]> git.ipfire.org Git - thirdparty/kernel/linux.git/commitdiff
Merge branch 'memory-observability' into x86/amd
authorJoerg Roedel <jroedel@suse.de>
Fri, 26 Apr 2024 10:54:13 +0000 (12:54 +0200)
committerJoerg Roedel <jroedel@suse.de>
Fri, 26 Apr 2024 10:54:13 +0000 (12:54 +0200)
1  2 
drivers/iommu/amd/amd_iommu.h
drivers/iommu/amd/init.c
drivers/iommu/amd/iommu.c
drivers/iommu/amd/ppr.c

Simple merge
index 7f9b5c710058ee1a1b6270a7391729ccea9f38a3,40b3b9cffade3c467ac34d6067a92777dd1d792c..fd3e76e43699233cded76be816810f09cb9b28bf
@@@ -834,14 -844,14 +833,14 @@@ static void iommu_disable_command_buffe
  
  static void __init free_command_buffer(struct amd_iommu *iommu)
  {
-       free_pages((unsigned long)iommu->cmd_buf, get_order(CMD_BUFFER_SIZE));
+       iommu_free_pages(iommu->cmd_buf, get_order(CMD_BUFFER_SIZE));
  }
  
 -static void *__init iommu_alloc_4k_pages(struct amd_iommu *iommu,
 -                                       gfp_t gfp, size_t size)
 +void *__init iommu_alloc_4k_pages(struct amd_iommu *iommu, gfp_t gfp,
 +                                size_t size)
  {
        int order = get_order(size);
-       void *buf = (void *)__get_free_pages(gfp, order);
+       void *buf = iommu_alloc_pages(gfp, order);
  
        if (buf &&
            check_feature(FEATURE_SNP) &&
@@@ -890,9 -900,44 +889,9 @@@ static void iommu_disable_event_buffer(
  
  static void __init free_event_buffer(struct amd_iommu *iommu)
  {
-       free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
+       iommu_free_pages(iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
  }
  
 -/* allocates the memory where the IOMMU will log its events to */
 -static int __init alloc_ppr_log(struct amd_iommu *iommu)
 -{
 -      iommu->ppr_log = iommu_alloc_4k_pages(iommu, GFP_KERNEL, PPR_LOG_SIZE);
 -
 -      return iommu->ppr_log ? 0 : -ENOMEM;
 -}
 -
 -static void iommu_enable_ppr_log(struct amd_iommu *iommu)
 -{
 -      u64 entry;
 -
 -      if (iommu->ppr_log == NULL)
 -              return;
 -
 -      iommu_feature_enable(iommu, CONTROL_PPR_EN);
 -
 -      entry = iommu_virt_to_phys(iommu->ppr_log) | PPR_LOG_SIZE_512;
 -
 -      memcpy_toio(iommu->mmio_base + MMIO_PPR_LOG_OFFSET,
 -                  &entry, sizeof(entry));
 -
 -      /* set head and tail to zero manually */
 -      writel(0x00, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET);
 -      writel(0x00, iommu->mmio_base + MMIO_PPR_TAIL_OFFSET);
 -
 -      iommu_feature_enable(iommu, CONTROL_PPRLOG_EN);
 -      iommu_feature_enable(iommu, CONTROL_PPRINT_EN);
 -}
 -
 -static void __init free_ppr_log(struct amd_iommu *iommu)
 -{
 -      iommu_free_pages(iommu->ppr_log, get_order(PPR_LOG_SIZE));
 -}
 -
  static void free_ga_log(struct amd_iommu *iommu)
  {
  #ifdef CONFIG_IRQ_REMAP
Simple merge
index 0463daa2d46b82761eb46c8e0e85358936e015be,0000000000000000000000000000000000000000..091423bb8aac88d2f6030bb88ac96e5084d11414
mode 100644,000000..100644
--- /dev/null
@@@ -1,286 -1,0 +1,288 @@@
-       free_pages((unsigned long)iommu->ppr_log, get_order(PPR_LOG_SIZE));
 +// SPDX-License-Identifier: GPL-2.0-only
 +/*
 + * Copyright (C) 2023 Advanced Micro Devices, Inc.
 + */
 +
 +#define pr_fmt(fmt)     "AMD-Vi: " fmt
 +#define dev_fmt(fmt)    pr_fmt(fmt)
 +
 +#include <linux/amd-iommu.h>
 +#include <linux/delay.h>
 +#include <linux/mmu_notifier.h>
 +
 +#include <asm/iommu.h>
 +
 +#include "amd_iommu.h"
 +#include "amd_iommu_types.h"
 +
++#include "../iommu-pages.h"
++
 +int __init amd_iommu_alloc_ppr_log(struct amd_iommu *iommu)
 +{
 +      iommu->ppr_log = iommu_alloc_4k_pages(iommu, GFP_KERNEL | __GFP_ZERO,
 +                                            PPR_LOG_SIZE);
 +      return iommu->ppr_log ? 0 : -ENOMEM;
 +}
 +
 +void amd_iommu_enable_ppr_log(struct amd_iommu *iommu)
 +{
 +      u64 entry;
 +
 +      if (iommu->ppr_log == NULL)
 +              return;
 +
 +      iommu_feature_enable(iommu, CONTROL_PPR_EN);
 +
 +      entry = iommu_virt_to_phys(iommu->ppr_log) | PPR_LOG_SIZE_512;
 +
 +      memcpy_toio(iommu->mmio_base + MMIO_PPR_LOG_OFFSET,
 +                  &entry, sizeof(entry));
 +
 +      /* set head and tail to zero manually */
 +      writel(0x00, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET);
 +      writel(0x00, iommu->mmio_base + MMIO_PPR_TAIL_OFFSET);
 +
 +      iommu_feature_enable(iommu, CONTROL_PPRINT_EN);
 +      iommu_feature_enable(iommu, CONTROL_PPRLOG_EN);
 +}
 +
 +void __init amd_iommu_free_ppr_log(struct amd_iommu *iommu)
 +{
++      iommu_free_pages(iommu->ppr_log, get_order(PPR_LOG_SIZE));
 +}
 +
 +/*
 + * This function restarts ppr logging in case the IOMMU experienced
 + * PPR log overflow.
 + */
 +void amd_iommu_restart_ppr_log(struct amd_iommu *iommu)
 +{
 +      amd_iommu_restart_log(iommu, "PPR", CONTROL_PPRINT_EN,
 +                            CONTROL_PPRLOG_EN, MMIO_STATUS_PPR_RUN_MASK,
 +                            MMIO_STATUS_PPR_OVERFLOW_MASK);
 +}
 +
 +static inline u32 ppr_flag_to_fault_perm(u16 flag)
 +{
 +      int perm = 0;
 +
 +      if (flag & PPR_FLAG_READ)
 +              perm |= IOMMU_FAULT_PERM_READ;
 +      if (flag & PPR_FLAG_WRITE)
 +              perm |= IOMMU_FAULT_PERM_WRITE;
 +      if (flag & PPR_FLAG_EXEC)
 +              perm |= IOMMU_FAULT_PERM_EXEC;
 +      if (!(flag & PPR_FLAG_US))
 +              perm |= IOMMU_FAULT_PERM_PRIV;
 +
 +      return perm;
 +}
 +
 +static bool ppr_is_valid(struct amd_iommu *iommu, u64 *raw)
 +{
 +      struct device *dev = iommu->iommu.dev;
 +      u16 devid = PPR_DEVID(raw[0]);
 +
 +      if (!(PPR_FLAGS(raw[0]) & PPR_FLAG_GN)) {
 +              dev_dbg(dev, "PPR logged [Request ignored due to GN=0 (device=%04x:%02x:%02x.%x "
 +                      "pasid=0x%05llx address=0x%llx flags=0x%04llx tag=0x%03llx]\n",
 +                      iommu->pci_seg->id, PCI_BUS_NUM(devid), PCI_SLOT(devid), PCI_FUNC(devid),
 +                      PPR_PASID(raw[0]), raw[1], PPR_FLAGS(raw[0]), PPR_TAG(raw[0]));
 +              return false;
 +      }
 +
 +      if (PPR_FLAGS(raw[0]) & PPR_FLAG_RVSD) {
 +              dev_dbg(dev, "PPR logged [Invalid request format (device=%04x:%02x:%02x.%x "
 +                      "pasid=0x%05llx address=0x%llx flags=0x%04llx tag=0x%03llx]\n",
 +                      iommu->pci_seg->id, PCI_BUS_NUM(devid), PCI_SLOT(devid), PCI_FUNC(devid),
 +                      PPR_PASID(raw[0]), raw[1], PPR_FLAGS(raw[0]), PPR_TAG(raw[0]));
 +              return false;
 +      }
 +
 +      return true;
 +}
 +
 +static void iommu_call_iopf_notifier(struct amd_iommu *iommu, u64 *raw)
 +{
 +      struct iommu_dev_data *dev_data;
 +      struct iopf_fault event;
 +      struct pci_dev *pdev;
 +      u16 devid = PPR_DEVID(raw[0]);
 +
 +      if (PPR_REQ_TYPE(raw[0]) != PPR_REQ_FAULT) {
 +              pr_info_ratelimited("Unknown PPR request received\n");
 +              return;
 +      }
 +
 +      pdev = pci_get_domain_bus_and_slot(iommu->pci_seg->id,
 +                                         PCI_BUS_NUM(devid), devid & 0xff);
 +      if (!pdev)
 +              return;
 +
 +      if (!ppr_is_valid(iommu, raw))
 +              goto out;
 +
 +      memset(&event, 0, sizeof(struct iopf_fault));
 +
 +      event.fault.type = IOMMU_FAULT_PAGE_REQ;
 +      event.fault.prm.perm = ppr_flag_to_fault_perm(PPR_FLAGS(raw[0]));
 +      event.fault.prm.addr = (u64)(raw[1] & PAGE_MASK);
 +      event.fault.prm.pasid = PPR_PASID(raw[0]);
 +      event.fault.prm.grpid = PPR_TAG(raw[0]) & 0x1FF;
 +
 +      /*
 +       * PASID zero is used for requests from the I/O device without
 +       * a PASID
 +       */
 +      dev_data = dev_iommu_priv_get(&pdev->dev);
 +      if (event.fault.prm.pasid == 0 ||
 +          event.fault.prm.pasid >= dev_data->max_pasids) {
 +              pr_info_ratelimited("Invalid PASID : 0x%x, device : 0x%x\n",
 +                                  event.fault.prm.pasid, pdev->dev.id);
 +              goto out;
 +      }
 +
 +      event.fault.prm.flags |= IOMMU_FAULT_PAGE_RESPONSE_NEEDS_PASID;
 +      event.fault.prm.flags |= IOMMU_FAULT_PAGE_REQUEST_PASID_VALID;
 +      if (PPR_TAG(raw[0]) & 0x200)
 +              event.fault.prm.flags |= IOMMU_FAULT_PAGE_REQUEST_LAST_PAGE;
 +
 +      /* Submit event */
 +      iommu_report_device_fault(&pdev->dev, &event);
 +
 +      return;
 +
 +out:
 +      /* Nobody cared, abort */
 +      amd_iommu_complete_ppr(&pdev->dev, PPR_PASID(raw[0]),
 +                             IOMMU_PAGE_RESP_FAILURE,
 +                             PPR_TAG(raw[0]) & 0x1FF);
 +}
 +
 +void amd_iommu_poll_ppr_log(struct amd_iommu *iommu)
 +{
 +      u32 head, tail;
 +
 +      if (iommu->ppr_log == NULL)
 +              return;
 +
 +      head = readl(iommu->mmio_base + MMIO_PPR_HEAD_OFFSET);
 +      tail = readl(iommu->mmio_base + MMIO_PPR_TAIL_OFFSET);
 +
 +      while (head != tail) {
 +              volatile u64 *raw;
 +              u64 entry[2];
 +              int i;
 +
 +              raw = (u64 *)(iommu->ppr_log + head);
 +
 +              /*
 +               * Hardware bug: Interrupt may arrive before the entry is
 +               * written to memory. If this happens we need to wait for the
 +               * entry to arrive.
 +               */
 +              for (i = 0; i < LOOP_TIMEOUT; ++i) {
 +                      if (PPR_REQ_TYPE(raw[0]) != 0)
 +                              break;
 +                      udelay(1);
 +              }
 +
 +              /* Avoid memcpy function-call overhead */
 +              entry[0] = raw[0];
 +              entry[1] = raw[1];
 +
 +              /*
 +               * To detect the hardware errata 733 we need to clear the
 +               * entry back to zero. This issue does not exist on SNP
 +               * enabled system. Also this buffer is not writeable on
 +               * SNP enabled system.
 +               */
 +              if (!amd_iommu_snp_en)
 +                      raw[0] = raw[1] = 0UL;
 +
 +              /* Update head pointer of hardware ring-buffer */
 +              head = (head + PPR_ENTRY_SIZE) % PPR_LOG_SIZE;
 +              writel(head, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET);
 +
 +              /* Handle PPR entry */
 +              iommu_call_iopf_notifier(iommu, entry);
 +      }
 +}
 +
 +/**************************************************************
 + *
 + * IOPF handling stuff
 + */
 +
 +/* Setup per-IOMMU IOPF queue if not exist. */
 +int amd_iommu_iopf_init(struct amd_iommu *iommu)
 +{
 +      int ret = 0;
 +
 +      if (iommu->iopf_queue)
 +              return ret;
 +
 +      snprintf(iommu->iopfq_name, sizeof(iommu->iopfq_name),
 +               "amdiommu-%#x-iopfq",
 +               PCI_SEG_DEVID_TO_SBDF(iommu->pci_seg->id, iommu->devid));
 +
 +      iommu->iopf_queue = iopf_queue_alloc(iommu->iopfq_name);
 +      if (!iommu->iopf_queue)
 +              ret = -ENOMEM;
 +
 +      return ret;
 +}
 +
 +/* Destroy per-IOMMU IOPF queue if no longer needed. */
 +void amd_iommu_iopf_uninit(struct amd_iommu *iommu)
 +{
 +      iopf_queue_free(iommu->iopf_queue);
 +      iommu->iopf_queue = NULL;
 +}
 +
 +void amd_iommu_page_response(struct device *dev, struct iopf_fault *evt,
 +                           struct iommu_page_response *resp)
 +{
 +      amd_iommu_complete_ppr(dev, resp->pasid, resp->code, resp->grpid);
 +}
 +
 +int amd_iommu_iopf_add_device(struct amd_iommu *iommu,
 +                            struct iommu_dev_data *dev_data)
 +{
 +      unsigned long flags;
 +      int ret = 0;
 +
 +      if (!dev_data->pri_enabled)
 +              return ret;
 +
 +      raw_spin_lock_irqsave(&iommu->lock, flags);
 +
 +      if (!iommu->iopf_queue) {
 +              ret = -EINVAL;
 +              goto out_unlock;
 +      }
 +
 +      ret = iopf_queue_add_device(iommu->iopf_queue, dev_data->dev);
 +      if (ret)
 +              goto out_unlock;
 +
 +      dev_data->ppr = true;
 +
 +out_unlock:
 +      raw_spin_unlock_irqrestore(&iommu->lock, flags);
 +      return ret;
 +}
 +
 +/* Its assumed that caller has verified that device was added to iopf queue */
 +void amd_iommu_iopf_remove_device(struct amd_iommu *iommu,
 +                                struct iommu_dev_data *dev_data)
 +{
 +      unsigned long flags;
 +
 +      raw_spin_lock_irqsave(&iommu->lock, flags);
 +
 +      iopf_queue_remove_device(iommu->iopf_queue, dev_data->dev);
 +      dev_data->ppr = false;
 +
 +      raw_spin_unlock_irqrestore(&iommu->lock, flags);
 +}