On 11.07.2025 02:15, Stefano Stabellini wrote:
> On Thu, 10 Jul 2025, Dmytro Prokopchuk1 wrote:
>> --- a/xen/common/memory.c
>> +++ b/xen/common/memory.c
>> @@ -773,7 +773,7 @@ static long
>> memory_exchange(XEN_GUEST_HANDLE_PARAM(xen_memory_exchange_t) arg)
>>
>> nrspin_lock(&d->page_alloc_lock);
>> drop_dom_ref = (dec_count &&
>> - !domain_adjust_tot_pages(d, -dec_count));
>> + !domain_adjust_tot_pages(d, (-1L) *
>> dec_count));
>
> ...but I don't understand this? It looks like it would also break 10.1
> and/or 10.3 as well?
>
> I would rather use casts if needed, which wouldn't change the behavior
> but would highlight the unsigned->signed conversion, making it more
> visible:
>
> !domain_adjust_tot_pages(d, -(int)dec_count));
Except that, as more often than not anyway, such casts are fragile.
>> --- a/xen/common/page_alloc.c
>> +++ b/xen/common/page_alloc.c
>> @@ -691,7 +691,7 @@ size_param("low_mem_virq_limit", opt_low_mem_virq);
>> /* Thresholds to control hysteresis. In pages */
>> /* When memory grows above this threshold, reset hysteresis.
>> * -1 initially to not reset until at least one virq issued. */
>> -static unsigned long low_mem_virq_high = -1UL;
>> +static unsigned long low_mem_virq_high = ~0UL;
>> /* Threshold at which we issue virq */
>> static unsigned long low_mem_virq_th = 0;
>> /* Original threshold after all checks completed */
>> @@ -710,7 +710,7 @@ static void __init setup_low_mem_virq(void)
>> * to ever trigger. */
>> if ( opt_low_mem_virq == 0 )
>> {
>> - low_mem_virq_th = -1UL;
>> + low_mem_virq_th = ~0UL;
>> return;
>> }
>>
>> @@ -778,7 +778,7 @@ static void check_low_mem_virq(void)
>> low_mem_virq_th_order++;
>> low_mem_virq_th = 1UL << low_mem_virq_th_order;
>> if ( low_mem_virq_th == low_mem_virq_orig )
>> - low_mem_virq_high = -1UL;
>> + low_mem_virq_high = ~0UL;
>> else
>> low_mem_virq_high = 1UL << (low_mem_virq_th_order + 2);
>> }
>> --- a/xen/common/time.c
>> +++ b/xen/common/time.c
>> @@ -84,7 +84,7 @@ struct tm gmtime(unsigned long t)
>> }
>> tbuf.tm_year = y - 1900;
>> tbuf.tm_yday = days;
>> - ip = (const unsigned short int *)__mon_lengths[__isleap(y)];
>> + ip = (const unsigned short int *)__mon_lengths[(int)__isleap(y)];
>
> __isleap return bool and we deviated bool conversions in logical
> operations but not here, so I understand why this is needed. OK.
No. If a cast is unavoidable here (which I question), it wants to be to
an unsigned type.
I'm glad I looked at the patch, btw, since these files - contrary to the
patch subject prefix - aren't Arm files.
>> --- a/xen/lib/strtol.c
>> +++ b/xen/lib/strtol.c
>> @@ -13,7 +13,7 @@
>> long simple_strtol(const char *cp, const char **endp, unsigned int base)
>> {
>> if ( *cp == '-' )
>> - return -simple_strtoul(cp + 1, endp, base);
>> + return (-1L) * simple_strtoul(cp + 1, endp, base);
>> return simple_strtoul(cp, endp, base);
>> }
>>
>> --- a/xen/lib/strtoll.c
>> +++ b/xen/lib/strtoll.c
>> @@ -13,7 +13,7 @@
>> long long simple_strtoll(const char *cp, const char **endp, unsigned int
>> base)
>> {
>> if ( *cp == '-' )
>> - return -simple_strtoull(cp + 1, endp, base);
>> + return (-1LL) * simple_strtoull(cp + 1, endp, base);
>> return simple_strtoull(cp, endp, base);
>> }
>>
Nor are these two.
As to the kind of change here - didn't we deviate applying unary minus to
unsigned types?
Jan
