Commit d9954a2d authored by James Smart's avatar James Smart Committed by Martin K. Petersen

scsi: lpfc: Fix oops when driver is loaded with 1 interrupt vector

The driver was coded expecting enough hardware queues and interrupt vectors
such that at least there was one per socket. In the case where there were
fewer than sockets, cpus were left unassigned thus null pointers.

Rework the affinity mappings. Map settings for the cpu's that are in the
irq cpu mask. For each cpu not in the mask, map to another cpu that does
have a mask. Choice of the "other" cpu will attempt to map to the same cpu
but differing hyperthread, or cpu within in same core, or cpu within same
socket, or finally cpu in the base socket.
Signed-off-by: default avatarDick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: default avatarJames Smart <jsmart2021@gmail.com>
Signed-off-by: default avatarMartin K. Petersen <martin.petersen@oracle.com>
parent b8e6f136
...@@ -5236,35 +5236,44 @@ lpfc_fcp_cpu_map_show(struct device *dev, struct device_attribute *attr, ...@@ -5236,35 +5236,44 @@ lpfc_fcp_cpu_map_show(struct device *dev, struct device_attribute *attr,
len += scnprintf( len += scnprintf(
buf + len, PAGE_SIZE - len, buf + len, PAGE_SIZE - len,
"CPU %02d hdwq None " "CPU %02d hdwq None "
"physid %d coreid %d ht %d\n", "physid %d coreid %d ht %d ua %d\n",
phba->sli4_hba.curr_disp_cpu, phba->sli4_hba.curr_disp_cpu,
cpup->phys_id, cpup->phys_id, cpup->core_id,
cpup->core_id, cpup->hyper); (cpup->flag & LPFC_CPU_MAP_HYPER),
(cpup->flag & LPFC_CPU_MAP_UNASSIGN));
else else
len += scnprintf( len += scnprintf(
buf + len, PAGE_SIZE - len, buf + len, PAGE_SIZE - len,
"CPU %02d EQ %04d hdwq %04d " "CPU %02d EQ %04d hdwq %04d "
"physid %d coreid %d ht %d\n", "physid %d coreid %d ht %d ua %d\n",
phba->sli4_hba.curr_disp_cpu, phba->sli4_hba.curr_disp_cpu,
cpup->eq, cpup->hdwq, cpup->phys_id, cpup->eq, cpup->hdwq, cpup->phys_id,
cpup->core_id, cpup->hyper); cpup->core_id,
(cpup->flag & LPFC_CPU_MAP_HYPER),
(cpup->flag & LPFC_CPU_MAP_UNASSIGN));
} else { } else {
if (cpup->hdwq == LPFC_VECTOR_MAP_EMPTY) if (cpup->hdwq == LPFC_VECTOR_MAP_EMPTY)
len += scnprintf( len += scnprintf(
buf + len, PAGE_SIZE - len, buf + len, PAGE_SIZE - len,
"CPU %02d hdwq None " "CPU %02d hdwq None "
"physid %d coreid %d ht %d IRQ %d\n", "physid %d coreid %d ht %d ua %d IRQ %d\n",
phba->sli4_hba.curr_disp_cpu, phba->sli4_hba.curr_disp_cpu,
cpup->phys_id, cpup->phys_id,
cpup->core_id, cpup->hyper, cpup->irq); cpup->core_id,
(cpup->flag & LPFC_CPU_MAP_HYPER),
(cpup->flag & LPFC_CPU_MAP_UNASSIGN),
cpup->irq);
else else
len += scnprintf( len += scnprintf(
buf + len, PAGE_SIZE - len, buf + len, PAGE_SIZE - len,
"CPU %02d EQ %04d hdwq %04d " "CPU %02d EQ %04d hdwq %04d "
"physid %d coreid %d ht %d IRQ %d\n", "physid %d coreid %d ht %d ua %d IRQ %d\n",
phba->sli4_hba.curr_disp_cpu, phba->sli4_hba.curr_disp_cpu,
cpup->eq, cpup->hdwq, cpup->phys_id, cpup->eq, cpup->hdwq, cpup->phys_id,
cpup->core_id, cpup->hyper, cpup->irq); cpup->core_id,
(cpup->flag & LPFC_CPU_MAP_HYPER),
(cpup->flag & LPFC_CPU_MAP_UNASSIGN),
cpup->irq);
} }
phba->sli4_hba.curr_disp_cpu++; phba->sli4_hba.curr_disp_cpu++;
......
...@@ -10653,24 +10653,31 @@ lpfc_find_hyper(struct lpfc_hba *phba, int cpu, ...@@ -10653,24 +10653,31 @@ lpfc_find_hyper(struct lpfc_hba *phba, int cpu,
static void static void
lpfc_cpu_affinity_check(struct lpfc_hba *phba, int vectors) lpfc_cpu_affinity_check(struct lpfc_hba *phba, int vectors)
{ {
int i, cpu, idx; int i, cpu, idx, new_cpu, start_cpu, first_cpu;
int max_phys_id, min_phys_id; int max_phys_id, min_phys_id;
int max_core_id, min_core_id; int max_core_id, min_core_id;
struct lpfc_vector_map_info *cpup; struct lpfc_vector_map_info *cpup;
struct lpfc_vector_map_info *new_cpup;
const struct cpumask *maskp; const struct cpumask *maskp;
#ifdef CONFIG_X86 #ifdef CONFIG_X86
struct cpuinfo_x86 *cpuinfo; struct cpuinfo_x86 *cpuinfo;
#endif #endif
/* Init cpu_map array */ /* Init cpu_map array */
memset(phba->sli4_hba.cpu_map, 0xff, for_each_possible_cpu(cpu) {
(sizeof(struct lpfc_vector_map_info) * cpup = &phba->sli4_hba.cpu_map[cpu];
phba->sli4_hba.num_possible_cpu)); cpup->phys_id = LPFC_VECTOR_MAP_EMPTY;
cpup->core_id = LPFC_VECTOR_MAP_EMPTY;
cpup->hdwq = LPFC_VECTOR_MAP_EMPTY;
cpup->eq = LPFC_VECTOR_MAP_EMPTY;
cpup->irq = LPFC_VECTOR_MAP_EMPTY;
cpup->flag = 0;
}
max_phys_id = 0; max_phys_id = 0;
min_phys_id = 0xffff; min_phys_id = LPFC_VECTOR_MAP_EMPTY;
max_core_id = 0; max_core_id = 0;
min_core_id = 0xffff; min_core_id = LPFC_VECTOR_MAP_EMPTY;
/* Update CPU map with physical id and core id of each CPU */ /* Update CPU map with physical id and core id of each CPU */
for_each_present_cpu(cpu) { for_each_present_cpu(cpu) {
...@@ -10679,13 +10686,12 @@ lpfc_cpu_affinity_check(struct lpfc_hba *phba, int vectors) ...@@ -10679,13 +10686,12 @@ lpfc_cpu_affinity_check(struct lpfc_hba *phba, int vectors)
cpuinfo = &cpu_data(cpu); cpuinfo = &cpu_data(cpu);
cpup->phys_id = cpuinfo->phys_proc_id; cpup->phys_id = cpuinfo->phys_proc_id;
cpup->core_id = cpuinfo->cpu_core_id; cpup->core_id = cpuinfo->cpu_core_id;
cpup->hyper = lpfc_find_hyper(phba, cpu, if (lpfc_find_hyper(phba, cpu, cpup->phys_id, cpup->core_id))
cpup->phys_id, cpup->core_id); cpup->flag |= LPFC_CPU_MAP_HYPER;
#else #else
/* No distinction between CPUs for other platforms */ /* No distinction between CPUs for other platforms */
cpup->phys_id = 0; cpup->phys_id = 0;
cpup->core_id = cpu; cpup->core_id = cpu;
cpup->hyper = 0;
#endif #endif
lpfc_printf_log(phba, KERN_INFO, LOG_INIT, lpfc_printf_log(phba, KERN_INFO, LOG_INIT,
...@@ -10711,6 +10717,12 @@ lpfc_cpu_affinity_check(struct lpfc_hba *phba, int vectors) ...@@ -10711,6 +10717,12 @@ lpfc_cpu_affinity_check(struct lpfc_hba *phba, int vectors)
eqi->icnt = 0; eqi->icnt = 0;
} }
/* This loop sets up all CPUs that are affinitized with a
* irq vector assigned to the driver. All affinitized CPUs
* will get a link to that vectors IRQ and EQ. For now we
* are assuming all CPUs using the same EQ will all share
* the same hardware queue.
*/
for (idx = 0; idx < phba->cfg_irq_chann; idx++) { for (idx = 0; idx < phba->cfg_irq_chann; idx++) {
maskp = pci_irq_get_affinity(phba->pcidev, idx); maskp = pci_irq_get_affinity(phba->pcidev, idx);
if (!maskp) if (!maskp)
...@@ -10728,6 +10740,119 @@ lpfc_cpu_affinity_check(struct lpfc_hba *phba, int vectors) ...@@ -10728,6 +10740,119 @@ lpfc_cpu_affinity_check(struct lpfc_hba *phba, int vectors)
cpu, cpup->hdwq, cpup->irq); cpu, cpup->hdwq, cpup->irq);
} }
} }
/* After looking at each irq vector assigned to this pcidev, its
* possible to see that not ALL CPUs have been accounted for.
* Next we will set any unassigned cpu map entries to a IRQ
* on the same phys_id
*/
first_cpu = cpumask_first(cpu_present_mask);
start_cpu = first_cpu;
for_each_present_cpu(cpu) {
cpup = &phba->sli4_hba.cpu_map[cpu];
/* Is this CPU entry unassigned */
if (cpup->eq == LPFC_VECTOR_MAP_EMPTY) {
/* Mark CPU as IRQ not assigned by the kernel */
cpup->flag |= LPFC_CPU_MAP_UNASSIGN;
/* If so, find a new_cpup thats on the the same
* phys_id as cpup. start_cpu will start where we
* left off so all unassigned entries don't get assgined
* the IRQ of the first entry.
*/
new_cpu = start_cpu;
for (i = 0; i < phba->sli4_hba.num_present_cpu; i++) {
new_cpup = &phba->sli4_hba.cpu_map[new_cpu];
if (!(new_cpup->flag & LPFC_CPU_MAP_UNASSIGN) &&
(new_cpup->irq != LPFC_VECTOR_MAP_EMPTY) &&
(new_cpup->phys_id == cpup->phys_id))
goto found_same;
new_cpu = cpumask_next(
new_cpu, cpu_present_mask);
if (new_cpu == nr_cpumask_bits)
new_cpu = first_cpu;
}
/* At this point, we leave the CPU as unassigned */
continue;
found_same:
/* We found a matching phys_id, so copy the IRQ info */
cpup->eq = new_cpup->eq;
cpup->hdwq = new_cpup->hdwq;
cpup->irq = new_cpup->irq;
/* Bump start_cpu to the next slot to minmize the
* chance of having multiple unassigned CPU entries
* selecting the same IRQ.
*/
start_cpu = cpumask_next(new_cpu, cpu_present_mask);
if (start_cpu == nr_cpumask_bits)
start_cpu = first_cpu;
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"3337 Set Affinity: CPU %d "
"hdwq %d irq %d from id %d same "
"phys_id (%d)\n",
cpu, cpup->hdwq, cpup->irq,
new_cpu, cpup->phys_id);
}
}
/* Set any unassigned cpu map entries to a IRQ on any phys_id */
start_cpu = first_cpu;
for_each_present_cpu(cpu) {
cpup = &phba->sli4_hba.cpu_map[cpu];
/* Is this entry unassigned */
if (cpup->eq == LPFC_VECTOR_MAP_EMPTY) {
/* Mark it as IRQ not assigned by the kernel */
cpup->flag |= LPFC_CPU_MAP_UNASSIGN;
/* If so, find a new_cpup thats on any phys_id
* as the cpup. start_cpu will start where we
* left off so all unassigned entries don't get
* assigned the IRQ of the first entry.
*/
new_cpu = start_cpu;
for (i = 0; i < phba->sli4_hba.num_present_cpu; i++) {
new_cpup = &phba->sli4_hba.cpu_map[new_cpu];
if (!(new_cpup->flag & LPFC_CPU_MAP_UNASSIGN) &&
(new_cpup->irq != LPFC_VECTOR_MAP_EMPTY))
goto found_any;
new_cpu = cpumask_next(
new_cpu, cpu_present_mask);
if (new_cpu == nr_cpumask_bits)
new_cpu = first_cpu;
}
/* We should never leave an entry unassigned */
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"3339 Set Affinity: CPU %d "
"hdwq %d irq %d UNASSIGNED\n",
cpu, cpup->hdwq, cpup->irq);
continue;
found_any:
/* We found an available entry, copy the IRQ info */
cpup->eq = new_cpup->eq;
cpup->hdwq = new_cpup->hdwq;
cpup->irq = new_cpup->irq;
/* Bump start_cpu to the next slot to minmize the
* chance of having multiple unassigned CPU entries
* selecting the same IRQ.
*/
start_cpu = cpumask_next(new_cpu, cpu_present_mask);
if (start_cpu == nr_cpumask_bits)
start_cpu = first_cpu;
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"3338 Set Affinity: CPU %d "
"hdwq %d irq %d from id %d (%d/%d)\n",
cpu, cpup->hdwq, cpup->irq, new_cpu,
new_cpup->phys_id, new_cpup->core_id);
}
}
return; return;
} }
......
...@@ -549,7 +549,9 @@ struct lpfc_vector_map_info { ...@@ -549,7 +549,9 @@ struct lpfc_vector_map_info {
uint16_t irq; uint16_t irq;
uint16_t eq; uint16_t eq;
uint16_t hdwq; uint16_t hdwq;
uint16_t hyper; uint16_t flag;
#define LPFC_CPU_MAP_HYPER 0x1
#define LPFC_CPU_MAP_UNASSIGN 0x2
}; };
#define LPFC_VECTOR_MAP_EMPTY 0xffff #define LPFC_VECTOR_MAP_EMPTY 0xffff
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment