Re: [QEMU][PATCHv2 7/8] xen: add map and unmap callbacks for grant region

[Vikram Garhwal <vikram.garhwal@xxxxxxx>]

On Wed, Oct 25, 2023 at 06:32:26PM -0700, Stefano Stabellini wrote:
> On Wed, 25 Oct 2023, Vikram Garhwal wrote:
> > From: Juergen Gross <jgross@xxxxxxxx>
> > 
> > Add the callbacks for mapping/unmapping guest memory via grants to the
> > special grant memory region.
> > 
> > Signed-off-by: Juergen Gross <jgross@xxxxxxxx>
> > Signed-off-by: Vikram Garhwal <vikram.garhwal@xxxxxxx>
> > ---
> >  hw/xen/xen-mapcache.c | 175 +++++++++++++++++++++++++++++++++++++++++-
> >  system/physmem.c      |  11 ++-
> >  2 files changed, 181 insertions(+), 5 deletions(-)
> > 
> > diff --git a/hw/xen/xen-mapcache.c b/hw/xen/xen-mapcache.c
> > index 8a61c7dde6..feb4a3b886 100644
> > --- a/hw/xen/xen-mapcache.c
> > +++ b/hw/xen/xen-mapcache.c
> > @@ -9,6 +9,8 @@
> >   */
> >  
> >  #include "qemu/osdep.h"
> > +#include "qemu/queue.h"
> > +#include "qemu/thread.h"
> >  #include "qemu/units.h"
> >  #include "qemu/error-report.h"
> >  
> > @@ -23,6 +25,8 @@
> >  #include "sysemu/xen-mapcache.h"
> >  #include "trace.h"
> >  
> > +#include <xenevtchn.h>
> > +#include <xengnttab.h>
> >  
> >  //#define MAPCACHE_DEBUG
> >  
> > @@ -385,7 +389,7 @@ uint8_t *xen_map_cache(hwaddr phys_addr, hwaddr size,
> >      return p;
> >  }
> >  
> > -ram_addr_t xen_ram_addr_from_mapcache(void *ptr)
> > +static ram_addr_t xen_ram_addr_from_mapcache_try(void *ptr)
> >  {
> >      MapCacheEntry *entry = NULL;
> >      MapCacheRev *reventry;
> > @@ -594,10 +598,178 @@ uint8_t *xen_replace_cache_entry(hwaddr 
> > old_phys_addr,
> >      return p;
> >  }
> >  
> > +struct XENMappedGrantRegion {
> > +    void *addr;
> > +    unsigned int pages;
> > +    unsigned int refs;
> > +    unsigned int prot;
> > +    uint32_t idx;
> > +    QLIST_ENTRY(XENMappedGrantRegion) list;
> > +};
> > +
> > +static xengnttab_handle *xen_region_gnttabdev;
> > +static QLIST_HEAD(GrantRegionList, XENMappedGrantRegion) 
> > xen_grant_mappings =
> > +    QLIST_HEAD_INITIALIZER(xen_grant_mappings);
> > +static QemuMutex xen_map_mutex;
> > +
> > +static void *xen_map_grant_dyn(MemoryRegion **mr, hwaddr addr, hwaddr 
> > *plen,
> > +                               bool is_write, MemTxAttrs attrs)
> > +{
> > +    unsigned int page_off = addr & (XC_PAGE_SIZE - 1);
> > +    unsigned int i;
> > +    unsigned int total_grants = 0;
> > +    unsigned int nrefs = (page_off + *plen + XC_PAGE_SIZE - 1) >> 
> > XC_PAGE_SHIFT;
> > +    uint32_t ref = (addr - XEN_GRANT_ADDR_OFF) >> XC_PAGE_SHIFT;
> > +    uint32_t *refs = NULL;
> > +    unsigned int prot = PROT_READ;
> > +    struct XENMappedGrantRegion *mgr = NULL;
> > +
> > +    if (is_write) {
> > +        prot |= PROT_WRITE;
> > +    }
> > +
> > +    qemu_mutex_lock(&xen_map_mutex);
> > +
> > +    QLIST_FOREACH(mgr, &xen_grant_mappings, list) {
> > +        if (mgr->idx == ref &&
> > +            mgr->pages == nrefs &&
> > +            (mgr->prot & prot) == prot) {
> > +            break;
> > +        }
> > +
> > +        total_grants += mgr->pages;
> > +    }
> > +
> > +    if (!mgr) {
> > +        if (nrefs + total_grants >= XEN_MAX_VIRTIO_GRANTS) {
> > +            return NULL;
> 
> missing qemu_mutex_unlock
Oops, thanks for catching this! Will fix it in v3.
> 
> 
> > +        }
> > +
> > +        mgr = g_new(struct XENMappedGrantRegion, 1);
> > +
> > +        if (nrefs == 1) {
> > +            refs = &ref;
> > +        } else {
> > +            refs = g_new(uint32_t, nrefs);
> > +            for (i = 0; i < nrefs; i++) {
> > +                refs[i] = ref + i;
> > +            }
> > +        }
> > +        mgr->addr = xengnttab_map_domain_grant_refs(xen_region_gnttabdev, 
> > nrefs,
> > +                                                    xen_domid, refs, prot);
> > +        if (mgr->addr) {
> > +            mgr->pages = nrefs;
> > +            mgr->refs = 1;
> > +            mgr->prot = prot;
> > +            mgr->idx = ref;
> > +
> > +            QLIST_INSERT_HEAD(&xen_grant_mappings, mgr, list);
> > +        } else {
> > +            g_free(mgr);
> > +            mgr = NULL;
> > +        }
> > +    } else {
> > +        mgr->refs++;
> > +    }
> > +
> > +    qemu_mutex_unlock(&xen_map_mutex);
> > +
> > +    if (nrefs > 1) {
> > +        g_free(refs);
> > +    }
> > +
> > +    return mgr ? mgr->addr + page_off : NULL;
> > +}
> > +
> > +static void xen_unmap_grant_dyn(MemoryRegion *mr, void *buffer, ram_addr_t 
> > addr,
> > +                                hwaddr len, bool is_write, hwaddr 
> > access_len)
> > +{
> > +    unsigned int page_off = (unsigned long)buffer & (XC_PAGE_SIZE - 1);
> > +    unsigned int nrefs = (page_off + len + XC_PAGE_SIZE - 1) >> 
> > XC_PAGE_SHIFT;
> > +    unsigned int prot = PROT_READ;
> > +    struct XENMappedGrantRegion *mgr = NULL;
> > +
> > +    if (is_write) {
> > +        prot |= PROT_WRITE;
> > +    }
> > +
> > +    qemu_mutex_lock(&xen_map_mutex);
> > +
> > +    QLIST_FOREACH(mgr, &xen_grant_mappings, list) {
> > +        if (mgr->addr == buffer - page_off &&
> > +            mgr->pages == nrefs &&
> > +            (mgr->prot & prot) == prot) {
> > +            break;
> > +        }
> > +    }
> > +    if (mgr) {
> > +        mgr->refs--;
> > +        if (!mgr->refs) {
> > +            xengnttab_unmap(xen_region_gnttabdev, mgr->addr, nrefs);
> > +
> > +            QLIST_REMOVE(mgr, list);
> > +            g_free(mgr);
> > +        }
> > +    } else {
> > +        error_report("xen_unmap_grant_dyn() trying to unmap unknown 
> > buffer");
> > +    }
> > +
> > +    qemu_mutex_unlock(&xen_map_mutex);
> > +}
> > +
> > +static ram_addr_t xen_ram_addr_from_grant_cache(void *ptr)
> > +{
> > +    unsigned int page_off = (unsigned long)ptr & (XC_PAGE_SIZE - 1);
> > +    struct XENMappedGrantRegion *mgr = NULL;
> > +    ram_addr_t raddr = RAM_ADDR_INVALID;
> > +
> > +    qemu_mutex_lock(&xen_map_mutex);
> > +
> > +    QLIST_FOREACH(mgr, &xen_grant_mappings, list) {
> > +        if (mgr->addr == ptr - page_off) {
> > +            break;
> > +        }
> > +    }
> > +
> > +    if (mgr) {
> > +        raddr = (mgr->idx << XC_PAGE_SHIFT) + page_off + 
> > XEN_GRANT_ADDR_OFF;
> > +    }
> > +
> > +    qemu_mutex_unlock(&xen_map_mutex);
> > +
> > +    return raddr;
> > +}
> > +
> > +ram_addr_t xen_ram_addr_from_mapcache(void *ptr)
> > +{
> > +    ram_addr_t raddr;
> > +
> > +    raddr = xen_ram_addr_from_mapcache_try(ptr);
> > +    if (raddr == RAM_ADDR_INVALID) {
> > +        raddr = xen_ram_addr_from_grant_cache(ptr);
> > +    }
> > +
> > +    return raddr;
> > +}
> > +
> > +static const struct MemoryRegionOps xen_grant_mr_ops = {
> > +    .map = xen_map_grant_dyn,
> > +    .unmap = xen_unmap_grant_dyn,
> > +    .endianness = DEVICE_LITTLE_ENDIAN,
> > +};
> > +
> >  MemoryRegion *xen_init_grant_ram(void)
> >  {
> >      RAMBlock *block;
> >  
> > +    qemu_mutex_init(&xen_map_mutex);
> > +
> > +    xen_region_gnttabdev = xengnttab_open(NULL, 0);
> > +    if (xen_region_gnttabdev == NULL) {
> > +        fprintf(stderr, "can't open gnttab device\n");
> > +        return NULL;
> > +    }
> > +
> >      memory_region_init(&ram_grants, NULL, "xen.grants",
> >                         XEN_MAX_VIRTIO_GRANTS * XC_PAGE_SIZE);
> >      block = g_malloc0(sizeof(*block));
> > @@ -612,6 +784,7 @@ MemoryRegion *xen_init_grant_ram(void)
> >      ram_grants.ram_block = block;
> >      ram_grants.ram = true;
> >      ram_grants.terminates = true;
> > +    ram_grants.ops = &xen_grant_mr_ops;
> >      ram_block_add_list(block);
> >      memory_region_add_subregion(get_system_memory(), XEN_GRANT_ADDR_OFF,
> >                                  &ram_grants);
> > diff --git a/system/physmem.c b/system/physmem.c
> > index 5db1b32823..155a8c05fb 100644
> > --- a/system/physmem.c
> > +++ b/system/physmem.c
> > @@ -2233,13 +2233,16 @@ RAMBlock *qemu_ram_block_from_host(void *ptr, bool 
> > round_offset,
> >  
> >      if (xen_enabled()) {
> >          ram_addr_t ram_addr;
> > +
> >          RCU_READ_LOCK_GUARD();
> >          ram_addr = xen_ram_addr_from_mapcache(ptr);
> > -        block = qemu_get_ram_block(ram_addr);
> > -        if (block) {
> > -            *offset = ram_addr - block->offset;
> > +        if (ram_addr != RAM_ADDR_INVALID) {
> > +            block = qemu_get_ram_block(ram_addr);
> > +            if (block) {
> > +                *offset = ram_addr - block->offset;
> > +            }
> > +            return block;
> >          }
> > -        return block;
> >      }
> >  
> >      RCU_READ_LOCK_GUARD();
> > -- 
> > 2.17.1
> >