\n",
> >              fastest->name, fastest->speed);
> >
> >
> > --
> > 2.48.1
> >
>