Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / misc / cxl / native.c
diff --git a/kernel/drivers/misc/cxl/native.c b/kernel/drivers/misc/cxl/native.c
new file mode 100644 (file)
index 0000000..29185fc
--- /dev/null
@@ -0,0 +1,704 @@
+/*
+ * Copyright 2014 IBM Corp.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+
+#include <linux/spinlock.h>
+#include <linux/sched.h>
+#include <linux/slab.h>
+#include <linux/sched.h>
+#include <linux/mutex.h>
+#include <linux/mm.h>
+#include <linux/uaccess.h>
+#include <asm/synch.h>
+#include <misc/cxl.h>
+
+#include "cxl.h"
+#include "trace.h"
+
+static int afu_control(struct cxl_afu *afu, u64 command,
+                      u64 result, u64 mask, bool enabled)
+{
+       u64 AFU_Cntl = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
+       unsigned long timeout = jiffies + (HZ * CXL_TIMEOUT);
+       int rc = 0;
+
+       spin_lock(&afu->afu_cntl_lock);
+       pr_devel("AFU command starting: %llx\n", command);
+
+       trace_cxl_afu_ctrl(afu, command);
+
+       cxl_p2n_write(afu, CXL_AFU_Cntl_An, AFU_Cntl | command);
+
+       AFU_Cntl = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
+       while ((AFU_Cntl & mask) != result) {
+               if (time_after_eq(jiffies, timeout)) {
+                       dev_warn(&afu->dev, "WARNING: AFU control timed out!\n");
+                       rc = -EBUSY;
+                       goto out;
+               }
+               pr_devel_ratelimited("AFU control... (0x%.16llx)\n",
+                                    AFU_Cntl | command);
+               cpu_relax();
+               AFU_Cntl = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
+       };
+       pr_devel("AFU command complete: %llx\n", command);
+       afu->enabled = enabled;
+out:
+       trace_cxl_afu_ctrl_done(afu, command, rc);
+       spin_unlock(&afu->afu_cntl_lock);
+
+       return rc;
+}
+
+static int afu_enable(struct cxl_afu *afu)
+{
+       pr_devel("AFU enable request\n");
+
+       return afu_control(afu, CXL_AFU_Cntl_An_E,
+                          CXL_AFU_Cntl_An_ES_Enabled,
+                          CXL_AFU_Cntl_An_ES_MASK, true);
+}
+
+int cxl_afu_disable(struct cxl_afu *afu)
+{
+       pr_devel("AFU disable request\n");
+
+       return afu_control(afu, 0, CXL_AFU_Cntl_An_ES_Disabled,
+                          CXL_AFU_Cntl_An_ES_MASK, false);
+}
+
+/* This will disable as well as reset */
+int cxl_afu_reset(struct cxl_afu *afu)
+{
+       pr_devel("AFU reset request\n");
+
+       return afu_control(afu, CXL_AFU_Cntl_An_RA,
+                          CXL_AFU_Cntl_An_RS_Complete | CXL_AFU_Cntl_An_ES_Disabled,
+                          CXL_AFU_Cntl_An_RS_MASK | CXL_AFU_Cntl_An_ES_MASK,
+                          false);
+}
+
+static int afu_check_and_enable(struct cxl_afu *afu)
+{
+       if (afu->enabled)
+               return 0;
+       return afu_enable(afu);
+}
+
+int cxl_psl_purge(struct cxl_afu *afu)
+{
+       u64 PSL_CNTL = cxl_p1n_read(afu, CXL_PSL_SCNTL_An);
+       u64 AFU_Cntl = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
+       u64 dsisr, dar;
+       u64 start, end;
+       unsigned long timeout = jiffies + (HZ * CXL_TIMEOUT);
+       int rc = 0;
+
+       trace_cxl_psl_ctrl(afu, CXL_PSL_SCNTL_An_Pc);
+
+       pr_devel("PSL purge request\n");
+
+       if ((AFU_Cntl & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
+               WARN(1, "psl_purge request while AFU not disabled!\n");
+               cxl_afu_disable(afu);
+       }
+
+       cxl_p1n_write(afu, CXL_PSL_SCNTL_An,
+                      PSL_CNTL | CXL_PSL_SCNTL_An_Pc);
+       start = local_clock();
+       PSL_CNTL = cxl_p1n_read(afu, CXL_PSL_SCNTL_An);
+       while ((PSL_CNTL &  CXL_PSL_SCNTL_An_Ps_MASK)
+                       == CXL_PSL_SCNTL_An_Ps_Pending) {
+               if (time_after_eq(jiffies, timeout)) {
+                       dev_warn(&afu->dev, "WARNING: PSL Purge timed out!\n");
+                       rc = -EBUSY;
+                       goto out;
+               }
+               dsisr = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
+               pr_devel_ratelimited("PSL purging... PSL_CNTL: 0x%.16llx  PSL_DSISR: 0x%.16llx\n", PSL_CNTL, dsisr);
+               if (dsisr & CXL_PSL_DSISR_TRANS) {
+                       dar = cxl_p2n_read(afu, CXL_PSL_DAR_An);
+                       dev_notice(&afu->dev, "PSL purge terminating pending translation, DSISR: 0x%.16llx, DAR: 0x%.16llx\n", dsisr, dar);
+                       cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
+               } else if (dsisr) {
+                       dev_notice(&afu->dev, "PSL purge acknowledging pending non-translation fault, DSISR: 0x%.16llx\n", dsisr);
+                       cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
+               } else {
+                       cpu_relax();
+               }
+               PSL_CNTL = cxl_p1n_read(afu, CXL_PSL_SCNTL_An);
+       };
+       end = local_clock();
+       pr_devel("PSL purged in %lld ns\n", end - start);
+
+       cxl_p1n_write(afu, CXL_PSL_SCNTL_An,
+                      PSL_CNTL & ~CXL_PSL_SCNTL_An_Pc);
+out:
+       trace_cxl_psl_ctrl_done(afu, CXL_PSL_SCNTL_An_Pc, rc);
+       return rc;
+}
+
+static int spa_max_procs(int spa_size)
+{
+       /*
+        * From the CAIA:
+        *    end_of_SPA_area = SPA_Base + ((n+4) * 128) + (( ((n*8) + 127) >> 7) * 128) + 255
+        * Most of that junk is really just an overly-complicated way of saying
+        * the last 256 bytes are __aligned(128), so it's really:
+        *    end_of_SPA_area = end_of_PSL_queue_area + __aligned(128) 255
+        * and
+        *    end_of_PSL_queue_area = SPA_Base + ((n+4) * 128) + (n*8) - 1
+        * so
+        *    sizeof(SPA) = ((n+4) * 128) + (n*8) + __aligned(128) 256
+        * Ignore the alignment (which is safe in this case as long as we are
+        * careful with our rounding) and solve for n:
+        */
+       return ((spa_size / 8) - 96) / 17;
+}
+
+static int alloc_spa(struct cxl_afu *afu)
+{
+       u64 spap;
+
+       /* Work out how many pages to allocate */
+       afu->spa_order = 0;
+       do {
+               afu->spa_order++;
+               afu->spa_size = (1 << afu->spa_order) * PAGE_SIZE;
+               afu->spa_max_procs = spa_max_procs(afu->spa_size);
+       } while (afu->spa_max_procs < afu->num_procs);
+
+       WARN_ON(afu->spa_size > 0x100000); /* Max size supported by the hardware */
+
+       if (!(afu->spa = (struct cxl_process_element *)
+             __get_free_pages(GFP_KERNEL | __GFP_ZERO, afu->spa_order))) {
+               pr_err("cxl_alloc_spa: Unable to allocate scheduled process area\n");
+               return -ENOMEM;
+       }
+       pr_devel("spa pages: %i afu->spa_max_procs: %i   afu->num_procs: %i\n",
+                1<<afu->spa_order, afu->spa_max_procs, afu->num_procs);
+
+       afu->sw_command_status = (__be64 *)((char *)afu->spa +
+                                           ((afu->spa_max_procs + 3) * 128));
+
+       spap = virt_to_phys(afu->spa) & CXL_PSL_SPAP_Addr;
+       spap |= ((afu->spa_size >> (12 - CXL_PSL_SPAP_Size_Shift)) - 1) & CXL_PSL_SPAP_Size;
+       spap |= CXL_PSL_SPAP_V;
+       pr_devel("cxl: SPA allocated at 0x%p. Max processes: %i, sw_command_status: 0x%p CXL_PSL_SPAP_An=0x%016llx\n", afu->spa, afu->spa_max_procs, afu->sw_command_status, spap);
+       cxl_p1n_write(afu, CXL_PSL_SPAP_An, spap);
+
+       return 0;
+}
+
+static void release_spa(struct cxl_afu *afu)
+{
+       cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0);
+       free_pages((unsigned long) afu->spa, afu->spa_order);
+}
+
+int cxl_tlb_slb_invalidate(struct cxl *adapter)
+{
+       unsigned long timeout = jiffies + (HZ * CXL_TIMEOUT);
+
+       pr_devel("CXL adapter wide TLBIA & SLBIA\n");
+
+       cxl_p1_write(adapter, CXL_PSL_AFUSEL, CXL_PSL_AFUSEL_A);
+
+       cxl_p1_write(adapter, CXL_PSL_TLBIA, CXL_TLB_SLB_IQ_ALL);
+       while (cxl_p1_read(adapter, CXL_PSL_TLBIA) & CXL_TLB_SLB_P) {
+               if (time_after_eq(jiffies, timeout)) {
+                       dev_warn(&adapter->dev, "WARNING: CXL adapter wide TLBIA timed out!\n");
+                       return -EBUSY;
+               }
+               cpu_relax();
+       }
+
+       cxl_p1_write(adapter, CXL_PSL_SLBIA, CXL_TLB_SLB_IQ_ALL);
+       while (cxl_p1_read(adapter, CXL_PSL_SLBIA) & CXL_TLB_SLB_P) {
+               if (time_after_eq(jiffies, timeout)) {
+                       dev_warn(&adapter->dev, "WARNING: CXL adapter wide SLBIA timed out!\n");
+                       return -EBUSY;
+               }
+               cpu_relax();
+       }
+       return 0;
+}
+
+int cxl_afu_slbia(struct cxl_afu *afu)
+{
+       unsigned long timeout = jiffies + (HZ * CXL_TIMEOUT);
+
+       pr_devel("cxl_afu_slbia issuing SLBIA command\n");
+       cxl_p2n_write(afu, CXL_SLBIA_An, CXL_TLB_SLB_IQ_ALL);
+       while (cxl_p2n_read(afu, CXL_SLBIA_An) & CXL_TLB_SLB_P) {
+               if (time_after_eq(jiffies, timeout)) {
+                       dev_warn(&afu->dev, "WARNING: CXL AFU SLBIA timed out!\n");
+                       return -EBUSY;
+               }
+               cpu_relax();
+       }
+       return 0;
+}
+
+static int cxl_write_sstp(struct cxl_afu *afu, u64 sstp0, u64 sstp1)
+{
+       int rc;
+
+       /* 1. Disable SSTP by writing 0 to SSTP1[V] */
+       cxl_p2n_write(afu, CXL_SSTP1_An, 0);
+
+       /* 2. Invalidate all SLB entries */
+       if ((rc = cxl_afu_slbia(afu)))
+               return rc;
+
+       /* 3. Set SSTP0_An */
+       cxl_p2n_write(afu, CXL_SSTP0_An, sstp0);
+
+       /* 4. Set SSTP1_An */
+       cxl_p2n_write(afu, CXL_SSTP1_An, sstp1);
+
+       return 0;
+}
+
+/* Using per slice version may improve performance here. (ie. SLBIA_An) */
+static void slb_invalid(struct cxl_context *ctx)
+{
+       struct cxl *adapter = ctx->afu->adapter;
+       u64 slbia;
+
+       WARN_ON(!mutex_is_locked(&ctx->afu->spa_mutex));
+
+       cxl_p1_write(adapter, CXL_PSL_LBISEL,
+                       ((u64)be32_to_cpu(ctx->elem->common.pid) << 32) |
+                       be32_to_cpu(ctx->elem->lpid));
+       cxl_p1_write(adapter, CXL_PSL_SLBIA, CXL_TLB_SLB_IQ_LPIDPID);
+
+       while (1) {
+               slbia = cxl_p1_read(adapter, CXL_PSL_SLBIA);
+               if (!(slbia & CXL_TLB_SLB_P))
+                       break;
+               cpu_relax();
+       }
+}
+
+static int do_process_element_cmd(struct cxl_context *ctx,
+                                 u64 cmd, u64 pe_state)
+{
+       u64 state;
+       unsigned long timeout = jiffies + (HZ * CXL_TIMEOUT);
+       int rc = 0;
+
+       trace_cxl_llcmd(ctx, cmd);
+
+       WARN_ON(!ctx->afu->enabled);
+
+       ctx->elem->software_state = cpu_to_be32(pe_state);
+       smp_wmb();
+       *(ctx->afu->sw_command_status) = cpu_to_be64(cmd | 0 | ctx->pe);
+       smp_mb();
+       cxl_p1n_write(ctx->afu, CXL_PSL_LLCMD_An, cmd | ctx->pe);
+       while (1) {
+               if (time_after_eq(jiffies, timeout)) {
+                       dev_warn(&ctx->afu->dev, "WARNING: Process Element Command timed out!\n");
+                       rc = -EBUSY;
+                       goto out;
+               }
+               state = be64_to_cpup(ctx->afu->sw_command_status);
+               if (state == ~0ULL) {
+                       pr_err("cxl: Error adding process element to AFU\n");
+                       rc = -1;
+                       goto out;
+               }
+               if ((state & (CXL_SPA_SW_CMD_MASK | CXL_SPA_SW_STATE_MASK  | CXL_SPA_SW_LINK_MASK)) ==
+                   (cmd | (cmd >> 16) | ctx->pe))
+                       break;
+               /*
+                * The command won't finish in the PSL if there are
+                * outstanding DSIs.  Hence we need to yield here in
+                * case there are outstanding DSIs that we need to
+                * service.  Tuning possiblity: we could wait for a
+                * while before sched
+                */
+               schedule();
+
+       }
+out:
+       trace_cxl_llcmd_done(ctx, cmd, rc);
+       return rc;
+}
+
+static int add_process_element(struct cxl_context *ctx)
+{
+       int rc = 0;
+
+       mutex_lock(&ctx->afu->spa_mutex);
+       pr_devel("%s Adding pe: %i started\n", __func__, ctx->pe);
+       if (!(rc = do_process_element_cmd(ctx, CXL_SPA_SW_CMD_ADD, CXL_PE_SOFTWARE_STATE_V)))
+               ctx->pe_inserted = true;
+       pr_devel("%s Adding pe: %i finished\n", __func__, ctx->pe);
+       mutex_unlock(&ctx->afu->spa_mutex);
+       return rc;
+}
+
+static int terminate_process_element(struct cxl_context *ctx)
+{
+       int rc = 0;
+
+       /* fast path terminate if it's already invalid */
+       if (!(ctx->elem->software_state & cpu_to_be32(CXL_PE_SOFTWARE_STATE_V)))
+               return rc;
+
+       mutex_lock(&ctx->afu->spa_mutex);
+       pr_devel("%s Terminate pe: %i started\n", __func__, ctx->pe);
+       rc = do_process_element_cmd(ctx, CXL_SPA_SW_CMD_TERMINATE,
+                                   CXL_PE_SOFTWARE_STATE_V | CXL_PE_SOFTWARE_STATE_T);
+       ctx->elem->software_state = 0;  /* Remove Valid bit */
+       pr_devel("%s Terminate pe: %i finished\n", __func__, ctx->pe);
+       mutex_unlock(&ctx->afu->spa_mutex);
+       return rc;
+}
+
+static int remove_process_element(struct cxl_context *ctx)
+{
+       int rc = 0;
+
+       mutex_lock(&ctx->afu->spa_mutex);
+       pr_devel("%s Remove pe: %i started\n", __func__, ctx->pe);
+       if (!(rc = do_process_element_cmd(ctx, CXL_SPA_SW_CMD_REMOVE, 0)))
+               ctx->pe_inserted = false;
+       slb_invalid(ctx);
+       pr_devel("%s Remove pe: %i finished\n", __func__, ctx->pe);
+       mutex_unlock(&ctx->afu->spa_mutex);
+
+       return rc;
+}
+
+
+static void assign_psn_space(struct cxl_context *ctx)
+{
+       if (!ctx->afu->pp_size || ctx->master) {
+               ctx->psn_phys = ctx->afu->psn_phys;
+               ctx->psn_size = ctx->afu->adapter->ps_size;
+       } else {
+               ctx->psn_phys = ctx->afu->psn_phys +
+                       (ctx->afu->pp_offset + ctx->afu->pp_size * ctx->pe);
+               ctx->psn_size = ctx->afu->pp_size;
+       }
+}
+
+static int activate_afu_directed(struct cxl_afu *afu)
+{
+       int rc;
+
+       dev_info(&afu->dev, "Activating AFU directed mode\n");
+
+       if (alloc_spa(afu))
+               return -ENOMEM;
+
+       cxl_p1n_write(afu, CXL_PSL_SCNTL_An, CXL_PSL_SCNTL_An_PM_AFU);
+       cxl_p1n_write(afu, CXL_PSL_AMOR_An, 0xFFFFFFFFFFFFFFFFULL);
+       cxl_p1n_write(afu, CXL_PSL_ID_An, CXL_PSL_ID_An_F | CXL_PSL_ID_An_L);
+
+       afu->current_mode = CXL_MODE_DIRECTED;
+       afu->num_procs = afu->max_procs_virtualised;
+
+       if ((rc = cxl_chardev_m_afu_add(afu)))
+               return rc;
+
+       if ((rc = cxl_sysfs_afu_m_add(afu)))
+               goto err;
+
+       if ((rc = cxl_chardev_s_afu_add(afu)))
+               goto err1;
+
+       return 0;
+err1:
+       cxl_sysfs_afu_m_remove(afu);
+err:
+       cxl_chardev_afu_remove(afu);
+       return rc;
+}
+
+#ifdef CONFIG_CPU_LITTLE_ENDIAN
+#define set_endian(sr) ((sr) |= CXL_PSL_SR_An_LE)
+#else
+#define set_endian(sr) ((sr) &= ~(CXL_PSL_SR_An_LE))
+#endif
+
+static int attach_afu_directed(struct cxl_context *ctx, u64 wed, u64 amr)
+{
+       u64 sr;
+       int r, result;
+
+       assign_psn_space(ctx);
+
+       ctx->elem->ctxtime = 0; /* disable */
+       ctx->elem->lpid = cpu_to_be32(mfspr(SPRN_LPID));
+       ctx->elem->haurp = 0; /* disable */
+       ctx->elem->sdr = cpu_to_be64(mfspr(SPRN_SDR1));
+
+       sr = 0;
+       if (ctx->master)
+               sr |= CXL_PSL_SR_An_MP;
+       if (mfspr(SPRN_LPCR) & LPCR_TC)
+               sr |= CXL_PSL_SR_An_TC;
+       /* HV=0, PR=1, R=1 for userspace
+        * For kernel contexts: this would need to change
+        */
+       sr |= CXL_PSL_SR_An_PR | CXL_PSL_SR_An_R;
+       set_endian(sr);
+       sr &= ~(CXL_PSL_SR_An_HV);
+       if (!test_tsk_thread_flag(current, TIF_32BIT))
+               sr |= CXL_PSL_SR_An_SF;
+       ctx->elem->common.pid = cpu_to_be32(current->pid);
+       ctx->elem->common.tid = 0;
+       ctx->elem->sr = cpu_to_be64(sr);
+
+       ctx->elem->common.csrp = 0; /* disable */
+       ctx->elem->common.aurp0 = 0; /* disable */
+       ctx->elem->common.aurp1 = 0; /* disable */
+
+       cxl_prefault(ctx, wed);
+
+       ctx->elem->common.sstp0 = cpu_to_be64(ctx->sstp0);
+       ctx->elem->common.sstp1 = cpu_to_be64(ctx->sstp1);
+
+       for (r = 0; r < CXL_IRQ_RANGES; r++) {
+               ctx->elem->ivte_offsets[r] = cpu_to_be16(ctx->irqs.offset[r]);
+               ctx->elem->ivte_ranges[r] = cpu_to_be16(ctx->irqs.range[r]);
+       }
+
+       ctx->elem->common.amr = cpu_to_be64(amr);
+       ctx->elem->common.wed = cpu_to_be64(wed);
+
+       /* first guy needs to enable */
+       if ((result = afu_check_and_enable(ctx->afu)))
+               return result;
+
+       add_process_element(ctx);
+
+       return 0;
+}
+
+static int deactivate_afu_directed(struct cxl_afu *afu)
+{
+       dev_info(&afu->dev, "Deactivating AFU directed mode\n");
+
+       afu->current_mode = 0;
+       afu->num_procs = 0;
+
+       cxl_sysfs_afu_m_remove(afu);
+       cxl_chardev_afu_remove(afu);
+
+       cxl_afu_reset(afu);
+       cxl_afu_disable(afu);
+       cxl_psl_purge(afu);
+
+       release_spa(afu);
+
+       return 0;
+}
+
+static int activate_dedicated_process(struct cxl_afu *afu)
+{
+       dev_info(&afu->dev, "Activating dedicated process mode\n");
+
+       cxl_p1n_write(afu, CXL_PSL_SCNTL_An, CXL_PSL_SCNTL_An_PM_Process);
+
+       cxl_p1n_write(afu, CXL_PSL_CtxTime_An, 0); /* disable */
+       cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0);    /* disable */
+       cxl_p1n_write(afu, CXL_PSL_AMOR_An, 0xFFFFFFFFFFFFFFFFULL);
+       cxl_p1n_write(afu, CXL_PSL_LPID_An, mfspr(SPRN_LPID));
+       cxl_p1n_write(afu, CXL_HAURP_An, 0);       /* disable */
+       cxl_p1n_write(afu, CXL_PSL_SDR_An, mfspr(SPRN_SDR1));
+
+       cxl_p2n_write(afu, CXL_CSRP_An, 0);        /* disable */
+       cxl_p2n_write(afu, CXL_AURP0_An, 0);       /* disable */
+       cxl_p2n_write(afu, CXL_AURP1_An, 0);       /* disable */
+
+       afu->current_mode = CXL_MODE_DEDICATED;
+       afu->num_procs = 1;
+
+       return cxl_chardev_d_afu_add(afu);
+}
+
+static int attach_dedicated(struct cxl_context *ctx, u64 wed, u64 amr)
+{
+       struct cxl_afu *afu = ctx->afu;
+       u64 sr;
+       int rc;
+
+       sr = 0;
+       set_endian(sr);
+       if (ctx->master)
+               sr |= CXL_PSL_SR_An_MP;
+       if (mfspr(SPRN_LPCR) & LPCR_TC)
+               sr |= CXL_PSL_SR_An_TC;
+       sr |= CXL_PSL_SR_An_PR | CXL_PSL_SR_An_R;
+       if (!test_tsk_thread_flag(current, TIF_32BIT))
+               sr |= CXL_PSL_SR_An_SF;
+       cxl_p2n_write(afu, CXL_PSL_PID_TID_An, (u64)current->pid << 32);
+       cxl_p1n_write(afu, CXL_PSL_SR_An, sr);
+
+       if ((rc = cxl_write_sstp(afu, ctx->sstp0, ctx->sstp1)))
+               return rc;
+
+       cxl_prefault(ctx, wed);
+
+       cxl_p1n_write(afu, CXL_PSL_IVTE_Offset_An,
+                      (((u64)ctx->irqs.offset[0] & 0xffff) << 48) |
+                      (((u64)ctx->irqs.offset[1] & 0xffff) << 32) |
+                      (((u64)ctx->irqs.offset[2] & 0xffff) << 16) |
+                       ((u64)ctx->irqs.offset[3] & 0xffff));
+       cxl_p1n_write(afu, CXL_PSL_IVTE_Limit_An, (u64)
+                      (((u64)ctx->irqs.range[0] & 0xffff) << 48) |
+                      (((u64)ctx->irqs.range[1] & 0xffff) << 32) |
+                      (((u64)ctx->irqs.range[2] & 0xffff) << 16) |
+                       ((u64)ctx->irqs.range[3] & 0xffff));
+
+       cxl_p2n_write(afu, CXL_PSL_AMR_An, amr);
+
+       /* master only context for dedicated */
+       assign_psn_space(ctx);
+
+       if ((rc = cxl_afu_reset(afu)))
+               return rc;
+
+       cxl_p2n_write(afu, CXL_PSL_WED_An, wed);
+
+       return afu_enable(afu);
+}
+
+static int deactivate_dedicated_process(struct cxl_afu *afu)
+{
+       dev_info(&afu->dev, "Deactivating dedicated process mode\n");
+
+       afu->current_mode = 0;
+       afu->num_procs = 0;
+
+       cxl_chardev_afu_remove(afu);
+
+       return 0;
+}
+
+int _cxl_afu_deactivate_mode(struct cxl_afu *afu, int mode)
+{
+       if (mode == CXL_MODE_DIRECTED)
+               return deactivate_afu_directed(afu);
+       if (mode == CXL_MODE_DEDICATED)
+               return deactivate_dedicated_process(afu);
+       return 0;
+}
+
+int cxl_afu_deactivate_mode(struct cxl_afu *afu)
+{
+       return _cxl_afu_deactivate_mode(afu, afu->current_mode);
+}
+
+int cxl_afu_activate_mode(struct cxl_afu *afu, int mode)
+{
+       if (!mode)
+               return 0;
+       if (!(mode & afu->modes_supported))
+               return -EINVAL;
+
+       if (mode == CXL_MODE_DIRECTED)
+               return activate_afu_directed(afu);
+       if (mode == CXL_MODE_DEDICATED)
+               return activate_dedicated_process(afu);
+
+       return -EINVAL;
+}
+
+int cxl_attach_process(struct cxl_context *ctx, bool kernel, u64 wed, u64 amr)
+{
+       ctx->kernel = kernel;
+       if (ctx->afu->current_mode == CXL_MODE_DIRECTED)
+               return attach_afu_directed(ctx, wed, amr);
+
+       if (ctx->afu->current_mode == CXL_MODE_DEDICATED)
+               return attach_dedicated(ctx, wed, amr);
+
+       return -EINVAL;
+}
+
+static inline int detach_process_native_dedicated(struct cxl_context *ctx)
+{
+       cxl_afu_reset(ctx->afu);
+       cxl_afu_disable(ctx->afu);
+       cxl_psl_purge(ctx->afu);
+       return 0;
+}
+
+static inline int detach_process_native_afu_directed(struct cxl_context *ctx)
+{
+       if (!ctx->pe_inserted)
+               return 0;
+       if (terminate_process_element(ctx))
+               return -1;
+       if (remove_process_element(ctx))
+               return -1;
+
+       return 0;
+}
+
+int cxl_detach_process(struct cxl_context *ctx)
+{
+       trace_cxl_detach(ctx);
+
+       if (ctx->afu->current_mode == CXL_MODE_DEDICATED)
+               return detach_process_native_dedicated(ctx);
+
+       return detach_process_native_afu_directed(ctx);
+}
+
+int cxl_get_irq(struct cxl_afu *afu, struct cxl_irq_info *info)
+{
+       u64 pidtid;
+
+       info->dsisr = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
+       info->dar = cxl_p2n_read(afu, CXL_PSL_DAR_An);
+       info->dsr = cxl_p2n_read(afu, CXL_PSL_DSR_An);
+       pidtid = cxl_p2n_read(afu, CXL_PSL_PID_TID_An);
+       info->pid = pidtid >> 32;
+       info->tid = pidtid & 0xffffffff;
+       info->afu_err = cxl_p2n_read(afu, CXL_AFU_ERR_An);
+       info->errstat = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
+
+       return 0;
+}
+
+static void recover_psl_err(struct cxl_afu *afu, u64 errstat)
+{
+       u64 dsisr;
+
+       pr_devel("RECOVERING FROM PSL ERROR... (0x%.16llx)\n", errstat);
+
+       /* Clear PSL_DSISR[PE] */
+       dsisr = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
+       cxl_p2n_write(afu, CXL_PSL_DSISR_An, dsisr & ~CXL_PSL_DSISR_An_PE);
+
+       /* Write 1s to clear error status bits */
+       cxl_p2n_write(afu, CXL_PSL_ErrStat_An, errstat);
+}
+
+int cxl_ack_irq(struct cxl_context *ctx, u64 tfc, u64 psl_reset_mask)
+{
+       trace_cxl_psl_irq_ack(ctx, tfc);
+       if (tfc)
+               cxl_p2n_write(ctx->afu, CXL_PSL_TFC_An, tfc);
+       if (psl_reset_mask)
+               recover_psl_err(ctx->afu, psl_reset_mask);
+
+       return 0;
+}
+
+int cxl_check_error(struct cxl_afu *afu)
+{
+       return (cxl_p1n_read(afu, CXL_PSL_SCNTL_An) == ~0ULL);
+}